(A.J. Culyer, J.P. Newhouse) Handbook of Health Ec

INTRODUCTION TO THE SERIES
The aim of the Handbooks in Economics series is to produce Handbooks for various
branches of economics, each of which is a definitive source, reference, and teaching
supplement for use by professional researchers and advanced graduate students. Each
Handbook provides self-contained surveys of the current state of a branch of economics
in the form of chapters prepared by leading specialists on various aspects of this branch
of economics. These surveys summarize not only received results but also newer devel-
opments, from recent journal articles and discussion papers. Some original material is
also included, but the main goal is to provide comprehensive and accessible surveys.
The Handbooks are intended to provide not only useful reference volumes for profes-
sional collections but also possible supplementary readings for advanced courses for
graduate students in economics.
KENNETH J. ARROW and MICHAEL D. INTRILIGATOR
PUBLISHER'S NOTE
For a complete overview of the Handbooks in Economics Series, please refer to the
listing on the last two pages of this volume.
v
CONTENTS OF THE HANDBOOK
VOLUME 1A
Introduction
The State and Scope of Health Economics
ANTHONY J. CULYER and JOSEPH P. NEWHOUSE
PART 1 - OVERVIEWS AND PARADIGMS
Chapter 1
International Comparisons of Health Expenditure
ULF-G. GERDTHAM and BENGT JONSSON
Chapter 2
An Overview of the Normative Economics of the Health Sector
JEREMIAH HURLEY
Chapter3
Medical Care Prices and Output
ERNST R. BERNDT, DAVID M. CUTLER, RICHARD G. FRANK, ZVI GRILICHES,
JOSEPH P. NEWHOUSE and JACK E. TRIPLETT
Chapter 4
Advances in CE Analysis
ALAN M. GARBER
Chapter 5
Information Diffusion and Best Practice Adoption
CHARLES E. PHELPS
Chapter 6
Health Econometrics
ANDREW M. JONES
PART 2 - DEMAND AND REIMBURSEMENT FOR MEDICAL SERVICES

Chapter 7
The Human Capital Model
MICHAEL GROSSMAN
Chapter 8
Moral Hazard and Consumer Incentives in Health Care
PETER ZWEIFEL and WILLARD G. MANNING
vii
viii Contents of the Handbook
Chapter 9
Physician Agency
THOMAS G. McGUIRE
Chapter 10
Insurance Reimbursement
MARK V. PAULY
PART 3 - INSURANCE MARKETS, MANAGED CARE, AND CONTRACTING

Chapter 11
The Anatomy of Health Insurance
DAVID M. CUTLER and RICHARD J. ZECKHAUSER
Chapter 12
Health Insurance and the Labor Market
JONATHAN GRUBER
Chapter 13
Managed Care
SHERRY GLIED
Chapter 14
Risk Adjustment in Competitive Health Plan Markets
WYNAND P.M.M. VAN DE VEN and RANDALL P. ELLIS
Chapter 15
Government Purchasing of Health Services
MARTIN CHALKLEY and JAMES M. MALCOMSON
VOLUME lB
PART 4 - SPECIFIC POPULATIONS
Chapter 16
Economics and Mental Health
RICHARD G. FRANK and THOMAS G. McGUIRE
Chapter 17
Long-Term Care
EDWARD C. NORTON
Chapter 18
The Economics of Disability and Disability Policy
ROBERT HAVEMAN and BARBARA WOLFE
Chapter 19
Child Health in Developed Countries
JANET CURRIE
Contents of the Handbook ix
PART 5 - THE MEDICAL CARE MARKET
Chapter 20
The Industrial Organization of Health Care Markets
DAVID DRANOVE and MARK A. SATTERTHWAITE
Chapter 21
Not-For-Profit Ownership and Hospital Behavior
FRANK A. SLOAN
Chapter 22
Economics of General Practice
ANTHONY SCOTT
Chapter23
Waiting Lists and Medical Treatment
JOHN G. CULLIS, PHILIP R. JONES and CAROL PROPPER
Chapter 24
Economics of Dental Services
HARRI SINTONEN and ISMO LINNOSMAA
Chapter 25
The Pharmaceutical Industry
EM. SCHERER
PART 6 - LAW AND REGULATION
Chapter26
Liability for Medical Malpractice
PATRICIA M. DANZON
Chapter 27
Antitrust and Competition in Health Care Markets
MARTIN GAYNOR and WILLIAM B. VOGT
Chapter 28
Regulation of Prices and Investment in Hospitals in the U.S.
DAVID S. SALKEVER
PART 7 - HEALTH HABITS
Chapter29
The Economics of Smoking
FRANK J. CHALOUPKA and KENNETH E. WARNER
Chapter30
Alcohol
PHILIP J. COOK and MICHAEL J. MOORE
x Contents of the Handbook
Chapter 31
Prevention
DONALD S. KENKEL
PART 8 - HEALTH
Chapter 32
The Measurement of Health-Related Quality of Life
PAUL DOLAN
Chapter 33
Economic Epidemiology and Infectious Diseases
TOMAS PHILIPSON
PART 9 - EQUITY
Chapter 34
Equity in Health Care Finance and Delivery
ADAM WAGSTAFF and EDDY VAN DOORSLAER
Chapter 35
Equity in Health
ALAN WILLIAMS and RICHARD COOKSON
ACKNOWLEDGMENTS
Several of the chapters in this Handbook were discussed at a conference at the Univer-
sity of Chicago sponsored by Pfizer and the NIA Center for Aging at the University of
Chicago. We are grateful for their having made this opportunity available.
Anthony J. Culyer
Joseph P. Newhouse
xi
Chapter 16
ECONOMICS AND MENTAL HEALTH*
RICHARD G. FRANK
HarvardMedical School
THOMAS G. McGUIRE
Boston University
Contents
Abstract 894
Keywords 894
1. Introduction 895
2. The institutional context 897
2.1. Mental illness and persons with mental illness 897
2.2. Who is treated for mental illness 900
2.3. Who pays for mental health care 901
2.4. The supply of mental health services 902
2.5. Managed behavioral health care 905
3. Private insurance markets, moral hazard, and mental health care 907
3.1. Evidence of moral hazard 908
3.2. Rationing in managed care 912
3.3. Managed care, supply side incentives and moral hazard: evidence 918
3.3.1. Research on hospital payment systems 920
3.3.2. Research on managed care and capitation 921
4. Insurance markets, adverse selection and mental health care 925
4.1. Evidence of selection in MH/SA 926
4.2. Policy responses to selection: fee-for-service-indemnity contracts 927
4.3. Selection and managed care: distorting "quality" 928
4.4. Policy responses and managed care 931
4.4.1. Risk adjustment 932
4.4.2. Behavioral health carve-outs 935
5. The public mental health and substance abuse treatment system 936
*We are grateful for financial support from the Robert Wood Johnson Foundation (#23498), the National
Institute of Mental Health (MH3703 and K05-MH01263), and the Alfred P. Sloan Foundation. We thank
Maggie Alegria, Randy Ellis, Jacob Glazer, Sherry Glied, Haiden Huskamp, Joe Newhouse, and Meredith
Rosenthal for comments on earlier versions of this chapter.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P.Newhouse

© 2000 Elsevier Science B. V All rights reserved
894 R. G. Frank and T G. McGuire
5.1. The technology of treatment for the severely mentally ill 937
5.2. Fiscal federalism and public mental health care 938
5.3. Externalities and public mental health care 941
6. Conclusions 943
References 945
Abstract
This paper is concerned with the economics of mental health. We argue that mental
health economics is like health economics only more so: uncertainty and variation in
treatments are greater; the assumption of patient self-interested behavior is more dubi-
ous; response to financial incentives such as insurance is exacerbated; the social con-
sequences and external costs of illness are more formidable. We elaborate on these
statements and consider their implications throughout the chapter. "Special character-
istics" of mental illness and persons with mental illness are identified and related to
observations on institutions paying for and providing mental health services. We show
that adverse selection and moral hazard appear to hit mental health markets with special
force. We discuss the emergence of new institutions within managed care that address
longstanding problems in the sector. Finally, we trace the shifting role of government in
this sector of the health economy.
Keywords
mental health, insurance, moral hazard, adverse selection, carve-out, risk adjustment
JEL classification:I10
Ch. 16: Economics and Mental Health 895
1. Introduction
In 1994, The Nobel Prize in Economics was awarded to the game theorist John Nash,
who, in the early 1950s, formulated elegant mathematical models for the strategic in-
teraction among small numbers of decision-makers in situations involving elements of
both conflict and cooperation. The "Nash equilibrium" remains the most widely used
equilibrium concept in game theory. Soon after his pioneering work was published,
it was discovered that Nash suffered from schizophrenia. In the last thirty-five years,
Nash has done little productive work, living in the care of hospitals, family, and friends.
Many doubted that Sweden's Royal Society would award the Nobel Prize to a person
with severe mental illness. When they did, Ariel Rubenstein, himself a prominent game
theorist, expressed in a New York Times interview his admiration for Nash's work and
his pleasure that the Royal Society acknowledged by their decision that there was noth-
ing disqualifying about mental illness. Schizophrenia was, in Rubenstein's words, "just
like cancer."
Public attitudes about mental illness have changed since the 1950s when Nash be-
came ill, and the mentally ill have in many ways been integrated into the mainstream of
the health care system. The fact remains, however, that in terms of public and private
policy in the US, mental illness and substance abuse are not treated the same as other
illnesses. In comparison to physical illness, governments pay for more of mental health
and substance abuse (MH/SA) care, and private insurance pays for less. Treatment for
mental and addictive disorders is often involuntary. This is rare for physical illness.
The public mental health care system has had quality problems that are regarded as
scandalous. The public system in mental health has a role as protector of public safety.
When efforts have been made to reform the health care sector in the US, mental health
and substance abuse care are usually handled separately. President Clinton's proposed
health care reform in 1993, to take the most recent example, did not include long-term
treatment for mental illness in the required services to be offered in health plans.
This chapter is concerned with the economics of mental illness and mental health
care. Following convention, we will use the term "mental illness" to include substance
abuse disorders such as drug or alcohol abuse and dependence. Mental health has been
an active and distinct subfield of health economics for some time. Though mental health
economics can claim no special methodology, it has its own conferences, training pro-
grams, and journals.1 Mental health economics is like health economics only more
so: uncertainty and variation in treatments are greater; the assumption of patient self-
interested behavior is more dubious; response to financial incentives such as insurance is
exacerbated; the social consequences and external costs of illness are more formidable.
We will elaborate on these statements and consider their implications throughout this
t Much of this is due to support from the National Institute of Mental Health in various forms. The late Carl
Taube was the official of the NIMH who was most responsible for promoting the field of economics of mental
health.
chapter. "Special characteristics" of mental illness, and the persons with mental illness,
will be identified and related to the observance of institutions paying for and providing
mental health care. When Pauly (1978) asked, "Is Medical Care Different?", he was
contrasting health care with the rest of the economy. Here we explore the question: "Is
mental health care different from health care?"
The first reaction for many people is to answer "yes" to this question, and give the
reason as stigma. Literally, a "mark" or a "stain", stigma sets persons with mental ill-
ness apart as undesirable. Nunnally (1961) found that regardless of the respondent's
education, the mentally ill were regarded as dangerous, unpredictable, and socially of
little value. In one of the few studies comparing attitudes over time, Matas et al. (1986)
analyzed treatment of mental illness in the press, concluding that in spite of some "mi-
2
nor, cosmetic changes", overall, "content and attitudes had changed little". One hopes
that national educational campaigns such as the NIMH's Depression Awareness initia-
tive have had some effect, but the degree is hard to judge. It seems safe to say that some
3
part of the public's fear of the mentally ill remains irrational and misplaced. The his-
torical importance of stigma calls attention to the salient point, important to the rational
as well as the irrational side of the story, that differential treatment of mental health
care for purposes of policy will be driven not just by differences in the disease and its
treatment (e.g., demand is more responsive), but differences in the people who have
the disease (e.g., they are more costly in other ways). This chapter, using methods of
economics, will be concerned with the more "rational" reasons why mental illness is
treated differently than other illnesses, without claiming that this is the full picture.
The core issues in mental health and health economics include:
* Adverse Selection and Moral Hazard: These are traditional concerns in health eco-
nomics. We will argue that these features of insurance markets apply with particular
force for mental health care.
* Non-contractable Provider Actions. The term "noncontractable" was not in use when
Arrow wrote his overview of health economics, but he clearly had this in mind when
he discussed the ways health markets adapt to ensure that physicians put sufficient
effort into caring for patients. Maintaining effort in a managed care environment is
a problem for all areas of care. We will argue that it may be especially problematic
in mental health due to the severity of selection-related incentives. At the same time
it appears that special institutions are arising (a la Arrow) that may be capable of
contending with some of these unwanted consequences.
* Externalities: Mental disorders are often chronic conditions that create substantial
disability and strike people early in life (ages 15 to 30). These illnesses are correlated
2 Prejudice against the mentally ill on the part of the nominating committee of the prestigious Economet-
ric Society thwarted Nash's induction in 1988, according to Nasar (1998). Two years later the nominating
committee was bypassed and Nash was elected directly by the members in an overwhelming vote.
3 Link and Cullen (1986) demonstrate that the more people have direct contact with persons with mental
illness, the less dangerous they are regarded, supporting this conclusion.
with other costly social problems: unemployment, crime, violence, and homeless-
ness. Many of these problems have consequences for others as well as the person
with the illness. For these reasons government has historically taken a large role in
provision and regulation of some forms of mental health care.
The mental health sector contains institutions, professions, and illnesses that are un-
familiar to policy analysts and even to specialists in health economics. As such, policy
has often been applied to the institutions of the mental health sector as an afterthought
to a broad health policy decision. We hope to introduce some of the special institutional
features in MH/SA to a health economics audience and to review some of the empiri-
cal research that has been focused on mental health issues. In addition, we will review
some of the policy trade-offs facing those interested in remedying market failures in the
mental health sub-sector.
This chapter is organized into six main sections. Following this introduction is a de-
scription of the institutional context within which mental health and substance abuse
care is provided. This includes a discussion of mental and addictive illnesses, their con-
sequences, and the organizations involved in financing and delivery of care. The third
section focuses on issues of moral hazard and rationing of care. Evidence on the impact
of benefit design and payment policies is assessed. A general framework for analyzing
rationing of treatment within managed care is presented, which allows us to evaluate
the case for "parity" for mental health in health care payment systems within a man-
aged care environment. Adverse selection is addressed in the fourth section. The ra-
tioning model developed in the previous section is applied to selection-related behavior
by managed health care plans. The reasons why mental health might be more vulnerable
to selection-driven market features are set out. Risk adjustment and behavioral health
carve-outs are examined as institutions aimed at countering selection-related incentives.
The fifth section focuses on externalities and the role of state government in the delivery
of mental health and substance abuse care. We offer some concluding observations in
the sixth and final section.
2. The institutional context
The nature of mental illness, the persons with mental illness, use of treatment services,
and the supply side of the market differ from the general health sector.
2.1. Mental illness and persons with mental illness
World-wide, mental illness is among the most prevalent and disabling illnesses. In the
US, approximately 30% of the population is estimated to experience some diagnosable
mental or addictive disorder in a 12-month period [Kessler et al. (1994)]. Some diseases
have been found to be roughly similar in high and middle income countries when the
same epidemiologic assessments are applied [Alegria, Bijl, Lin, Walter and Kessler (in
press)]. However, comparative epidemiological studies show considerable variation in
898 R.G. Frankand TG. McGuire
illness patterns across nations. For example, rates of depression are estimated to be
considerably lower in Puerto Rico than in the US or Switzerland. Substance abuse is
much higher (roughly 4 times higher) in the US and Puerto Rico than in Switzerland
[Swendsen et al. (1998)].
The most severe mental disorders, schizophrenia, manic depression, and some forms
of major depression affect about 4% of the population each year and are very disabling.
Within this group are the psychoses, illnesses associated with severe disturbances in
cognitive functions. These disorders are persistent illnesses that tend to have initial on-
sets relatively early in life (ages 15 to 30). Individuals with these serious illnesses most
often suffer for relatively long periods [APA (1995)]. Severe forms of mental illness
reduce an individual's ability to function in the consumer role, and interfere with the
maintenance and creation of social networks (family and friends) weakening the con-
nection with others who might serve as caregivers or proxy decision-makers [Segal,
Silverman and Baumohl (1989)].
Effective treatments, primarily drug therapy, reduce symptoms and improve function-
ing for all of these illnesses. Existing treatments contend with but do not "cure" mental
illness, and persons with these illnesses require long-term monitoring and treatment as
well as periods of intensive services such as hospitalization. Many require extended pe-
riods of assistance with housing and social support, contributing to the need for a public
role in caring for this most severely ill group.
Mental and addictive disorders are costly to society both in terms of direct spending
on treatment as well as in terms of the losses sustained as a consequence of the disorders.
Spending on mental health and substance abuse care in (MH/SA) 1995 was estimated
to be about $75 billion [Triplett (1998)], amounting to about 8.3% of personal health
expenditures. Overall spending on MH/SA in the US (including specialty and general
care) grew at 12.6% per annum during the 1963-1972 period, at 13.9% per year during
1972-1980, and 9.3% in the 1980-1995 time span. Total health care spending grew at
yearly rates of 11.5%, 13.7%, and 9.9% for the three time intervals, respectively. This
suggests that mental health spending tracked overall spending quite closely over the
1963-1995 period [Triplett (1998)].
While the United States spends more on MH/SA in absolute terms than do other west-
ern nations, it spends a lower proportion of personal health outlays than Great Britain
(16.6%), Canada (11.4%), and Australia (8.4%) [Triplett (1998)]. Mental disorders im-
pose costs on the individual, his or her family, and on society as a whole. In aggregate,
Rice et al. (1990) found indirect costs to be twice the direct costs of care. Studies of
individual behavior have documented the impact of mental health on employment, pro-
ductivity and earnings [Bemrdt et al. (1997), Ettner et al. (1997), Bartel and Taubman
(1986), Mullahy and Sindelar (1993)], criminal activity [Steadman et al. (1998), Link
(1992)], motor vehicle accidents [Rice et al. (1990)], child abuse and neglect [Kelleher
et al. (1994)], homelessness Jencks (1994)], and divorce [Bartel and Taubman (1986)].
The indirect cost of mental illness has also been studied in terms of employment
and earnings. A common finding in psychiatric epidemiology is an inverse correla-
tion between income and rates of illness in a population [Bruce, Takeuchi and Leaf
(1991)]. However, the causal connection between mental illness and income and other
labor market outcomes is complex. Economic stress, such as involuntary unemploy-
ment, may aggravate illness. Some difficult-to-measure personal characteristics which
make a positive contribution to earnings are correlated with some illnesses. Creativity,
energy, and attention to detail may each be more common among persons who have
mania or obsessive-compulsive disorders. Furthermore, persons who have diseases and
are successful enough to remain in the labor force may have atypical values of some
other unmeasured labor market characteristics. Estimates of the impact of mental dis-
orders on labor market outcomes have used longitudinal data, or instrumental variables
and cross-sectional data to find substantial reductions in earnings and other measures of
productivity associated with illness. Bartel and Taubman (1986), and Frank and Gertler
(1991) use longitudinal data with information on prior illness to estimate reductions in
earnings of men of between 20% and 25% for conditions that are thought to produce
the most impairment such as psychotic disorders and major depression. Neuroses and
other mental disorders had smaller but significant negative impacts on earnings (5% to
15%). Ettner, Frank and Kessler (1997) used cross sectional data from a national epi-
demiological survey of the US population to examine the effect of mental and addictive
illnesses on employment and earnings. Information on the family history of mental ill-
ness and the timing of the onset of symptoms of mental illness enabled the authors to
use instrumental variable techniques to estimate the impact of mental illness on income
taking account of possible endogeneity of mental illness. They found that the presence
of a diagnosable mental illness reduced employment by about 11% for both males and
females and for those who worked, the estimated loss of income attributable to mental
illnesses was about 20% for women and 10% for men.
The connection between mental disorders and violence and crime has been contro-
versial, since it is tied so closely to the sensitive issue of social stigma. Recent research
has found clear evidence that mental illness and substance abuse are associated with
higher rates of criminal activity. A study by Link and colleagues (1992) illustrates the
tenor of a larger set of research findings. When psychiatric patients are compared to
a control population, matched according to the neighborhood in which they live and
socioeconomic characteristics, psychiatric patients exhibit significantly higher rates of
weapon use and violent behavior. Torrey (1994), an advocate for the seriously mentally
ill, arrived at a similar conclusion based on a review of the literature, stating: "Although
the vast majority of individuals with serious mental illness are not more dangerous than
members of the general population, recent findings suggest the existence of a sub-group
that is more dangerous" (p. 653). According to Steadman et al. (1998), the subgroup in-
cludes those individuals with a co-occurring substance abuse disorder. Individuals who
suffer from both mental illness and substance abuse problems are more likely to be in-
volved in violence than otherwise similar people with mental illness only or without
mental illness. (It is notable that individuals with a mental disorder are significantly
more likely to abuse substances than are people without such illnesses.)
Addictive disorders have also been tied to auto accidents [see Rice et al. (1990) for
a review], unsafe sexual practices and child abuse and neglect [Kelleher et al. (1994)].
900 R. G. Frankand TG. McGuire
Alcohol and drug abuse in mothers has also been linked to poor birth outcomes [Sec-
retary of DHHS (1993), ASPE (1994)]. Even at levels of use that do not qualify as
"abuse", alcohol and drug use can be problematic, especially among the young. Lowry
et al. (1994) report that high school students were more likely to engage in risky sexual
behavior if they used alcohol (relative odds of 2.7), marijuana (odds of 9.2) or cocaine
(odds of 26.8) relative to non-users.
Finally, homelessness has often been linked to mental illness and substance abuse
in both recent literature [Jencks (1994)] and the popular press. Many mental hospital
patients were shifted to nursing homes, jails, and streets as a result of the reduction in
public mental hospital capacity during the 1960s and 1970s [O'Flaherty (1996)]. Jencks
contends that the "deinstitutionalization" movement was a prime cause of the growth of
homelessness during the 1980's, while other observers [e.g., O'Flaherty (1996)] have
questioned the strength of the causal connection.
This brief profile of mental and addictive disorders and their social consequences
leads to two conclusions useful for an economic analysis of mental health and men-
tal health care. First, mental and addictive disorders are prevalent and associated with a
variety of social costs not incurred by the affected individuals. The existence of external-
ities means that decentralized market decision-making will tend to undervalue effective
treatments for mental and addictive illnesses. Second, the social costs of mental and ad-
dictive disorders are concentrated in the 4% of the population that experience the more
severe forms of the disorders. This sub-group of people displays a series of character-
istics that make them "undesirable" to insurers as clients, employers as workers and
to significant segments of the general population as neighbors. Thus, in the absence of
some compensatory factor, there are selection-related private benefits linked to avoiding
employing and insuring those at risk for mental illness.
2.2. Who is treatedfor mental illness
One of the dilemmas in formulating policy towards mental health care is that "under-
treatment" and "overtreatment" can coexist within the same payment system. Of the
30% of the population that has a mental illness at some point during a year, only 17.3%
get some treatment in the health care sector with an additional 7.4% getting their only
treatment from a human service agency or a self-help group (e.g., Alcoholics Anony-
mous). 4 Thus, only about 25% of those with a diagnosable condition get some form of
treatment over a twelve-month period. Low rates of treatment conditional on having a
disorder also characterize the more severely ill. Only 36% of those with manic depres-
sion or major depression in a year are treated in any sector, and only 25% of those with
substance abuse are treated. (Approximately 57% of individuals with schizophrenia do
get some health care treatment.)
4 All the data cited in this paragraph are based on information from the National Comorbidity Survey.
Detailed discussion of these findings is in from Kessler et al. (1997).
At the same time, 4.5% of individuals with no disorder (assessed with a diagnostic
instrument) receive mental health care from a health care provider and 8.2% get treat-
ment from some human service organization or self-help group. Those individuals with
no diagnosed condition that obtain treatment make almost the same number of visits as
those with at least one diagnosed condition: 7.9 visits per year compared to 9.3 visits
[Kessler et al. (1997)]. Putting this together with the figures in the earlier paragraph, we
can say that nearly 38% of all users and 28% of all visits for mental health care are not
associated with a diagnosable disorder. 5 Moreover, a large segment of that group report
that they are in either good or excellent emotional health [see Kessler et al. (1997)1.6 It
is interesting to note that in Canada the portion of users of MH/SA care who have no
disorders is considerably smaller than in the US [Kessler et al. (1997)].
Spending on treatment is concentrated on those people with the most disabling con-
ditions. It was estimated that in 1990 nearly 30% of spending on mental health and sub-
stance abuse care was accounted for by 5% of the users of care [Frank et al. (1994)]. For
example, the mean level of spending on treatment of mental health and substance abuse
care in a large insured population for 1993 was $8 per enrollee per month, while the
mean cost of treating someone with a diagnosis of manic depression was about $6,700.
People with a history of mental health care use also tend to incur higher levels of
general health expenditures than do others. For example, data from Michigan Medicaid
indicate that the average person had expenditures of $1,873 per year in health care over
a three-year period (1991-1993) compared to $3,722 (including mental health care) for
individuals with any treatment for mental illness during that time period. Roughly 66%
of the difference is accounted for by mental health costs. Cuffel and Goldman (1998)
report that mental health care users spend nearly 90% more on general medical care
than did non-users.
A large segment of individuals who receive the highest intensity of care do so, in part,
because they are compelled to by the legal system. For example, in 1994, approximately
43% of days of care provided in specialty psychiatric hospitals and specialty psychiatric
units of general hospitals were accounted for by individuals involuntarily admitted. In
addition, 27% of days of care in residential facilities and 8% of patients in ambulatory
settings were involuntarily placed in care.7
2.3. Who pays for mental health care
Spending on mental health and substance abuse care displays a different pattern than
that found in the health sector overall. Table 1 reports the composition of spending on
5 The share of the total population that are both users of care and have a diagnosable condition is 0.05
(0.17 x 0.30). The share of the total population that use services but do not have a diagnosable condition is
0.03 (0.045 x 0.70). Thus 38% of the users do not have a diagnosable condition.
6 Results from an epidemiological survey in Ontario suggest that people with the most disabling illnesses
are somewhat more likely to get treatment in Ontario [Kessler et al. (1997)]. The US may perform relatively
badly on this score.
7 These data are based on unpublished information from the Inventory of Mental Health Organizations
collected by the Substance Abuse and Mental Health Services Administration of the USDHHS, 1996.
Table 1
Mental health and substance abuse spending by source
of payment US, 1996
Payer MH/SA %* All health %*
Private insurance 25.8 31.0

Private out-of-pocket 15.1 18.1
Medicare 14.0 21.0
Medicaid 18.8 14.8
Other federal 3.9 4.3
Other state/local 19.4 7.2
Other private 2.7 3.3
Total 100 100
*Percentages may not sum due to rounding.

Source: McKusick et al. (1998).
health and MH/SA care. Among the most important differences reported in the table is
the role of government as a direct funder of care. The other federal (block grants) and
other state and local categories comprise 23.3% of all MH/SA spending compared to
11.5% of overall healthcare spending. In addition, Medicaid plays a somewhat larger
role in MH/SA spending, 18.8% vs. 14.8%. Finally Medicare plays a considerably
smaller part in funding MH/SA than it does for all health services. Thus state and lo-
cal government generally allocated more resources for MH/SA, 42.1% (summing block
grant, state/local and Medicaid) than for health services generally, 26.3%. This high-
lights the differing division of labor between federal and state government. Whereas the
federal government funds over 25% of health spending, it accounts for less than 20% of
MH/SA expenditures.
Table 1 shows that private health insurance accounts for a smaller share of spending
in MH/SA than in all health care (25.8% vs. 31%). The table also suggests that private
out-of-pocket spending makes up a smaller share of non-public funding in MH/SA than
in overall health care. As we shall see below, this finding is inconsistent with data on
insurance coverage and may be an artifact of using insurance claims to estimate private
spending.
2.4. The supply of mental health services
There are a great variety of organizations and professionals that supply mental health
services, not limited to traditional medical care providers. For example, in the state of
Massachusetts one could receive office-based psychotherapy from any of the following
licensed providers: primary care physicians, psychiatrists, psychologists, social work-
ers, counselors, and nurses. The services provided by each of these professions are paid
for by public and private insurance plans. Similarly, inpatient psychiatric care might be
provided in the medical-surgical section of a general hospital, a general hospital psy-
chiatric unit, a private psychiatric hospital or a state mental hospital. In some cases the
Table 2
Spending on specialty care organizations, US (nominal dollars)
1969 1994
State mental hospitals $1,814 million $7,824 million

General hospital psychiatric units 298 5,344
Private psychiatric hospitals 450 6,468
VA medical centers 122 1,386
Residential treatment centers 186 2,360
Free standing clinics 202 6,046
Community based treatment programs 220 3,706
Total $3,293 million $33,136 million
Source: Mental Health, United States (1998).
functions of these various suppliers are differentiated and in others they offer services
that appear to be close substitutes [Goldman and Skinner (1989)].
Table 2 describes the types of organizations that offer specialty mental health ser-
vices in the US and their nominal levels of spending in 1969 and 1994. Note the dra-
matic shifts in patterns of spending within the hospital sector and between hospital and
non-hospital organizations over time. For example, state mental hospitals accounted for
about 55% of all specialty mental health spending in 1969 compared to roughly 25%
in 1994. General hospitals accounted for 9% of spending in 1969 compared to 16%
in 1994. Community based treatment programs (which include mental health centers)
accounted for 7% of spending in 1969 and 11% in 1994. These changes in the roles of
suppliers of mental health care also reflect the emergence of markets that has taken place
over the past 40 years. Even public funds are increasingly allocated via markets. Insur-
ance mechanisms like Medicaid have grown in importance over time. Public managed
care programs turn over operations of systems of care to private organizations. Whereas
publicly owned and operated mental hospitals dominated the supply of care in the 1950s
and 1960s, privately owned non-profit hospitals, mental health centers and clinics now
play a central role, as do private for-profit organizations. These new treatment sites em-
phasize outpatient therapy, management of drug treatments, partial hospital services,
rehabilitation and case management services.
Psychiatrists and psychologists account for less than half of mental health profes-
sionals. While there were about 33,500 psychiatrists and nearly 70,000 psychologists
in 1995, social workers, counselors, and family therapists accounted for 94,000, 61,000
and 46,000 practitioners, respectively. Individuals with higher income and private in-
surance receive mental health care from physicians, psychologists, or social workers
specializing in mental health services. Shifting patterns of supply correspond to altered
approaches to treating mental illnesses.
The treatment of mental illnesses has changed dramatically over the past 40 years, in
part due to scientific changes in treatment technology such as pharmaceutical innova-
tion [Berndt, Cockburn and Griliches (1997), Grob (1991)], new methods of organizing
elements of treatment [Stein and Test (1980)], and improved approaches to brief psy-
chotherapy. Changes in the organization and financing of mental health services have
also contributed to changing treatment patterns [Mechanic (1987)]. In particular, limits
on insurance coverage for inpatient days or outpatient visits, financial incentives to re-
duce hospital stays and payment arrangements which reward health plans for reducing
overall health care spending have contributed to shifts in treatment of mental and ad-
dictive disorders [Harrow and Ellis (1992), Frank and McGuire (1996)]. Various types
of mental health services have been posited to be substitutes. Research based in HMOs
and publicly funded treatment programs provides evidence indicating that community
based outpatient treatments are substitutes in production for inpatient hospital care [see
Weisbrod (1983), Stein and Test (1980), Hoult et al. (1983), Finch et al. (1992), Calla-
han et al. (1995)]. Similarly for psychotherapy services, outcomes evaluations suggest
that there are a range of professions trained in psychotherapy that produce comparable
clinical gains to patients, implying that these professions are substitutes in production
of psychotherapy treatment [Knesper (1989)].
There is also mounting evidence that for certain specific illnesses pharmaceutical
treatments can substitute for psychotherapy [Elkin et al. (1989), Kupfer et al. (1992)].
Berndt, Frank and McGuire (1997) offer evidence that drugs and psychotherapy are also
substitutes in demand. Empirical analyses of cross-price responses of demand for social
workers' and psychologists' services relative to psychiatrists' indicates that a substantial
segment of the services delivered by each profession are also close substitutes in demand
[Fairbank (1989), Frank (1985a)].
Data on medical practice patterns indicate that there is considerable disagreement
about how medical treatments of all types, health and mental health care, should be used.
Holding other factors constant, the likelihood of receiving a particular treatment can
vary dramatically based on a patient's residence [Phelps (1998)]. Phelps and Mooney
(1993) contend that much of the variation in practice is due to beliefs, information,
and learning at the individual physician level. If variation is unrelated to differences in
patient need or benefit, it will cause significant welfare loss. In the mental health area,
variation in treatment tends to be greater than in medical care overall [Phelps (2000)].
This may, in part, be explained by the wide range of professions (with their theoretical
orientations) and modalities (with their patterns of costs and incentives) that supply
treatment to individuals with a particular illness. It may also reflect greater clinical
uncertainty. With more disagreement among clinicians about proper treatment, welfare
losses from treatment not related to benefits may as a result be larger in the mental
health area. 8
8 The variation and welfare loss must take into account other factors. Even granting that variation is around
an "optimal point", the welfare loss depends on the shape of the total benefit schedule. A flat marginal benefit
function, associated with an elastic demand will imply a relatively low welfare loss.
2.5. Managed behavioralhealth care
Managed care is transforming the health care sector generally, and may be having more
of an impact on mental health than in health care. Employers, government and other
purchasers are bargaining for lower prices and monitoring treatment patterns. The re-
sponse to the new spirit of prudent purchasing of health care services has been an accel-
eration in the growth of managed care organizations. Preferred Provider Organizations
(PPOs), Point of Service (POS) plans and Health Maintenance Organizations (HMOs)
accounted for 73 % of the insured population in 1995 [PPRC (1997)]. State governments
have moved to strengthen the bargaining power of buyers of health care by encourag-
ing the creation of purchasing alliances that enable smaller purchasers of group health
insurance to command more choice at more advantageous prices. State Medicaid plans
and the federal Medicare program are all experiencing rapid growth in enrollment of
beneficiaries in managed care organizations that bear significant financial risk.
A striking development in mental health and substance abuse has been the devel-
opment of so-called managed behavioral health care carve-outs. Traditionally, the pur-
chaser, usually an employer, contracted with a single insurance plan to cover a full range
of health risks. Increasingly, however, purchasers of health insurance are offering bene-
ficiaries a range of plans. Purchasers may also "carve out" certain benefits, which means
that they separate the health insurance function by disease or service category and con-
tract separately for the management of those risks. This carve-out in insurance need not
be associated with "managed care", but it virtually always is.
There are three forms of carve-outs found in the MHSA health sub-sector, with poten-
tially distinct economic impacts. They are: (a) payer specialty carve-outs from all health
plans; (b) payer specialty carve-outs from only indemnity and PPO type arrangements;
and (c) individual health plan carve-outs to specialty vendors.
The two forms of payer carve-outs are illustrated in Figures 1 and 2. In Figure 1,
enrollees have a choice among a traditional indemnity insurance plan and managed
care plans (e.g., an HMO and a PPO). The payer in this case writes a separate contract
Emoyer
Traditional- Carve ' '

Plan Out |
Enrollees / managed care plan 1 . -- ,

managed care plan 2 - -
Health Behavioral
Health
Figure 1.
906 R. G. Fankand G. Mc~uire
I Plan 1
Carve
Enrollees Out
Health Behavioral
Health
Figure 2.
with a specialty vendor, for the carved out service (e.g., behavioral health), to manage
a segment of the risk in the traditional plan. Some well-known carve-out programs are
of this type, the Massachusetts Medicaid [Callahan et al. (1995), Frank and McGuire
(1997) and Beinecke et al. (1997)] and the Massachusetts Group Insurance Commission
[Huskamp (1997), Ma and McGuire (1998a), Merrick (1997)]. In this case, a carve-
out eliminates a traditional indemnity plan for behavioral health care, ensuring that all
behavioral health care will be managed. The payer also intervenes in the competitive
process by preventing the traditional plan from competing on the basis of the behavioral
health benefit (or other service carve-outs).
The payer may entirely remove the carved-out service from the market for compet-
ing (otherwise) integrated health plans. Figure 2 shows the case where enrollees choose
among competing health plans for all of their healthcare except for the carved-out ser-
vice. Behavioral healthcare for the State of Ohio employees and for employees and
dependents of Pacific Bell are organized in such a fashion [Goldman, McColloch and
Sturm (1998)]. Enrollees are not given a choice of plan for the carved out service, al-
though a payer would typically use a competitive process to choose the carve-out ven-
dor.
The third major form of carve-out arrangement is illustrated by Figure 3. In this
case, enrollees have a choice of health plan for all services. Health plans choose to
manage certain services such as mental health or cancer care by sub-contracting with a
specialized managed care organization. In this case, the carve-out is an element of the
competitive strategy adopted by a health plan. The payer may set general requirements
for plans to meet, but does not specify organizational form. For example, an employer
might contract with the Prudential HMO in St. Louis, which in turn carves out MH/SA
to Merit Behavioral Health, Inc.
At present, mental health carve-out contracts are a rapidly growing feature of health
insurance. According to Oss (1995), approximately 53 million people are enrolled in
carve-out programs of all types. Between 20 and 25 million people are enrolled in so-
called risk-based carve-out contracts (whereby the carve-out vendor assumes some or
Ch. 16: Economics andMental Health 907
Employer
=
arve Out
Enrollees veOut
Plan 3
Health Behavioral
Health
Figure 3.
all of the financial risk for claims), which account for about 60% of the total revenue
of firms that manage MHSA benefits. Carve-outs are more common among larger firms
than smaller firms. Umland (1995) reports that 35% of employers with 5,000 or more
employees were contracting with a specialty MHSA carve-out vendor compared with a
rate of about 3% for firms with fewer than 500 employees.
Carve-outs are an important new institutional feature in mental health and substance
abuse. Although carve-outs exist for other conditions, they are most common in the care
of MH/SA. To understand why carve-outs are part of the MH/SA landscape and what
their effects are, we need to turn to issues of moral hazard and adverse selection covered
in the next two sections.
3. Private insurance markets, moral hazard, and mental health care
Private insurance markets have long offered insurance for mental health coverage on
much more limited terms than for general health care coverage. Most individuals who
obtain health insurance coverage through their employers have some coverage for men-
tal health and substance abuse treatment, but it is rarely on the same terms as for other
medical care. Over 90% of all employees covered by employer sponsored health insur-
ance have mental health coverage. This is true of employees of both large and small
firms [BLS (1996)]. 9 In 1993, large employers offered insurance plans with more re-
strictive coverage for mental health care than other services 86% of the time for inpatient
benefits and 97% of the time for outpatient benefits. The corresponding figures for small
employers were 85% and 99% respectively. The 1993 data offer evidence suggesting an
9 In 1993-1994, 82% of employees of large firms and 66% of employees of small firms participated in
employer sponsored health plans (unpublished data from the Employee Benefit Survey of the BLS).
erosion in coverage from the 1980s [Buck and Umland (1997)]. It is important to note
that the mere fact that more health plans establish limits for MH/SA coverage that they
did in previous years does not mean that financial protection is less overall. Establishing
that would require a more complete consideration of the terms of coverage. Typically,
private insurance limits the number of reimbursable days of inpatient mental health care
to 30, and the number of outpatient visits to between 20 and 30. Outpatient care gener-
ally carries 50% cost sharing (in about 54% of all policies). Lifetime spending limits are
common for both inpatient and outpatient mental health care (in 40% of health plans).l°
The upper limit on plan cost sharing is usually defined as a limit on reimbursable visits
or plan spending.11
In sum, mental health and substance abuse insurance coverage provides some cov-
erage for low ranges of spending but leaves households unprotected against more ex-
pensive and potentially financially ruinous treatment. It is these observations along with
the history of stigmatization that have led mental health advocacy groups to focus so
much effort on obtaining "parity" in the terms of benefit design in private insurance
for MH/SA and general medical care. The first principle of optimal insurance is that
insurance ought to cover high-end expenses (where the marginal utility of money is
greater) [Arrow (1963)]. A higher demand response for mental health might imply dif-
ferent cost sharing arrangements for mental health services in a "second best" world,
but this does not explain the presence of coverage limits. Optimal insurance in principle
would still imply better high-end coverage, a different pattern of coverage than what is
observed. The traditional explanations for differences in coverage and apparent failures
in the insurance market are moral hazard and adverse selection.
3.1. Evidence ofmoral hazard
Since the 1950s, coverage of treatment for mental illness under private insurance has
been controversial because of perceptions that psychotherapy was discretionary and its
use would be greatly affected by insurance [McGuire (1981)]. Insurers argued that equal
coverage for health and mental health services would create a "cost control" problem.
The moral hazard argument for special treatment of mental health care is based on the
proposition that demand response to insurance coverage for mental health services is
greater than that of other medical services and therefore the welfare loss from coverage
is larger while the risk spending benefits are similar [Pauly (1968), Zeckhauser (1970)].
During the 1960s and 1970s, the demand response of mental health services to the terms
of insurance was studied by examining the experiences of large insured populations.
The federal employees health benefit program was frequently studied because claims
data were available for this population and because coverage under the Blue Cross/Blue
10 All data except the level of day and visit limits reported in this paragraph are from the Employee Benefit
Survey of the BLS.
11 In 1996, Congress passed a so-called mental health parity law requiring dollar limits on mental health
coverage to be the same as for other conditions. Visit and day limits remain permissible.
P-
fi '%
CP
I \
Di 2
\'D3
~~~~ ~ ~~~~~~~~~~~~~~~~~~~~~
Q1 Q2
Figure 4. Non-linear price schedules and demand for MH/SA care. Source: Frank and Manning (1992).
Shield high option plan was quite generous [Reed (1975), Von Korff and Kramer (1978),
Hustead and Sharfstein (1978)]. Other large insured populations studies are summarized
in Frank and McGuire (1986). These included the United Mine Workers, Blue Cross of
Michigan and Massachusetts, Group Health of Puget Sound, the United Auto Workers
Health Plan, CHAMPUS and the State of Washington employees among others. The
empirical problem of separating adverse selection effects (persons more likely to use
coverage choose better coverage) from the moral hazard problem (better coverage leads
to demand response) was not addressed in this early work.
In the 1980s a number of econometric analyses made use of cross sectional surveys
to investigate the magnitude of demand response for ambulatory mental health services,
the services thought to be most responsive to cost sharing [McGuire (1981), Horgan
(1986), Taube, Kessler and Bums (1986), Watts, Scheffler and Jewel (1986)].12 This
first generation of econometric models focused on estimating the demand response of
ambulatory mental health care use to differences in the cost-sharing provisions across
private insurance plans. The empirical models of demand were built on simple assump-
tions about the price schedule and consumer expectations (Manning and Frank, 1992).
Annual number of visits (or dollars) were assumed to be the relevant decision unit, con-
sumers were assumed to face a constant price. Consumers formulated their demand at
the beginning of the annual decision-period. Insurance coverage for mental health is
rarely described by a single price block, such as constant 50% coinsurance. Much more
frequently there are two or even three blocks. Figure 4 shows a three-block schedule
that would result from a deductible, a covered region and a limit on coverage. Empirical
studies during the 1980s used an "average" price, and related this to quantity used. The
12 There is a more limited literature on the demand for inpatient psychiatric care. See, for example, Scheffler
and Watts (1986).
910 R.G. Frankand T G. McGuire
block structure of pricing builds in a relation between use and average price unrelated
to demand response (e.g., with a declining block price, average price and use are nega-
tively correlated). In principle, instrumental variables might deal with this, but this is a
highly imperfect way to address this measurement issue.
The models estimated during the 1980's under a variety of approaches to measuring
out-of-pocket costs were so-called two part models of demand [Manning et al. (1981)].
In the first part, the impact of cost sharing on the probability that an individual would
use any mental health services was estimated using a logit or a probit model. The second
part estimated the effect of cost sharing on the level of utilization (often subject to a log-
arithmic transformation) of mental health care conditional on some use of services. The
second stage was usually estimated using ordinary least squares. The first generation of
econometric research resulted in similar findings across studies: that ambulatory men-
tal health services were highly responsive to cost sharing. Studies generally could not
make a direct comparison to responsiveness in health care, but they typically arrived
at a conclusion that demand for ambulatory mental health care was more responsive
to cost sharing than ambulatory medical services. Those results were based on non-
experimental assignment of individuals to insurance plans, and were subject to selection
problems. 13 Contending with the bias introduced by selection of insurance condition in
general health as well as mental health was a primary rationale for mounting the RAND
Health Insurance Experiment (HIE).
The RAND HIE improved upon earlier studies of demand response by randomly as-
signing families to insurance conditions, minimizing the problems of the correlation
of insurance and use introduced by unobserved variables. Over and above this central
virtue of an experimental design, the HIE made dramatic improvements in the mea-
surement of key variables that plagued earlier studies. Rather than relying on patient or
provider reports about use, the HIE (RAND functioning as a third-party payer) directly
observed what was used and when. Prices charged and paid were also directly observed,
obviating the need to use the self-reported ranges and averages from earlier work. Fi-
nally, measures of health and mental health status, along with other covariates likely to
influence demand, were better measured in the HIE.
Random assignment and better measurement were great helps, but the problems intro-
duced by the non-constant price schedules discussed around Figure 4 remained. Every
HIE plan included a stop loss that limited a family's out-of-pocket spending to $1,000
or less. A family in a 50% plan with a $750 stop loss would, for example, have all care
completely paid for, once they spent $1,500 during a year. A family with any high cost
treatment early in a year, or any foreseeable treatment (such as long-term psychother-
apy) would rationally treat the marginal price of care as free, just like a family in the
free care plan. Interpreting differences in use between the 50% plan and the free care
plan, as well as other plans, requires confronting the issue of family's expectations about
spending.
13 McGuire (1981) used an instrumental variable approach to correct for endogenous insurance coverage.
Other studies had no method for dealing with selection.
Data collection for the HIE took place in the mid-1970s. Early empirical work on
the HIE [Newhouse et al. (1981), Manning et al. (1984), Wells et al. (1982)] compared
plans on the basis of coinsurance only. A "plan response" as in Manning et al. (1984)
compared users in the 50% plan to the free care plan. This does not yield a simple price
elasticity estimate because the price change is averaged over stop losses. Later research
also deals explicitly with expectations, which the omnibus response to plan treated only
implicitly. With the addition of some structural assumptions about how expectations
were formed, estimates of the demand response could be derived.
Keeler, Manning and Wells (1988) examined the demand of a subset of users in the
RAND HIE who began mental health (or general health) treatment while far away (in
dollar terms) from the limit on out-of-pocket expenses. They assumed that individuals
would foresee, at the initiation of care, all the care that would eventually be used in a
given episode of care, but that individuals did not foresee that they might exceed the
out-of-pocket limit. They show individuals in families with full insurance (free care)
coverage used about four times more ambulatory mental health care than do those with
virtually no coverage (95% cost sharing). This is roughly double the response reported
using a similar methodology for ambulatory medical care. Research on other data dealt
with the issue of block pricing and demand. Ellis (1986) proposed an empirical approach
which builds on the model developed by Keeler, Newhouse and Phelps (1977). He stud-
ied a mental health benefit where there was an increasing block price (no deductible,
cost sharing up to a limit on outpatient spending). Ellis assumed that consumers would
base consumption decisions on their "expected" end of year price. Ellis and McGuire
(1986a) applied this model to estimate price elasticities of demand. This research also
showed mental health services to be relatively price elastic. In sum, nearly all the avail-
able evidence, experimental or observational, points in the direction of greater price
response for ambulatory mental health than other health care services.
Table 3 summarizes the cumulative evidence on the demand response of ambulatory
mental health services to cost sharing provisions in insurance. Note that the magnitudes
of the price responses vary considerably. The relative response compared to ambulatory
medical care is however, quite consistent when comparison was possible. For example,
Taube and colleagues (1986), using the NMCUES survey, report price elasticity esti-
mates for mental health care that are four times those estimated for general ambulatory
Table 3
Research on demand response
Study Type of elasticity MH/SA estimate General health estimate
McGuire (1981) Point -1.00

Taube et al. (1986) Point (level of use) -0.54 -0.13
Horgan (1986) Point (level of use) -0.44 -0.16
Ellis and McGuire (1986a) Point (level of use) -0.37
Manning et al. (1989) Arc -0.80 -0.30
Source: McGuire (1989) and authors' calculations.

912 R.G. Frankand T.G. McGuire
health care. Similarly, Horgan (1986) obtained elasticity estimates from the NMCES
data set for mental health care that were 2.75 times those for ambulatory medical care.
The relative elasticity estimate for mental health and medical care found in the RAND
HIE was 2.66. The main policy implication of the empirical literature on the demand
response of ambulatory mental health care to cost sharing is that there is an efficiency
rationale for psychotherapy (the predominant form of ambulatory treatment) to be cov-
ered at a higher level of cost sharing than other types of ambulatory health care. Thus,
in the absence of other forms of rationing or cost control, the strong evidence showing
relatively high demand response implies higher cost sharing for psychotherapy. 14 Based
on the empirical results from the HIE (and some assumptions about risk aversion), Man-
ning and Marquis (1992) estimate that 50% cost sharing for psychotherapy is optimal
(second best).
The moral hazard issue continues to be debated in the context of managed care
[Scheffler et al. (1994)]. While there exists a growing body of research showing that
managed care arrangements result in substantial savings in mental health and substance
abuse (together referred to as behavioral health) spending paid through insurance, this
evidence pertains primarily to the effect of "managed care" on levels of spending, not
the response of spending to the terms of coverage [Christianson et al. (1995), Goldman,
McCulloch and Sturm (1998), Ma and McGuire (1998b), Calahan et al. (1995), Brisson
et al. (1998)]. Demand response in managed care is an important area for study. The
reason the subject of demand response must be looked at afresh is that the control of
moral hazard in managed in managed care is done with other mechanisms in addition
to demand-side cost sharing.
3.2. Rationing in managed care
Assessing the cost control or moral hazard problem in the context of managed care re-
quires one to change the conceptions of rationing that have been employed in studying
fee-for-service and indemnity insurance arrangements. Managed care in general and
managed behavioral health care (MBHC) in particular address the moral hazard prob-
lem with tactics that ration care without relying on money prices paid by the consumer
[Mechanic, Schlesinger and McAlpine (1995)]. Managed behavioral health care organi-
zations (MBHO) must often allocate treatment resources subject to a prospectively set
budget for serving a defined number of people. Features of rationing within managed
behavioral health care organizations include:
* Establishment of a network of selected providers to furnish specialized services to a
defined population of enrollees.
* Directing individuals to levels of care (e.g., inpatient hospital, residential, outpatient)
based on clinical criteria about appropriate matches of clinical circumstances and
provider capabilities.
14 This conclusion follows if a consumer demand curve is given a normative interpretation as a marginal
benefit schedule.
* Writing contracts to providers that include financial incentives to limit care.

* Application of concepts of medical necessity to determine the need and benefits from
continuing treatment at differing levels of care.
* Feedback of information to clinicians on treatment patterns relative to peers and clin-
ical norms.
The MBHO makes these rationing choices in the context of either: (a) markets where
potential enrollees may choose among health plans based on quality indicators, price
and other aspects of reputation and/or (b) regulatory standards set by payers that require
health plans to achieve certain specified levels of performance. 15 From the very begin-
ning of research on the economics of health insurance [e.g., Pauly (1968)], when the
moral hazard problem was addressed exclusively by demand-side cost sharing, it was
assumed that rationing took place in an economically rational fashion. As a price was
introduced, the units of quantity that were "rationed out" were those that were the least
valuable to the consumers. Consumers/patients were assumed to be utility-maximizing
price takers in the face of the prices presented to them by their insurance coverage, im-
plying that the consumer could buy as little or as much health care as he or she wished at
the going price, and did so to maximize utility. The validity of these assumptions have,
of course, generated thousands of academic journal pages. For example, Rice (1998)
has argued that the rationing mechanism economists employ is not correct in a positive
or descriptive sense, in that low value units are not necessarily those rationed out after
prices rise.
With managed care, the assumption of the price-taking consumer can no longer be
maintained. The essence of managed care - its very rationale - is the ability to ration
care without imposing financial risk on consumers. Whatever one's view on the work-
ability of the assumption of price-taking consumers in times gone by,16 rationing by
price-taking consumers certainly is indefensible in a health plan with managed care.
Recent research on the economics of managed care has begun to characterize rationing
within managed care, working along two tracks in modeling the effect of managed care.
One approach views managed care organizations as setting quantities for individuals
who may be heterogeneous with respect to severity of illness and demand [Baumgardner
(1991), Glazer and McGuire (2000), Ramsey and Pauly (1997)]. This approach views
managed care as specifying what a person, given a demand curve, would get in terms of
services under managed care. Note that if heterogeneous patients get the same quantity
of services, managed care rations in an inefficient manner. In this case, shifting resources
from consumers with low valuation to those with high valuation could improve welfare.
A second line of research proposes that managed care organizations, ration by "shadow
prices" [Keeler, Carter and Newhouse (1998), Frank, Glazer and McGuire (1998)]. In
this approach, it is "as if" consumers were charged a price, and the managed care plan
15 For example, it is common to find employers requiring MBHOs to achieve certain access standards, to
facilitate entry into treatment and to leave patients largely satisfied with the treatment process [see IOM
(1997)].
16 See McGuire (1998) for discussion of the price-taking assumption in health care markets.
914 R. G. Frank and T.G. McGuire
gave them all services that were valued above the shadow price and denied care for all
uses for which the value was below the shadow price. In this characterization of man-
aged care the shadow price determines a "need" or benefit threshold that a patient must
attain in order to qualify for treatment. Rationing by shadow prices is efficient in the
sense that this form of rationing analyzes the marginal benefit of services to all users.
The fact that managed care rations without demand prices has important implications
for the discussion of "parity" for mental health coverage and generally for the efficiency
of benefit designs for mental health versus general medical care. In our discussion of
moral hazard above we noted that the empirical evidence showing greater demand re-
sponse to cost sharing for mental health care relative to general medical care in the
context of fee-for-service-indemnity insurance arrangements implies that it would be
efficient to cover ambulatory mental health care differently from general health care.
This is an argument against parity for MH/SA. Normative conclusions about coverage
change are altered under managed care. For example, Ramsey and Pauly (1997) con-
sider the roles for quantity-type managed care rationing and demand-side cost sharing
in a model in which quantity received by the consumer is the minimum of what would
be demanded, or what the managed care firm would supply. They are concerned with
the optimal combination of the two rationing instruments. Some quantity rationing is
always part of the optimal policy, and it is unclear how different demand responses fit
into the story.
In the case in which rationing is only by shadow prices, we can show clearly the
potentially major impact of superimposing a new rationing mechanism for quantity de-
termination. In this case, the higher demand elasticity of mental health services does
not imply a higher shadow price, where this form of rationing is used, is in contrast
to the usual optimal insurance result. We first consider the problem diagrammatically.
Figures 5a and 5b characterize the demand curves for general medical care and mental
health care in accord with empirical findings from the literature. The demand for mental
5a 5b
General Health Mental Health
demand
\<mand
$1
.... .... 1...
q'
......
q
I
g' g m' m
Figure 5. Managed care rationing by shadow prices.
health services is more price elastic than is the demand for medical care. We begin by
imposing a shadow price of q to ration both types of services. At this initial shadow
price the managed care plan would provide g dollars in general medical care spending
and m dollars of mental health care for an individual with these two demand curves. If
we increase the shadow price to q' for both areas of care, which represents stricter ra-
tioning, the result is g' and m' levels of spending on general medical and mental health
care, respectively. The higher elasticity for mental health services means that the cut
back in services for mental health will be greater than that in other service areas.
The efficiency implications of a change in shadow price under managed care are
quite different from comparable changes in benefit design under indemnity insurance.
Because the shadow price is an "as if" price consumers do not actually bear more finan-
cial risk when it rises. There is no risk/inefficiency trade-off taking place to determine
the optimal shadow price. The only efficiency issue is the allocation of resources be-
tween health care and mental health care, which is done efficiently when the shadow
prices are equal, irrespective of the relative demand responses.
This can be shown more formally within a model that will first be used to define ef-
ficient rationing, and later be used to compare efficient rationing with that which may
occur in the presence of selection incentives. 17 Consider a planner seeking to maxi-
mize net benefits subject to a budget, for a given set of enrollees in a plan. The planner
chooses the level of rationing q, for each of s services. Each service quantity is ex-
pressed in dollars, assuming price is normalized to $1. Use of each service for each
enrollee is a function of its own shadow price only.
Specifically, suppose a plan has a membership consisting of Nt members of type t,
Et Nt = N, where a type of person corresponds to a valuation that person places on
health care services. Person of type t has a benefit function Bt(mlt(ql), m2t(q2), . .,
mst(qs)), where qs is the shadow price for service s and mst(qs) is the spending type t
can expect if the shadow price is q,.18
Note that qs is set in common for all users of service s. What each person gets is
dependent on their "need" or marginal benefit. Then Bt() is the person's valuation of
all the elements of expected spending. With this notation, total benefits to all enrollees
are
E NtBt(mlt(ql), m2t(q2), ... , mst(qs)) (1)
with a total cost of
E Nt ( mst (s)). (2)
17 The discussion in this section has benefited greatly from extensive discussions with Jacob Glazer and is
closely related to the discussion in Frank, Glazer and McGuire (1998).
18 Consumers value spending on the various services. Spending is determined by the shadow-price rationing
and expressed as mst (qs). Diagrammatically, the relation between q and spending can be seen in Figure 5.
916 R.G. Frankand T G. McGuire
Maximizing benefits less costs with respect to each qS implies
Nt(Btsmt -mt) = 0 Vs. (3)

t
Bts is the derivative of Bt (-) with respect to the sth argument, and m't = (dmt (qs))/dq,
Rewriting (3) yields
NtBts
Nt Btsm't N mst, or (4)
t t
Since rationing by shadow price implies Bts is the same for all types, (4) implies that
Bts = 1 Vs. (4')
All shadow prices should be set equal to one. All services would be subject to the same
level of rationing and that every dollar of spending would generate at least a dollar in
health care benefits. The implication of the first best is that the managed care organi-
zation should have a budget sufficient to cover the costs of all services implied by the
efficient level of rationing.
It should be recognized that the condition given in Equation (4) is based on the as-
sumption that the demand response for a service (e.g., cardiac care) is the same for each
person in the patient type category t. The demand response may, of course, vary across
services within a patient type category. This will be important for the selection analyses
presented in Section 4.3.
This normative problem could be transformed into a second-best analysis by maxi-
mizing the value of spending to the members of the plan subject to a given budget, rather
than maximizing benefits less cost. The second-best analysis is more general and covers
the case in which the goal is to maximize welfare subject to given levels of public funds
or perhaps a set of capitation payments. It is straightforward to show that the result in
(4') is altered so that the marginal benefit of spending in all services will be equalized
to a new shadow price, the value of the marginal dollar given the budget. This shadow
price is one, of course, at the first-best level of spending. It is above (below) one when
the health budget is set below (above) the first-best level.
Two points flow from this analysis that are important for the economics of mental
health. First, if managed care rations by shadow price, and does so efficiently, providing
full coverage (parity) for mental health services can improve efficiency by increasing
financial protection with no additional loss due to moral hazard, a point that has been
anticipated in some policy discussions [Frank, Koyanagi and McGuire (1997), National
Advisory Mental Health Council (1997), Sturm (1997)]. Second, efficient rationing im-
plies rationing equally across service areas, in the sense of setting the same shadow
price across services. Efficiency conditions imply equality in shadow prices irrespective
of demand elasticities. (Note that the demand-response term, mst, drops out of (4).)
The normative implications of this model of rationing are parallel to those that un-
derlie models used in cost-effectiveness analysis [Weinstein (1995)]. The manner in
which cost-effectiveness information has been proposed for use in resource allocation
is consistent with the simple model proposed above. Weinstein and Zeckhauser (1972)
propose an approach to efficient rationing within a budget using estimates of Quality
Adjusted Life Years (QALYs) derived from specific health programs. QALYs were ini-
tially designed as a means of developing a common metric for evaluating heterogeneous
health outcomes in a fashion which takes account of individual preferences for different
health states [see Kaplan (1995) for a review]. While that work focuses on rank order-
ing health treatment programs in terms of the QALYs they yield in order to find the
incremental intervention, the point here is that budget allocation process across classes
of "incremental" interventions should allocate funds so that the health benefit per dollar
(perhaps as measured by QALYs) is equal. 19
In the analysis just conducted, we have assumed that shadow prices were the only
rationing device, whereas, in fact, demand-side price rationing is likely to be going on
at the same time. A short-side model in which utilization is the minimum of demand
and supply is the approach noted above by Pauly and Ramsey. Another is a bargaining
model in which utilization is a compromise between what the patient wants at the price
he or she pays and what the seller wants to supply given the incentives and constraints it
faces [Ellis and McGuire (1990)]. Any supply-side instrument will have the virtue that
rationing is being accomplished without imposing financial risk on consumers, and,
therefore, is likely to be used "first" in putting together an optimal payment system
[Ellis and McGuire (1993)]. Supply-side policies, especially "one size fits all" types,
may run the risk of not respecting consumer preferences. Ramsey and Pauly's (1997)
approach suggests one way to capture this. Supply side policies also do not address
initial visits very well. The literature in health economics on optimal combinations of
demand and supply-side policies is thin, and reasonable incorporation of managed care
tactics in normative models is only beginning [Ma and McGuire (1997)].
Considerations of optimal payment systems in mental health encounter the same is-
sues about patient and provider motivation as do such discussions in the general field
of health economics. The assumption of rational, utility-maximizing, price-taking con-
sumers with fixed preferences seems, if anything, less convincing in demand for mental
health care than for general health.
Interpretation of Demand: The normative interpretation of mental health care de-
mand in the fee-for-service indemnity insurance world has long been problematic. Long
standing issues of asymmetric information and imperfect agency relationships that were
noted early in the study of health care markets [Arrow (1963)] temper the normative
19 The analysis would be altered if we assumed, Baumgardner-like, that managed care plans ration by quan-
tity setting, instead of assuming, Keeler, Newhouse, and Carter-like, that plans ration by a shadow price.
Maximization of benefits less costs with respect to ms with the implied constraint that mt = ms for all t
yields a public-goods like condition in (4). The weighted marginal benefits must be equalized across services.
In this altered analysis, the "expected" marginal benefits are equalized across services.
918 R.9G. Frank and TG. McGuire
interpretation of all health care demand functions as marginal benefit schedules. The
demand for mental health care has a special set of constraints on consumer informa-
tion. Demand for mental health services may be influenced by fear of stigma, Veblen's
"bandwagon" effects [McGuire (1981)] and unclear information about efficacy of spe-
cific treatments. In addition, many mental disorders affect the capacity of individuals to
make decisions in their own best interests [Rubin (1978)]. For these illnesses, placing a
strong normative interpretation on observed demand behavior is unlikely to be justified.
The relation between behavior and normative marginal benefit schedules under man-
aged care arrangements may differ from that in fee-for-service indemnity insurance for
many reasons related to new forms of rationing, including creation of managed care
provider networks, using payment systems that depart from simple fee-for-service ar-
rangements, and the use of utilization management techniques and information feed-
back. Managed care also may alter the traditional agency relationships between patient
and clinician [Blomqvist (1991), Mechanic (1998)]. Providers seeking to be part of a
managed care network must balance patient, managed care organization, and their own
economic interest differently than under FFS. If provider actions influence demand, the
relationship between demand-side price and quantity consumed must be constructed
with a whole new set of factors in the ceterisparibus under managed care. In general,
since managed care mechanisms are intended to affect rationing, we can expect that
the positive demand curve will change. Conducting research on these changes is an im-
portant element of the current research agenda in mental health, as well as in health
services.
We can speculate about what changes we might expect in demand behavior. If forces
to ration care in addition to demand-side prices are in play, the change in quantity
utilized with respect to a change in price is probably smaller. This emerges from a
short-side model [Ramsey and Pauly (1997)], because some price changes for some
consumers do not affect quantity since the supply constraint is binding. It would also
emerge from a bargaining model because the quantity used is a compromise, and even
if "desired demand" changes by the same amount, actual use is a weighted average of
desired demand and desired supply (which does not change). Thus, here too, alternative
rationing dampens demand response.
The normative view of such changes must be approached cautiously for reasons stated
above. Nevertheless, at the very least the "cost control problem" that has been such a
central concern in mental health policy is potentially attenuated with managed care. The
welfare loss from equal coverage, or parity, would also likely be reduced.
3.3. Managedcare, supply side incentives and moral hazard: evidence
Alternative methods of organizing and paying for mental health and substance abuse
services within private insurance, Medicaid and Medicare have received increasing at-
tention from researchers and policy makers in the US since the mid-1980s. Similarly,
Britain has recently begun to use supply side incentives to affect resource allocation in
the primary care and specialty mental health sectors [Knapp (1997)]. To date most of
the evidence on the impact of payment and organizational arrangements on utilization

is based on the US experience.
Research and policy debates about supply side payment incentives have primarily
been focused on hospital and health plan payment schemes. Prior to the mid- 1980s pay-
ments by insurers in the US were commonly made on the basis of retrospectively in-
curred costs or charges for hospital care and charges or customary fees for professional
services [IOM (1993)]. Since the mid-1980s payments have increasingly been made
prospectively and are based on days or discharges for hospitals and on a per capita basis
for health plans and groups of health care professionals.
Prospective payment systems were introduced in order to give providers incentives
to control costs. The basic idea is to separate the level of payment, determined prospec-
tively, from the costs the provider incurs. In practice, payment systems are not always
completely prospective, and may include some features, which, at least partially, reim-
burse costs. In Medicare's Prospective Payment System for hospital discharges, there
are outlier features for some costs. Medicare's TEFRA system for paying psychiatric
hospitals and psychiatric units of general hospitals has explicit features that increase
payments if a facility's cost rises above a certain level [Cromwell et al. (1992)]. State
payment systems for Medicaid may include rebasing features which, in effect, reim-
burse part of costs. To capture elements of prospective and retrospective payment that
characterize actual payment systems, we can characterize a general payment system in
the following way:
P =a+,C, (5)
where a is the prospective portion of payment that is unrelated to incurred costs and can
be paid in connection to any of the above mentioned units of payment (days, cases, per
capita); B is what has been referred to as the supply side cost sharing parameter [Ellis
and McGuire (1986b), Newhouse (1996)] indicating the portion of incurred costs, C,
that are reimbursed in the payment system. Cost based reimbursement can be charac-
terized by setting a = 0 and P = 1. A pure form of per case prospective payment for
hospitals (similar to Medicare's Prospective Payment System or PPS) pays on the basis
of admissions to hospitals, sets a = the average cost of a case, and B = 0.20 A "mixed"
system is in the general case, where a > 0 and 0 < < 1. One can use this categoriza-
tion for contracts to managed care organizations.
Under managed care arrangements in the mental health and substance abuse area
there are three dominant approaches to payment, all based on the cost of care per capita
for a defined population: pure capitation where the per capita payment is set such that
a = the average MH/SA costs per enrollee, and B = 0; mixed payment or soft capitation
20 The average cost is typically based on actual costs in a previous year and is usually adjusted for the case
mix of the users of the service. Case mix will be discussed in the context of risk adjusted capitation rates
below.
where a > 0 and 0 < < 1; and Administrative Services Only (ASO) contracts where
claims are reimbursed on a fee-for-service basis but the managed care organization is
judged and contract renewal is related to its performance in controlling spending.
Representation of payment systems with a prospective and retrospective component,
in a mixed system, has also been useful in research on "optimal payment systems",
which may feature only partial supply side cost sharing (with 0 < < 1) for reasons of
agency [Ellis and McGuire (1986b)], or information issues [Newhouse (1996)].21
3.3.1. Research on hospitalpayment systems
The impact of payment arrangements on inpatient psychiatric care has been the sub-
ject of many studies. 2 2 Analyses of natural experiments within Medicare, Medicaid and
several states which regulated hospital payments have formed the foundation for the re-
search. In Medicaid research indicated that there were strong responses to various forms
of prospective payment relative to cost based reimbursement [Frank and Lave (1989),
Lave and Frank (1990)]. Studies of Medicaid also suggest that the responses to per case
prospective payments appear to be more complex than one might have expected. For
example, relative to cost based payment fewer psychiatric patients paid under per case
prospective payment had very short hospital stays (1 to 10 days) and fewer also had
long stays (over 30 days) resulting in a lower average length of stay under prospec-
tive payment. Ellis and McGuire (1996) using data from the state of New Hampshire
found hospitals respond both to the level of payment ("a") as well as to the degree of
supply-side cost sharing ("i")in terms of quality competition.
Studies of Medicare data used implementation of the PPS to quantify general hospital
response to the new payment arrangement [Frank et al. (1987), Freiman et al. (1989),
Lave et al. (1988)].23 The estimated impacts of the new payment system showed length
of stay reductions for inpatient psychiatric care in the range of 17% to 25%. The es-
timated effects of the initial implementation of PPS for general medical care was to
reduce length of stay by between 9% and 12% [PROPAC (1986)].
Several states implemented all-payer rate setting programs in the 1970s and 1980s
in an effort to control hospital costs. Rupp, Steinwachs and Salkever (1984) exam-
ined the impact of the per case prospective payment method introduced in Maryland,
finding evidence for decreases in length of stay, and increases in re-admission rates.
Frank and Jackson (1989) studied the introduction of prospectively set hospital budgets
(a = historical budget adjusted for general inflation and / = 0) in two regions of New
York State during the 1980s. Compared to "control" hospitals paid using a prospectively
21 The average cost is typically based on actual costs in a previous year adjusted for inflation and is usually
adjusted for the case mix of the users of the service. Case mix will be discussed in the context of risk adjusted
capitation rates below.
22 See Harrow and Ellis (1992) for a review of the early literature.
23 Most specialty mental health providers were exempt from PPS and paid under the mixed system, TEFRA.
set per diem payment, prospectively set budgets led to reduced admissions of between
16% and 22%.
In summary the evidence from the application of prospective payment methods to
hospitals suggests strong hospital responses to incentives to reduce costs and utilization
of psychiatric services.
3.3.2. Research on managed care and capitation
The general form of a contract as a "mixed system" contract can be applied to capitation
and managed care. Figure 6 illustrates forms of contracts that are based on capitation
arrangements that are commonly found in connection with managed behavioral health
contracts. The horizontal axis measures actual costs incurred by the managed care orga-
nization. The vertical axis measures payments made to the managed care organization.
The 45 ° line represents cost based reimbursement, that is payments equal incurred costs.
The horizontal line at C represents "pure" capitation ( = C and = 0). The managed
care organization bears 100% of the financial risk for incurred costs per enrollee. For
managing care of a population at spending levels to the left of T the managed care or-
ganization collects a profit equal to the vertical distance between the line at C and the
45 degree line. If spending exceeds T the managed care organization incurs a loss equal
to the vertical distance between the line at C and the 45 ° line. The financial incentive
to restrain spending is clear. Most HMOs and some carve-out programs use this type of
payment arrangement.
The line NEB represents a mixed or soft capitation payment system. Under these
types of arrangements the payer sets a capitation target at T. Thus, if the managed care
organization incurs costs per enrollee equal to T it will break even. If the managed
care organization incurs costs between points L and M on Figure 6, the payer and the
paymen
Figure 6. Payment systems for managed care.

managed care organization will share the profit or loss according to the slope of the
line segment EB (or P in our characterization of payment above). At levels of spend-
ing above M and below L the payers holds all the risk. Profit and losses are capped
at spending level L and M. This means that the managed care organization's maxi-
mum profit is EF and the maximum loss is AB. This is illustrated in Figure 6 by the
fact that line segment NE is parallel to the 45 ° line. Under the soft capitation payment
method there are incentives to reduce spending but they are weaker because the link
between payment and incurred costs has not been fully severed ( > 0). One rationale
for such arrangements is concern that managed care organizations may "over manage"
the mental health benefit (possibly due to selection incentives discussed below). A sec-
ond rationale relates to the fact that most payers are large relative to specialty managed
behavioral care organizations and thus are in a better position to bear risk.
3.3.2.1. Evidence: the early experiences. Risk contracting for managed mental health
care predates the specialty industry of managed behavioral health care (MBHC). Pre-
paid group practices and health maintenance organizations (HMOs) have accepted risk
contracts for MH/SA services, along with other health care, for many years. In general,
these contracts are capitation contracts in which all the cost or claims risk is born by
the prepaid group. As part of the Health Insurance Experiment (HIE), Manning and col-
leagues (1989) compared the cost and use of care by families assigned to a prepaid group
practice, the Group Health Cooperative of Puget Sound, with that of families assigned to
receive free care in the fee-for-service sector. Although enrollees with the two types of
insurance sought care at the same rate, the fee-for-service population had mental health
expenditure levels almost three times greater than enrollees in the prepaid health plan
($69.70 vs. $24.60 in 1977 dollars). Paula Diehr and colleagues compared the use of
outpatient mental health care in a fee-for-service unmanaged benefit plan, a staff model
HMO, and an individual practice association (IPA) prepaid plan for Washington State
employees, with results that were consistent with Manning's [Diehr et al. (1984)]. How-
ever, because the Washington employees chose their plan and were not assigned to an
insurance plan as in the Health Insurance Experiment, the Diehr findings may at least
partly reflect differences in each study group's needs and not just an effect of the plan.
Prepaid groups can exert direct managerial authority over the supply of mental health
services. Indeed, by controlling the number of therapist hours available, they can almost
directly ration the volume of care to be provided. Managed behavioral health care com-
panies, however, may have weaker incentives to reduce costs than prepaid groups do,
and they typically have much less direct control over their contracted providers. Thus,
the cost reductions from managed behavioral health care should be expected to be more
modest than from prepaid groups.
The literature on specific MBHC programs is relatively recent. The initial experi-
ences reported by employers include some instances of large reductions in the costs
of MH/SA care. Hodgkin (1992) reviewed the early literature on the effects of utiliza-
tion management, finding very few studies that offered evaluation methods that could
produce convincing results. The lone study that was methodologically sound showed
savings in the neighborhood of 10% to 15% on total claims costs, a faint indication of
what was to come.
3.3.2.2. More recent evidence. The CHAMPUS program experimented with an "at
risk" PPO for behavioral health services during the late 1980s in the Tidewater, Virginia
area. That area was known to be a high cost region with regard to MH/SA outlays.
The demonstration showed significant savings (about 31% below expected costs in the
absence of the program) stemming largely from reduced use of inpatient care. In spite
of the reported savings, there were clearly areas of considerable waste in expenditures
and difficulties in effectively running the program [Coulam and Smith (1990)].24
A number of private corporations have adopted specialty MBHC carve-out programs.
It is fairly common to see reports of reductions in claims costs of 40% to 50%. The in-
terpretation of these changes is, of course, quite difficult. Often, more than one change
is made and attributing cause and effect is difficult. Within a plan, there can be consid-
erable year-to-year variation for unknown reasons [Dickey and Azeni (1992), McGuire
(1994)]. Finally, a version of the "file-drawer" problem in research may be at work; only
"good" (read "publishable") experience may see the light.
With these limitations in mind, we review the reports of the performance of Managed
Behavioral Health Care (MBHC) carve-out programs. Key aspects of these studies are
summarized in Table 4. As the table indicates each of these natural experiments has
taken place in the context of different institutional arrangements. Some carve-outs were
implemented within State Medicaid (Massachusetts and Utah) programs while others in
privately insured populations (the GIC and Pacific Bell). The risk-sharing arrangements
also varied considerably. In Utah a "pure" capitation contract was phased in, while in
Massachusetts, the GIC and Medicaid contracts shifted only a small amount of risk to
the MBHC vendor. The Pacific Bell ASO contract involved no financial risk at all to the
MBHC vendor.
Table 4 reports impressive reductions in mental health spending relative to fee-for-
service arrangements (the comparison condition for all the studies). The estimated re-
ductions in spending range from - 17% to -43%. The reductions for the most part took
place in the context of programs that had historically experienced high levels of spend-
ing on mental health services, with the exception of the Utah study. While the observed
24 The CHAMPUS program has continued to experiment with managed care. The program is now subject to
three differing forms of managed care arrangements. The National Utilization Management program works
under a CHAMPUS contract with a specialty MBHC vendor to provide pre-admission certification and con-
current review on a nation wide basis. The contract with the MBHC vendor does not place the vendor at any
financial risk related to utilization of MH/SA care. The CRI program under CHAMPUS is a fixed price "at
risk" contract that is in place in Hawaii and California. Finally, the Tidewater CPA arrangement continues
to be in operation. In recent years the CRI and the general MBHC arrangement have realized the largest re-
ductions in costs. The Tidewater plan reported a small increase in costs. The absence of cost reductions in
Tidewater during the 1989 to 1992 time period may be due to the substantial savings that were realized during
the early years of the program. A 9% increase over 4 years that was reported for the CRI program is quite
small for any health plan during that time period.
924 R.G. Frank and T.G. McGuire
Table 4
Carve-out impacts
Case Risk Impact Studies

arrangement
Mass GIC RS Pre-post outpatient spending Ma and McGuire (1998a)

(-25% -- -16%) Huskamp (1997)
Total spending (-32%)
Mass Medicaid RS Pre-post spending Callahan et al. (1995)
(-26% -22%) Frank et al. (1996)
Frank and McGuire (1997)
Utah Medicaid 1989-1992 Cap Pre-post spending (-17%) Christianson (1995)
Pacific Bell 1988-1995 ASO Pre-post spending (-43%) Goldman et al. (1998)
ASO = Administrative Service Only; RS = Risk Sharing; Cap = Capitation.
savings across studies were in many respects achieved by similar shifts in services uti-
lization patterns, there are some important differences. Savings were primarily realized
by (1) reductions in use of inpatient hospital care (all studies), (2) reductions in nominal
prices paid to providers [Goldman et al. (1998), Ma and McGuire (1998b), Callahan
et al. (1995)], and (3) reduced duration of outpatient treatment [Goldman et al. (1998),
Huskamp (1997)].
The third column of Table 4 reports changes in spending on mental health and sub-
stance abuse care for the insured populations under study. These data reflect impor-
tant differences in the utilization patterns observed across studies. The Massachusetts
Medicaid experience saw an initial increase in use of behavioral health care following
introduction of the carve-out program [Callahan et al. (1995)]. After three years the in-
crease in use had largely vanished [Frank, McGuire, Notman and Woodward (1996)].
In the Pacific Bell study a significant increase in the percentage of enrollees using any
behavioral health care was estimated (a 17% rise). In contrast, the Massachusetts GIC
experienced very large reductions in the percentage of the population using behavioral
health care (20% to 30% reductions). It is interesting to note that some companies such
as Sterling-Winthrop report dramatic increases in access to care (50% increase in rates
of utilization), due to expanded use of outpatient care, at the same time that claims costs
were falling. Reductions in rates of use create concern because they may indicate re-
ductions in access to care for individuals that may benefit substantially from treatment.
Managed care programs are quite complicated and use many methods of rationing to
control use. The studies discussed also reflect heterogeneous populations and differ-
ences in other institutional features. For this reason there are as yet no clear explanations
for why the response to managed care arrangements might vary so strongly in terms of
the percentage of the population using care.
Theory implies that the more high-powered incentives associated with pure capita-
tion should lead to greater cost reductions in comparison to lower-powered incentives
with risk sharing or ASO contracts. As Sturm (1997) pointed out, this pattern has not
materialized in the experience so far: large reductions have occurred even without high-
powered incentives. The high/lower power of a contract is one dimension, but actual
contracts can be quite complex, especially with regard to their dynamic incentives. The
first Massachusetts Medicaid contract made a fixed payment for administrative costs
(giving incentives not to spend on administration and managing care), weak incentives
for cost reduction, and in some years ratchet effects which create a link between targets
in future years and performance in past years [Frank and McGuire (1997)]. A powerful
incentive in this new and growing industry is the role contract performance will have on
future business. This may be the dominant incentive with respect to all current contracts
[Ma and McGuire (1998a)], perhaps serving as an explanation of why large cost sav-
ings emerge in many forms of contracts. Relating experience to the form of the contract
seems particularly treacherous on the basis of current data, given the rich set of incen-
tives that are probably operative. Case studies are building an empirical base on which
conclusions ultimately can be drawn.
Although much work remains to be done about the magnitude of savings that can be
expected in particular circumstances and the connections between savings and contract
features, it seems clear that managed care can substantially reduce costs in MH/SA.
Some research has taken place on the quality impact of managed care. Generally, in
comparing fee-for-service to capitated managed care plans does not reveal a uniform
quality impact one way or another [Miller and Luft (1997)]. In mental health, two
studies have found that quality may be adversely affected in HMO-style managed care
[Wells et al. (1996), Lurie et al. (1992)]. Merrick (1997) studied the pattern of claims
for persons hospitalized for major depression prior to and after the carve-out plan in the
GIC plan noted above. Her results pointed to more appropriate patterns of care under the
carve-out. Readmissions did not rise, and contact with outpatient providers following
discharge improved under managed care.
There is as yet very little research on responses to differences in risk sharing arrange-
ments across MBHC plans. Sorting out these explanations requires careful measurement
of contractual features and market circumstances facing MBHC vendors. Thus, while
there has been considerable progress in estimating the gross spending and utilization
responses to MBHC contracts, we have a long way to go to understand the specific con-
tractual and market mechanisms that generate such changes in the delivery of mental
health and substance abuse care.
4. Insurance markets, adverse selection and mental health care
The special effort needed to control moral hazard in MH/SA has been put forward as
one reason why MH/SA services are organized and paid for differently than other types
of health care. Managed care represents a new set of institutions that appear to change
the terms of Zeckhauser's (1970) dilemma, allowing moral hazard to be controlled with-
out reduction in risk spreading [Mechanic (1997)]. The speculation above that managed
care can substitute for demand-side cost sharing as a cost control device would sug-
gest that insurance coverage for MH/SA should improve with managed care, as the goal
of risk spreading could be pursued with less moral hazard cost. Nevertheless, in the
early and mid-1990s, when managed care was emerging, we observed two significant
developments in insurance markets related to MH/SA. First, there was some evidence
of erosion of insurance coverage for MH/SA [Buck and Umland (1997)]. In partic-
ular, the portion of health plans which imposed tight limits on coverage of MH/SA
care appears to have grown during the 1990s. This is puzzling given the rapid expan-
sion of enrollment in managed care plans [PPRC (1997)]. A second development has
been the growth of specialized behavioral health carve-out programs [Frank, Huskamp,
McGuire and Newhouse (1996)]. Appearance of coverages and insurance arrangements
more generally reflect the profit-driven considerations of adverse selection, as well as
concerns for moral hazard. As in the case of moral hazard, evidence suggests that the
forces of adverse selection may work more powerfully in mental health than in health
care.
In the context of insurance coverage for mental health services, conventional wisdom
is that high cost enrollees are attracted by relatively generous coverage provisions for
mental health and substance abuse care. Competition among indemnity insurance plans
may have resulted in inefficiently low levels of coverage for behavioral health care.
This was the basis of argument in the 1980s that justified federal and state "mandated
coverage" legislation requiring private insurance to cover minimum levels of mental
health care [McGuire and Montgomery (1982), Frank (1989)].
In the sections that follow, we review the evidence on selection in mental health
and substance abuse, discuss the policy responses by government in the context of fee-
for-service indemnity insurance contracts and then examine selection in the context
of managed care. This discussion will point to explanations for the new institutional
arrangements that are arising in the MH/SA sub-sector.
4.1. Evidence of selection in MH/SA
The Federal Employees Health Benefit Program (FEHBP) during the 1960's and 1970's
provided an early example of how concerns about selection drove competing insurers
to lower benefits for MH/SA services. Plans offering more generous benefits quickly
attracted individuals who wanted to avail themselves of these services. The generous
coverage of MH/SA lost viability as people not expecting to use services enrolled in
plans with more limited coverage [Reed (1975)]. Use of mental health care has been
found to be two to three times higher in the Blue Cross/Blue Shield "high option" plan
compared to the low option plan, even though the actual coverage differences are quite
small [Padgett et al. (1993)], implying that the differences in use were due to selection
rather than demand response (moral hazard). Further evidence for adverse selection in
the FEHBP comes from comparing responses to the price of MH/SA care under the
FEHBP and The RAND Health Insurance Experiment [Newhouse et al. (1993)]. In the
RAND experiment, individuals were randomly assigned to health insurance plans, and
the observed price response to differential coverage was substantially lower than what
was observed in FEHBP [Newhouse et al. (1993)]. The differences in price response
suggest that where plan choice was possible (under FEHBP), the "high option" (lower
priced) plan differentially attracted poorer risks making it appear as if the plan with
slightly more generous coverage induced much higher utilization of MH/SA care.
Adverse selection is an issue for all of health insurance, but may be especially se-
rious in the mental health area. Deb, Rubin and Wilcox-Gok (1996) found that indi-
viduals with a family member with a mental illness were more likely than otherwise
similar members of the US population to choose coverage with more generous men-
tal health care provisions. Sturm and his colleagues (1994) analyzed the treatment of
depression across health plans as part of the Medical Outcomes Study (MOS), finding
that depressed individuals receiving care from specialists were more likely to migrate
from prepaid to fee-for-service plans. They also found that individuals switching from
prepaid to fee-for-service plans were at risk for poorer outcomes. Ellis (1988) exam-
ined the persistence of spending over time and its implications for health plan choice.
Individuals with a history of mental health care utilization had persistently higher levels
of spending than did otherwise similar insured individuals. He also found that a history
of mental health care utilization had a significant impact on an individual's choice of
health plan. Higher levels of prior year mental health spending increased the likelihood
that an enrollee chooses a low deductible plan. This suggests choice based on antici-
pated spending such that the expected deductible payments exceed the differences in
plan premium differentials.
Perneger and colleagues (1995) found evidence of adverse selection related to mental
health care in the context of insurance markets in Switzerland. They analyzed a situation
where one indemnity plan among several health plans was changed to a managed care
plan. The managed care plan introduced gatekeepers and limits on insurance coverage
for psychiatric services. Those who remained in the indemnity plan made on average 2.3
more visits for mental health care in the previous year and were more likely to receive
a prescription for a psychoactive medication than those who chose managed care.
Taken together, these results suggest that users of mental health care may have greater
subsequent year health care spending than otherwise similar people, putting plans at-
tracting mental health users at a financial disadvantage. Persistent levels of above-
average spending for the individuals with severe mental disorders within the Medicaid
program was recently reported by Kronick et al. (1996). In sum, there is both direct and
indirect evidence suggesting that the mentally ill and substance abuse users are associ-
ated with higher levels of health care spending and that they systematically select health
plans that offer more generous coverage for behavioral health treatment. Such behavior
creates economic incentives for health plans to adopt strategies that will reduce their
attractiveness to users of mental health care.
4.2. Policy responses to selection:fee-for-service-indemnity contracts
During the 1970s and 1980s competition to avoid "bad risks" was channeled into limit-
ing coverage for treatment of mental and addictive disorders. Approximately 22 states
counteracted adverse selection by mandated benefit statutes which specified minimum
928 R.G. Frank and T G McGuire
levels of coverage for MH/SA care [McGuire and Montgomery (1982), Frank (1989)].
These statutes generally specified coverage minimums in terms of coinsurance, limits
on outpatient visits and hospital days, and deductibles. Since benefit design features
were the key provisions of an insurance contract determining coverage, regulation of
these components of coverage was potentially effective in limiting market failure asso-
ciated with adverse selection. The impact of mandated benefit statutes was limited due
to exemption of self-insured employers under ERISA. It is worth noting that most large
self-insured employers (often with populations in several states) typically offered their
employees health insurance plans that complied or exceeded the terms of most state
mandated benefit statutes.
This strategy towards "fixing" difficulties in the insurance market continues today. In
1993 and 1994, a debate took place regarding mandated benefits in insurance as pro-
posed under President Clinton's Health Security Act. The inclusion of MH/SA as part
of the benefit mandate was especially contentious, primarily because of concerns over
the costs of such provisions. The same argument reappeared in 1996 in the form of pro-
posed legislation that would call for parity in benefit design provisions between health
benefits and those for MHISA care. Again, concern over the costs of such mandates
and the uncertainty around predicted impacts strictly limited the scope of the legislation
that eventually passed [Appropriation Authorization for the Department of Veterans Af-
fairs (1996)]. Attenuating selection-related incentives is the main efficiency argument
supporting policies to mandate insurance benefits.
4.3. Selection and managed care: distorting "quality"
As competition among managed care plans becomes the predominant form of mar-
ket interaction in health care, adverse selection takes a new form which may actually
be harder to address in policy, relative to traditional forms of health insurance con-
tracts discussed above. That is, as health insurance moves away from traditional fee-
for-service-indemnity arrangements, where enrollees have free choice of providers, and
becomes managed care, the mechanisms a health insurance plan uses to effectuate se-
lection change from readily regulated coinsurance, deductibles, limits and exclusions,
to more difficult to regulate internal management processes which ration treatment in
managed care plans.
Researchers on the economics and payment and managed care are well aware of
the issue. Ellis (1998) labels under provision of care to avoid bad risks as "skimping".
Newhouse et al. (1997) call it "stinting". As Miller and Luft (1997) put it:
"Under the simple capitation payments that now exist, providers and plans face strong disincentives
to excel in care for the sickest and most expensive patients. Plans that develop a strong reputation for
excellence in quality of care for the sickest will attract new high cost enrollees ...
The flip side, of course, is that in response to selection incentives the plan might
provide too many of the services used to treat the less seriously ill, in order to attract
good risks. A plan, motivated by selection, might provide so many of certain services
that enrollees may not benefit in accord to what it costs the plan to provide them [New-
house et al. (1997)]. Hence, in the presence of selection-related incentives, capitation
and managed care market forces will generate too little care in some areas and too much
in others.
This set of observations points to the likelihood that competition in the context of
managed care health plans will create strong incentives for rationing rules to be based
not just on the relative benefits provided by a service given an overall health care bud-
get, as was implied by the second best equilibrium among health plans described in
Section 3.2 above. Instead, the nature of competition between health plans forces plans
to take account of both the direct cost containment impacts of rationing (e.g., setting
a shadow price at a given level) as well as indirect effects associated with the types of
enrollees that are attracted to a plan under different patterns of rationing across services.
The classic asymmetry of information between insurer and enrollees of Rothschild and
Stiglitz (1976), along with these market forces may create distortions in rationing rules
that result in service competition to attract profitable enrollees.
We demonstrate this latter point by returning to the model of the planner's prob-
lem of setting shadow prices for managed care plans given in Equations (1)-(4) above.
Consider now profit maximization, and how this condition compares to the condition
for efficiency. Profit maximization is used to describe the objectives of the plan. Ear-
lier, in (4), we described the conditions for social efficiency in regard to managed care
rationing. If selection were not an issue, a plan seeking to attract enrollees would have
incentives to offer efficient insurance [Zeckhauser (1970)]. Otherwise, another plan with
the efficient combination of premium and rationing would attract the business. With the
introduction of selection problems, however, the close relation between the normative
(efficient) and positive (profit-maximizing) plan will be disturbed.
Here we characterize the nature of the distortion introduced. We also introduce
risk adjustment at this point, since the purpose of risk adjustment is to contend with
selection-related incentives. 25 Define Rt to be the risk-adjusted payment a plan gets for
enrolling a person of type t.2 6 Profits are then:
= NtR, t- N ( st(qs))
Recognize that the number of persons of type t joining a plan, Nt, is a positive function
of the benefits they anticipate, Nt(Bt). Define Ct = Ys mst(qs). Ct is the cost of a
person of type t. The first order condition for profit maximization with respect to q is:
NBstmstRt - N BstmsCt- E Ntmst =0 Vs.

t
25 See also Van de Ven and Ellis (2000) on risk adjustment.

26 This could be regarded as averaged over the characteristics of persons of type t.
930 R.G. Frank and TiG. McGuire
Rewriting, we have
N'Bstmst(Rt - C) E Ntm't;
I t
with the assumption that the demand elasticity for each type of person is the same for
any service, this implies:
CNtBst(Nf/Nt)(Rt - Ct)mst
=1. (6)
N
Comparing conditions (6) and (4) reveals that the efficiency condition is embedded
in the profit maximizing condition, allowing us to characterize the distortion caused
by selection. The new term is: (Nf/Nt)(Rt - Ct)mt in the numerator of (6). The two
parts of the term due to the selection distortion have to do with the responsiveness of
membership to a change in the anticipated benefits of membership in the plan, 2 7 Nt/Nt,
and to the profit and loss consequences of membership of a person of type t, and to the
level of spending on service s, mst.
Suppose all types are equally responsive to benefits, so Nt/Nt is the same for all t.
Then, the term that will create distortions is the relation of risk adjustment to cost for
persons of type t and its correlation with spending for a service. Consider first what
happens without risk adjustment. Then, Rt = R, and a common payment is made for
all enrollees. The term R - Ct will be smaller for persons of a "high cost" type. If the
cross-product of this term with mt tends to be large, that is, if people of this high cost
type tend to put a high value on service s, then the numerator of (6) will be large, and the
shadow price for this service will be set "too high" (relative to the social optimum) to
discourage membership by this high cost type. If the difference Rt - Ct is the same for
each type t, that is, if risk adjustment compensates for type differences in expected cost
in a way to equalize Rt - Ct, then risk adjustment will be effectively dealing with the
incentive to distort just described. Equation (6) describes a situation where the profit-
maximizing plan sets q "too high" (rationing too tightly) for services that are valued by
persons for whom risk adjustment "underpays" and sets q "too low" for services valued
by those for whom risk adjustment pays generously. 2 8
27 For presentational purposes, we disregard here the nature of persons' and plans' expectations about ben-
efits and costs. We treat these as common knowledge. Frank, Glazer and McGuire (1998) analyze a similar
model where the benefit functions are expected benefits by the consumer. Plans set q's on the basis of their
expectations of the distribution of consumers' expected benefits in the population. Mental health care is rela-
tively predictable, making mental health possibly more vulnerable to selection incentives.
28 Frank, Glazer and McGuire (1998) develop this line of argument and propose a distortion index stemming
from the selection-related incentives in the context of profit-maximizing health plans paid by risk-adjusted
capitation. Based on an equation like (6), they show that services that are rationed tightly in managed care
are those that are predictable by the individual, and those with a positive correlation with other (predicted)
spending. The Frank, Glazer and McGuire (1998) index is illustrated using Medicaid data for AFDC-eligible
All services are potential candidates for selection-driven distortions under managed
care. This is another way of saying that incentives to under- or over-provide mental
health services within a capitated plan must be considered in relation to the incentives
to supply other services. Mental health may be one of the services most distorted, but
there will be others, and mental health may not be the most in need of economic rescue.
Characterizing the incentives and monitoring the actions of managed care plans is a
central issue in the economics of health and mental health.
4.4. Policy responses and managed care
Suppose it has been determined that some service, say mental health, needs special
protection in a health insurance market with managed care. One implication of our
analysis of managed care is that policies that focus primarily on the nominal insurance
benefit will not be sufficient to ameliorate the inefficiencies created by selection-related
incentives. Managed care insurance contracts, with their complex rationing devices, are
more remote from regulation than traditional fee-for-service-indemnity contracts. Many
of the instruments that are used to ration care under managed care are difficult for a
regulator to observe and require clinical judgments about individual cases.
An example is the application of the concept of "medical necessity". Most managed
health plans cover medically necessary services. Medical necessity and therefore effec-
tive coverage depends on a complex set of interactions involving features of the benefit
package, the structure of the provider network organized by the health plan, financial
incentives facing providers and the administrative mechanisms used to assign patients
to specialty care and manage quality assurance. Determination of medical necessity
occurs on a case-by-case basis, thereby conferring discretion on those making the deci-
sions such as primary care physicians, plan clinical staff, specialists, and case managers.
In a word, the management of care within a health plan has become increasingly non-
contractible.The nominal insurance benefit has become one part of a complex contract
which rations care and provides protection against the financial risks of treating illness.
For many years, advocates for mental health and substance abuse have sought to
achieve "parity" in insurance benefits. The analysis presented above suggests that such
efforts, if they are successful, will not be sufficient to guarantee equality in access to ser-
vices in mental health [Frank and McGuire (1998)]. If managed care rationing devices
cannot be directly controlled, what options are available to a regulator?
Risk adjustment of capitation payments and carve-out arrangements are two re-
sponses to selection-related incentives. Purchasers of managed care services are making
adults from the State of Michigan. They calculate the selection-related distortion index for eight major classes
of services. Mental health expenditures are relatively predictable, largely because of the high year-to-year
correlation. They are not, however, different in their correlation with other costs, at least in this predominantly
young, female population. Applying the index does reveal that mental health is a service more subject to
selection problems, though the results are sensitive to the informational assumptions used.
use of each approach to deal with biased selection in the case of mental health and sub-
stance abuse. For example, the State of Maryland has chosen to integrate substance
abuse services for Medicaid enrollees with all other medical care. Selection related in--
centives are being addressed by using risk adjustment to adjust capitation rates for dif-
ferences in enrollee health care risk. In contrast, the State of Arizona carves out (as in
Figure 2) all mental health and substance abuse care from its general Medicaid HMOs
and contracts separately with one specialized managed behavioral health care organi-
zation (MBHO) in each region of the state. A third configuration is being proposed in
New York, where mental health is carved out of the Medicaid HMO program. However,
multiple MBHOs would be permitted to compete to enroll individuals for their mental
health care. Each competing MBHO is slated to be paid a flat capitation fee [Office of
Mental Health, State of New York (1996)]. How well can risk adjustment and carve-outs
be expected to do in countering selection incentives?
4.4.1. Risk adjustment
Managed care plans can engage in various activities designed to select good (profitable)
risks from an insurance pool [Cutler and Zeckhauser (2000)]. They may prevent or
discourage high-cost individuals from joining their plan (sometimes called "dumping")
even if this is prohibited under "open enrollment" regulations. They may also distort the
services they provide in order to attract the good and deter the bad risks, a perfectly legal
activity. Risk adjustment is intended to counter incentives to engage in activities which
may lead to inefficient health plan services and unequal access for potential enrollees.
Risk adjustment of capitation rates makes use of information about the characteris-
tics of individuals to align payments with expected costs. The rationale is that the closer
payments track costs, the less services will be the subject to inefficiencies such as the
ones just mentioned. For example, age, sex, welfare status, and county of residence have
traditionally been used to adjust Medicare's capitation to HMOs enrolling program ben-
eficiaries. If those over the age of 75 years are found to cost more, premium payments
on the behalf of those older beneficiaries are adjusted upward by an estimate of their
higher average cost. Most risk adjustment systems rely on demographic factors and clin-
ical information on individuals from past time periods. The clinical information usually
consists of diagnoses and procedures arranged in clusters based on clinical judgments
regarding the complexity and intensity of past treatment [Ellis et al. (1996), Weiner et
al. (1996)].
The empirical research used to develop risk adjusters can be viewed in the context
of an empirical model of health care spending that relies on pooled time series and
cross section data [Newhouse (1996), Newhouse et al. (1989)]. Equation (7) is a simple
characterization of such a model
Sit = a + XitB + Ii +-sit. (7)
Sit is spending for individual i in period t, Xit represents a set of characteristics of in-
dividuals that are included in the risk adjustment system, /i is a time invariant individual
Ch. 16: Economics andMental Health 933
effect, 8 it is a possible auto-regressive error with mean zero, and a and B are param-
eters. Most evaluations of risk adjustment rely on the ability of models such as that
given in Equation (7) to explain variation in individual spending as measured by an R2
statistic. Newhouse and colleagues (1989) pointed out that a more appropriate standard
for judging the ability of a risk adjusted payment system to attenuate selection related
incentives is to measure the portion of the "explainable" variance accounted for by the
risk adjustment system. Individuals (or plans) can only select a health plan (or deter
enrollment) based on spending they can predict. If one makes the assumption that indi-
viduals know the information contained in a set of X's and their past use, the explainable
variance consists of variation associated with the Xs, it, and any auto-correlation in
[Newhouse (1996)].
Risk adjustment can be thought of as a tax-subsidy scheme [Diamond (1992)], in-
tended to correct selection-created inefficiencies. Selection problems can take two gen-
eral forms: individual-based discrimination and plan-wide actions such as service dis-
tortions. In light of these multiple objectives, empirical risk adjusters will not be able
to achieve a first-best. Thus, to evaluate alternatives from a welfare standpoint, a more
explicit welfare criterion than an R-squared is necessary. In a new literature on op-
timal risk adjustment, weights on risk adjusters (such as age) are variables that are
solved for within an explicit market structure and an explicit welfare framework [Glazer
and McGuire (2000), Encinosa (1998), Shen and Ellis (1998)]. In general, the opti-
mal risk adjusters are not regression coefficients that maximize explainable variance in
individual-level health care costs. The economic performance of risk-adjustment can
be improved if risk adjusters are regarded as taxes and subsidies, and not simply as
statistical results.
Classificationsystems for mental health and substance abuse: In the development of
risk adjustment systems, little attention has been paid to MH/SA, partly because initial
development of the existing risk adjustment systems proceeded first in the Medicare
context, where MH/SA is a very small part of total spending. Continued applied research
on the systems, including use of younger populations, is leading to more attention to
MH/SA. A consistent finding in the research so far, is that however past diagnostic
information is configured, it has little predictive power in behavioral health [Ettner et
al. (1998)].
Classification of MH/SA patients has posed a difficult problem for policy mak-
ers since the initial introduction of prospective payment policies in the early 1980s.
The development of Medicare's Prospective Payment System (PPS) required the fed-
eral government to determine whether psychiatric and substance abuse Diagnosis Re-
lated Groups (DRGs) constituted an adequate patient classification system. Horgan and
Jencks (1987) and Jencks and Goldman (1987) reviewed competing patient classifica-
tion systems for grouping psychiatric inpatients. Their conclusion, expressing the notion
of a low R-square in lay language, was: "In general, research has not provided a robust
explanation of differences in costs between psychiatric facilities. In particular, research
has not developed classification systems that class together inpatient episodes with simi-
lar costs or that have substantial differences in costs between classes" [Jencks and Gold-
man (1987, p. S42)]. The low explanatory power of the DRGs for MH/SA was not the
most serious problem. The unexplained variation in cost was systematically related to
certain classes of facilities (conditional on the prior reimbursement system). Even after
risk adjustment, simulation analyses (summarized in Jencks and Goldman) showed that
more specialized psychiatric facilities drew a more costly case mix than general hos-
pitals without specialized facilitates. Thus the initial effect of putting MH/SA into the
PPS would have conferred windfall gains (on non-specialized facilities) and losses (on
specialized facilities). Responses of facilities to the new system would have modified
these loses and gains, but the fundamental unfairness of the PPS in this case, which,
we emphasize could only be evaluated with a conception of how the equilibrium would
look, could not be avoided [Freiman et al. (1987)].
The inadequacies of inpatient discharge level, risk-adjustment raises concerns about
the potential of per-person level risk conventional adjustment to adjust capitation rates
for mental health care. Other research in health services suggests that the variation in
rates of use of MH/SA care might be especially large and difficult to capture with the
routinely available risk adjusters. Research on demand for mental health services sel-
dom offers models with explanatory power comparable to those found in general health
services. In the RAND Health Insurance Experiment, for example, Keeler et al. (1986,
1988) were able to group general outpatient medical care into "episodes" and explain
the occurrence and extent of these episodes statistically. A similar effort met with much
less success in the case of outpatient mental health care.
Two initial evaluations of risk adjusters for MH/SA have been completed using Med-
icaid and private insurance data sets. Ettner and Notman (1997) evaluated the predictive
power of the ACG classification system, a set of diagnostic clusters and age and sex
groupings using data on approximately 30,000 Medicaid enrollees in New Hampshire
for fiscal years 1993 and 1994. New Hampshire-specific weights for the classification
systems were constructed using fiscal year 1993 data to predict 1994 expenditures. The
authors evaluated the explanatory power of the classification systems for predicting:
(1) total individual health care spending, and (2) individual MH/SA spending. The re-
sults reveal several key points. First, none of the classification systems studied explained
more than 4% of variance in total health spending, with the percent of variance ex-
plained ranging from 2% to 4%. Second, in the MH/SA area the maximum explanatory
power was 13% of the variance. Third in the analysis of MH/SA spending the results
suggest that including age and sex along with a set of variables indicating whether an
individual had 1, 2 or 3+ separate MH/SA disorders indicated in claims during the
previous year provided greater explanatory power than any other method. 29
In the second analysis Ettner et al. (1998) examined risk adjusters within a larger data
set of privately insured employees and their dependents for 1992 and 1993. In that study,
2
29 Dunn et al. (1995) show that with stratified data and use of ADG, adjusted R for total health charges can
be as high as 0.20. This exceeds the explanatory power found in MH/SA. The ADGs are aggregate of ACGs,
which include inpatient diagnoses.
data from a population of approximately 450,000 enrollees were examined to evaluate

several conventional risk adjustment systems. To study the predictive ability of each
system for total MH/SA spending, the authors made use of actual health plan choices
of employees to assess how well each classification system would account for natu-
rally occurring selection into plans (i.e. selection with no risk adjustment in place). The
1992 patterns of illness were used to classify enrollees and to predict 1993 spending.
As in the case of New Hampshire, no classification system displayed strong predictive
ability. The results from the analysis of naturally occurring selection across plans for
two large employer groups contained in the data illuminated the weaknesses of all the
classification systems. 30 When the payments that would have been made under each
classification scheme were compared with payments based on the simple average for
all enrollee (across all plan choices), little meaningful improvement was contributed by
the risk adjusted payment mechanisms. The results suggest that little of the systematic
risk between plans was accounted for by the classification methods examined. Thus, the
condition implied by Equation (6) to minimize selection is unlikely to be met under any
regression-based risk adjustment system.
Risk adjustment policy can be combined with payment system changes. By paying
partly on cost as in a mixed system, incentives to select are reduced [Newhouse (1996)].
In mental health care, mixed payment systems are in use which resemble the one de-
picted in Figure 6.
4.4.2. Behavioralhealth carve-outs
Behavioral health carve-outs have become central to payment and delivery of MH/SA
services under managed care. Carve-outs are usually regarded as cost control devices.
Carving out MH/SA from an indemnity plan or in an indemnity/managed care choice
plan ensures that all MH/SA care will be managed. (See Figure 1.) Carve-outs may also
have a role in diminishing selection-related incentives. The economic role of carve-out
programs can differ significantly depending on the specific form of the carve-out. For
example, the carve-outs shown in Figure 3 are chosen by the health plan and can be
viewed as an organizational structure that helps the health plan implement its desired
rationing scheme. Viewed in this manner carve-outs which are simple sub-contractors
of health plans are not expected to have any impact on selection because consumers
continue to choose among integrated health plans where the implementation of rationing
rules across services can affect enrollment patterns. The incentives to ration MH/SA
care to the organization are present with and without a carve-out contract.
In contrast, the type of carve-out program depicted in Figure 2 separates MH/SA ser-
vices from overall health care and as such removes it as a dimension of competition
30 The apparent selection across 3 plans may have been quite large. For example, in comparing plans with
similar deductibles, differences in annual visit limits of 50, 50 and 25, and similar copayments for outpa-
tient care, per person per year costs ranged from $6 to $105. Since the cost differences were unlikely to be
attributable to differences in limits we interpret cost variation to be largely due to selection.
among health plans for enrollees. This can have potentially large impacts on the incen-
tive to provide services. Carving-out a service, MH/SA in this case, isolates MH/SA
from selection-related incentives. Rationing will be determined by the contract between
the payer and the specialty MBHO. It is also important to note that the rationing for any
one service depends on all the other service demands. Thus, carving-out any one service
will affect the degree of rationing for all others.
Carve-out programs have other pros and cons that must be considered along with
the potential welfare gains related to selection. One controversial question relates to
whether MH/SA care is more effectively delivered in a fashion that is integrated with
medical care via a primary care physician. In theory, "integrated" care is better than
"fragmented" care. In practice, a separate mental health system has some advantages.
Primary care physicians tend to overlook mental illness in their patients [Jencks (1985),
Morlock (1989)]. When mental illnesses are recognized primary care physicians often
fail to provide appropriate treatment [Shapiro et al. (1987), Wells et al. (1996)].
Carve-out programs add administrative costs. Estimates of the additional administra-
tive costs associated with carve-out arrangements range from 8% to as much as 20%
of MH/SA benefit costs. Finally, new boundaries between payers' responsibility creates
opportunities for cost-shifting and strategic behavior. Anecdotal evidence suggests that
carve-out plans are especially prone to adopt pharmacotherapeutic strategies because
the drug benefit represents an "off budget" set of treatments. Brisson et al. (1998) find
a higher propensity for individuals with histories of substance abuse treatment to be
hospitalized in a general medical setting following introduction of a carve-out plan.
5. The public mental health and substance abuse treatment system
A unique feature of mental health care delivery in most western nations is the large role
assigned to public hospitals and clinics for the care of individuals with mental and ad-
dictive disorders. Direct public services for the most seriously ill persons is a common
feature of health systems that may otherwise be organized to provide and pay for other
health care with a range of approaches Hollingsworth (1992)]. The public mental health
systems in the United States, Germany, Canada, Great Britain and France are quite simi-
lar, despite vast differences in their general approaches to health care financing. Each of
these nations assigns responsibility for mental health care to sub-national government
(e.g., states in the US, Idnder in Germany, provinces in Canada and local authorities
in the United Kingdom) and tends to rely on local tax funding more than either central
government or premium-based funding. The role of local government-provided mental
health care predates the development of either public or private insurance arrangements
and tends to emphasize providing care to the poor and disabled. These public mental
health systems have been the source of public dissatisfaction with respect to the qual-
ity of care and horizontal equity [Mangen and Rao (1985)]. Yet they continue to play
central roles in the care of severe mental and addictive illnesses.
There are three main factors that explain the commonality. First, the technology of
treating severe mental disorders calls for different organizational and financial arrange-
ments from other medical conditions. Second, public mental health systems predate
modem insurance arrangements, and these established systems of public provision may
have inhibited other forms of financing for mental health and substance abuse. Third,
the externalities created by mental illnesses means that provision of mental health care
has a public safety component and therefore the state has a greater interest in assuming
more direct control over the delivery of certain forms of mental health care to fulfill its
obligation of public protection.
5.1. The technology of treatmentfor the severely mentally ill
The acute health care delivery system is ill-equipped to deal with the array of problems
associated with severe mental disorders. Mechanic (1987) describes the complexity of
services required for treatment of individuals with severe mental disorders this way:
"Effective community care for the most seriously disabled patients requires performance of many of
the same functions as the mental hospital, ranging from assuring appropriate shelter to managing seri-
ous medical and psychiatric problems. To do so in the community context requires some influence over
areas of responsibility involving different sectors (housing, medical care, social services, welfare) .. "
(p. 492).
This characterization suggests important differences in the technology required for

treating the sickest and mostly costly people with mental illnesses from those with most
medical conditions. The observation that medical care delivery systems have difficulty
accommodating individuals with chronic diseases has been widely acknowledged in
other contexts [Moon (1996), Wallack and Levine (1996)]. It appears that this point
applies with special force in the case of severe mental disorders.
Treatment programs aimed at people with severe mental illnesses must concern them-
selves with the acute care of the symptoms of illness, as does the rest of medicine. In
addition, however, mental health care must address the housing, income support, reha-
bilitation, social contacts, and social control of the affected individuals. The nature of
severe mental illnesses and their treatment requires that mental health providers coordi-
nate these services. Many of the most successful innovations in the treatment of severe
mental disorders involve new ways of organizing and coordinating the various elements
of care for individuals with severe mental disorders [Stein and Test (1980), Dill and
Rochefort (1993), Taube, Morlock, Burns and Santos (1990)]. The health systems in
the US, Germany, Britain and Canada all confine responsibility of the medical care sys-
tem for mental health services and treatments to those that roughly correspond to acute
care medical treatments. Such actions are consistent with recognition of a different tech-
nology of treatment that may call for a different set of institutions.
In European contexts, the organizational response to differences in technology for
treating people with severe mental disorders in great measure reflects the political
choices of planners or the results of political bargaining between different levels of
government. In the US context, the organizational response to the needs of persons with
serious mental illness represents a different set of forces involving market failure, per-
haps due to adverse selection and externalities (discussed below), and the economics of
a federalist system.
5.2. Fiscalfederalismandpublic mental health care
In assessing the lessons from the history of mental health policy, Grob (1994) points to
the evolving nature of fiscal federalism as a central force in shaping the role of govern-
ment in mental health care delivery. He goes so far as to assert that for some payers,
including states, cost shifting has been as important in designing policies as the im-
petus to construct a "rational" system. "Deinstitutionalizing" the mentally ill refers to
shrinking or closing state hospitals, and transferring the care of previously hospitalized
patients to a diverse set of care providers. While a rationale for deinstitutionalization can
be made on the basis of single system cost-effectiveness, cost-shifting was also a mo-
tive. State government in the US paid the costs in state hospitals, and only a share of the
costs of care given by alternative providers, courtesy of the federal Medicaid program.
Since states (and other regional governments outside the US) make the majority of
spending decisions about public funds for care of mental disorders, the literature has nat-
urally directed attention towards state choices regarding mental health policy. A simple
starting point for analyses of state mental health policy is to view state policy deci-
sions as being the result of choices by a social planner seeking to coordinate mental
health services for the poor. State policy makers have two major instruments: direct
state spending on services, and the insurance-like Medicaid program. Michael (1980)
and Frank (1985a) regard the "state" as a single decision-maker with an objective func-
tion containing: welfare for the poor and state budget costs. The choices of the state
planner are constrained by factors such as the income in the state, the size of the public
mental hospital system, the availability of alternatives to state funded providers (e.g.,
nursing homes), federal rules governing Medicaid especially the federal matching rate
on spending,31 and the amount of private insurance coverage in a state. Direct state
spending on mental health care will be reduced, according to this approach, by generous
federal matching provisions,3 2 the availability of care in settings funded by Medicaid
(nursing homes), and expansion of private insurance coverage for mental health care.
These all point to rational cost shifting responses to exogenous changes in regulation
and market structure.
Although few formal analyses of mental health financing for countries other than
the US have appeared in the literature, similar observations have been made about the
relation of central government financing to local funds. For example, Britain has exper-
31 The Medicaid program in the US matches state spending according to a formula based on the inverse of
per capita income. The federal government constrains minimum participation to 50% and maximum matching
rates are about 78%. This formula favors high-income states.
32 It is possible that the state's elasticity of demand would be high enough to reverse this result, but this is
very unlikely.
imented with central government matching grants to local authorities [Knapp (1997),
Yellowlees (1990)]. In general, these schemes have not resulted in the desired effort by
localities. In Germany, the sickness funds have resisted expanding benefits to include
long-term care for mental disorders due to concerns that costs would be shifted from
the Iinderto the sickness funds [Cooper and Bauer (1987)].
Existing empirical evidence from the US is broadly consistent with the single
decision-maker model predictions. For example, Frank (1985a) using a cross section
of states for the year 1976 found that the state share of the Medicaid program was pos-
itively related to per capita state spending for direct mental health services. The size of
state mental hospital systems (relative to population size) was estimated to have a large
positive impact on spending, while the presence of a mandated mental health care insur-
ance statute (creating another destination for cost-shifting) reduced state mental health
spending. Michael (1980) found that the availability of nursing home services led to re-
duced use of state mental hospitals. He also shows that the number of mental hospitals
per state senatorial district was estimated to have a positive and significant impact on
the use of public mental hospitals. Gronfein (1985) found that the introduction of Medi-
caid, and its opportunity for shifting mental health costs from state budgets was the most
important factor determining the rate of deinstitutionalization in states. The Medicaid
impact dominated the effects of innovation in drug treatment and the creation of com-
munity mental health centers. During the period 1955 to 1965, the populations of public
mental hospitals were reduced by about 1.5% a year. Following the implementation of
Medicaid the rate of population reduction increased to 6% a year.
The prominence of cost shifting in state mental health policy explains tensions be-
tween federal (or central) governments and local governments in the United States and
other western nations in mental health policy. For example, when the Medicaid and
Medicare programs were implemented in the United States, Congress included provi-
sions which prohibited Medicaid reimbursement for care provided in an "Institution for
Mental Disease" (IMD) and limited Medicare hospital payments to 190 days over an in-
dividual's lifetime. These rules were aimed at preventing states from shifting the costs
of state mental hospitals onto the federal government's budget (which turned out to be
partially and temporarily successful). Thus, once the federal government appreciates the
responses it will get in a cost shifting game with the states, the result will be a division of
labor that may leave states with their historical responsibility for public mental hospitals
that care for the most impaired and difficult indigent people with mental disorders.
Cost shifting has also been reported in Canada. Following adoption of the National
Health Insurance plan in 1968, Nova Scotia and other provinces attempted to reorganize
mental health services in order to capture health insurance payments for services which
were previously the responsibility of the provincial government [Rochefort (1993)]. The
National Health Insurance plan guarded against cost shifting by (1) strictly limiting the
range of providers who could be reimbursed for supplying mental health care (MDs
only), and (2) limiting the scope of services to acute care treatment, thereby excluding
day treatment and rehabilitation services.
While single decision-maker models are adequate to explain cost shifting behavior
between federal and state (regional) government, this approach may not offer a fully
satisfactory explanation for observed patterns of mental health policy. Within a state,
agencies (Medicaid, mental health, substance abuse) may behave as independent (and
possibly competing) organizations. Casual observations have noted policies which shift
responsibility for care of individuals with mental disorders from state mental health
agencies to social welfare agencies, school systems and criminal justice programs. Such
observations suggest the limits of characterizing state government behavior in terms of
a single planner that coordinates policy [Rochefort (1993), Mangen and Rao (1985),
Cooper and Bauer (1987)]. One way to investigate the empirical importance of a states'
organizational choices is to see if different states' approaches to dividing up adminis-
trative responsibility for Medicaid, mental health, substance abuse, and related services
affect the level of state spending on those services. Using a time-series of the fifty states
in the United States, Jacobsen, Notman and McGuire (1996) found that organizational
changes, such as putting responsibility for substance-abuse services in a mental health
department had, predictable, but small, effects on levels of spending. If intrastate di-
vision of labor among agencies matters for state fiscal outcomes, the single decision-
maker model is contradicted.
A natural way to regard intergovernmental relations is as a sequential game consisting
of the federal government, state government, local government and participants in pri-
vate markets [Frank and McGuire (1996)]. The federal government makes policy given
the historical role of the states which in turn sets the stage for state policy action. The
state and federal policies determine the outlines of the market where private parties buy
and sell insurance against the consequences of mental and addictive illnesses.
Within this framework, the federal government sets rules for public insurance pro-
grams (Medicaid) that seek to limit the shifting of the costs of state mental hospitals
from state to federal budgets. Marmor and Gill (1989) propose a political model along
these lines that applies to the US and Britain. The IMD rule associated with the US
Medicaid program, noted above, is one example of a policy adopted to reduce cost
shifting by states. States in turn responded to matching provisions under Medicaid by
orienting program design towards Medicaid. For example, state governments shifted
large numbers of elderly residents in state mental hospitals to nursing homes following
the introduction of Medicaid in the late 1960s. This meant that the costs of treating one
segment of the elderly population were moved entirely off budget for the state mental
health agency. State governments retained some financial responsibility via matching
provisions, of roughly 30% to 50%, assigned to the state under the Medicaid program.
States also set general rules within which individual agencies and local governments
operate. Within state governments, funding for mental health care and substance abuse
treatment is administered and often supplied separately. Nevertheless, there is consider-
able co-occurrence of these classes of disorders. Current clinical thought suggests that
organizational and payment arrangements stand in the way of effective treatment for
this class of expensive and disabling conditions. Policy regarding utilization of state
mental hospitals offers an example of the response by local mental health systems to of
state payment rules. For many years state mental health agencies provided public men-
tal hospital services to local public mental health programs "free of charge" [Frank and
Gaynor (1995)]. One result was a tendency of local programs to "overuse" state mental
hospitals. This set of institutional arrangements has been pointed to as leading to distor-
tions in spending towards state mental hospitals and to inefficiently low levels of effort
aimed at treating people with severe mental illness in community programs [McGuire
and Riordan (1995)].
Private insurance markets offer insurance designs in the context of a public mental
health system that will provide hospital and outpatient care for mentally ill individuals
without insurance coverage. Approximately 55% of individuals admitted to public psy-
chiatric hospitals in 1994 had no insurance coverage. In addition, a significant number
of people in state hospitals with coverage had exhausted their mental health benefit.
The presence of a public mental health system along with market forces associated with
selection incentives discussed above serve to undermine the provision of private insur-
ance for mental health and substance abuse care. The availability of publicly funded
and provided mental health care allows employers to strictly limit insurance coverage
for mental health care while at the same time giving their employees recourse should
a catastrophic mental illness strike. As noted earlier, mandated mental health insurance
statutes represent a policy response by states to market failures stemming from adverse
selection. Mandates also represent a means of shifting costs from state government bud-
gets to private employers [Frisman, McGuire and Rosenbach (1985)]. Mental health
mandates have been politically contentious in state legislature and have been limited in
their effect by federal policy. The federal Employment Retirement Income Security Act
(ERISA) allows self-insured firms to be exempt from state laws which govern the busi-
ness of insurance. During the 1980s and 1990s an increasing number of mid and large
size firms have chosen to become self-insured. The consequences are that in most states
only a modest portion of the population is subject to mandated benefit statutes. Thus,
federal policy has limited the ability of states to (1) address market failure in insurance
and (2) shift costs onto private employers.
Mental health policy can be viewed as a cascading cost-shifting game. Each of the
players across levels and within levels of the game makes choices subject to rules set at
a higher level. Players are presumed to be aware of the behavior of other players at the
same level and below and develop policies accordingly. These ideas appear also to apply
in nations with federal types of systems in mental health, including Britain, Germany,
and Canada [Yellowlees (1990), Marmor and Gill (1989)].
5.3. Externalitiesand public mental health care
Prior to the 1820s, mental problems were not so clearly part of the medical domain.
Care for "lunatics" or "distracted" persons was provided through a variety of informal
mechanisms [Grob (1994)]. Poor houses and alms houses were settings for the support
of people with disabling mental problems. During the first part of the 19th century a
new institution, the asylum, became the focal point for treatment of more clearly recog-
nized mental illnesses. Initially these institutions were the shared responsibility of state
942 R.G. Frank and TG. McGuire
and local governments. State government in the 1820s and 1830s typically provided
capital financing for asylums and localities paid for operating expenses. Initially, the
priority populations for treatment in asylums were individuals viewed as "dangerously
insane" [Grob (1973)]. The requirement that local government pay for costs of treating
mentally ill people in asylums created an incentive for localities to continue housing the
mentally ill in alms houses which had very low per diem costs. Aware of the problems
created by divided responsibility for asylums, states began to take over all financial
and operational responsibility for asylums. As financial responsibility shifted to state
governments, localities developed new enthusiasm for the use of asylums to treat the
mentally ill. Not only did localities transfer responsibility to state asylums for individ-
uals that were "chronically" mentally ill, they also redefined senility as a psychiatric
condition and shifted responsibility and the costs of caring for the senile elderly to the
state. The state has had the paramount role in mental health policy ever since.
Thus from early in the 19th century to the present the state mental hospital system
(and later state mental health systems) has played two roles. First, it has served as an
institution for involuntary treatment and confinement of individuals who were viewed
as dangerous due to a mental disorders. Second, it has served as a safety net institution
for housing and care of other disabled and vulnerable populations (initially the senile
elderly, later chronic brain damaged alcoholics). Two roles of the state have been used
to justify the presence and persistence of a public mental health system that supplies
involuntary treatment [Rubin (1978)]. These are the parenspatriaedoctrine and the po-
lice power of the state. The parens patriaedoctrine claims that when an individual is
mentally incapable of taking care of him or herself, the state may serve as an agent of the
individual and institutionalize the individual in order to care adequately for them. The
police power of the state justifies institutionalization of individuals who are a danger to
themselves or others in order to protect the individual or society at large. In the latter
case the state can choose between separating an individual from society and offering
treatment via the mental health system or through incarceration in the criminal justice
system.
The presence of external effects provides an efficiency underpinning to both justifi-
cations for assigning the state a role in involuntary commitment for mental health care.
The parents patriae doctrine substitutes judgments of the state for individual choice
based on the notion that there is a collective interest in seeing that individuals who are
so impaired by mental disorders receive sufficient care so as to survive and possibly
improve their ability to function or recover. This collective interest is a consumption
externality, the polity benefiting from consumption of mental health care by those too
impaired to make such choices on their own. The police power rationale also involves
an externality. Protecting affected individuals intent on harming themselves or protect-
ing the general public from individuals whose mental disorders make them dangerous
to others through treatment and confinement also confers external benefits. Publicly
funded mental hospitals that care for these individuals generate some non-excludable
benefits and therefore have a public good feature.
All US states have laws and regulations that govern the process by which individu-
als can be involuntarily committed to mental institutions. Since the early 1970s there
has also been judicially imposed regulation concerning the conditions of the institutions
to which these individuals are committed. Over the past 30 years there has been con-
siderable flux in public policy regarding involuntarily commitment to mental hospitals
[Rubin (1978)]. Since the early 1970s state psychiatric hospitals have been required to
provide active treatment and to have physical facilities and staffs that are consistent with
what the courts have viewed as the ability to provide active treatment.
Empirical analyses have shown that commitment laws and regulations, as well as
court-dictated facility and staffing guidelines, have had a significant impact on the size
of the populations served in public mental hospitals as well as the budgets of those insti-
tutions. Lambrinos and Rubin (1981) used a simultaneous equations model to estimate
the impact of commitment laws and regulations on average daily census and spending
on public mental hospitals with data from 1974 and 1975. Their results showed that
states that institute mechanisms to protect patient rights such as appointment of an at-
torney and requiring a formal hearing tend to reduce use of mental institutions. States
that clearly define "risk" and disability in their commitment statutes also tend to have
lower rates of institutionalization. Rubin (1980) and Lambrinos and Rubin (1981) of-
fer evidence suggesting that the regulations setting out standards for staffing and the
physical plants of state hospitals have increased spending on public mental health care.
The state responsibility for public protection and the care of those who cannot fend
for themselves is a long-established basis for the public mental hospital, a role predating
the development of either public or private insurance mechanisms. Most states continue
to have laws in place that tend to favor society's interests (externalities) over individ-
ual liberties of the mentally ill. This in part explains the persistence of the state role in
provision of mental health care. Litigation and court decisions during the 1970s have
forced states to provide treatment in public mental institutions which, in turn, increased
spending on those institutions in the 1970s and 1980s. 3 3 At the same time the division
of labor has shifted whereby private organizations have increasingly served individu-
als with insurance (public and private) who are involuntarily committed. State mental
hospitals have become facilities that serve the most impaired, dangerous, and indigent
people with mental illness. Privatization of production in this area of service has been
quite limited.
6. Conclusions
Economic analyses of the cost of illness and other assessments of the global burden of
disease testify to the disabling effect of mental disorders. Mental illnesses often persist,
33 For an overview of the troubling history of the state mental hospital in the United States see Grob (1973,
1994).
and can be accompanied by a range of social pathologies, generating public misun-

derstanding and fear. Public safety is therefore viewed as a key product of the mental
health care system, one factor in explaining the major role assigned to government in
the financing and delivery of mental health services relative to what is found in general
medical care. There is an unresolved tension between the insurance-medical care de-
livery aspects of mental health care and the public safety role. For most of this century
the public safety role featuring direct government provision appears to have dominated
the organization and financing of this system. Nevertheless, over the past thirty years,
private institutions and markets have ascended to important positions in the provision
of mental health services. Markets for health insurance and health services have devel-
oped against a background of large publicly funded and managed mental health systems,
though downsizing has been one key force affecting the development of private insur-
ance markets in the US and the design and implementation of public insurance in the
US, Britain and Canada among other nations.
Concerns about insurance arrangements in mental health relative to general medical
care have dominated policy debates in the mental health arena during the 1980s and
1990s. Here the traditional factors which threaten the efficient functioning of insurance
markets, adverse selection and moral hazard, are important for understanding the per-
formance of the market with regard to mental health care. Mental health advocates are
troubled by the "special treatment" of the mentally ill in the structure of private in-
surance. Advocates have tended to understand the special treatment as discrimination
rather than as a response to economic fundamentals. Payers have been deeply trou-
bled by the cost control problems associated with coverage of mental health care under
indemnity-fee for service arrangements.
In economic terms, the behavior of providers and consumers in the market for mental
health services reflects the characteristics of mental health disorders and their treat-
ments. Adverse selection and moral hazard appear to hit mental health markets with
special force. In some of the first economic research on mental health, the demand re-
sponse was found to be high. The persistence of many mental disorders and the fact
that people most likely to suffer from mental disorders are most likely to choose health
plans with generous mental coverage suggests strong incentives for plans to limit mental
health coverage. Sorting out the roles of moral hazard and selection in explaining ob-
served market outcomes, and in terms of the implications for optimal policy, has been
an ongoing theme in research in the economics of mental health.
The basis for the institutions governing coverage for mental health care have been
dramatically altered by managed care, which changes the terms of Zeckhauser's
dilemma, wherein coverage should balance losses from risk and moral hazard. Mental
health services are now rationed by means other than demand-side prices. Even though
little is known about internal rationing methods used within managed care organiza-
tions, the ability to control costs in this way means that the main efficiency argument
against parity in mental health coverage is potentially weakened under managed care.
At the same time, parity in demand-side cost sharing delivers less because expanded
nominal coverage is no longer a guarantee of access. Demand-side cost sharing may no

longer be the binding constraint on use.
Managed care has also continued to treat mental health differently than other ill-
nesses, the most important example being the emergence of behavioral health carve-out
programs. Are these institutions a fiscal discrimination against the mentally ill, or a re-
sponse to market failure? Economic research understands these institutions in terms of
key functions that they serve. The first relates to economies of specialization and con-
trol of moral hazard, while a second involves attenuation of selection related incentives
common in competitive insurance markets. Payer carve-outs may increase market ef-
ficiency by removing mental health from competition for "good risks" among health
plans.
As the focus of government policy towards mental health has shifted towards treat-
ment and away from public safety, maintenance of large publicly supported institutions
has been harder to justify. The public mental hospital is increasingly being consigned
to care for involuntarily committed individuals and the severely impaired, narrowing
its public safety function. The emergence of private providers of managed behavioral
health care that will assume some risk, has proved to be an attractive alternative for a
number of states. The result is a shifting of functions for many state governments from
payer (via grants) and provider to payer (via contracts) and regulator. A key dilemma
facing the states is that while insurance markets continue to be subject to failures stem-
ming from selection related incentives, state Medicaid programs and the related federal
disability programs are becoming increasingly restricted. States remain nevertheless re-
sponsible for the indigent mentally ill, yet they are reducing the infrastructure available
to attend to the mental health needs of these vulnerable populations. How states will
choose to address their traditional role as provider of last resort in the context of the
new private market for mental health care is unclear. What is clear is that in the past
fifty years resource allocation in mental health in dealing with state government has
moved steadily out of the realm of public administration and into the domain of eco-
nomics and mental health.
References
Alegria, M., R. Bijl, E. Lin, E. Walter and R. Kessler (in press), "Income differences in persons seeking out-
patient treatment of mental disorders: a comparison of the US with Ontario and the Netherlands", Archives
of General Psychiatry.
American Psychiatric Association (1995), Diagnostic and Statistical Manual IV (APA Press, Washington,
DC).
Appropriation Authorization for the Department of Veterans Affairs (1996).
Arrow, K.J. (1963), "Uncertainty and the welfare economics of medical care", American Economic Review
53:941-969.
Assistant Secretary for Planning and Evaluation (ASPE) (1994), Substance Abuse Among Women and Parents
(USDHHS, Washington, DC).
Barr, N. (1992), "Economic theory and the welfare state: a survey and interpretation", Journal of Economic
Literature 30(2):741-763.
946 R.G. Frank and TG. McGuire
Bartel, A., and P. Taubman (1986), "Some economic and demographic consequences of mental illness", Jour-
nal of Labor Economics 4(2):243-256.
Baumgardner, J. (1991), "The interaction between forms of insurance contract and types of technical change
in medical care", RAND Journal of Health Economics 22(1):36-53.
Beinecke, R.H., D.S. Shepard, M. Goodman and M. Rivera (1997), "Assessment of the Massachusetts man-
aged behavioral health program: year three", Administration and Policy in Mental Health 24(3):191-204.
Berndt, E.R., I. Cockburn and Z. Griliches (1996), "Pharmaceutical innovations and market dynamics: track-
ing effects on price indexes for antidepressant drugs", Brookings Papers on Economic Activity: Microeco-
nomics 1996 (Brookings Institute, Inc.) 133-188.
Berndt, E.R., R.G. Frank and T.G. McGuire (1997), "Altemate insurance arrangements and the treatment of
depression: what are the facts?" American Journal of Managed Care 3(2):243-252.
Blomqvist, A. (1991), "The doctor as double agent: information asymmetry, health insurance and medical
care", Journal of Health Economics 10:411-432.
Brisson, A.S., R.G. Frank et al. (1998), "A case study in the impact of managed care on utilization and costs
of behavioral health care in an HMO", NBER Working Paper #6242.
Bruce, M.L., D. Takeuchi and P.J. Leaf (1991), "Poverty and psychiatric status", Archives of General Psychi-
atry 48:470-474.
Buck, J.A., and B. Umland (1997), "Trends: covering mental health and substance abuse services", Health
Affairs 16(4):120-126.
Bureau of Labor Statistics (1996), Tabulations from Employee Benefits in Medium and Large Private Estab-
lishments (USGPO, Washington).
Callahan, J.J., D.S. Shepard, R.H. Beinecke, M.H. Larson and D. Cavanaugh (1995), "Mental
health/substance abuse treatment in managed care: the Massachusetts Medicaid experience", Health Af-
fairs 14(3):173-184.
Center for Mental Health Servicies (1998), Mental Health US (USGPO, Washington, DC).
Chang, C.F., L.J. Kiser, J.E. Bailey, M. Martins, W.C. Gibson, K.A. Schaberg, D.M. Mirvis and W.B. Apple-
gate (1998), "Tennessee's failed managed care program for mental health and substance abuse services",
JAMA 279(11):864-869.
Christianson, J.B., et al. (1995), "Utah's prepaid mental health plan: the first year", Health Affairs 14(3):160-
172.
Christianson, J.B., D. Wholey and M.S. Peterson (1997), "Strategies for managing service delivery in HMOs:
an application to mental health care", Medical Care Research and Review 54(2):200-222.
Cooper, B., and M. Bauer (1987), "Developments in mental health care and services in the Federal Republic
of Germany", International Journal of Mental Health 16:78-93.
Coulam R., and J. Smith (1990), "Evaluation of the CPA-Norfolk Demonstration: Final Report", US Depart-
ment of Defense, MDA 907-87-C-0003, Abt Associates.
Comes, R. and T. Sandler (1986), The Theory of Externalities, Public Goods and Club Goods (Cambridge
University Press, Cambridge, England).
Cromwell, J., R.P. Ellis, B. Harrow and T.G. McGuire (1992), "A modified TEFRA system for Medicare
discharges from psychiatric facilities", in: R.G. Frank and Manning, eds., Economics and Mental Health
(Johns Hopkins University Press).
Cuffel, B., W. Goldman and H. Schlesinger (1998), "The effects of increased mental health benefits and a
MBHC carve-out on general medical costs", Working Paper.
Cutler, D., and R. Zeckhlauser (2000), "The anatomy of health insurance", in: J.P. Newhouse and A.J. Culyer,
eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 11.
Deb, P., J. Rubin, V. Wilcox-Gok and A. Holmes (1996), "Choice of health insurance by families of the
mentally ill", Health Economics 5(1):61-76.
Dickey, B., and H. Azeni (1992), "Impact of managed care on mental health services", Health Affairs
11(3):197-226.
Diamond, P. (1992), "Organizing the health insurance market", Econometrica 60(2):1233-1254.
Diehr, P., et al. (1984), "Ambulatory mental health services utilization in three provider plans", Medical Care
22(1):1-13.
Dill, A.P., and D. Rochefort (1993), "The challenge of coordination", in: D. Rochefort, ed., From Poor Houses
to Homelessness (Auburn House, Westport, CT) 133-148.
Duan, N., W.G. Manning, C.N. Morris and J.P. Newhouse (1983), "A comparison of alternative models of the
demand for medical care", Journal of Business and Economic Statistics 1(2):115-126.
Dunn, D., et al. (1995), "A comparative analysis of methods of health risk assessment: Final Report", Society
of Actuaries.
Elkin, I., T. Shea, J.T. Watkins, S.D. Imber, S.M. Sotsky, J.F Collins, D.R. Glass, P.A. Pilkonis, W.R. Leber,
J.P. Docherty, S.J. Fiester and M.B. Parloff (1989), "National Institute of Mental Health Treatment of
Depression Collaborative Research program: general effectiveness of treatments", Archives of General
Psychiatry 46:971-982.
Ellis, R.P. (1986), "Rational behavior in the presence of coverage ceilings and deductibles", Rand Journal of
Economics 17:158-175.
Ellis, R.P. (1988), "The effect of prior-year health expenditures on health coverage plan choice", in:
R.M. Scheffler and L.F Rossiter, eds., Advances in Health Economics and Health Services Research (JAI
Press, Greenwich, CT) 6:149-170.
Ellis, R.P. (1998), "Creaming, skimping and dumping: provider competition on the intensive and extensive
margins", Journal of Health Economics 17(5):537-556.
Ellis, R.P., and T.G. McGuire (1986a), "Cost sharing and patterns of mental health care utilization", Journal
of Human Resources 21(3):359-380.
Ellis, R.P., and T.G. McGuire (1986b), "Provider behavior under prospective reimbursement: cost sharing and
supply", Journal of Health Economics 5(2):129-151.
Ellis, R.P., and T.G. McGuire (1988), "Insurance principles and design of prospective payment systems",
Journal of Health Economics 7:215-237.
Ellis R.P., and T.G. McGuire (1990), "Optimal payment systems for health services", Journal of Health Eco-
nomics 9:375-396.
Ellis, R.P., and T.G. McGuire (1993), "Supply-side and demand-side cost sharing in health care", Journal of
Economic Perspectives 7(4):135-151.
Ellis, R.P., and T.G. McGuire (1996), "Hospital response to prospective payment: moral hazard, selection,
and practice-style effects", Journal of Health Economics 15:257-277.
Ellis, R.P., G.C. Pope, L.I. Iezzoni, J.Z. Ayanian et al. (1996), "Diagnosis-based risk adjustment for Medicare
capitation payments", Health Care Financing Review 17(3):101-128.
Encinosa, W. (1998), "Risk adjusting health insurance", unpublished manuscript (Agency for Health Care
Policy and Research).
Ettner, S.L., and E.H. Notman (1997), "How well do ambulatory care groups predict expenditures of mental
health and substance abuse patients?", Journal of Administration and Policy in Mental Health 24(4):339-
358.
Ettner, S.L., R.G. Frank and R.C. Kessler (1997), "The impact of psychiatric disorder on labor market out-
comes", Industrial and Labor Relations Review 51(1):64-81.
Ettner, S.L., R.G. Frank, T.G. McGuire, J.P. Newhouse and E.H. Notman (1998), "Risk adjustment of mental
health and substance abuse payments", Inquiry 35(2):223-239.
Fairbank, A. (1989), "Expanding insurance coverage to alternate types of psychotherapists: demand and sub-
stitution effects", Inquiry 26(2):170-181.
Finch, M., et al. (1992), "The treatment of alcohol and drug abuse among mentally ill Medicaid enrollees:
the utilization of services in prepaid plans versus fee-for-service care", in: R. Frank and W. Manning, eds.,
Economics and Mental Health (Johns Hopkins Press, Baltimore).
Frank, R.G., and M.S. Gaynor (1995), "Incentives, optimality and publicly provided goods: the case of mental
health services", Public Finance Quarterly 23(2):167-192.
Frank, R.G., and P. Gertler (1991), "An assessment of measurement error bias for estimating the impact of
mental distress on earnings", Journal of Human Resources 26(1):154-164.
Frank, R.G., and C.A. Jackson (1989), "The impact of prospectively set hospital budgets on psychiatric ad-
missions", Social Science and Medicine 28(8):861-867.
Frank, R.G., and J.R. Lave (1989), "A comparison of hospital responses to reimbursement policies for Medi-
caid psychiatric patients", Rand Journal of Economics 20(4):588-600.
Frank, R.G., and W.G. Manning (1992), Economics and Mental Health (Johns Hopkins University Press,
Baltimore).
Frank, R.G., and T.G. McGuire (1986), "A review of studies of the impact of insurance on the demand and
utilization of specialty mental health services", Health Services Research 21(2):241-266.
Frank, R.G., and T.G. McGuire (1996), "Federal health financing reform and state mental health policy", in:
R. Rich and W. White, eds., Health Policy, Federalism and the American States (Urban Institute Press).
Frank, R.G., and T.G. McGuire (1997), "Savings from a carve-out program for mental health and substance
abuse in Massachusetts Medicaid", Psychiatric Services 48(9):1147-1152.
Frank, R.G., and T.G. McGuire (1998), "The economic functions of carve-out in managed care", The Ameri-
can Journal of Managed Care 4(SP):SP31-SP39.
Frank, R.G., J.R. Lave, C.A. Taube et al. (1987), "The impact of Medicare's prospective payment system on
psychiatric patients treated in scatterbeds", in: R. Scheffier and T.G. McGuire, eds., Advances in Health
Economics Health Services Research (JAI Press, Greenwich) 1-21.
Frank, R.G., J. Glazer and T.G. McGuire (1998), "Measuring adverse selection in managed health care",
Working Paper.
Frank, R.G., H.H. Goldman and T.G. McGuire (1992), "A model mental health benefit", Health Affairs
11(3):98-117.
Frank, R.G., H.A. Huskamp, T.G. McGuire and J.P. Newhouse (1996), "Some economics of a mental health
carve out", Archives of General Psychiatry 53:933-957.
Frank, R.G., C. Koyanagi and T.G. McGuire (1997), "Political economy of 'parity' for mental health in
insurance", Health Affairs 16(4):108-119.
Frank, R.G., T.G. McGuire and J.P. Newhouse (1995), "Risk contracts in managed mental health care", Health
Affairs 14(3):50-64.
Frank, R.G., T.G. McGuire, J. Bae and A. Rupp (1997), "Solutions for adverse selection in behavioral health
care", Health Care Financing Review 18(3):109-122.
Frank, R.G., T.G. McGuire, E.H. Notman and R.M. Woodward (1996), "Developments in Medicaid behav-
ioral health care", in: R. Manderscheid and M.A. Sonnerscheid, eds., Mental Health US 1996 (USGPO,
Washington).
Frank, R.G., T.G. McGuire, D.A. Regier, T. Manderscheid and A. Woodward (1994), "Paying for mental
health and substance abuse care", Health Affairs 13(1):337-342.
Frank, R.G., D.S. Salkever and S.S. Sharfstein (1991), "A look at rising mental health insurance costs", Health
Affairs 10(2):116-124.
Frank, R.G. (1985a), "A model of state expenditures on mental hospital services", Public Finance Quarterly
13(3):319-338.
Frank, R.G. (1985b), "Pricing and location of physician services in mental health", Economic Inquiry 23:115-
133.
Frank, R.G. (1989), "Regulatory policy and information deficiencies in the market for mental health services",
Journal of Health Politics, Policy and Law 14(3):477-503.
Freiman, M.P., R.P. Ellis and T.G. McGuire (1989), "Provider response to Medicare's PPS: reductions in
length of stay for psychiatric patients treated in scatter beds", Inquiry 26(2):192-201.
Freiman, M.P., J.B. Mitchell and M.L. Rosenbach (1987), "Modifications of the prospective payment sys-
tem and payments for Medicare psychiatric admissions", in: T.G. McGuire and R.M. Scheffler, eds., The
Economics of Mental Health Services: Advances in Health Economics and Health Services Research (JAI
Press, Greenwich, CT and London, UK) 8:23-47.
Frisman, L.K., T.G. McGuire and M.L. Rosenbach (1985), "Costs of mandates for outpatient mental health
care in private health insurance", Archives of General Psychiatry 42:558-561.
Glazer, J., and T.G. McGuire (2000), "Optimal risk adjustment in markets with adverse selection: an applica-
tion to managed health care", American Economic Reviews.
Goldman, H.H., and A. Skinner (1989), "Specialty mental health services: research on specialization and
differentiation", in: C. Taube, D. Mechanic and A. Hohmann, eds., The Future of Mental Health Services
Research (USGPO, Washington).
Goldman, W., J. McCulloch and R. Sturm (1998), "Costs and utilization of mental health services before and
after managed care", Health Affairs 17:40-52.
Goulam, G., and J. Smith (1990), "Evaluation of the CPA-Norfolk demonstration", Final Report, Department
of Defense Contract MDA 907-87-C-0003 (Abt Associates, Cambridge, MA).
Grob, G.N. (1973), Mental Institutions in America: Social Policy to 1875 (Free Press, New York).
Grob, G.N. (1994), "Government and mental health policy: a structural analysis", The Milbank Memorial
Fund Quarterly 72(3):471-499.
Grob, G.N. (1991), From Asylum to Community: Mental Health Policy in Modem America (Princeton Uni-
versity Press, Princeton, NJ).
Gronfein, W. (1985), "Incentives and intentions in mental health policy: a comparison of the Medicaid and
Community Mental Health Programs", Journal of Health and Social Behavior 26:192-206.
Haas, P.J. (1989), "A comparative analysis of state mental health policy", Journal of Management Science &
Policy Analysis 6(4):8-27.
Harrow, B., and R.P Ellis (1992), "Mental health provider response to the reimbursement system", in:
R. Frank and W. Manning, eds., Economics and Mental Health (Johns Hopkins Press, Baltimore) 19-39.
Hodgkin, D., C.M. Horgan and D.W. Garnick (1997), "Make or buy: HMOs' contracting arrangements for
mental health care", Administration and Policy in Mental Health 24(4):359-376.
Hodgkin, D. (1992), "The impact of private utilization management of psychiatric care: a review of the liter-
ature", Journal of Mental Health Administration 19(2):143-157.
Hollingsworth, E.J. (1992), "Falling through the cracks: care of the chronically mentally ill in the United
States, Germany, and the United Kingdom", Journal of Health Politics, Policy and Law 17(4):899-928.
Horgan, C.M., and S. Jencks (1987), "Research on psychiatric classification and payment systems", Medical
Care 25(9):522-536.
Horgan, C.M. (1986), "The demand for ambulatory mental health services from specialty providers", Health
Services Research 21(2):291-320.
Hoult, J., et al. (1983), "A controlled study of psychiatric hospital versus community treatment: the results of
a randomized trial", Australia and New Zealand Journal of Psychiatry 37:26-31.
Huskamp, H.A. (1997), "The impact of a managed behavioral health care carve-out and benefit expansion
on outpatient spending for mental health and substance abuse services", Doctoral dissertation (Harvard
University).
Hustead, E., and S.S. Sharfstein (1978), "Utilization and costs of mental illness coverage in FEHBP, 1973",
American Journal of Psychiatry 135(3):315-319.
Institute of Medicine (1993), Employment and Health Benefits (NAS Press, Washington).
Institute of Medicine (1997), Managing Managed Care (NAS Press, Washington).
Jacobsen, K., T.G. McGuire and E.H. Notman (1996), "Organizational structure and state mental health ex-
penditures", Administration and Policy in Mental Health 23(6):475-492.
Jencks, C. (1994), The Homeless (Harvard University Press, Cambridge).
Jencks, S.F. (1985), "Recognition of mental distress and diagnosis of mental disorder in primary care", JAMA
253(13):1903-1907.
Jencks, S.F., and H.H. Goldman (1987), "Implications of research on prospective payment", Medical Care
25(9):S42-S51.
Kaplan, R.M. (1995), "Utility assessment for estimating quality adjusted life years", in: E Sloan, ed., Valuing
Health Care (Cambridge University Press, Cambridge, UK).
Kapp, M. (1990), "Economic barriers to innovation in mental health care: community care in the United
Kingdom", in: I.M. Marks and R. Scott, eds., Mental Health Care Delivery (Cambridge University Press,
Cambridge).
Keeler, E.B., et al. (1986), "The demand for episodes of mental health services", Rand Report R-3432-NIMH
(Santa Monica, CA).
Keeler, E.B., W.G. Manning and K.B. Wells (1988), "The demand for episodes of mental health services",
Journal of Health Economics 7(2):369-392.
Keeler, E.B., J.P. Newhouse and C.E. Phelps (1977), "Deductibles and demand: a theory of the consumer
facing a variable price schedule under uncertainty", Econometrica 45(3):641-655.
Keeler, E.B., G.M. Carter and J.P. Newhouse (1998), "A model of the impact of reimbursement schemes on
health plan choice", Journal of Health Economics 17(2):297-326.
Kelleher, K., M. Chaffin, J. Hollenberg and E. Fischer (1994), "Alcohol and drug disorders among physically
abusive and neglectful parents in a community-based sample", American Journal of Public Health 84:1586-
1590.
Kenkel, D.S., and D.C. Ribar (1994), "Alcohol consumption and young adults' socioeconomic status", Brook-
ings Papers in Microeconomics.
Kessler, R.C., K. McGonagle, S. Zhao, C. Nelson, M. Hughes, S. Eshleman, H. Wittehen and K. Kendler
(1994), "Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States",
Archives of General Psychiatry 51(1):8-19.
Kessler, R.C., R.G. Frank, M. Edlund et al. (1997), "Differences in the use of psychiatric outpatient services
between the United States and Ontario", New England Journal of Medicine 336(8):551-557.
Knapp, M. (1997), "Cost effectiveness, accountability and their relationship to alternative fiscal mechanisms:
results from Great Britain and the United States", Proceedings of Mental Health and Fiscal Reform (The
Clarke Institute, Toronto, Canada).
Knesper, D.J. (1989), "Substitution in production as the basis for research and policy relevant to mental health
specialists", in: C. Taube, D. Mechanic and A. Hohmam, eds., The Future of Mental Health Services
Research (USGPO, Washington).
Kouzis, A.C., and W.W. Eaton (1994), "Emotional disability days: prevalence and predictors", American
Journal of Public Health 84(8): 1304-1307.
Kronick, R., et al. (1996), "Diagnostic risk adjustment for Medicaid: the disability payment system", Health
Care Financing Review 17(3):7-34.
Kupfer, D.J., et al. (1992), "Five year outcomes for maintenance therapies in recurrent depression", Archives
of General Psychiatry 49:769-773.
Lambrinos, J., and J. Rubin (1981), "The determinants of average daily census in public mental hospitals:
a simultaneous model", Medical Care 19(9):895-906.
Lave, J.R., and R.G. Frank (1988), "Factor affecting Medicaid patients' length of stay in psychiatric units: the
effect of provider incentives", Health Care Financing Review 10(2):57-66
Lave, J.R., and R.G. Frank (1990), "The effect of the structure of hospital payment on length of stay", Health
Lave, J.R., R.G. Frank, C.A. Taube et al. (1988), "PPS and psychiatry: the first year", Inquiry 25(3):354-363.
Link, B.G., H. Andrews and F.T. Cullen (1992), "The violent and illegal behavior of mental patients recon-
sidered", American Sociological Review 57:275-292.
Link, B.G., et al. (1986), "Contact with the mentally ill and perceptions of how dangerous they are", Journal
of Health and Social Behavior 27(4):289-302.
Lowry, R., D. Holtzman, B.I. Truman, L. Kann, J.L. Collins and L.J. Kolbe (1994), "Substance use and HIV-
related sexual behaviors among US high school students: are they related?", American Journal of Public
Health 84:1116-1120.
Lurie, N., I.S. Moscovice, M. Finch, J.B. Christianson and M.K. Popkin (1992), "Does capitation affect the
health of the chronically mentally ill? Results from a randomized trial", Journal of the American Medical
Association 267(24):3300-3304.
Ma, C.A. (1995), "Health care payment systems: cost and quality incentives", Journal of Economics &Man-
agement Strategy 3(1):93-112.
Ma, C.A., and TG. McGuire (1997), "Optimal health insurance and provider payment", American Economic
Review 87(4):685-704.
Ma, C.A., and T.G. McGuire (1998a), "Cost and incentives in a behavioral health carve-out", Health Affairs
17:53-67.
Ma, C.A., and T.G. McGuire (1998b), "Network effects in managed health care", Working Paper (Boston
University).
Mangen, S.P., and B. Rao (1985), "United Kingdom: socialised system-better services?", in: S.P. Mangen,
ed., Mental Health Care in the European Community (Croom Helm Books, London).
Mangen, S.P. (1985), "Germany: the psychiatric enquete and its aftermath", in: S.P. Mangen, ed., Mental
Health Care in the European Community (Croom Helm Books, London).
Manning, W.G., et al. (1981), "The two-part model of the demand for medical care: preliminary results from
the health insurance study", in: J. Van der Gaag and M. Perlman, eds., Health, Economics and Health
Economics (North-Holland, Amsterdam).
Manning W.G., et al. (1984), "Cost sharing and the use of ambulatory mental health services", American
Psychologist 89(10):1077-1089.
Manning, W.G., K.B. Wells, J.L. Buchanan, E.B. Keeler, R.B. Valdez and J.P. Newhouse (1989), "Effects of
mental health insurance: evidence from the health insurance experiment", RAND R-3015-NIMH/HCFA.
Manning, W.G., and R.G. Frank (1992), "Econometric issues in the demand for mental health care under
insurance", in: R. Frank and W. Manning, eds., Economics and Mental Health (Johns Hopkins Press, Bal-
timore).
Manning, W.G., and M.S. Marquis (1992), "Effect of mental health insurance: evidence from the HIE", Work-
ing Paper (University of Michigan).
Marmor, T.R., and K.C. Gill (1989), "The political and economic context of mental health care in the United
States", Journal of Health Politics, Policy, and Law 14(3):459-475.
Matus, M., et al. (1986), "Mental illness and the media: II, Context analysis of press coverage of mental health
topics", Canadian Journal of Psychiatry 31(5):431-433.
McKusick, D., T. Mark, E. King et al. (1998), "Spending for mental health and substance abuse treatment
1996", Health Affairs 17(5):147-157.
McGuire, T.G. (1989), "Financing and reimbursement of mental health services", in: C. Taube, D. Mechanic
and A. Hohmann, eds., The Future of Mental Health Services Research (National Institute of Mental
Health).
McGuire, T.G. (1994), "Predicting the costs of mental health benefits", Milbank Memorial Fund Quarterly:
Health And Society 72(1):3-23.
McGuire, T.G. (1996), "Commentary" [on Rizzo and Blumenthal, "Is the target income hypothesis an eco-
nomic heresy?"], Medical Care Research and Review 53(3):267-273.
McGuire, T.G., and R.P. Ellis (1990), "Optimal payment systems for health services", Journal of Health
McGuire, T.G., and J. Montgomery (1982), "Mandated mental health benefits in private health insurance
policies", Journal of Health Politics, Policy and Law 7(2):380-406.
McGuire, T.G., and M.H. Riordan (1995), "Incomplete information and optimal market structure: public
purchases from private providers", Journal of Public Economics 56:125-141.
McGuire, T.G. (1981), Financing Psychotherapy: Costs Effects and Public Policy (Cambridge, MA).
McGuire, T.G. (1998), "Physician agency", in: J.P. Newhouse and A.Y. Culyer, eds., Handbook of Health
Economics (North-Holland Press, Amsterdam).
Mechanic, D. (1987), "Correcting misconceptions in mental health policy: Strategies for improved care of the
seriously mentally ill", Milbank Memorial Fund Quarterly 65(2):203-230.
Mechanic, D., M. Schlesinger and D. McAlpine (1995), "Management of mental health and substance abuse
services: state of the art and early results", Milbank Memorial Fund Quarterly 73:19-55.
Mechanic, D. (1997), "Muddling through elegantly: finding the proper balance in rationing", Health Affairs
16(5):83-92.
Mechanic, D. (1998), "The functions and limitations of trust in provision of medical care", Journal of Health
Politics, Policy and Law 23(4):661-686.
MedPac (1998), Report to Congress: Medicare Payment Policy (USGPO, Washington).
Mental Health, United States (1996) (USGPO, Washington).

Merrick, E.L. (1997), "Impact of a behavioral health carve out on treatment of major depression", Doctoral
dissertation (Barndeis University).
Michael, R.J. (1980), "Bureaucrats, legislators and the decline of the state mental hospital", Journal of Eco-
nomics and Business 32(3):198-205.
Miller, R.H., and H.S. Luft (1997), "Does managed care lead to better or worse quality of care?", Health
Affairs 16(5):7-25.
Moon, M. (1996), "The special health care needs of the elderly", in: S. Altman and U. Reinhardt, eds., Strate-
gic Choices for a Changing Health Care System (Health Administration Press, Chicago).
Morlock, L.L. (1989), "Recognition and treatment of mental health problems in the general sector", in:
C. Taube, D. Mechanic and A. Hohmann, eds., The Future of Mental Health Services Research (USGPO,
Washington).
Mullahy, J., and J.L. Sindelar (1993), "Alcoholism, work and income", Journal of Labor Economics
11(3):494-520.
Mullahy, J., and J. Sindelar (1990), "Gender differences in the effects of mental health on labor force partic-
ipation", in: I. Sirageldin, A. Sarkin and R.G. Frank, eds., Research in Human Capital and Development
(JAI Press, Greenwich, CT) 125-145.
Mulvey, E.P. (1994), "Assessing the evidence of a link between mental illness and violence", Hospital and
Community Psychiatry 45(7):663-668.
Nasar, S. (1998), A Beautiful Mind: A Biography of John Forbes Nash Jr. (Simon-Schuster, New York).
National Advisory Mental Health Council (1997).
Newhouse, J.P., and C.E. Phelps (1976), "New estimates of price and income elasticities for medical care
services", in: R. Rosett, ed., The Impact of Health Insurance on the Health Services Sector (National Bureau
of Economic Research, New York).
Newhouse, J.P., et al. (1981), "Some interim results from a controlled trial of cost sharing and health insur-
ance", New England J. Medicine 303:1501-1507.
Newhouse, J.P., and the Insurance Experiment Group (1993), Free for All? Lessons from the RAND Health
Insurance Experiment (Harvard University Press, Cambridge).
Newhouse, J.P., M. Beeuwkes Buntin and J.D. Chapman (1997), "Risk adjustment and Medicare: taking a
closer look", Health Affairs 16:26-43.
Newhouse, J.P., W.G. Manning, E.B. Keeler and E. Sloss (1989), "Adjusting capitation rates using objective
health measures and prior utilization", Health Care Financing Review 10(3):41-54.
Newhouse, J.P., C.E. Phelps and M.S. Marquis (1980), "On having your cake and eating it too: econometric
problems in estimating the demand for health services", Journal of Econometrics 13(3):365-390.
Newhouse, J.P. (1994), "Patients at risk: health reform and risk adjustment", Health Affairs 13(1):132-146.
Newhouse, J.P. (1996), "Reimbursing health plans and health providers: selection versus efficiency in produc-
tion", Journal of Economic Literature 34(3): 1236-1263.
Nunnally, J. (1961), Popular Conceptions of Mental Health: Their Development and Change (Holt, Rinehart
and Winston, New York).
O'Flaherty, B. (1996), Making Room: The Economics of Homelessness (Harvard University Press, Cam-
bridge).
Office of Mental Health, State of New York (1996), "Managed behavioral health care for medicaid", unpub-
lished memo.
Oss, M. (1995), Managed Behavioral Health Market Share in the United States (Open Minds, Gettysburg,
PA).
Padgett, D., et al. (1993), "The effect of insurance benefit changes and use of child and adolescent outpatient
mental health services", Medical Care 31(2):96-110.
Pauly, M.V. (1968), "The economics of moral hazard: comment", American Economic Review 58:531-537.
Pauly, M.V. (1978), "Is medical care different", in: W. Greenberg, ed., Competition in the Health Sector
(Aspen, Germantown, MD).
Pauly, M.V. (1985), "What is adverse about adverse selection?", in: R.M. Scheffler and L.E Rossiter eds.,
Advances in Health Economics and Health Services Research: Biased Selection in Health Care Markets
(JAI Press, Greenwich, CT) 281-286.
Perneger, TV., A.F Allaz, J.E Etter and A. Rougemont (1995), "Mental health and choice between managed
care and indemnity health insurance", American Journal of Psychiatry 52(7):1020-1025.
Phelps, C.E., and C. Mooney (1993), "Variation in medical practice use: causes and consequences", in:
R. Arnold and W. White, eds., Competitive Approaches to Health Care Reform (Urban Institute Press,
Washington, DC).
Phelps, C.E. (1998), "Comments on price index papers", in: Helms and Triplett, eds.
Phelps, C.E. (2000), "The role of information in the supply and demand for health care", in: J.P. Newhouse
and A.Y. Culyer, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 5.
Physician Payment Review Commission (PPRC) (1997), Annual Report (Washington, DC).
PROPAC (1986), Report to Congress (Washington, DC).
Ramsey, E (1927), "A contribution to the theory of taxation", Economic Journal 37:41-61.
Ramsey, S., and M.V. Pauly (1997), "Structure incentives and adoption of medical technologies in HMO and
fee-for-service health insurance plans", Inquiry 34:228-236.
Reed, L. (1975), "Coverage and utilization of care for mental health conditions under health insurance",
Various Studies 1973-74 (APA, Washington).
Rice, D., et al. (1990), The Cost of Alcohol and Drug Abuse and Mental Illness (USGPO, Washington).
Rice, T. (1998), The Economics of Health Reconsidered (Health Administration Press, Ann Arbor).
Rochefort, D.A. (1993), From Poor Houses to Homelessness (Auburn House, Westport).
Rothschild, M., and J. Stiglitz (1976), "Equilibrium in competitive insurance markets: an essay in the eco-
nomics of imperfect information", Quarterly Journal of Economics 90:629-649.
Rubin, J. (1978), Economics, Mental Health, and the Law (D.C. Heath and Company, Lexington, MA).
Rubin, J. (1980), "Judicial standards and the financing of mental health services", Inquiry 17(2):165-171.
Rupp, A., D.M. Steinwachs and D.S. Salkever (1984), "The effect of hospital payment methods on patterns
and cost of mental health care", Hospital and Community Psychiatry 35(5):450-458.
Scheffier, R.M., and C.A. Watt (1986), "Inpatient mental health use among the heavily insured", Journal of
Human Resources 21(3):338-358.
Scheffier, R.M., C. Grogan, B. Cuffel and S. Penner (1994), "A specialized mental health plan for persons
with severe mental illness under managed competition", Hospital and Community Psychiatry 44:937-942.
Secretary of the Department of Health and Human Services (1993), Alcohol and Health (USGPA, Washing-
ton, DC) ch. 9.
Segal, S., C. Silverman and J. Baumohl (1989), "Seeking person-environment fit in community care place-
ment", Journal of Social Issues 45(3):49-64.
Shapiro, S., P. German, A. Skinner et al. (1987), "An experiment to change detection and management of
mental health morbidity in primary care", Medical Care 25(4):327-339.
Shen, Y., and R.P. Ellis (1998), "Cost minimizing risk adjustment", Working Paper (Boston University).
Steadman, H.J., E.P. Mulvey, J. Monohan, P.C. Robbins, PS. Appelbaum, T. Grisso, L.H. Roth and E. Silver
(1998), "Violence by people discharged from acute psychiatric inpatient facilities and by others in the same
neighborhoods", Archives of General Psychiatry 55:1-9.
Stein, L.I., and M.A. Test (1980), "Alternative to mental hospital treatment", Archives of General Psychiatry
37:392-397.
Sturm, R. (1997), "How expensive is unlimited mental health care coverage under managed care?", JAMA
278:1533-1537.
Sturm, R., et al. (1994), "Provider choice and continuity for the treatment of depression", RAND Paper DRU-
692-AHCPR.
Swendsen, J.D., K.R. Merikangas, G.J. Canino, R.C. Kessler, M. Rubio-Stipec and J. Angst (1998), "The
comorbidity of alcoholism with anxiety and depressive disorders in four geographic communities", Com-
prehensive Psychiatry 39:176-184.
954 R. G. Frankand T G. McGuire
Taube, C.A., L.G. Kessler and B.J. Burns (1986), "Estimating the probability and level of ambulatory mental
health services use", Health Services Research 21(2):321-340.
Taube, C.A., L. Morlock, B.V. Burns and A.B. Santos (1990), "New directions in research on assertive com-
munity treatment", Hospital and Community Psychiatry 41(6):642-647.
Torrey, EE. (1994), "Violent behavior by individuals with serious mental illness", Hospital and Community
Psychiatry 45(7):653-662.
Triplett, J.E. (1998), "What's different about health? Human repair and care repair in national accounts",
Health Affairs (in press).
Umland, B. (1995), "Behavioral healthcare benefit strategies of self-insured employers", Behavioral Health-
care Tomorrow vol:65-70.
van de Ven, W.P.M.M., and R.P. Ellis (2000), "Risk adjustment in competitive health plan markets", in:
A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 14.
von Korff, M., and M. Kramer (1978), Mental and Nervous Disorders Utilization and Cost Survey (NIMH
and OPM, Washington).
Wallack, S.S., and H.J. Levine (1996), "Managed care for people with disabilities: caring for those with the
greatest need - overview", in: S. Altman and U. Reinhardt, eds., Strategic Choices for a Changing Health
Care System (Health Administration Press, Ann Arbor).
Watts, C.A., R.M. Scheffler and N.P. Jewell (1986), "Demand for outpatient mental health services in a heavily
insured population: the case of the Blue Cross and Blue Shield Association's Federal Employees Health
Benefits Program", Health Services Research 21(2):267-290.
Weiner, J.P. (1996), personal communication.
Weiner, J.P., A. Dobson, S.L. Maxwell, K. Coleman, B.H. Starfield and G.F. Anderson (1996), "Risk-adjusted
capitation rates using ambulatory and inpatient diagnoses", Health Care Financing Review 17(3):77-99.
Weinstein, M.C., and R.J. Zeckhauser (1972), "Critical ratios and efficient allocation", Journal of Public
Economics 2(2):147-157.
Weinstein, M.C. (1995), "From cost-effectiveness ratios to resource allocation: where to draw the line", in:
F. Sloan, ed., Valuing Health Care (Cambridge University Press, Cambridge, UK).
Weisbrod, B.A. (1983), "A guide to benefit-cost analysis as seen through a controlled experiment in treating
the mentally ill", Journal of Health Politics, Policy and Law 7(4):88-845.
Wells, K.B., W.G. Manning, N. Duan, J. Ware and J.P Newhouse (1982), Cost Sharing and the Demand for
Ambulatory Mental Health Services, R-2960-HHS (The RAWD Corporation, Santa Monica, CA).
Wells, K.B., et al. (1996), Caring for Depression (Harvard University Press, Cambridge).
World Health Organization (1994), Global Comparative Assessments in the Health Sector: Disease Burden
Expenditures and Intervention Packages (WHO, Geneva).
Yellowlees, H. (1990), "Administrative barriers to implementation and diffusion of innovative approaches
to mental health care in the United Kingdom", in: I.M. Marks and R.A. Scott, eds., Mental Health Care
Delivery: Innovations, Impediments and Implementation (Cambridge University Press, Cambridge).
Zeckhauser, R. (1970), "Medical insurance: a case study of the trade-off between risk-spreading and appro-
priate incentives", Journal of Economic Theory 2:10-26.
Chapter 17
LONG-TERM CARE*
EDWARD C. NORTON
University of North Carolinaat ChapelHill
Contents
Abstract 956
1. Introduction 957
2. Taxonomy of long-term care 958
2.1. Nursing home care 958
2.2. Other forms of long-term care 959
2.3. International comparisons 962
3. Supply of long-term care 964
3.1. Models of bed supply 964
3.2. Models of access 966
3.3. Models of quality of care 968
3.4. Models of supply of informal care 972
4. Demand for long-term care 975
4.1. Empirical evidence of demand factors 975
4.2. Models to explain why private insurance is rarely purchased 978
4.3. Models of how spend-down affects savings 981
5. Demographics 983
5.1. Demographics trends 983
5.2. Why demographic changes may not exacerbate problems 984
6. Conclusion 987
References 988
*This research was funded by a grant from the Agency for Health Care Policy and Research and the National
Institute on Aging (R01-HS09515). Alan Garber, Sherry Glied, Alvin Headen, Joseph Newhouse, Jonathan
Skinner, Frank Sloan, Sally Steams, and Peter Zweifel provided useful comments, and Ying-Chun Li provided
research assistance.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P. Newhouse
956 E.C. Norton
Abstract
This chapter summarizes recent theoretical and empirical economic research on long-
term care. Long-term care differs from acute medical care in four fundamental ways.
Long-term care is care for chronic illness, the nursing home industry is dominated by
for-profit facilities sometimes facing excess demand, long-term care is often provided
by unpaid caregivers, and little private long-term care insurance is purchased. This chap-
ter starts with a taxonomy of long-term care, from expensive formal nursing home care
to the vast provision of informal care provided in the home. The taxonomy is followed
by a review of the supply of and demand for long-term care. Competition between pre-
dominantly for-profit nursing homes is an issue for industrial organization. Raising the
public reimbursement rate to nursing homes paradoxically may actually lower quality.
The supply of informal care, usually by daughters, is an issue for labor economics.
Women who provide informal care may reduce their hours in the labor force. Private
insurance for such large and uncertain health expenditures is an issue for risk and insur-
ance. Few elderly purchase long-term care insurance, despite the high expenditure risk,
for a variety of sound reasons. Savings behavior, whether for precautionary motives or
bequests, is an issue for life-cycle behavior. A means-tested public insurance program
combined with uncertainty about future expenditures may cause a separating equilib-
rium in which people either save for precautionary reasons or save nothing. Finally, a
look to the future shows that demographic trends will greatly increase the number and
percentage of elderly in industrialized nations. Future long-term care expenditures are
not likely to be as burdensome as many fear. Although the majority of citations are
from research in the United States, the issues presented here are usually pertinent to
other industrialized nations.
JEL classification:Ill, 112, 118, J14

Ch. 17: Long-Term Care 957
1. Introduction
Long-term care has become an important area of health economics. Its importance lies
not only in its share of Gross National Product, which is about 1 percent in the United
States, but in how long-term care affects economic decisions for individuals over a
lifetime and across generations. For example, an elderly widow anticipating need for
long-term care may decrease savings or increase bequests to qualify for means-tested
public insurance, or may demand informal care from a working daughter, even though
she ultimately never enters a nursing home.
Long-term care differs from acute medical care in four fundamental ways. First, long-
term care is care for chronic illness or disability instead of treatment of an acute illness.
Caring for a chronic illness lasts as long as a person is alive so that medical expenses
accumulate unrelentingly. The market has developed many forms of long-term care in
response to demand from persons with different family and financial situations.
Second, the nursing home industry is dominated by for-profit facilities sometimes fac-
ing excess demand, in contrast to the hospital industry which is dominated by nonprofit
facilities with an excess supply of beds. Certificate-of-need regulation has constrained
the nursing home bed supply in many states. Waiting lists contribute to inefficient pro-
duction of nursing home care and may adversely affect quality of care because facilities
do not have to improve quality of care to remain competitive.
Third, long-term care is often provided by unpaid caregivers, instead of always being
provided by paid professionals. Many elderly people receive informal care from friends
or family, often a spouse or child, in the home. Informal care may affect the caregiver's
labor supply or may influence bequests, if such bequests are used to elicit attention and
informal caregiving by children.
Fourth, in contrast to relatively comprehensive acute care insurance for elderly
through Medicare and Medigap policies, little private long-term care insurance is pur-
chased and most public insurance is means-tested with high copayments. Thus, long-
term care is usually the greatest expenditure risk faced by the elderly. The potential
magnitude of long-term care costs suggests that means-tested public insurance may af-
fect lifetime savings.
In addition, the population is expected to age rapidly over the next several decades,
which will affect the long-term care industry even more than acute care. The falling
birth rate combined with longer life expectancy means that the elderly around the world
are expected to comprise a higher share of the population in the future. Who will care
for the elderly, or pay for their care? Policymakers are concerned that the rapidly aging
population will increase public expenditures on long-term care.
The economic issues in long-term care lie at the intersection of many other branches
of economics. Competition between predominantly for-profit nursing homes is an issue
for industrial organization. Raising the public reimbursement rate to nursing homes
paradoxically may actually lower quality. The supply of informal care, usually by
daughters, is an issue for labor economics. Women who provide informal care may
reduce their hours in the labor force. Private insurance for such large and uncertain
958 E.C. Norton
health expenditures is an issue for risk and insurance. Few elderly purchase long-term
care insurance, despite the high expenditure risk, for a variety of sound reasons. Sav-
ings behavior, whether for precautionary motives or bequests, is an issue for life-cycle
behavior. A means-tested public insurance program combined with uncertainty about
future expenditures may cause a separating equilibrium in which people either save
for precautionary reasons or save nothing. The potential future increase in demand for
long-term care by an aging society is an issue for demography. Future long-term care
expenditures are not likely to be as burdensome as many fear.
This chapter summarizes recent theoretical and empirical economic research on long-
term care. Nursing home care dominates the literature because of its visibility and cost
compared to other forms of long-term care. Long-term care, however, is anything but
a homogeneous good. This chapter starts with a taxonomy of long-term care, from ex-
pensive formal nursing home care to the vast provision of informal care provided in
the home. In between is a continuous spectrum of intermediate forms of care that are
close substitutes. The taxonomy is followed by a review of the supply of and demand
for long-term care. These two sections present theoretical models and empirical results.
They also focus primarily on nursing homes. Finally, a look to the future shows that
demographic trends will greatly increase the number and percentage of elderly in indus-
trialized nations. The implications for the supply of and demand for care and the cost of
care are discussed. Although the majority of citations are from research in the United
States, the issues presented here are usually pertinent to other industrialized nations.
2. Taxonomy of long-term care

2.1. Nursing home care
Long-term care covers a continuous spectrum from infrequent informal care provided
by a neighbor to institutional care with around-the-clock nursing. The nursing home
industry is an appropriate starting point for a review of long-term care because of its
size and cost. Many elderly, and a few disabled nonelderly, enter a nursing home when
they are no longer able to live independently. In the United States there were 1.56 mil-
lion residents in 16,840 nursing homes in 1996 [Rhoades, Potter, and Krauss (1998)].
The average number of beds in a nursing home is 104, though the sizes vary widely.
Unlike hospitals, most nursing homes are for-profit (65.9 percent), with 26.2 percent
private nonprofit and the remaining 7.9 percent government owned [Rhoades, Potter,
and Krauss (1998)]. Arrow (1963) hypothesized that nonprofit providers are common
in markets for complex persond services because they have less incentive than for-profit
providers to underprovide quality to poorly informed consumers (Hirth, 1999). As of
1994, 46 states had either certificate-of-need regulation or moratoria policy [Harring-
ton et al. (1997a)]. The overall occupancy rate is 88.8 percent [Rhoades, Potter, and
Krauss (1998)], but is generally lower in states without-certificate-of-need regulation.
Many nursing homes maintain waiting lists. They have empty beds only briefly after a
resident dies or is discharged.
Over a lifetime the probability of nursing home use is high [Murtaugh, Kemper, and
Spillman (1990), Kemper and Murtaugh (1991), Liang et al. (1996), Murtaugh et al.
(1997)]. Twenty-seven percent of all persons over 25 will use a nursing home at some
time in their life. Those who do will spend an average of 2.4 years in a nursing home
[Murtaugh et al. (1997)]. Not surprisingly, the risk of use rises with age. About 70
percent of the elderly who will be admitted to a nursing home are first admitted after age
75, and nearly half are first admitted after age 85 [Dick, Garber, and MaCurdy (1994)].
A person who reaches age 65 has a 43 percent chance of being admitted to a nursing
home prior to death [Kemper and Murtaugh (1991)]. For women, the probability is
greater than 50 percent, and for men it is one-third. On any given day five percent of the
persons age 65 and older are nursing home residents. Although some research papers
like to categorize residents as being either long-stayers or short-stayers, such a polar
view is not supported by the continuous distribution of lengths of stay. Some residents
stay only for days or weeks. These residents are more likely to return home, especially
if they are recovering from an acute medical problem such as a hip fracture. Many
people have multiple stays over their lifetime, and between 15 and 20 percent of nursing
home residents will have cumulative use greater than five years [Murtaugh et al. (1997)].
A few residents stay for as long as 20 years. Not surprisingly, with such wide variation
in length of stay, prospective payment per episode, which Medicare uses for inpatient
care, is not used for nursing home care.
Nursing home care is expensive, and insurance is far from complete in the United
States. One year in a nursing home can cost more than $47,000 for a private pay resi-
dent (numbers on nursing home expenditures in this paragraph are for year 1996 in the
United States, based on Levit et al. (1997)). The reimbursement rate is typically 10 to 30
percent less for a Medicaid resident. About half of residents are covered by Medicaid.
Those covered by Medicaid must spend down their assets and contribute most of their
monthly income. Medicare pays only 11.4 percent of nursing home revenues because
it only covers nursing home care immediately following an inpatient stay and only up
to 100 days [Norton and Newhouse (1994), Murtaugh, Kemper, and Spillman (1995)].
Private long-term care insurance is limited, for reasons discussed in Section 4.2, and
pays only about 5.2 percent of revenues. The wide variation in length of stay and lack
of complete insurance mean that the elderly face substantial out-of-pocket costs.
Expenditures on intermediate care facilities for the mentally retarded and develop-
mentally disabled were among the fastest growing portion of the Medicaid budget
throughout the 1980s. Most residents are nonelderly. The number of these facilities grew
ten-fold, from 547 in 1977 to 5,405 in 1990 [Congressional Research Service (1993)].
2.2. Otherforms of long-term care

Nursing homes are not the only setting for long-term care, just the most expensive and
thoroughly studied. People choose their care based on their physical and mental health,
finances, and family situation. This section provides an overview of some of the more
common forms of long-term care, but is not an exhaustive treatment. For example, it
does not cover social support services that may enable people in need of long-term care
960 E. C.Norton
to remain in the community, such as meals on wheels, adult faster care Nyman, Finch,
and Kane (1997)], and adult day care [Weissert et al. (1991)]. Nor does it cover hospice
care, which although not usually considered long-term care is an alternative form of
care for elderly with a terminal illness [Hamilton (1993)].
Despite the visibility of nursing homes, most care for the elderly is provided infor-
mally. Quantifying the extent of informal care is extremely difficult because it is not
reimbursed in a market. A general rule of thumb is that about two-thirds of care for
the elderly is informal care. Spouses are the first line of defense, especially for married
men. Because long-term care is most often needed near the end of life, which tends to
come sooner for men than their wives, more married women than married men provide
informal care. The other most common informal caregiver is a child, usually a daughter.
More than three-quarters of children caregivers are daughters, due in part to the lower
average opportunity cost of time [Stone, Cafferata, and Sangl (1987)]. The remaining
caregivers are siblings, sons- or daughters-in-law, grandchildren, neighbors, and other
nonrelatives.
Time spent providing informal care is time taken from other activities, namely work
or leisure. Many women must consider the tradeoff between providing care for a par-
ent and working in the labor force. Among caregivers, about 11 percent of daughters
and 5 percent of sons quit a job in order to provide that care [Stone, Cafferata, and
Sangl (1987)]. The increase in job opportunities for women has made this tradeoff per-
tinent for many women. Although the issues are potentially similar for men, in practice
care is more likely to be provided by women, and thus labor economists have studied
the theoretical and empirical relationship between informal care and the labor supply
of working-aged women [Chang and White-Means (1995), Stern (1995), Nocera and
Zweifel (1996), Sloan, Hoerger, and Picone (1996), Sloan, Picone, and Hoerger (1997)].
This is explored further in Section 3.4.
An alternative to informal care is formal home care provided by paid help. Home
care can range from periodic help with shopping and cleaning to full-time nursing help.
In 1982 an estimated 1.1 million received paid care in the United States [Liu, Manton,
and Liu (1985)1. Since then home health care services have expanded rapidly, amouting
to $30.2 billion of spending in 1996 [Levit et al. (1997)]. Persons who are older and
more physically disabled are more likely to receive home care, while living in an urban
or rural location and cognitive status seem to be unrelated [Coughlin et al. (1992)].
Public home care insurance has an obvious moral hazard problem. Who would not want
some paid help at no out-of-pocket cost with household chores? The average annual
percent change in home health care expenditures exceeded 20 percent in the early 1990s,
but that rapid growth has slowed as concerns about fraud and abuse grew [Levit et al.
(1997)]. The estimated cost of expanding public home care coverage depends critically
on whether elderly use the public home care in addition to or instead of other services
[Coughlin et al. (1992)]. Paid home care has not been shown to be a cost-saving way
of keeping elderly out of nursing homes and improving their quality of life. One quasi-
experimental study found that elderly who received home delivered meals were less
likely to be admitted to intermediate care facilities and that quality of life improved
[Hughes et al. (1987)]. However, there was no difference in admission rates to skilled
nursing facilities, and expenditures on the treatment group were 25 percent higher than
the control.
Board and care homes are residential settings that provide more supportive services
than boarding or rooming houses, but less medical care than a nursing home [Phillips et
al. (1997)]. Board and care homes provide lodging, meals, protective oversight, activi-
ties, and some assistance with medication, personal care, and activities of daily living.
Board and care homes are known by many other names, including personal care homes,
domiciliary care homes, residential care homes, homes for the aged, adult foster homes,
and assisted living facilities. They have two advantages over nursing homes. First, resi-
dents find them more homelike. Also, Medicaid finds them less expensive than nursing
homes, although comparisons are difficult because board and care residents tend to be
healthier than nursing home residents.
Board and care homes are smaller and more numerous than nursing homes [Phillips
et al. (1997)]. There are about 35,000 licensed board and care homes with a total of
600,000 beds in the United States. In addition, there are between 20,000 and 30,000
unlicensed homes. Most board and care homes have ten or fewer beds and serve the
elderly. A few have hundreds of beds, and many also serve younger persons with chronic
mental illness or developmental disabilities. More than 80 percent of the licensed homes
and 60 percent of the unlicensed homes are for-profit facilities.
More than 220,000 wealthy elderly are residents in a continuing care retirement com-
munity (CCRC) [Conover and Sloan (1995)]. CCRCs essentially combine a residential
community with long-term care insurance, although there is considerable variation in
the extent of this insurance [Ruchlin (1988)]. Residents pay a large initial fee upon entry
and rent an apartment for an additional monthly fee in a community setting designed
for elderly. If their health declines, residents receive some long-term care, often in an
onsite nursing home for additional charges. CCRCs also provide many activities, fine
food, assistance with living, and other amenities. One potential benefit of the support-
ive services and capitation of CCRCs is that nursing home utilization may be reduced.
Sloan, Shayne, and Conover (1995) found that CCRCs reduced nursing home care by 13
percent and personal care by 5 percent. CCRCs are not the general solution to providing
appropriate care to the elderly because they are prohibitively expensive for a majority
of elderly. CCRCs are financed by housing equity and by monthly fees for rent and
services. After rapid growth during the 1970s, some well-publicized bankruptcies at-
tracted congressional scrutiny and slowed industry growth [Conover and Sloan (1995)].
The policy concern was that elderly risk losing their life savings. Residents in CCRCs
live longer than the population average, which may be due to selective admission of
residents, CCRCs providing good care, or to longevity related to higher lifetime income
[Feinstein (1993)].
Two well-studied programs have tried to coordinate long-term care services. The first
is the Program of All-inclusive Care for the Elderly, or PACE, which provided integrated
care and financing at 11 sites in 9 states in 1997, and is adding up to 20 new sites per year
[Branch, Coulam, and Zimmerman (1995), Eng et al. (1997)]. PACE is managed care
962 E. Norton
for a combination of acute and long-term care. Medicaid and Medicare pay a monthly
capitated rate to PACE, which then assumes full risk for all care. Participants pay no
copayment or deductibles for more services than Medicare and Medicaid combined. An
interdisciplinary team of physicians, nurses, and therapists coordinate a personalized
care plan. PACE emphasizes providing care in a single health center, because it is most
cost-effective. The early results show that PACE may reduce expenditures by between
5 and 15 percent, although this program takes much effort to develop, has failed to
attract middle-income elderly persons who are not dually-eligible, and has trouble hiring
primary care physicians [Eng et al. (1997)].
The second program was the National Long-Term Care Demonstration, known as
Channeling, which provided home and community based long term care management
at 10 sites between 1981 and 1985. Numerous studies of Channeling have been pub-
lished [e.g., the entire special issue of Health Services Research, April 1988; Ra-
biner, Steams, and Mutran (1994), Kemper and Pezzin (1996), Pezzin, Kemper, and
Reschovsky (1996)]. The results from the Channeling studies largely confirmed that
providing home and community-based care is not cost-saving. It is not cost-saving be-
cause it is hard to target persons most at risk of institutionalization, where the potential
savings lie. Studies showed small benefits to caregivers. Home care did not substitute for
informal care by family members. The benefits of targeting home and community-based
care are limited and not worth the expense, according to these studies.
In summary, the market has developed a variety of solutions to the problem of giving
care to chronically ill persons with widely varying physical and mental health status,
finances, and family situations. Economic models of long-term care need to capture the
full range of services available.
2.3. Internationalcomparisons
Other countries also have a wide variety of forms of long-term care [Doty (1988)]. Most
countries rely on a combination of public and private financing, and only Germany
now mandates public long-term care insurance [Doty (1990), Schulte (1996)]. Most
countries have a range of services from informal care provided by children to formal
care in institutions, according to the most recent figures from the OECD (1996). The
percentage of elderly living with their own offspring varies widely among countries
from less than 10 percent in Denmark, Netherlands, and Sweden, to 65 percent in Japan
(see Table 1 for statistics on types of care in 17 countries). Home health care rates
ranges from one percent in Italy and New Zealand to 24 percent in Finland. The rates
of institutionalization are all in the range of five to ten percent. Ikegami and colleagues
(1997) found a wide variation in the percentage of low-care residents, those without
major limitations in activities of daily living, across six countries.
The cross-sectional statistics in Table 1 mask the rapid changes in the organization
and financing of long-term care. It would be pointless to describe in detail the current
institutional arrangements in each country because they are changing too quickly. Sev-
eral trends, however, are worth noting. First, there is a trend away from elderly living
with their offspring in intergenerational families. For example, the percentage living
Ch.17: Long-Term Care 963
Table 1
Percentage of elderly by type of long-term care services
Percentage of elderly Percentage of Percentage of elderly

living with offspring elderly receiving living in an institution
home help
% Year % Year % Year
Australia 7 1988 10.4 1986

Austria 25 1987 3 1991 4.7 1992
Canada 2 1990 8.1 1991
Denmark 4 1988 17 1991 5.7 1992
Finland 14 1987 24 1990 7.2 1986
France 17 1990 7 1985
Germany 14 1987 1-3 1992
Ireland 43 1979 3 1990
Italy 39 1990 1 1988
Japan 65 1985 2 1993
Netherlands 8 1987 8 1990
New Zealand 18 1977-8 1 1993 6.3 1991
Norway 11 1988 14 1991 7.1 1992
Spain 30 1988 2 1994
Sweden 5 1986 13 1990 5.4 1990
United Kingdom 16 1980 13 1991 5.1 1990
United States 15 1987 4 1990 5.4 1990
Table lists the percentage of the population age 65 and above that lived with own off-
spring, received home help, and lived in an institution. Based on data from OECD (1996,
Tables 1.A.4, 3.6, and 3.2). The percentage in an institution includes any communal estab-
lishment, not just nursing homes.
with offspring fell in Finland from 55 percent in 1950 to 14 percent in 1987, and in
Japan from 80 percent in 1953 to 65 percent in 1985 [OECD (1996)]. Second, there
are relatively fewer women to supply informal care, and many more of these women
now work. From 1960 to 1990 the proportion of women aged 46 to 69 relative to the
number of persons age 70 and over shrank from 2.26 to 1.53 in eight OECD countries
[OECD (1996)]. Third, there is a trend away from providing long-term care in hospitals,
because less expensive alternatives such as nursing homes are becoming more available
[Doty (1993)].
Although the issues in long-term care cut across countries, the published research
does not. There are surprisingly few economic studies written in English that focus on
issues specific to non-US countries, or use non-US data. In addition to the papers cited
above that provide international comparisons, Cambois and Robine (1996) compare
trends in disability-free life expectancy across countries. Several studies focus on recent
or impending expansion of long-term care insurance. Ikegami (1997) discusses the fund-
ing, benefits, and potential problems with the proposed public long-term care insurance
program in Japan, due to start in 2000. Naon (1996) shows how expansion of long-term
care insurance to all disabled elderly in Israel in 1988 has affected families, institutional-
964 E.C.Norton
ization, and expenditures. Schulte (1996) discusses the economic issues surrounding the
recent long-term care insurance reform in Germany, that extended coverage and benefits
for both institutional and noninstitutional care. A few other studies address interesting
aspects of long-term care in specific countries. Bartlett and Phillips (1996) discuss the
rise in private long-term care facilities in the United Kingdom, which mirrors the rise
in private sector involvement in acute health care. Forder (1997) studies incentives in
contracts between purchasers and providers of residential care in the United Kingdom.
Lane and colleagues (1987) use a Markov model to forecast the use of various home and
facility placements in British Columbia's long-termn care program. Nocera and Zweifel
(1996) study the determinants of informal care using data from Switzerland.
3. Supply of long-term care

3.1. Models of bed supply
The nursing home market has many properties of a competitive market [Bishop (1988)].
Barriers to entry are low, at least when there is no certificate-of-need regulation. Capital
costs per bed are much lower for a nursing home than for a hospital, and new nursing
homes can enter with little owner equity [Baldwin and Bishop (1984)]. Nursing homes
hire relatively unskilled labor and do not need highly specialized equipment. Adminis-
trative and licensing costs are also low. Furthermore, there are few, if any, economies
of scale. Nursing homes can enter with few beds. Therefore, barring regulation of en-
try, nursing homes of all sizes should be able to enter the market easily, based on entry
costs.
The competitive nature of the market also depends on the extent to which consumers
can make informed choices [Nyman (1989)]. If, prior to admission, potential residents
can search many nursing homes at low cost and observe the price and quality of care,
then the market will be competitive. Unlike acute medical care, the demand for nursing
home care is often not time sensitive. Potential residents may have weeks or months
in which to search. When choosing a nursing home potential residents may also ob-
tain help from hospital discharge planners, relatives, and social workers. Nursing home
services are not technical and can be evaluated more easily by consumers than, say,
surgical skills. Potential residents or family members can observe and assess the odor,
food, room size, price, social activities, attitude of other residents, and staff demeanor.
The wide range of close substitutes, often at lower prices, should place nursing homes
in a competitive environment. The low barriers to entry and pool of informed consumers
should make the nursing home industry competitive and responsive to consumers' de-
mands.
Despite these attributes of a competitive market, the nursing home market is not com-
petitive in many ways. Many nursing homes have waiting lists and operate at, or near,
full capacity. The waiting lists may imply that demand exceeds supply, which would
not happen in a freely competitive market in equilibrium. In a competitive equilibrium
new nursing homes would enter the market until waiting lists were eliminated, unless
the waiting lists served a productive purpose such as indicating quality. Testing for the
extent of excess demand, if any, is difficult, and has lead to mixed findings [Nyman
(1993)].
Although consumers can be informed and selective, most are not. Elderly who need
nursing home care are disproportionally the ones with no close family to help them
search, and end up in a nursing home because they have fewer options than other elderly.
Elderly with close family often postpone searching for a nursing home because the
thought of institutionalization is unpleasant. Then when a decision becomes necessary,
location is often the overriding criteria, not quality of food or activities. Elderly may
have no choice if there are waiting lists and they are covered by Medicaid which pays
a lower reimbursement rate. However, the limits on consumer choice alone do not fully
explain why waiting lists are so prevalent.
The supply of nursing homes in the United States has been partially restricted by
certificate-of-need regulation at the state level. The National Health Planning and Re-
sources Development Act of 1974 required state approval for all new construction or
expansion of health facilities, including nursing homes [Harrington et al. (1997b)]. By
the end of the 1970s every state except Arizona had implemented certificate-of-need
regulation [Harrington et al. (1997a)]. This federal law was repealed in 1986, although
by 1994 only 9 states had repealed their certificate-of-need regulation. Meanwhile 16
states had added moratoria on bed construction, leaving only 5 states with neither form
of regulation.
States passed certificate-of-need regulation to limit rapidly expanding Medicaid ex-
penditures on the elderly. A nursing home could not enter the market or construct new
beds without first demonstrating, either by the occupancy rate or the bed-to-aged pop-
ulation ratio, that more bed supply was needed. By capping total nursing home beds,
legislators hoped also to cap beds allocated to Medicaid beneficiaries, and thus Medi-
caid long-term care expenditures. Certificate-of-need regulation was therefore intended
to combat the moral hazard problem of Medicaid long-term care insurance by impos-
ing supply-side constraints. However, the economic justification for certificate-of-need
regulation rests on whether an unregulated market would result in an excess of capital
expenditure and capacity. This depends largely on how responsive demand is to price
given third-party insurance. For the roughly half of the residents who pay for nursing
home care out of pocket, moral hazard is not a concern. For the other half who are cov-
ered by Medicaid, the substantial copayment of virtually all income and the disutility
from entering a nursing home also call into question this assumption.
The effects of certificate-of-need regulation were immediate, wide ranging, and often
unintended [Feder and Scanlon (1980)]. First, nursing home operators acted quickly to
expand capacity before restrictive regulations were in force. Some states saw increases
in the bed supply of up to one-fourth between the passage of legislation and the start
date. Second, regulated supply combined with Medicaid rates below private pay rates
caused inefficiencies in the allocation of nursing home care. Medicaid beneficiaries, par-
ticularly those with heavy care needs, faced access problems. This limited their choice
of provider, and forced many to stay in hospitals long past the appropriate time of dis-
966 E.C Norton
charge. The state then had to pay the higher hospital rate for administratively necessary
days rather than the lower nursing home rate. Third, constrained supply may have led
to lower quality of care. Although granting new licenses could be used as a carrot for
nursing homes to provide high quality of care, quality of care may still have suffered
because the state would be reluctant to close a nursing home with very low quality of
care for lack of another nursing home to place the residents.
Fourth, certificate-of-need regulation does not necessarily restrict the supply in the
long run. An important policy question is to what extent did the certificate-of-need and
moratorium regulation slow the growth in the number of beds. The theoretical liter-
ature cited above assumes that the number of beds is fixed, at least in the short run.
However, between 1981 and 1991 the number of beds rose in all but one state, and
the total increased from 1.4 million to 1.7 million. Harrington and colleagues (1997a,
1997b) tested whether the state certificate-of-need and moratorium regulation slowed
the growth of nursing home beds between 1981 and 1993. They found that the change
in the number of beds per elderly was smaller in states with either certificate-of-need or
moratorium regulation, after controlling for sociodemographic, economic, and political
factors.
3.2. Models of access
Reimbursement policies affect nursing homes' decisions about whom to admit. Cur-
rently, most state Medicaid programs pay nursing homes a constant per diem reimburse-
ment rate that depends on historical costs [Congressional Research Service (1993)].
Such a payment system provides undesirable incentives. Because the rate is indepen-
dent of a resident's health, a nursing home prefers to admit healthier residents, who on
average use fewer resources and cost less. Furthermore, as residents recuperate, they
require less care and, with a fixed per diem, become financially more attractive. Thus, a
fixed per diem can contribute to longer stays among residents who may be less in need
of care than those waiting for admission.
Scanlon (1980) modeled access by showing how demand from two segments of the
population affect supply. His demand model has implications for the supply of nurs-
ing home beds, and it influenced subsequent models of the supply of quality of care.
Scanlon's analysis is based on the dual-market model, which has been applied to other
markets in economics. He assumed that each nursing home faces two types of demand
curves. Private residents have a downward sloping demand curve because they pay the
full marginal cost. In contrast, he assumed that Medicaid residents pay nothing out of
pocket, and so have unlimited demand for nursing home benefits. In the model presented
here, there is always excess Medicaid demand at the regulated Medicaid price (Scanlon
also presented a model without excess demand). The nursing home has a fixed supply
of beds.
Formally, assume that the nursing home maximizes profits 7r from two types of resi-
dents. Private residents pay p and have demand x (p). The nursing home receives reim-
bursement rate r for each Medicaid resident. The total bed supply is x. Costs c(x) are
the same for private and Medicaid residents. Therefore, as long as the nursing home is
full, total costs are fixed. Nursing homes maximize profits with respect to private price
max7r = px(p) + r(x - x(p)) - c(X). (3.1)
Nursing homes maximize profits by setting the private price such that marginal pri-
vate revenue equals marginal Medicaid revenue. In his paper, Scanlon describes his
results in terms of setting the marginal private revenue equal to marginal cost, but if
the bed supply is fixed, as it is in the short run for the nursing home, then the model
presented here is more appropriate. Although Scanlon also derives results for nonprofit
nursing homes we focus on the for-profit results because nonprofits are in the vast mi-
nority in this industry.
Scanlon's model has several implications for the supply of nursing home beds relative
to the demand. Scanlon concludes that Medicaid beneficiaries may face more limited
access to nursing home beds than private-pay residents. Limited bed supply and low
reimbursement rates cause excess demand for Medicaid beneficiaries. The number of
private residents depends on the Medicaid reimbursement rate because the nursing home
sets marginal private revenue equal to marginal Medicaid revenue. A higher Medicaid
reimbursement rate raises the marginal private revenue in equilibrium, and fewer private
residents are admitted.
Excess demand causes longer lengths of stay in the hospital for people dually-eligible
for Medicaid waiting for admission to a nursing home because of differences in both
the expected reimbursement and cost of care. On the reimbursement side, the nursing
home reimbursement rate is lower for Medicaid residents than for private residents.
On the cost side, the cost of care is higher for disabled persons than that for the more
able-bodied, and many dually-eligible person have more disabilities [Kenney and Ho-
lahan (1991)]. Therefore, when there is excess demand nursing homes may to wait for
a private-pay resident. Gruenberg and Willemain (1982) found that the length of stay
in Massachusetts hospitals was longer not only for Medicaid patients waiting for place-
ment in a nursing home but also for anyone disabled. Ettner (1993) also found evidence
that Medicaid patients have more nursing home access problems than private patients.
She found that Medicaid patients in areas with relatively low bed supply and in areas
with greater demand from private patients were more likely to be on a waiting list for
admission to a nursing home. However, Weissert and Cready (1988) showed that there
was only a one-day difference between Medicaid and private patients discharge delays
in North Carolina, and that many heavy-care requirements were not associated with de-
lays. Norton (1992a) found that nursing homes in San Diego, California, were more
likely to admit persons with disabilities when they were provided a reimbursement rate
adjusted for health status rather than a flat reimbursement rate. Therefore the bulk of the
empirical evidence supports the hypothesis that Medicaid residents have more restricted
access to nursing homes than do private pay residents.
Not every Medicaid program pays a flat per-diem rate with these perverse incentives.
Some programs provide positive incentives to admit sicker people, although their quan-
968 E.C. Norton
titative effects have not yet been evaluated. Several states assign residents to one of sev-
eral health status categories using the Resource Utilization Group system, and vary the
reimbursement rate by category [Fries et al. (1994)]. The New York experience with the
Resource Utilization Group payment system, which increases reimbursement rates for
residents in worse health status, has been that nursing homes admitted more heavy-care
residents and reduced days of care to lighter-care residents [Thorpe, Gertler, and Gold-
man (1991)]. Reimbursement methods in countries other than the United States vary
widely, but are generally not keyed to health status or outcomes. The Resource Utiliza-
tion Group payment system has been validated in many countries, including Sweden,
Spain, Netherlands, the United Kingdom, and Japan [Ikegami et al. (1994)]. It will be
implemented in Canada and Barcelona, and has been proposed for use in the United
Kingdom.
3.3. Models of quality of care
The quality of care in nursing homes is recognized as being low. Nursing homes have
been accused for years of financial fraud, providing low quality medical care, and ig-
noring residents' personal needs [Mendelson (1974)]. The Institute of Medicine (1986)
criticized the regulatory process as focusing too much on structure while ignoring care
and the effects of that care on outcomes. State regulations generally pertain to civil
rights and safety, such as fire codes, rather than broader measures of quality.
Efforts to control costs may have unintended side-effects on the quality of care. The
public sector uses three policy instruments to control costs. Medicaid can restrict eligi-
bility, restrict the reimbursement rate, and restrict the supply of nursing home beds. The
first policy reduces the demand by Medicaid residents, the second policy reduces the de-
mand for Medicaid residents by nursing homes, and the third policy reduces the supply
of Medicaid services. Several authors expanded Scanlon's model to include quality of
care, and these models enable analysis of the effects of cost control policies on quality
of care [Nyman (1985), Dusansky (1989), Gertler (1989), Gertler (1992), Gertler and
Waldman (1992)]. The following model is a synthesis of work by these authors, who
assume that private residents care about quality, and that the cost function depends on
quality.
The nursing home takes Medicaid reimbursement r and its own bed supply as
givens, and chooses private price p and quality of care q to maximize profits r:
max 7r = px(p, q) + r (x - x(p, q)) - c(q lx). (3.2)

p,q
The number of private residents is assumed to be decreasing in the private price and
increasing in quality, while cost is increasing in quality. Thus, the signs of the first
derivatives are: Xp < 0, xq > 0, and Cq > O0.The number of private residents and the
cost are assumed to be twice continuously differentiable. The first-order conditions for
maximization of profits with respect to private price and quality of care are:
7rp = O (p - r)xp + x = 0, (3.3)

rq = O = (p - r)xq - Cq = 0. (3.4)
The first-order conditions lead to the standard interpretation that the nursing home
sets the marginal revenue from a change in either private price or quality of care to
be equal to the marginal cost. Rewriting these conditions offers other insights. From
Equation (3.3):
dp = p
( X p
r=P+ d-x 1+- , p (3.5)
dx Exp XP(x/p)
The nursing home sets the marginal revenue from an additional private resident equal to
the Medicaid rate, holding the quality of care constant. The marginal revenue from an
additional private resident is composed of two parts. One part is due to the additional
revenue received from having another resident, and the other part is due to the required
adjustment in revenue from all current residents to compensate for the small decline in
private price from taking in one more resident.
The private price is a function of the elasticity of private demand with respect to price,
or xp . This is shown by solving Equation (3.5) for the private price:
p = ( + xp ) r (3.6)
If demand is elastic (xp much less than -1), then the private price is only slightly
greater than the Medicaid rate because the nursing home cannot raise the private price
above the Medicaid rate without losing private demand. In contrast, as demand becomes
less elastic (xp approaches -1), then the private price greatly exceeds the Medicaid
rate. Nyman (1989)] estimated the price elasticity to be -1.7.
The elasticity of demand with respect to quality of care, or exq, is proportional to the
elasticity of demand with respect to price Esp, but has the opposite sign. This result is
found by rewriting Equation (3.4) and using Equation (3.5) for (p - r):
exq - (x/q)(p - ) Exp. (3.7)
Thus, both elasticities in general will be either high or low in absolute value. The elas-
ticity of demand with respect to quality of care increases with quality q.
The Medicaid rate affects more than just the profit per Medicaid resident, because it
also affects the relative profitability of Medicaid and private residents. If the Medicaid
rate is increased, then not only will nursing homes admit more Medicaid residents, but
the decisions about the private price and quality of care will also change as nursing
homes maximize profits under the new policy. The effect of a change in the Medicaid
970 E.C. Norton
reimbursement rate on the private price and the quality of care is found by totally differ-
entiating the first-order conditions in Equations (3.3) and (3.4) with respect to private
price, quality of care, and Medicaid rate and applying Cramer's Rule to get the follow-
ing system of equations:
dp 7rprTrqq - 7tqrTpq _ (+)(- - (-(?)

dr IHI (+) (
dq 7rpp,7qr - qp7pr _ (-)) - (+) (39)
dr IH] (+)
The changes in private price and quality of care will be positive if the expressions in
Equations (3.8) and (3.9) are positive, which is determined by examining the first-order
and second-order conditions. According to Equations (3.3) and (3.4), rpr = -p > 0,
and rqr = -xq < 0. The Hessian IHI is positive, and the second derivatives with respect
to private price and quality of care are negative, by the second-order conditions. Neither
numerator can be signed, though, because although intuition suggests that as quality of
care increases the change in profit with respect to private price 7pq is negative, it can be
positive. However, if J7pq is small or negative, then an increase in Medicaid rate raises
the private price but lowers the quality of care.
Quality of care may decline with an increase in the Medicaid reimbursement rate be-
cause such care is a public good. All residents, including those covered by Medicaid,
benefit when the quality of care rises. However, only private residents pay for quality be-
cause the Medicaid rate is set independent of quality. Therefore, as Medicaid residents
replace private residents, the pool of people able to pay for quality shrinks. This reduc-
tion raises the marginal cost of quality to the remaining private residents and results
in less quality. Nyman (1988) found that in markets where excess demand was likely,
increased percentage of Medicaid residents was associated with increased violations
(lower quality). In contrast, where excess demand was unlikely, increased percentage of
Medicaid residents was unrelated to violations.
Medicaid programs limit Medicaid expenditures not only by setting the Medicaid
reimbursement rate but also by restricting the supply of beds by using certificate-of-
need regulation. An increase in the number of nursing homes n will also affect the
private demand faced by each nursing home, since it must now compete with more
nursing homes for the same pool of private residents. For simplicity, assume all nursing
homes are identical. As more nursing homes enter the market, each nursing home will
perceive the demand as becoming more elastic. It is assumed that a nursing home cannot
capture the entire private market if its private price is slightly lower than all other nursing
homes. This assumption implies that certain second derivatives of profit are not zero,
namely that rp, < 0 and WTqn > 0. A change in the number of nursing homes affects
only the private demand and not the cost.
The effect of a change in the number of nursing homes is found in an analogous
manner as for a change in the Medicaid rate. The following system of equations is
found by totally differentiating the first-order conditions with respect to private price,
quality of care, and number of nursing homes, and then applying Cramer's Rule:
7
dp nrpnqq - rqnrpq _ (-)(-) - (+)(?) (3.10)
dn IHI (+)
dq rpp7qn - pn (-)(+) - (?)(-.11)
(3.11)
dn IHI (+)
The changes in private price and quality of care will be positive if the expressions
marked with a question mark in Equations (3.10) and (3.11) are positive. By a similar
reasoning process as before, if 7rpq is small or negative, then an increase in the number
of nursing homes lowers the private price but raises the quality of care.
The private price will typically fall as the number of nursing homes increases. How-
ever, quality of care will increase as the number of nursing homes increases. This result
is again explained by quality being a public good. As demand becomes more elastic,
an increase in quality of care will increase private demand. This increase in quality will
pay for itself if the increase in private demand is large enough. Thus, the marginal ben-
efit of raising the quality of care is higher than before, and nursing homes will raise that
quality to attract private demand.
Medicaid can also influence the proportion of Medicaid residents directly, by mak-
ing the proportion another instrument of public policy instead of indirectly through
certificate-of-need regulation. For mathematical ease the policy variable in this model
is the number of Medicaid residents, not the proportion. Suppose that a binding rule on
the maximum number of Medicaid residents is relaxed slightly, but there is still excess
Medicaid demand. Then the number of private residents is simply the total number of
beds less the number of Medicaid residents. The nursing home will offset the increase in
Medicaid residents one-for-one with a decrease in private residents. It is not necessary
to solve a system of equations to find the effect of a change in the number of Medicaid
residents on private price and quality of care when the constraint is binding.
dp dp
- d > 0, (3.12)
dm dx
dq - dq 0. (3.13)
dm dx
These findings show that if a binding constraint on the number, or proportion, of Med-
icaid residents is relaxed, the private price increases and the quality of care decreases.
Quality of care could be increased, therefore, by limiting the number or proportion of
Medicaid residents in a nursing home. The idea is to raise quality by exposing public
residents to as many private residents as possible. Although the government has not re-
stricted the number of Medicaid residents in nursing homes, it has used this approach
to improve quality in two other markets. The Veterans' Administration requires its hos-
pitals to be affiliated with a teaching hospital, and Health Maintenance Organizations
were limited in the number of Medicare patients they can accept (until this was repealed
in the recent Balanced Budget Agreement).
972 E.C. Norton
Although most states reimburse nursing homes in a way that does not provide in-
centives to provide high-quality care or care for more disabled residents, a better reim-
bursement system would provide incentives for higher quality and improved outcomes.
One method for doing so was tested in a controlled experiment in San Diego, Cali-
fornia, that showed that changing the basis of reimbursement could improve the level
of care and admission mix with little change in expenditures [Norton (1992a), Norton
(1992b)]. Eighteen nursing homes in the experimental group were paid on a schedule
that depended on the classification of residents into one of nine health status categories.
Additionally, reimbursement was linked to outcomes, as proposed by Kane and col-
leagues [Kane (1986), Kane and Kane (1988)]. When contrasted with a control group of
eighteen nursing homes, the experimental nursing homes admitted sicker residents, who
were then more likely to improve their health and return home. Contracts with prices
contingent on resident characteristics may lead nursing homes to misrepresent informa-
tion, requiring regulation and appropriate contractual incentives [Forder (1997)].
Cohen and Spector (1996) found empirical evidence that the reimbursement type and
level affects the quality of nursing home care. They used a nationally representative
sample of nursing home residents from the National Medical Expenditure Survey Insti-
tutional Population Component to study how Medicaid reimbursement type and level
affects quality of care, measured as the case-mix adjusted staff to resident ratio, and
how quality affects resident outcomes, measured as mortality and bedsores. They found
that reimbursement affects staffing, and staffing affects outcomes, although they found
no direct effect of reimbursement on outcomes.
3.4. Models of supply of informal care
Informal care is critical for persons who need long-term care but are not in a nursing
home or other residential setting. The models of supply of informal care involve trade-
offs between work, leisure, and the supply of informal care [Chang and White-Means
(1995), Stern (1995), Nocera and Zweifel (1996), Sloan, Hoerger, and Picone (1996),
Sloan, Picone, and Hoerger (1997)]. The following discussion is an abbreviated presen-
tation of the model developed by Nocera and Zweifel, which addresses the same issues
as in other papers on supply of informal care, but in a more formal context. They derive
the reservation caregiver's wage as a function primarily of the wage rate, the marginal
rates of substitution between consumption, leisure, and caregiving, and the productivity
of caregiving.
Nocera and Zweifel (1996) assume that the potential caregiver determines the level of
informal care, not the elderly parent. Because informal caregivers are usually not paid,
they must derive utility from their caregiving or would provide no care. The caregiver's
utility therefore depends on consumption services C, leisure L, and informal care Z.
Consumption services depend on both leisure and consumption goods X. A person with
a high value of Cx is productive in his or her use of consumption goods. The price p of
consumption goods is normalized to one. Marginal utility is increasing in consumption
and leisure, but may be increasing or decreasing in informal care. The caregiver's utility
is:
U = U(C(L, X), L, Z). (3.14)
The total time T may be spent on labor, leisure, and caregiving. The caregiver may
devote time A to informal care, but cannot purchase formal care for her parent. Total
income Y is derived from wage earnings w and a compensating lump sum payment for
informal care M. The total income is:
Y = w(T - L- A) + M. (3.15)
Finally, the amount of caregiving Z is a function only of the time devoted to caregiv-
ing A. However, some persons are more efficient at caregiving, so that ZA may be large
or small.
Nocera and Zweifel solve their model to find the reservation wage dM/dA for care-
giving. Although they assume that the lump sum payment M is small relative to in-
come, making changes in hours worked negligible (dL/dA = -1), that assumption is
not made here. Much of the empirical work in the United States is aimed at determin-
ing the effect of caregiving on labor force participation, not just the reservation wage,
assuming no crowding out. Comparative statics yield the following expression for the
reservation wage:
dM UZZA dL UL (3.16)
dA UCCx dA UcCx (3.16)
The reservation wage equals the wage rate less adjustments for other factors. The
reservation wage is generally positive. A sufficient condition for a positive reservation
wage is that marginal utility from caregiving be nonpositive, Uz < 0, the other three
derivatives in the second term being positive. The third term is always negative because
dL/dA is negative and the derivatives in parentheses are positive, so subtracting a nega-
tive increases the reservation wage. Only a person who derives an extraordinary amount
of utility from caregiving would not require a reservation wage.
The reservation wage therefore depends on the wage rate, the marginal rate of substi-
tution between consumption and caregiving, the marginal rate of substitution between
consumption and leisure, the productivity of caregiving, the productivity of goods in
consumption, and the marginal effect of time spent caregiving on leisure time. The
reservation wage will be higher for a person who has a high reservation wage for work,
measured by the marginal rate of substitution between consumption and leisure. Produc-
tive caregivers who have large ZA require a lower reservation wage than less productive
caregivers. This model does not allow for the substitution of time spent working for
time spent providing informal care (Nocera and Zweifel assume that dL/dA is negative
one). However, it does show what factors determine the opportunity cost of providing
care, and could be modified to allow for changing labor force participation.
974 E.C.Norton
Because few informal caregivers are paid explicitly, the empirical work has focused
on three related questions. First, are bequests used to elicit caregiving? If caregivers
are not paid while they work, then perhaps the bequest is used as payment after the
elderly parent dies. This question is addressed further in Section 4.2 because the bequest
literature is closely related to the purchase of long-term care insurance.
Second, does the market wage affect the amount of caregiving? The theoretical model
above implies that higher paid persons have a higher reservation wage, and would be less
likely to provide informal care for free. The empirical evidence is mixed. In a study of
Swiss households, Nocera and Zweifel (1996) found little relationship between market
wages and the reservation wage of caregiving. In an American study, caregivers with
high wages, estimated in a standard wage equation, are more likely to be employed and
work longer hours than caregivers with low wages [White-Means and Chollet (1996)].
Third, does providing informal care reduce hours worked in the labor force? This
question is difficult to answer both because the decisions to provide informal care and
to work in the labor force are endogenous, and because a large sample of potential
caregivers is rare. Three studies found a negative effect of caregiving on hours worked
after controlling for endogeneity in samples not restricted to caregivers [Stern (1990),
Ettner (1995), Wolf and Soldo (1995)]. However, the estimated effect size and level of
significance varied considerably, in part due to the different methods of controlling for
endogeneity. In an alternative approach, Stem (1995) estimated informal caregiving as a
function of working in the labor force. When he controlled for the endogeneity of work
status, Stem found work status to have a statistically insignificant effect on caregiving.
However, his instrumental variables were lagged endogenous variables, which do not
become exogenous when decisions about future caregiving are made far in advance.
Although conceptually it is easier to model a working woman's decision to provide
care or not, empirically it is easier to identify and survey a group of caregivers and then
estimate the probability of working. Many factors determine labor force participation
by caregivers. Chang and White-Means (1995) found that caregivers in good health,
with higher education, and not living with an elder are more likely to work in the labor
force. Surprisingly, they also found that gender has no effect, and those with higher
wages (estimated by a standard wage equation) are less likely to work. The number
of hours worked in the labor force also depends negatively on wages and nonwage
income. This result suggests that income effects are quite strong. Working in the labor
force reduces informal caregiving. Boaz (1996) found that among caregivers, those who
worked full-time provided between 22 and 25 fewer hours of care per week. The fraction
of caregivers who worked full-time ranged from 15 to 19 percent. Neither result changed
significantly from 1982 to 1989, indicating small changes in the supply of informal
caregivers during a rapid rise in the labor force participation of women.
Publicly provided home care may decrease the supply of informal care. Formal care
is a close substitute for informal care. If the public sector expands home health care to
assist elderly in the community, that formal care may crowd out informal care given by
family or friends. The availability of public formal care is a moral hazard problem for
families of frail elderly. The few studies of this issue have found that formal care does
substitute for informal care, but at a much lower rate than one-for-one [Christianson
(1988), Moscovice, Davidson, and McCaffrey (1988), Pezzin, Kemper, and Reschovsky
(1996)].
4. Demand for long-term care
4.1. Empirical evidence of demandfactors
The theory of demand for long-term care is straightforward. The most important factors
are health status, which determines need and the out-of-pocket price relative to the price
of close substitutes. Those in worse health demand more long-term care. Those with
fewer substitutes, or whose substitutes are higher-priced, demand more long-term care.
Demand curves slope downward, and health shocks shift the demand curve outward.
This section summarizes the empirical studies of long-term care demand. As usual,
these studies focus on the demand for nursing home care as opposed to other forms
of long-term care. Although these studies are motivated by trying to understand what
affects demand, few empirically try to separate demand from supply. Therefore, any
issues of differential access are ignored in the empirical work, and what is interpreted as
demand is actually a reduced form combination of demand and supply. One exception
is the study of demand by Reschovsky (1996) that used bivariate probit models with
partial observability.
Empirical research on demand has looked at cross-sectional studies of demand for
nursing home care [e.g., Headen (1993), Bauer (1996), Reschovsky (1996)], longitu-
dinal studies of care over a lifetime [e.g., Murtaugh et al. (1997), Murtaugh, Kemper,
and Spillman (1990)], and length of stay once a person is in the nursing home [e.g.,
Garber and MaCurdy (1993), Morris, Norton, and Zhou (1994)]. The studies generally
find that similar factors affect the probability of any nursing home use, length of stay,
and lifetime use.
The primary determinant of demand for nursing home care is health status - both
physical and mental health. Persons in worse health status are more likely to go to a
nursing home. As physical or mental health deteriorates, a person is less able to care
for himself or herself, and less able to perform basic activities that most persons take
for granted. For nursing home care, the best measure of health status is a measure of
basic function called activities of daily living. This scale measures whether a person
needs help with each of six activities - eating, bathing, toileting, transferring into and
out of bed, dressing, and continence [Katz et al. (1963)]. The exact definition of ADLs
varies slightly across surveys. ADLs are an excellent determinant of demand. Those
with more difficulties with ADLs are more likely to need nursing home care [Garber
and MaCurdy (1990), Liu, McBride, and Coughlin (1994), Hoerger, Picone, and Sloan
(1996), Reschovsky (1996)]. However, the length of stay may be shorter for persons
with more difficulties with ADLs because of a higher mortality rate [Morris, Norton,
and Zhou (1994)]. Some residents go to a nursing home to recover from an acute care
976 E.C. Norton
episode in the hospital and are expected to be discharged home after a relatively brief
recovery period. These short-stay residents are often covered by Medicare. Residents
with cancer, fractures, or stroke are more likely to be short-stay residents [Liu, McBride,
and Coughlin (1994)]. Specific medical conditions associated with nursing home care
are chronic conditions such as Alzheimer's disease [Bauer (1996)].
Demand for long-term care is also related to other demographic characteristics, such
as age, gender, and race, but probably because these variables are proxies for health
status. Age is correlated with demand for nursing home care because age is negatively
correlated with health status and because older persons have a longer exposure period
[Murtaugh, Kemper, and Spillman (1990)]. Gender is related to nursing home use, but
much of the effect of gender is due to health status and marital status. Men have worse
health status at admission than women because women tend to outlive their husbands.
Men are typically able to receive more care at home, and thus can avoid going to a
nursing home for longer.
Race is significant in nearly every empirical study of nursing home use. Whites are
more likely to use nursing home care than blacks, Hispanics, or Asians. Blacks are more
likely than whites to be on Medicaid, have severe illness, and not have long-term care in-
surance coverage - all factors that hinder admission to a for-profit nursing home [White-
Means (1997)]. Differences persist in empirical work even after controlling for observ-
able differences in insurance and health status. The difference in nursing home use may
be related to cultural differences in preference for location of care, differences in health
status, or to differences in access due to racial discrimination [Headen (1992)]. Race
encompasses social, psychological, biological, and genetic influences [White-Means
(1995)]. Race therefore pervades socioeconomic status, attitudes, and family culture,
implying that empirical work should include not merely a dummy variable for race but
a fully-interacted model. The effect of race may also be related to the opportunity cost
of informal care and nursing home care [Headen (1992)]. For example, if the wage rates
of blacks are lower than whites, and the nursing home price is the same, then the oppor-
tunity cost of informal care is lower for blacks. Headen (1992) found evidence that the
opportunity cost of time - measured by labor force participation, education, age, and
social support - is lower for black informal caregivers than white informal caregivers.
The financial determinants of nursing home demand are the price, the relative price
of close substitutes, and the person's income and assets. Clearly nursing home demand
will increase when the price falls, or when the price of close substitutes rises. Private in-
surance lowers the out-of-pocket cost of nursing home care, but few elderly have private
insurance, and those who do may still face substantial copayments and deductibles. In-
come and assets do not affect nursing home demand in a straightforward way. A poor el-
derly person who receives Supplemental Security Income (SSI) is eligible for Medicaid.
In contrast, an elderly woman of modest means would have to pay for care privately un-
til her wealth dropped below the eligibility threshold. Then she would contribute nearly
all her income as copayment, with Medicaid paying the remainder. The copayment and
deductibles increase nearly one for one at the margin with income and assets. Therefore,
wealthier persons may demand less nursing home care than poor persons because of the
out-of-pocket cost. Although there is some variation across states in the income and
asset rules, there is even more variation within states among elderly persons in income
and assets.
A few studies of nursing home utilization include economic variables such as income
and wealth or the price of close substitutes [for a review of the economic status of the el-
derly, see Hurd (1990)]. Headen (1992) found that persons with higher housing wealth,
a pension income, and rental income were less likely to enter a nursing home. He esti-
mated a negative price elasticity. Bbrsch-Supan and colleagues (1992) also found that
those with higher incomes were less likely to go to a nursing home, possibly because the
income can be used to purchase ambulatory care or to make transfer payments to chil-
dren. Garber and MaCurdy (1990) showed that owning a home decreases the probability
of going to a nursing home, but income was not an important factor. Hoerger, Picone,
and Sloan (1996) found that public subsidies through Medicaid affected the choice of
living arrangement. They used a multinomial probit model of choice between living
independently, with others, or in a nursing home. Elderly persons were more likely to
enter a nursing home when Medicaid eligibility criteria were less restrictive and when
Medicaid reimbursement rates were more generous. Raising subsidies for home health
services increased the probability of living independently relative to living with oth-
ers, but had no effect on the probability of institutionalization, even reducing nursing
home use is often cited as the goal of expanding home health services. Reschovsky
(1998) found in a national sample of disabled elderly that demand is more elastic with
respect to price, income, and wealth among married persons compared to nonmarried
persons, and also for less disabled compared to more disabled. In contrast, using data
from the Chanelling demonstration Reschovsky (1996) found almost no association be-
tween economic variables - income, wealth, and nursing home price - and demand.
Medicare policy is another source of variation in price, because of the increasing co-
payment required over time. Garber and MaCurdy (1993) exploited temporal variation
in Medicare reimbursement to determine the hazard rate of discharges from nursing
homes. Medicare pays the full cost of nursing home care for only the first 20 days,
following an inpatient stay. Then from day 21 through 100, Medicare requires a sub-
stantial copayment, which may be mitigated by private Medigap insurance. Medicare
ceases payment after 100 days, as do many Medigap policies. Garber and MaCurdy
used semiparametric duration models to estimate a flexible discharge hazard function
for residents covered by Medicare for at least part of their stay. They found large peaks
in the hazard rate for Medicare residents at about 20 and 100 days, with the latter hazard
rate being about four times the hazard rate before and after 100 days. They concluded
that nursing home use is price sensitive.
The other major financial determinant of nursing home care is the availability of close
substitutes. For example, informal care, adult day care, and board and care homes are
all substitutes for institutional care. Therefore, married persons are much less likely to
go to a nursing home than unmarried persons because a married person receives care
from a spouse [e.g., Murtaugh et al. (1997)]. Married persons are also more likely to
have children, another important source of informal care. Married persons have shorter
978 E.C. Norton
lengths of stay, conditional on admission, either because of the ability to return home, or
because of the higher risk of mortality because of entering in worse health status [Garber
and MaCurdy (1990), Morris, Norton, and Zhou (1994)]. Surprisingly, Bauer (1996)
found little evidence that weak support systems, measured primarily as availability of
informal care, were associated with risk of nursing home entry for persons who received
home- and community-based care in a capitated long-term care system.
Finally, a strong bequest motive may also influence the demand for long-term care
[for literature on bequests, see Hurd (1987, 1989)]. If a parent uses the bequest strate-
gically to elicit care from children, then demand for formal long-term care would be
lower. A parent who prefers care from children to care from a nursing home could use
the promise of a bequest to induce children to visit, call, and provide help. When the
child responds with the desired care, demand for nursing home care is reduced. The em-
pirical evidence for the bequest motive is discussed more in Section 4.2 on long-term
care insurance.
4.2. Models to explain why private insurance is rarely purchased
A risk-averse person facing an uncertain and expensive risk of needing long-term care
should demand insurance. The risk of needing nursing home care is high, as explained
in Section 2.1. The cost of care often exceeds $40,000 per year. Medicare and Medicaid
pay for only about half of all expenses. Rice (1989) found that 82.5 percent of out-of-
pocket medical expenses paid by elderly with at least $3,000 in expenses went towards
nursing home care. Therefore the elderly face substantial financial risk of long-term
care expenses. In addition to reducing financial risk, the desire to leave a bequest to
spouse and children may be a major motive for purchasing long-term care insurance
[Bernheim, Shleifer, and Summers (1985), Bernheim (1991)].
Yet, only a few percent of elderly have purchased private long-term care insurance in
the United States. Private long-term care insurance was first sold in the 1970s, but was
not sold aggressively until the mid- 1980s when many states encouraged it as an alterna-
five to public insurance [Somers and Merrill (1991), Cutler (1996)]. In 1993 there were
only about 3.4 million private policies sold. This accounts for five to six percent of the
elderly, with private insurance paying for about two to three percent of nursing home ex-
penses [Norton and Newhouse (1994), Murtaugh, Kemper, and Spillman (1995), Cohen
and Kumar (1997)].
The reasons for the lack of private long-term care insurance has been the subject of
much theoretical and empirical research. The reasons studied include not only the usual
problems of adverse selection and moral hazard, but also Medicaid crowding out, high
administrative costs, and nondiversifiable intertemporal risk.
Adverse selection makes insurance most attractive to persons most likely to use it. Se-
lection is more serious in marketing private insurance to elderly than nonelderly because
the elderly are more heterogeneous in their risk and more likely to know something
about their risk. Insurance companies have necessarily developed techniques to screen
bad risks; they typically deny ten to twenty percent of elderly applicants. Applications
almost always ask about prior health service use, physical and mental functioning, and
serious medical conditions, but not lifestyle. One study estimated that between 12 and
23 percent of 65-year olds would be rejected for private insurance due to health prob-
lems, with these numbers rising to 20 to 31 percent at age 75 [Murtaugh, Kemper, and
Spillman (1995)]. One study found that an individual's expectation of being in a nurs-
ing home was highly positively correlated with purchasing long-term care insurance,
even after controlling for observable expenditure risks such as health status [Sloan and
Norton (1997)]. This result held for both elderly and near elderly, and implies that se-
lection is a problem. Among the nonelderly, by contrast, the problem of selection does
not seem nearly as serious, because individuals are unlikely to know whether they are
more or less likely than others to be admitted to a nursing home later in life.
Insurers combat moral hazard in part by requiring deductibles of 20 to 30 days for
institutional care, and of 90 to 100 days for home health care [Cutler (1996)]. In addi-
tion to the usual problem of moral hazard, Pauly (1990) suggested a particular form of
moral hazard that may apply to long-term care insurance. The desire to leave a bequest
larger than one would otherwise give should increase demand for insurance [Bernheim,
Shleifer, and Summers (1985), Bernheim (1991), Pauly (1996)]. However, the elderly
may fear that if they purchase insurance children may institutionalize them when they
are unable to act on their own. Insurance may provide this incentive because it reduces
the price to the children of institutionalizing a parent, relative to the cost of giving per-
sonal attention. An elderly person who prefers attention from children over purchased
help may therefore decide not to buy long-term care insurance. Zweifel and Str0we
(1996a, 1996b) extend this analysis to simultaneous decisions to purchase long-term
care insurance and the terms according to which bequests are structured to provide an
incentive for children to care for their elderly parents. If long-term care insurance is
purchased, it is necessary for parents to increase the sensitivity of the bequest to care-
giving in order to elicit attention from children. They concluded that with the bequest
as an instrument to elicit caregiving, demand for private long-term care insurance will
be very limited.
The availability of Medicaid as a payer of last resort may crowd out demand for
private long-term care insurance. Medicaid was designed to provide coverage only for
the very poor. To qualify, one must have virtually no nonhousing wealth. Medicaid is a
low-priced substitute for private long-term care insurance. In particular, like private in-
surance, Medicaid reduces the expenditure risk that relatives of nursing home residents
would otherwise bear. But Medicaid is not good at preserving assets which the person
would want upon leaving the nursing home. In this sense Medicaid and private long-
term care insurance are imperfect substitutes. The price of Medicaid coverage depends
on a person's income and wealth. The deductible equals the person's wealth minus a
small amount that Medicaid programs allow recipients to keep. The copayment is the
person's income minus allowances for personal needs and home maintenance. In some
states, however, individuals with more than $1020 of income per month are not eligible
for nursing home benefits through Medicaid irrespective of their medical expenditure.
To combat spousal impoverishment, the 1989 Medicare Catastrophic Coverage Act in-
980 E.C. Norton
creased substantially the amount of income and assets the community spouse could
retain [Norton and Kumar (2000)]. Therefore, Medicaid long-term care insurance is
more expensive for the affluent and unmarried. It is plausible that Medicaid discourages
private long-term care insurance purchases, especially by less affluent individuals who
have few nonhousing assets to protect. Individuals may avoid this tax on assets by trans-
ferring assets to others or setting up irrevocable trusts. If the transfer is made far enough
in advance of admission to a nursing home, now 30 months, the transfer is excluded
from countable assets by Medicaid.
An additional reason for the lack of private long-term care insurance sold is the high
administrative cost. The price of insurance is the expected benefit plus the loading fac-
tor, which may be substantial. The administrative load is typically from half to two-
thirds of the total cost [Cutler (1996)]. Ninety-seven percent of private long-term care
insurance is sold to individuals as opposed to groups, and selling to an individual is far
more expensive. Purchasing a group policy, say through employment, would save ad-
ministrative costs. However, few companies offer long-term care insurance as a fringe
benefit to workers. For acute care health insurance, the contract is for care rendered
while the employee is employed. For long-term care insurance, the contract must nec-
essarily entail financial obligations long after the person has left the employer. Employ-
ers are likely to be more adverse to providing long-term care insurance than pensions
because, for the former, there are likely to be demands that benefits be tied to increases
in the price of long-term care services.
A final reason for the lack of private insurance relates to the long time period over
which coverage is typically purchased [Cutler (1996)]. Private insurance bought by
someone age 65 may not pay benefits for twenty years or more. Whereas risk for acute
care depends on cross-sectional heterogeneity within a single year, risk for long-term
care depends in part on changes in the average cost over time. This risk is common to
everyone in the cohort and cannot be diversified. If the logarithm of costs in year t + 1
is serially correlated with the logarithm of costs in year t, then the variance of the log-
arithm of future costs rises with the correlation. Suppose that Ct+l = pct + t/t+l, then
the variance after k periods is:
k-I -
var(c) = (ph)] a2 (4.1)
which increases as p increases from zero. There are two implications from this simple
model. First, insurers will offer indemnity benefits instead of service benefits. Indemnity
benefits, which pay a fixed sum, place more risk on the consumer than service benefits,
which pay the full cost of care. A policy that offers a maximum daily benefit of $100
may seem reasonable by today's prices, but will be inadequate in 30 years. Cutler found
that out of a sample of 73 policies investigated in 1991, only one was a service plan
and it went out of business by the following year. Second, insurers will require a higher
rate of return on long-term care insurance than less risky investments. A higher rate
of return raises the premium. Cutler found evidence that both indemnity benefits and a
higher rate of return will reduce demand.
In summary, there are many economic reasons for why persons do not purchase pri-
vate long-term care insurance. When combined, adverse selection, moral hazard, Med-
icaid crowding out, high administrative costs, and nondiversifiable intertemporal risk
may present an overwhelming case. Additional demand-side reasons may also be im-
portant. Elderly may systematically underestimate the probability of needing long-term
care, reducing their demand although one study found no evidence to support this [Lind-
rooth, Hoerger, and Norton (1999)]. Also, long-term care insurance, unlike other health
insurance, is not subsidized by federal tax breaks. Researchers should perhaps instead
ask why anyone would buy long-term care insurance.
4.3. Models of how spend-down affects savings
A means-tested insurance program such as Medicaid may do more than affect the de-
mand for private insurance. It may also affect financial decisions related to Medicaid el-
igibility. Since Medicaid requires beneficiaries to have almost no nonhousing financial
assets, economists have explored how Medicaid may affect savings, trusts, and housing
stock.
Recent theoretical work has shown that the Medicaid program could have substan-
tial negative effects on personal savings of the elderly [Hubbard, Skinner, and Zeldes
(1995), Sloan, Hoerger, and Picone (1996)]. Since Medicaid pays for nursing home care
for elderly with low savings, there is a disincentive to save. Medicaid policy thus creates
a moral hazard problem for saving. Hubbard, Skinner, and Zeldes (1995) develop a dy-
namic model of consumption and savings to show the effect of Medicaid. The household
maximizes expected future utility U(C) which depends on consumption C:
r DsU(Cs) \
max E (1 + 6)-t (4.2)
where D is a state variable that is one if the person is alive and zero otherwise, and future
utility is discounted by a rate of time preference . Each period the household receives
income, consumes, and may save for the future. Medical expenses M may be incurred.
These medical expenses are not insurable, except through the means-tested program.
Therefore, they assume that there is no market for private long-term care insurance.
Most importantly, the household receives transfer payments if its assets are below a
threshold. The threshold C sets a floor on consumption after medical expenses. A per-
son with very low income, or large medical expenses relative to income, will receive a
transfer payment that raises consumption to C. Hubbard, Skinner, and Zeldes show in a
two-period model that the usual tradeoff between consumption now and later depending
on the interest rate only holds if there is no means-tested transfer payment TR in the sec-
982 E.C. Norton
ond period. If the household receives a transfer, the marginal change in second-period
consumption with first-period consumption is zero.
dC2 0 if TR2 > 0,

(4.3)
dC 1 -( + r) otherwise.
The budget constraint is not a straight line, but is kinked. At low levels of wealth
the household will consume all wealth, save nothing, and depend on transfer payments.
As wealth rises, eventually a critical level of wealth is hit where consumption falls,
savings rise, and transfer payments cease. Hubbard, Skinner, and Zeldes therefore find
the unusual result that the marginal rate of consumption with respect to wealth may
decrease over certain ranges of wealth.
They extend their basic model to include uncertain earnings and many periods, al-
though the intuition remains. A means-tested insurance program discourages savings
for those households with few assets and near the eligibility threshold because of the
large implicit tax rate on savings. In their dynamic model, households may separate
into households that save and those that do not. The effect of a means-tested program is
weak for households with high wealth because the consumption floor is a smaller per-
centage of income, and the uninsured risk of medical spending is a smaller percentage
of wealth. Hubbard, Skinner, and Zeldes find that their model is in concordance with
observed patterns of lifetime household saving and wealth. Gruber and Yelowitz (1999)
found evidence supporting the hypothesis that Medicaid reduces savings in two data
sets from 1983-1993. Households on Medicaid reduce savings more if there is an asset
test, and they estimate that Medicaid expansions reduced wealth holdings over the study
period by 8.2 percent.
Researchers have looked for empirical evidence of spend-down behavior over shorter
time intervals, usually after a person enters the nursing home. Thus, this behavior is
short-term rather than over the entire life-cycle. Empirical research has not found spend-
down to be rampant during the shorter time scale of being in a nursing home. The styl-
ized fact is that very few persons spend-down to Medicaid eligibility after entering the
nursing home. Although roughly 40 percent of new admissions are covered by Medi-
caid, and there is a perception in the popular press that spend-down is widespread, less
than 20 percent of persons who are private-pay at admission actually spend-down after
admission [Rice (1989), Mor, Intrator, and Laliberte (1993), Norton (1995), Spillman
and Kemper (1995)].
Norton (1995) found that instead of Medicaid long-term care insurance reducing sav-
ings, the opposite may happen through welfare aversion [Moffitt (1983)]. Norton used
data from two different samples of the elderly to predict the distribution of time until
spend-down according to a model of spend-down absent of behavioral effects. These
distributions were then compared that to the actual distribution of the time until spend-
down for nursing home residents. Contrary to expectations, it appears that the elderly
avoid Medicaid eligibility. This result cannot be explained away by sample selection,
demographics, or uncertainty about prices. It is possible that the elderly receive transfers
from family members in order to avoid Medicaid, but this result remains unexplained.
5. Demographics
5.1. Demographicstrends
Over the next several decades three important demographic trends may profoundly
change the long-term care market in every country. First, the baby boom generation
is approaching retirement and old age. As a result, the sheer number of elderly will in-
crease dramatically. Second, mortality rates have fallen. The higher life expectancy for
both men and women is another reason that the absolute number of elderly is increas-
ing. Third, fertility rates have also fallen. The lower number of young persons implies
that the support ratio, the number of working-age adults to children and elderly, will
fall. The changes in mortality and fertility rates started many years ago, and they will
accentuate the increase in elderly due to the post-war baby boom generation. Therefore,
both the number and proportion of elderly is expected to rise rapidly in all countries.
In particular, the number and percentage of the oldest old, age 85 and older, will in-
crease rapidly, forcing the public and private sectors to question whether the supply of
long-term care will be adequate to meet the demand.
The percentage of the population over age 65 and over age 80 will increase several
fold between now and 2050, according to demographic projections for seven industrial-
ized countries (see Table 2). While the percentage of the population in these countries
over age 65 will be between 12 percent and 18 percent in the year 2000, those numbers
are expected to grow to between 20 percent and 36 percent in fifty years. The percent-
age of those age 80 and older are expected to increase even more rapidly. In these seven
countries the percentage will grow from less than 4 percent to at least 7 percent and as
high as 14 percent. Germany, Italy, and Japan will see the largest percentages of elderly,
while Canada and the United States will have the least, among these seven countries.
These demographic trends are projected to be qualitatively similar in all other industri-
alized nations. The numbers in the table are the most recent United Nations projections,
and are higher than earlier projections because of revised estimates of falling birth rates.
Table 2
Population projections as a percentage of total population, by age and year
2000 2025 2050

65+ 80+ 65+ 80+ 65+ 80+
Canada 12.6 2.9 20.5 4.6 24.5 9.0
France 16.2 4.0 22.5 5.9 26.4 9.8
Germany 15.9 3.4 21.8 5.7 29.2 10.5
Italy 17.7 3.6 25.6 7.4 35.7 13.9
Japan 16.5 3.3 25.9 7.7 30.4 10.8
United Kingdom 15.8 4.0 20.3 5.0 23.2 8.2
United States 12.4 3.2 18.3 4.0 21.2 7.3
Table lists the projected percentage of the population that will be age 65 and
above or age 80 and above. Based on data from United Nations (1997).
984 E.C. Norton
The demographic trends raise concerns about the ability to care for elderly. How
much care will the elderly demand? Who will provide the care? What will it cost?
Who will pay? Projections based on models of current use paint a bleak picture. The
longer life expectancy and post-war population bulge will raise both demand for long-
term care and its price. The reduced support ratio will increase the tax burden on the
nonelderly. The total cost to governments, which finance much of the formal long-term
care, will also rise. Many articles and books have sounded the warning and proposed
creative solutions to the impending financial and social crisis [e.g., Rivlin and Wiener
(1988), for the United States, and Doty (1993), for a discussion of proposals in many
industrialized countries].
5.2. Why demographic changesmay not exacerbateproblems
Several recent economic studies have explored the economic implications of an aging
population from both a theoretical and empirical viewpoint, and have reached surprising
conclusions [Lubitz, Beebe, and Baker (1995), Zweifel, Felder, and Meier (1996), Lee
and Skinner (1997), Lakdawalla and Philipson (1998)]. Although conventional wisdom
says that the three demographic trends will greatly increase the burden of paying for
health care in the future, these studies argue that the burden will not be as great as feared.
Health care expenditures will definitely increase because of the increased number of
baby boomers. These articles argue, however, that increased longevity by itself is not
likely to contribute substantially to increased health expenditures.
The extent of the problem depends on how longevity is associated with health care ex-
penditures. The relationship may be different for acute care and for long-term care, and
so are discussed separately. For acute care, if health care expenditures depend primarily
on age, with higher expenditures for older persons, then increased longevity necessarily
will increase per person expenditures. If health care expenditures depend primarily on
time until death, with higher expenditures prior to death, then increased longevity will
not increase per person expenditures, but will push expenditures further into the future.
More likely, health care expenditures depend on both age and time until death, with
expenditures before death and the composition of expenditures changing with age.
Health care expenditures increase as death approaches. In a study of a cohort of Medi-
care enrollees from 1976 through 1978, Lubitz and Prihoda (1984) found that only 6
percent of the enrollees died in 1978 but their Medicare expenditures during their last
year of life accounted for 28 percent of total Medicare expenditures over a one-year
period for the cohort. The largest component (77 percent) of their Medicare expenses
were for hospital care. Similarly, using data from the 1980 National Medical Care Uti-
lization and Expenditure Survey, Kovar (1986) found that among people age 65 or older
who lived in the community at the beginning of the year, only 5 percent died or were
institutionalized during the year, but that they accounted for 22 percent of total medical
care expenditures during that year. McCall (1984) showed that over 60 percent of the
Medicare expenditures during the last year of life occurred during the last three months
of life as use of hospital services intensified, and Lubitz and Riley (1993) found that
Ch. 17. Long-Tern Care 985
40 percent of Medicare payments during the last year of life occurred during the last
30 days. Using data on a sample of decedents from the Longitudinal Study of Aging,
Stearns and colleagues (1996) showed that almost three-quarters of Medicare beneficia-
ries age 70 and older were hospitalized during that year.
Higher expenditures prior to death says little about how expenditures depend on
longevity. The concern about future health care costs for the elderly is based on the
empirical finding that per capita health expenditures are higher for older persons than
younger persons. An increase in the number of elderly, so the argument goes, should
raise costs based on the results of these cross-sectional studies.
A careful review of the indirect evidence suggests a different conclusion. Newhouse
(1992) calculated that the changing proportion of age groups was responsible for only
about a 15 percent rise in real medical expenditures per capita from 1950 through 1987,
whereas real expenditures increased by more than a factor of 5. That empirical find-
ing, however, is based on age-specific expenditures. Per-capita health expenditures are
higher for 85-year olds than 65-year olds because mortality increases with age, and
health expenditures are highest near the end of life. Therefore, predictions based on
current age-specific expenditure rates may be biased if mortality rates are falling [Fuchs
(1990)]. One way around this problem is to analyze cohort data and see whether age
matters in a regression of expenditures controlling for time until death. If age positively
affects costs after controlling for time until death, then the aging of the population will
increase costs, assuming that changes in technology are not related to age. Alternatively,
if aging does not affect costs then aging per se does not drive health care expenditures.
The direct empirical evidence is limited. Lubitz, Beebe, and Baker (1995) analyzed
17 years of Medicare claims data to estimate lifetime Medicare payments as a func-
tion of age and time until death. They found that expenditures rise rapidly in the three
years prior to death, but less so for older persons. Lifetime Medicare expenditures rise
with age, but at a diminishing rate. They conclude that increased longevity may not in-
crease Medicare spending. Their data were limited to Medicare claims, and so excluded
HMO enrollees, all out-of-pocket payments, and most nursing home expenses. Steams
and colleagues (1996) found that the likelihood of non-terminal hospital use increased
with age until age 81 but then decreased. Once hospitalized, the total number of nights
hospitals was not significantly associated with age. Zweifel, Felder, and Meier (1996)
analyzed individual data from two cohorts of Swiss elderly, and found that in most spec-
ifications age did not matter, after controlling for time until death. What matters is time
until death. They conclude that health care expenditures depend far more on time until
death than calendar age, and so the aging of the population will have a negligible ef-
fect on health expenditures. The work of Zweifel, Felder, and Meier has a very limited
sample size (N < 400), did not include long-term care expenses, and does not estimate
interactions between age and time until death. Their measure of health expenditures
included hospital, physician, prescription drug, and only part of long-term care.
Scitovsky (1988) used data from patients at the Palo Alto Medical Clinic to explore
whether people with poorer functional status receive fewer hospital services during the
last year of life. Scitovsky found high levels of long term care expenditures relative to
986 E.C. Norton
acute care expenditures for the frail elderly and functionally impaired. Scitovsky con-
cluded that sophisticated life-extending hospital services may already be allocated in
a more rational manner than might generally be assumed; her results were consistent
with a situation in which age and functional status are taken into account, implying that
implicit rationing may be occurring and that concerns about excessive use of hospi-
tal services by persons with poor prognoses may be unfounded. However, Scitovsky's
analysis is limited to patients at a single medical clinic, and she used a measure of func-
tional impairment based in part on the duration of impairment during the last year of
life, which is problematic since functional status can change a great deal during the last
year of life.
For long-term care, which is a large fraction of health care expenses for the oldest old,
the time until death is less relevant because long-term care expenditures cannot be run
up quickly like acute care expenditures. Long-term care expenses depend not merely on
how long someone lives but the fraction of life spent in a healthy state compared to the
fraction spent in a frail state in need of long-term care, and the health of their spouse
[Manton, Stallard, and Corder (1998)]. If the additional years of life are spent in a frail
state with high demand for expensive care, then the dire predictions may come true.
Under this scenario each additional year of life is expensive. However, if elderly stay
healthy for longer but spend the same amount of time in a frail state as they do now,
then future costs will not increase nearly as much as feared. Under this scenario each
additional year of life is not expensive, although the expensive years are pushed further
into the future.
In terms of the macroeconomic effect of aging on long-term care expenditures, Lak-
dawalla and Philipson (1998) argue that increased life expectancy may actually reduce
the price of long-term care. Their theoretical model hinges on two things. First, it de-
pends on whether people who live longer spend more time in good health or poor health.
They assume that demand for long-term care depends on the number of persons in poor
health, while the supply depends on the number in good health, in a model with only
two health states. If people live longer and spend a higher fraction of time in good
health then the supply of long-term care increases faster than the demand, and price
falls. Long-term care is labor intensive, so its cost depends primarily on the supply of
labor, hence the quantity of healthy persons. The price is inversely related to the sup-
port ratio, defined as the ratio of number of persons in good health to the number of
persons in poor health. In addition, the change in total expenditures depends on how
demand responds to price. If demand is price elastic, then a change in longevity will
have a smaller effect on total expenditures than that predicted solely by the change in
demand because of price effects. These price effects may also affect public expenditures
differently than private expenditures. If the government offers a means-tested insurance
program for long-term care, such as Medicaid in the United States, then a price increase
raises a person's desire to spend-down, and public expenditures will increase.
Second, Lakdawalla and Philipson show that marital status is important because a
spouse is a close substitute for formal long-term care. If a frail person goes to a nursing
home only if they are single, but not if married, then aggregate long-term care expen-
Ch. 17: Long-TermCare 987
ditures will depend on the differential frailty and mortality rates of men and women.
They argue that demand for long-term care grew more slowly during the last 20 years
when male mortality rates decreased faster than female rates. When men live longer
they decrease the demand for long-term care from women, who are not widows as long.
The policy implication is that future long-term care expenditures depend on frailty.
The period of disability and chronic morbidity for the elderly has become more com-
pressed, resulting in more healthy years of life [Manton, Stallard, and Liu (1993)]. Other
analysis of longitudinal data on disability-free life expectancy shows that the decline in
mortality rate has led to an increase in light or moderate disabilities, but not severe dis-
abilities [Cambois and Robine (1996)]. Laditka (1998) used empirical estimates from
two national data sets of how nursing home demand depends on functional status to
simulate how improvements in morbidity would alter time spent in a nursing home.
She concluded that increased morbidity would increase life expectancy in such a way
that the fraction of spent in a nursing home compared to the community would remain
unchanged.
Given policymakers' great concern about future expenditures, much more research
is needed in this area. In particular, most empirical work has focused on acute medical
care, or total health expenditures, but not specifically on long-term care expenditures. It
is possible that long-term care expenses increase with age but acute medical expenses
do not. In summary, there is hope that additional years alive are spent healthy and that
the financial burden will be less than feared.
6. Conclusion
This chapter covers the major theoretical and empirical economic research in long-term
care. Past research has focused on answering questions related to the four fundamental
ways in which long-term care differs from acute care. Future research needs to address
the many remaining unanswered questions.
Long-term care is care for chronic illness or disability instead of treatment of an acute
illness. The market has created a wide spectrum of types of care in response to vary-
ing needs and ability to pay. Much long-term care is provided by informal caregivers
because of the expense of formal care and lack of complete insurance due to moral
hazard. Home care, board and care homes, and other intermediate forms of formal care
give elderly many choices. How will the market develop to provide coordinated care for
elderly whose health and needs change over the years?
The nursing home industry is dominated by for-profit facilities often facing excess de-
mand due to constrained bed capacity, in contrast to the hospital industry which is dom-
inated by nonprofit facilities with an excess supply of beds. We know that certificate-
of-need regulation has sometimes limited the supply of nursing home beds and reduces
access. How can nursing home market structure be changed to improve competition and
quality of care?
Long-term care is often provided by unpaid caregivers, instead of always being pro-
vided by paid professionals. Informal caregivers, often women, must trade off providing
988 E.C. Norton
care and labor force participation. We know that women reduce their labor force partic-
ipation in order to give informal care. How will informal care change as more women
with frail parents enter the labor force?
There is little private long-term care insurance and most public insurance is means-
tested with high copayments, compared with relatively comprehensive acute care insur-
ance coverage. We know that few elderly purchase long-term care insurance, despite
the enormous financial risk. What is the role of public insurance and regulation in the
financing of long-term care? We know that means-tested welfare programs can have
strong incentive effects. Whlat deleterious unintended effects will public long-term care
insurance have?
Finally, we know that the number of elderly will increase over the next several
decades. How will the expected increase in the number and proportion of elderly af-
fect the cost, quality, and access to long-term care?
References
Arrow, K.J. (1963), "Uncertainty and the welfare economics of medical care'. American Economic Review
53:941-973.
Baldwin, C.Y., and C.E. Bishop (1984), "Return to nursing home investment: issues for public policy", Health
Care Financing Review 5(4):43-52.
Bartlett, H.P., and D.R. Phillips (1996), "Policy issues in the private health sector: examples from long-term
care in the U.K.", Social Science and Medicine 43(5):731-737.
Bauer, E.J. (1996), "Transitions from home to nursing home in a capitated long-term care program: the role
of individual support systems", Health Services Research 31(3):309-326.
Bernheim, B.D. (1991), "How strong are bequest motives? Evidence based on estimates of the demand for
life insurance and annuities", Journal of Political Economy 99(5):899-927.
Bernheim, B.D., A. Shleifer and L.H. Summers (1985), "The strategic Bequest motive", Journal of Political
Economy 93(6): 1045-1076.
Bishop, C.E. (1988), "Competition in the market for nursing home care", Journal of Health Politics, Policy
and Law 13(2):341-360.
Boaz, R.E (1996), "Full-time employment and informal caregiving in the 1980s", Medical Care 34(6):524-
536.
B6rsch-Supan, A., V. Hajivassiliou, L.J. Kotlikoff and J.N. Morris (1992), "Health, children, and elderly living
arrangements: a multiperiod-multinomial probit model with unobserved heterogeneity and autocorrelated
errors", in: D.A. Wise, ed., Topics in the Economics of Aging (University of Chicago Press, Chicago)
79-104.
Branch, L.G., R.F. Coulam and Y.A. Zimmerman (1995), "The PACE evaluation: initial findings", The Geron-
tologist 35(3):349-359.
Branch, L.G., D.J. Friedman, M.A. Cohen, N. Smith and E. Socholitzky (1988), "Impoverishing the elderly:
a case study of the financial risk of spenddown among Massachusetts elderly people", The Gerontologist
28(5):648-652.
Cambois, E., and J.-M. Robine (1996), "An international comparison of trends in disability-free life ex-
pectancy", in: R. Eisen and F.A. Sloan, eds., Long-Term Care: Economic Issues and Policy Solutions
(Kluwer Academic Publishers, Boston) 11-24.
Chang, C.F., and S.I. White-Means (1995), "Labour supply of informal caregivers", International Review of
Applied Economics 9(2): 192-205.
Christianson, J.B. (1988), "The effect of channeling on informal caregiving", Health Services Research
23(1):99-117.
Cohen, M.A., and A.K.N. Kumar (1997), "The changing face of long-term care insurance in 1994: profiles
and innovations in a dynamic market", Inquiry 34:50-61.
Cohen, M.A., and W.D. Spector (1996), "The effect of Medicaid reimbursement on quality of care in nursing
homes", Journal of Health Economics 15(1):23-48.
Congressional Research Service (1993), Medicaid Source Book: Background Data and Analysis (A 1993
Update) (US Government Printing Office, Washington, DC).
Conover, C.J., and EA. Sloan (1995), "Bankruptcy risk and state regulation of continuing care retirement
communities", Inquiry 32:444-456.
Coughlin, T.A., T.D. McBride, M. Perozek and K. Liu (1992), "Home care for the disabled elderly: predictors
and expected costs", Health Services Research 27(4):453-479.
Cutler, D.M. (1996), "Why don't markets insure long-term risk?", Working Paper (Harvard University and
National Bureau of Economic Research).
Dick, A., A.M. Garber and T. MaCurdy (1994), "Forecasting nursing home utilization of elderly Americans",
in: D.A. Wise, ed., Studies in the Economics of Aging (University of Chicago Press, Chicago) 365-394.
Doty, P.A. (1988), "Long-term care in international perspective", Health Care Financing Review (Annual
Suppl.): 145-155.
Doty, PA. (1990), "A comparison of long-term care financing in the US and other developed nations: dis-
pelling some myths", Generations 14(2):10-14.
Doty, P.A. (1993), "International long-term care reform: a demographic, economic, and policy overview",
Journal of Cross-Cultural Gerontology 8:447-461.
Dusansky, R. (1989), "On the economics of institutional care of the elderly in the US: the effects of change in
government reimbursement", Review of Economics 56:141-150.
Eng, C., J. Pedulla, G.P. Eleazer, R. McCann and N. Fox (1997), "Program of All-inclusive Care for the
Elderly (PACE): an innovative model of integrated geriatric care and financing", Journal of the American
Geriatric Society 45(2):223-232.
Ettner, S.L. (1993), "Do elderly Medicaid patients experience reduced access to nursing home care?", Journal
of Health Economics 12(3):259-280.
Ettner, S.L. (1995), "The impact of 'parent care' on female labor supply decisions", Demography 32(1):63-
80.
Feder, J., and W. Scanlon (1980), "Regulating the bed supply in nursing homes", Milbank Memorial Fund
Quarterly/Health and Society 58(1):54-88.
Feinstein, J.S. (1993), "The relationship between socioeconomic status and health: a review of the literature",
The Milbank Quarterly 71(2):279-322.
Forder, J. (1997), "Contracts and purchaser-provider relationships in community care", Journal of Health
Economics 16(5):517-542.
Fries, B.E., D. Schneider, W.J. Foley, M. Gavazzi, R. Burke and E. Cornelius (1994), "Refining a case-mix
measure for nursing homes: Resource Utilization Groups (RUG-III)", Medical Care 32:668-685.
Fuchs, V.R. (1990), "The health sector's share of the gross national product", Science 247:534-538.
Garber, A.M., and T.E. MaCurdy (1990), "Predicting nursing home utilization among the high-risk elderly",
in: D.A. Wise, ed., Issues in the Economics of Aging (University of Chicago Press, Chicago) 173-200.
Garber, A.M. and T.E. MaCurdy (1993), "Nursing home discharges and exhaustion of Medicare benefits",
Journal of the American Statistical Association 88(423):727-736.
Gertler, P.J. (1989), "Subsidies, quality, and the regulation of nursing homes", Journal of Public Economics
38:33-52.
Gertler, P.J. (1992), "Medicaid and the cost of improving access to nursing home care", Review of Economics
and Statistics 74(2):338-345.
Gertler, P.J., and D.M. Waldman (1992), "Quality-adjusted cost functions and policy evaluation in the nursing
home industry", Journal of Political Economy 100(6):1232-1256.
Gruber, J., and A. Yelowitz (1999), "Public health insurance and private savings", Journal of Political Econ-
omy 107(6):1249-1274.
990 E. C. Norton
Gruenberg, L.W., and T.R. Willemain (1982), "Hospital discharge queues in Massachusetts", Medical Care
20(2):188-200.
Hamilton, V.H. (1993), "The Medicare hospice benefit: the effectiveness of price incentives in health care
policy", RAND Journal of Economics 24(4):605-624.
Harrington, C.A., M. Curtis, H. Carrillo, B. Bedney, J.H. Swan and J.A. Nyman (1997a), "State regulation of
the supply of long term care providers", The Journal of Applied Gerontology 16(1):5-30.
Harrington, C.A., J.H. Swan, J.A. Nyman and H. Carrillo (1997b), "The effects of certificate of need and
moratoria policy on change in nursing home beds in the United States", Medical Care 35(6):574-588.
Headen Jr., A.E. (1992), "Time costs and informal social support as determinants of differences between
Black and White families in the provision of long-term care", Inquiry 29:440-450.
Headen Jr., A.E. (1993), "Economic disability and health determinants of the hazard of nursing home entry",
The Journal of Human Resources 28(1):80-110.
Hirth, R.A. (1999), "Consumer information and competition between nonprofit and for-profit nursing homes",
Hoerger, T.J., G. Picone and F.A. Sloan (1996), "Public subsidies, private provision of care and living ar-
rangements", Review of Economics and Statistics 78(3):428-440.
Hubbard, R.G., J. Skinner and S.P. Zeldes (1995), "Precautionary savings and social insurance", Journal of
Political Economy 103:360-399.
Hughes, S.L., L.M. Manheim, P.L. Edelman and K.L. Conrad (1987), "Impact of long-term home care on
hospital and nursing home use and cost", Health Services Research 22(1):19-47.
Hurd, M.D. (1987), "Savings of the elderly and desired bequests", American Economic Review 77(3):298-
312.
Hurd, M.D. (1989), "Mortality risk and bequests", Econometrica 57(4):779-813.
Hurd, M.D. (1990), "Research on the elderly: economic status, retirement, and consumption and saving",
Journal of Economic Literature 28:565-637.
Ikegami, N. (1997), "Public long-term care insurance in Japan", Journal of the American Medical Association
278(16):1310-1314.
Ikegami, N., B.E. Fries, Y. Takagi, S. Ikeda and I. Toshiko (1994), "Applying RUG-III in Japanese long-term
care facilities", The Gerontologist 34(5):628-639.
Institute of Medicine (1986), Improving the Quality of Care in Nursing Homes (National Academy Press,
Washington, DC).
Kane, R.A. (1988), "The noblest experiment of them all: learning from the national channeling evaluation",
Health Services Research 23(1):189-198.
Kane, R.A., and R.L. Kane (1988), "Long-term care: variations on a quality assurance theme", Inquiry
25:132-146.
Kane, R.L. (1986), "Outcome-based payment: a new beginning?", Health Progress (April):36-70.
Katz, S., A.B. Ford, R.W. Moskowitz, B.A. Jackson and M.W. Jaffe (1963), "Studies of illness in the aged: the
index of ADL, a standardized measure of biological and psychosocial function", Journal of the American
Medical Association 185(12):914-919.
Kemper, P. (1988), "The evaluation of the national long term care demonstration", Health Services Research
23(l):161-174.
Kemper, P., and C.M. Murtaugh (1991), "Lifetime use of nursing home care", New England Journal of
Medicine 324:595-600.
Kemper, P., and L.E. Pezzin (1996), "The effect of public provision of home care on living and care arrange-
ments: evidence from the channeling experiment", in: R. Eisen and F.A. Sloan, eds., Long-Term Care:
Economic Issues and Policy Solutions (Kluwer Academic Publishers, Boston) 125-146.
Kenney, G., and J. Holahan (1990), "The nursing home market and hospital discharge delays", Inquiry 27:73-
85.
Kenney, G., and J. Holahan (1991), "Nursing home transfers and mean length of stay in the prospective
payment era", Medical Care 29(7):589-609.
Kovar, M.G. (1986), "Expenditures for the medical care of elderly people living in the community in 1980",
The Milbank Quarterly 64(1):100-132.
Laditka, S.B. (1998), "Modeling lifetime nursing home use under assumptions of better health", Journal of
Gerontology: Social Sciences 53B(4):S177-S187.
Lakdawalla, D., and T. Philipson (1998), "The rise in old age longevity and the market for long-term care",
Working Paper (University of Chicago).
Lane, D., D. Uyenuo, A. Stark, G. Gutman and B. McCashin (1987), "Forecasting client transitions in British
Columbia's long-term care program", Health Services Research 22(3):671-706.
Lee, R., and J. Skinner (1997), "Will aging baby boomers bust the federal budget?", Working Paper (Univer-
sity of California at Berkeley).
Levit, K.R., H.C. Lazenby, B.R. Braden, C.A. Cowan, A.L. Sensenig, P.A. McDonnell, J.M. Stiller, D.K. Won,
A.B. Martin, L. Sivarajan, C.S. Donham, A.M. Long and M.W. Stewart (1997), "National health expendi-
tures", Health Care Financing Review 19(1):161-200.
Liang, J., X. Liu, E. Tu and N. Whitelaw (1996), "Probabilities and lifetime durations of short-stay hospital
and nursing home use in the United States, 1985", Medical Care 34(10):1018-1036.
Lindrooth, R.C., T.J. Hoerger and E.C. Norton (1999), "Instrumental variable estimation of error components
models: the case of expectations about nursing home entry", Working Paper (Northwestern University).
Liu, K., K.G. Manton and B.M. Liu (1985), "Home care expenses for the disabled elderly", Health Care
Financing Review 7(2):51-58.
Liu, K., T. McBride and T. Coughlin (1994), "Risk of entering nursing homes for long versus short stays",
Medical Care 32(4):315-327.
Lubitz, J., J. Beebe and C. Baker (1995), "Longevity and Medicare expenditures", New England Journal of
Medicine 332(15):999-1003.
Lubitz, J., and R. Prihoda (1984), "The use and cost of Medicare services in the last two years of Life", Health
Care Financing Review 5(3): 117-131.
Lubitz, J., and G.F. Riley (1993), "Trends in Medicare payments in the last year of life", The New England
Journal of Medicine 328(15):1092-1096.
Manton, K.G., E. Stallard and L. Corder (1998), "Economic effects of reducing disability", The American
Economic Review: Papers and Proceedings 88(2):101-105.
Manton, K.G., E. Stallard and K. Liu (1993), "Forecasts of active life expectancy: policy and fiscal implica-
tions", The Journals of Gerontology 48(Special Issue): 11-26.
McCall, N. (1984), "Utilization and costs of Medicare services by beneficiaries in their last year of life",
Medical Care 22(4):329-342.
Mendelson, M.A. (1974), Tender Loving Greed (Vintage Books, New York).
Moffitt, R. (1983), "An economic model of welfare stigma", The American Economic Review 73(5): 1023-
1035.
Mor, V., O. Intrator and L. Laliberte (1993), "Factors affecting conversion rates to Medicaid among new
admissions to nursing homes", Health Services Research 28(1):1-25.
Morris, C.N., E.C. Norton and X.H. Zhou (1994), "Parametric duration analysis of nursing home usage", in:
N. Lange, L. Ryan, L. Billard, D. Brillinger, L. Conquest and J. Greenhouse, eds., Case Studies in Biometry
(John Wiley and Sons, New York) 231-248.
Moscovice, I., G. Davidson and D. McCaffrey (1988), "Substitution of formal and informal care for the
community-based elderly", Medical Care 26(10):971-981.
Murtaugh, C.M., P. Kemper and B.C. Spillman (1990), "The risk of nursing home use in later life", Medical
Care 28(10):952-962.
Murtaugh, C.M., P. Kemper and B.C. Spillman (1995), "Risky business: long-term care insurance underwrit-
ing", Inquiry 32:271-284.
Murtaugh, C.M., P. Kemper, B.C. Spillman and B.L. Carlson (1997), "The amount, distribution, and timing
of lifetime nursing home use", Medical Care 35(3):204-218.
Newhouse, J.P. (1992), "Medical care costs: how much welfare loss?", Journal of Economic Perspectives
6(3):3-21.
992 E.C. Norton
Naon, D. (1996), "The impact of the community long-term care insurance law on services for the elderly in
Israel", in: R. Eisen and F.A. Sloan, eds., Long-Term Care: Economic Issues and Policy Solutions (Kluwer
Academic Publishers, Boston) 149-170.
Nocera, S., and P. Zweifel (1996), "Women's role in the provision of long-term care, financial incentives,
and the future financing of long-term care", in: R. Eisen and F.A. Sloan, eds., Long-Term Care: Economic
Issues and Policy Solutions (Kluwer Academic Publishers, Boston) 79-102.
Norton, E.C. (1992a), "Incentive regulation of nursing homes", Journal of Health Economics 11(2):105-128.
Norton, E.C. (1992b), "Incentive regulation of nursing homes: specification tests of the Markov model", in:
D.A. Wise, ed., Topics in the Economics of Aging (The University of Chicago Press, Chicago) 275-303.
Norton, E.C. (1995), "Elderly assets, Medicaid policy, and spend-down in nursing homes", Review of Income
and Wealth 41(3):309-329.
Norton, E.C., and V. Kumar (2000), "The long-run effects of the Medicare Catastrophic Coverage Act",
Inquiry (forthcoming).
Norton, E.C., and J.P. Newhouse (1994), "Policy options for public long-term care insurance", Journal of the
American Medical Association 271(19):1520-1524.
Nyman, J.A. (1985), "Prospective and cost-plus Medicaid reimbursement, excess Medicaid demand, and the
quality of nursing home care", Journal of Health Economics 4(3):237-259.
Nyman, J.A. (1988), "Excess-demand, the percentage of Medicaid patients, and the quality of nursing home
care", Journal of Human Resources 23(1):76-92.
Nyman, J.A., M. Finch and R.A. Kane (1997), "The substitutability of adult foster care for nursing home care
in Oregon", Medical Care 35(8):801-813.
Nyman, J.A. (1989), "The private demand for nursing home care", Journal of Health Economics, 8(2):209-
231.
Nyman, J.A. (1993), "Testing for excess demand in nursing home care markets", Medical Care 31(8):680-
693.
OECD (1996), "Caring for frail elderly people - Policies in evolution", Social Policy Studies No. 19, Paris.
Pauly, M.V. (1990), "The rational nonpurchase of long-term care insurance", Journal of Political Economy
98:153-168.
Pauly, M.V. (1996), "Almost optimal social insurance for long-term care", in: R. Eisen and F.A. Sloan, eds.,
Long-Term Care: Economic Issues and Policy Solutions (Kluwer Academic Publishers, Boston) 307-329.
Pezzin, L.E., P. Kemper and J. Reschovsky (1996), "Does publicly provided home care substitute for family
care?", The Journal of Human Resources XXXI(3):650-676.
Phillips, C.D., E.C. Norton, J. Han, R. Green and C. Hawes (1997), "Comparing Medicare utilization among
nursing home and board and care residents", Draft Report for The Office of the Assistant Secretary for
Planning and Evaluation (US Department of Health and Human Services).
Rabiner, D.J., S.C. Steams and E. Mutran (1994), "The effect of channeling on in-home utilization and subse-
quent nursing home care: a simultaneous equation perspective", Health Services Research 29(5):605-622.
Reschovsky, J.D. (1996), "Demand for and access to institutional long-term care: the role of Medicaid in
nursing home markets", Inquiry 33:15-29.
Reschovsky, J.D. (1998), "The roles of Medicaid and economic factors in the demand for nursing home care",
Rhoades, J., D.E.B. Potter and N. Krauss (1998), "Nursing homes - structure and selected characteristics,
1996", MEPS Research Findings No. 4, AHCPR Publication No. 98-0006 (Agency for Health Care Policy
and Research, Rockville, MD).
Rice, T. (1989), "The use, cost, and economic burden of nursing-home care in 1985", Medical Care
27(12):1133-1147.
Rivlin, A.M., J.M. Weiner, R.J. Hanley and D.A. Spence (1988), Caring for the Disabled Elderly: Who Will
Pay? (The Brookings Institution, Washington, DC).
Ruchlin, H.S. (1988), "Continuing care retirement communities: an analysis of financial viability and health
care coverage", The Gerontologist 28(2):156-162.
Scanlon, W.J. (1980), "A theory of the nursing home market", Inquiry 17:25-41.
Schulte, B. (1996), "Social protection for dependence in old age: the case of Germany", in: R. Eisen and
EA. Sloan, eds., Long-Term Care: Economic Issues and Policy Solutions (Kluwer Academic Publishers,
Boston) 149-170.
Scitovsky, A. (1988), "Medical care in the last 12 months of life: the relation between age, functional status,
and medical care expenditures", The Milbank Quarterly 66(4):640-660.
Sloan, EA., T.J. Hoerger and G. Picone (1996), "Effects of strategic behavior and public subsidies on families'
long-term care decisions", in: R. Eisen and EA. Sloan, eds., Long-Term Care: Economic Issues and Policy
Solutions (Kluwer Academic Publishers, Boston) 45-78.
Sloan, EA., and E.C. Norton (1997), "Adverse selection, bequests, crowding out, and private demand for
insurance: evidence from the long-term care insurance market", Journal of Risk and Uncertainty 15:201-
219.
Sloan, EA., G. Picone and T.J. Hoerger (1997), "The supply of children's time to disabled elderly parents",
Economic Inquiry 35:295-308.
Sloan, EA., M.W. Shayne and C.J. Conover (1995), "Continuing care retirement communities: prospects for
reducing institutional long-term care", Journal of Health Policy, Politics, and Law 20(1):75-98.
Somers, S.A., and J.C. Merrill (1991), "Supporting states' efforts for long-term care insurance", Health Affairs
177-179.
Spillman, B.C., and P. Kemper (1995), "Lifetime patterns of payment for nursing home care", Medical Care
33(3):280-296.
Steams, S.C., M.G. Kovar, K. Hayes and G.G. Koch (1996), "Risk indicators for hospitalization during the
last year of life", Health Services Research 31(1):49-69.
Stem, S. (1990), "Measuring the effect of disability on labor force participation", Journal of Human Resources
24(3):361-395.
Stern, S. (1995), "Estimating family long-term care decisions in the presence of endogenous child character-
istics", Journal of Human Resources 30(3):551-580.
Stone, R., G.L. Cafferata and J. Sangl (1987), "Caregivers of the frail elderly: a national profile", The Geron-
tologist 27(5):616-626.
Taylor, Jr., D.H., EA. Sloan and E.C. Norton (1999), "Formation of trusts and spend down to Medicaid",
Journal of Gerontology: Social Sciences 54B(4):S194-S201.
Thorpe, K.E., P.J. Gertler and P. Goldman (1991), "The Resource Utilization Group system: its effect on
nursing home case mix and costs", Inquiry 28:357-365.
United Nations (1997), The Sex and Age Distribution of the World Populations: The 1996 Revisions (United
Nations, Department for Economics and Social Affairs, Population Division, New York).
Weissert, W.G. (1988), 'The national channeling demonstration: what we knew, know now, and still need to
know", Health Services Research 23(1):175-187.
Weissert, W.G., and C.M. Cready (1988), "Determinants of hospital-to-nursing home placement delays: a pilot
study". Health Services Research 23(5):619-647.
Weissert, W.G., J.M. Elston, M.C. Musliner and E. Mutran (1991), "Adult day care regulation: deja vu all
over again?", Journal of Health Politics, Policy and Law 16(1):51-65.
White-Means, S.I. (1995, Part II), "Conceptualizing race in economic models of medical utilization: a case
study of community-based elders and the emergency room", Health Services Research 30(1):207-223.
White-Means, S.I. (1997), "The continuing significance of race in meeting health care needs of Black elderly",
in P. Mason and R. Williams, eds., Race, Markets, and Social Outcomes (Kluwer Academic Publishers,
Norwell, MA).
White-Means, S.I., and D. Chollet (1996), "Opportunity wages and workforce adjustments: understanding the
cost of in-home elder care", Journal of Gerontology 58B(2):S82-S90.
Wolf, D.A., and B.J. Soldo (1995), "Married women's allocation of time to employment and care of elderly
parents", Journal of Human Resources 29(4):1259-1276.
Zweifel, P., S. Felder and M. Meier (1996), "Demographische Alterung und Gesundheitskosten: Eine Fehlin-
terpretation", in: P. Oberender, ed., Alter und Gesundheit. Gesundheitsbikonomische Beitrage 26, Baden-
Baden (Germany) Nomos, 29-46.
994 E.C. Norton
Zweifel, P., and W. Striiwe (1996a), "Long-term care insurance and bequests as instruments for shaping
intergenerational relationships", Journal of Risk and Uncertainty 12:65-76.
Zweifel, P., and W. Striwe (1996b), "Long-term care insurance and trust saving in a two-generation model",
in: R. Eisen and FA. Sloan, eds., Long-Term Care: Economic Issues and Policy Solutions (Kluwer Aca-
demic Publishers, Boston) 225-250.
Chapter 18
THE ECONOMICS OF DISABILITY AND DISABILITY

POLICY*
ROBERT HAVEMAN
Department of Economics University of Wisconsin-Madison
BARBARA WOLFE
Departmentof Economics University of Wisconsin-Madison
Contents
Abstract 996
Keywords 996
1. Introduction and overview 997
2. The disabled population: how many and who are they? 997
2.1. The definition of disability 998
2.2. The disabled population: prevalence and its trend 999
2.3. The demographic composition of the disabled 1002
2.4. Disease patterns underlying disability 1006
2.5. Employment patterns among the working-age disabled population 1006
2.6. The economic well-being of the disabled 1008
2.7. The economic costs of disability 1010
3. Disability policy toward working-age people: the case of the United States 1012
3.1. The Social Security Disability Income (SSDI) program 1013
3.2. The Supplemental Security Income (SSI) program 1015
3.3. The Workers' Compensation program and job-related handicaps 1017
3.4. Accident and Injury Prevention programs 1017
3.5. Vocational Training, Rehabilitation, and Return-to-Work programs 1018
3.6. Health Insurance Aspects of Disability programs 1019
4. The economics of policies toward working-age people with disabilities 1020
4.1. Disability policy and its economic effects: a road map 1020
4.2. The labor supply effects of disability benefits 1023
4.3. Cash benefits, economic well-being, and poverty reduction 1028
4.4. Training, rehabilitation, and the return to work 1030
*The authors gratefully acknowledge the helpful comments of Leo Aarts, Monroe Berkowitz, Philip de Jong,
Richard Disney, Brent Kreider. and Joseph Newhouse, and the assistance of Elizabeth Evanson and Dawn
Duren, the financial support of the Netherlands Institute for Advanced Study (NIAS), the Graduate School of
the University of Wisconsin-Madison, and the Institute for Research on Poverty.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P. Newhouse
996 R. Haveman and B. Wolfe
4.5. Public health care and in-kind benefits 1031

4.6. Antidiscrimination legislation 1032
5. Children's disability and public policy 1033
5.1. SSI benefits for children 1033
5.2. Special education in the public schools 1035
6. Disability policy across OECD countries 1037
7. Unresolved issues and research needs 1040
Appendix A. United States income support programs for working-age disabled
people 1042
A. 1. The Social Security Disability (SSDI) program 1042
A.2. The Supplemental Security Income (SSI) program 1044
A.3. The Workers' Compensation program and job-related handicaps 1045
References 1047
Abstract
We discuss and critique the main lines of economic research that address the economic
status and behavior of the working-age population of people with disabilities. We define
this population as those with physical or mental limitations that impede their daily ac-
tivities or their productivity on the job. Using this definition, we assess the prevalence,
trend, and composition of the population of disabled working-aged people in the United
States and other Western societies, and document the extent of market work among this
population. Such market work contributes to the economic well-being of the working-
age disabled, but for most of them, income from public transfers and from the earnings
of other household members are crucial in determining the level of family economic
well-being. Relative to the nondisabled, those with disabilities have substantially lower
levels of economic well-being in spite of public income support programs. While public
income support is important in sustaining the level of well-being of the disabled, these
policies also have serious incentive effects, especially labor supply disincentives. We
document these incentive effects in US policy, and review the research studies that esti-
mate the response of disabled people to these incentives. In addition to income support
policy, we also describe public policy toward disabled people associated with antidis-
crimination legislation, rehabilitation and training programs, income support for poor
disabled children, and public regulations and financial support for special education in
schools. We conclude by comparing US disability policy with that in other Western in-
dustrialized countries and identifying research issues that are relevant to all societies
with advanced policies toward working-age people with disabilities.
Keywords
disability, labor force participation, transfer policy, incentives, rehabilitation
JEL classification:1120, I1180, J700, J280, J290, H550

Ch. 18: The Economics ofDisability and Disability Policy 997
1. Introduction and overview
In this review, we describe the main lines of research and analysis addressing the eco-
nomic status and behavior of the population of working-age people with disabilities.
After presenting our definition of this population and assessing the prevalence and
composition of disability in Western societies, we discuss a variety of economic issues
regarding people with disabilities: their employment and work effort, their economic
well-being relative to the nondisabled population, and the economic costs which their
limitations impose on the society.
The ability of the disabled population to work and to earn is the central economic
issue in this research area. Because the level of work effort by the disabled population
is closely tied to the alternative sources of income support which they have available
to them, we first outline the nature of public policy (especially, income support pol-
icy) toward disabled persons in the United States and other developed countries. We
then assess the labor supply response of disabled workers to the incentives implicit in
these programs and their effectiveness in maintaining the level of living of the disabled
population.
Disability programs also compensate workers for injuries sustained on the job, reha-
bilitate disabled people so as to enable them to return to work, and regulate workplaces
in an effort to reduce the incidence of work-related disabilities. We describe these pro-
grams and assess what is known about their economic effectiveness.
In recent decades, increased attention has been devoted to children with disabilities,
who have accounted for rapidly increasing costs both for income support and for special
schooling arrangements, especially in the United States. We discuss these new develop-
ments and raise questions regarding their purpose and effectiveness.
Throughout our review, we have attempted to set the main economic questions re-
garding persons with disabilities in a broader international context. To accomplish that,
we leaven our discussion of both policy and research in the United States with the find-
ings of studies done in other countries. We also summarize the major differences in
disability policy across the OECD countries.
Finally, we conclude our review by presenting an assessment of the remaining unre-
solved issues and the research needs and priorities that follow from them.
2. The disabled population: how many and who are they?
Discussion of the economics of disability requires a common understanding of the

meaning of "disability" and an overview of both the extent and demographic and eco-
nomic composition of the disabled population. We begin by offering a definition of
disability that we will maintain throughout the paper; we then present measures of the
extent and composition of the disabled population in the United States over time. We
1 We present estimates only for the United States; the compositional patterns there are similar to those in
other Western industrialized countries, although trends in the prevalence of disability in the working-age
focus on the working-age population, defined as individuals aged 18-65; because of

recent interest in disabled children and policy toward them, we also include them in our
discussion.
2.1. The definition of disability
Establishing a meaningful concept of disability is difficult. At the limit, any personal

characteristic that leads one person to be less proficient in performing some worthwhile
function than another otherwise identical person might be described as a handicap, or
a disabling condition. Yet not all characteristics related to proficiency would be con-
sidered in defining disability; required sleep time or ability to work cooperatively with
others are two such attributes. If two otherwise identical and normally functioning peo-
ple were to differ in only these respects, we would be unlikely to consider the less
proficient person "disabled".
In most economic (as opposed to medical) definitions of disability, the characteristics
referred to in defining disability involve mental and physical characteristics that either
constrain normal daily activities or cause a substantial reduction in productivity on the
job. Personality traits, tastes, and appearance have been excluded. 2
Given these accepted characteristics for forming a judgment on disability status, it
is necessary to establish a norm or standard to which the observed characteristic can
be compared. The norm most commonly adopted concerns the ability to carry out the
tasks of a commonly accepted occupation - that is, the ability to engage in sufficient
work to "earn a living". In the United States, for example, one norm is stated as the
ability to engage in "substantial gainful activity"; another norm is that of carrying out
standard housekeeping and personal maintenance functions without assistance. In both
cases, the criterion is a performance criterion, and not a structural one relating to the
characteristics of the person. 3
This focus on performance seems appropriate, given the multidimensional character
of physical or mental conditions, the difficulty of accurately measuring many of them,
and the complex ways in which they relate to the demands of the labor market. When a
paraplegic can be more successful in the labor market (or self-care) than a person with a
combination of varicose veins, back problems, and low education, measuring disability
through assessing characteristics seems less satisfactory than observing performance
directly.
A third element is important in defining disability, namely, establishing the magnitude
of the deviationfrom a norm that is necessary before a person is classified as disabled.
population vary across countries. Differences in compositional and trend patterns until the mid-1980s are
shown in Haveman, Halberstadt, and Burkhauser (1984).
2 It should be noted that if the "inability to cooperate" characteristic were to reach such intensity that the
person is judged to have emotional problems, it could be included as a disabling condition.
3 In practice, however, certain structural characteristics are accepted as proof that performance cannot be
adequate. Thus, a quadriplegic is typically classified as disabled, even though he or she may have perfectly
adequate labor market performance and earnings.
If the norm were to be the ability to work a forty-hour week, classifying all people who
are unable to meet this goal as disabled would seem inappropriate. However, one might
classify as disabled all people who, because of physical and mental limitations, are able
to work at most ten hours per week.
This brief discussion is sufficient to convey the multidimensional character of
disability, even though we have dealt with only the two-way dichotomy of dis-
abled/nondisabled. Additional issues arise when the objective is to distinguish degrees
of disablement among the disabled. In all of these decisions, some judgment regarding
the deviation from a norm must be made, but the basis for that judgment - expected
duration of condition, age, education, labor market conditions - is not clear. Different
societies, or the same society at different points in time, have quite different expectations
regarding the level of performance of an individual, and in what dimension shortfalls
from a norm would qualify the person as being disabled.
However, once a definition is agreed upon there is still an issue of accuracy of deter-
mination of who is, and who is not disabled. This issue arises whenever a decision must
be made to determine eligibility for a program that provides benefits. This "tagging" of
individuals may not be perfect with errors of both omission and commission. 4
This economic approach to disability measurement, then, emphasizes the interactions
of a person's physical or mental limitations with a variety of nonmedical characteristics
of that person, such as age, occupation, and work experience. The focus is on the ability
of persons with physical or mental limitations to adjust to their work environment.5 This
definition is employed to facilitate measurement of the extent and character of disability
in a society.
2.2. The disabledpopulation:prevalence and its trend
In spite of the need for judgment, establishing the magnitude and composition of the dis-
abled working-age population is essential in assessing the economic and public policy
domains of the problem.
The primary technique employed in identifying the disabled population is the use of
individual reports of self-assessed disability status through surveys. Several measure-
ment issues are relevant in assessing this approach, and the judgments made on each of
them will affect the resulting estimates of the size, composition, and growth over time
of the disabled population. Assuming reliance on answers to survey questions, issues
regarding design of the questions, including assessment of the durationof the disability
(permanent, of long duration, temporary), the severity of the disability (total, severe,
4 Boadway, Marceau, and Sato (1997) cite two examples of 18-20 percent error rates in accepting and
rejecting of applicants.
5 This definition is based on what is referred to as the impairment-limitation disability model, as opposed to
the medical model of disability. See Berkowitz, Johnson, and Murphy (1976) and Nagi (1969).
partial), and the norms from which the shortfall is measured (restrictions on work time,
constraints on productivity, limits on normal activities) are all relevant. 6
In most industrialized countries, estimates of the size of the disabled working-age
population have relied on survey-based self-reports of "functional limitations", restric-
tions on "activities of daily living", and the presence of "impairments or chronic dis-
abling conditions". Because of differences in both the criteria and the nature of the
survey question asked, estimates of the prevalence of disability differ across countries.
A standard question in the United States defines a disability as "a limitation in the kind
or amount of work (or housework) resulting from a chronic health condition or im-
pairment lasting three or more months".' In the United Kingdom, estimates rely on a
cross-sectional survey and responses to a question on labor force participation, which
includes a category of "sick, not seeking work". In the Netherlands, estimates rely on
surveys that ask whether a person receives disability benefits, or on administrative data.
For the years 1989-1990, Burkhauser and Daly (1996) estimated the prevalence of
disability (the percentage of the working-age population with disabilities) in the United
States at between 8 and 12 percent. These estimates are based on three different data
sets, but employ a similar definition of disability. 8 Using a definition of disability that
relies on a two-part criterion (the presence of work limitations and/or the receipt of
government benefits reserved for the disabled) and data from the Current Population
Survey (CPS), Haveman et al. (1995) found 7.6 percent of the working-age popula-
6 The World Health Organization (1980) defines disability as: "a restriction on or lack (resulting from an
impairment) of an ability to perform an activity in the manner or within the range considered normal". For
the Americans with Disabilities Act (ADA), "Disability means with respect to an individual (1) a physical or
mental impairment that substantially limits one or more of the major life activities of such individual, (2) a
record of such an impairment, or (3) being regarded as having such an impairment". Note that this definition
requires that the terms "impairment", "limits", and "major life activities" all be defined.
7 This question served as the basis of the Social Security estimate of the size of the disabled population from
its Survey of Disability and Work, taken in 1966, 1970, 1972, 1974, and 1978. For the final year, 16.8 percent
of the working-age population was classified as disabled, of which about one-half were described as "severely
disabled", meaning "... an inability to work altogether or regularly". An alternative definition that has been
used in studying the labor supply of persons with health limitations is subsequent mortality. See, for example,
Parsons (1980a, 1980b).
8 Using the Michigan Panel Study of Income Dynamics (PSID), the percentages for males and females aged
25-61 were 9.2 and 10.6, respectively. [Those who responded affirmatively in two consecutive years to the
question: "Do you have any nervous or physical condition that limits the type of the amount of work you can
do?"] Using the Current Population Survey (CPS), the two percentages were 8.1 and 7.8, respectively. [Those
who responded affirmatively to either of the questions: "Do you have a health problem or disability which
prevents you from working or which limits the kind or the amount of work you can do?" or "Main reason
did not work in 1989 was ill or disabled"; or who stated that the "Current activity reason for not looking for
work is ill or disabled."] Estimates from the Survey of Income and Program Participation (SIPP) indicated
that the two percentages were 11.7 and 11.6. [Those who responded affirmatively to the question: "Do you
have a physical, mental, or other health condition which limits the kind or amount of work you can do?"]
[SIPP estimates from McNeil (1993)].
Ch. 18: The Economics of Disabilityand Disability Policy 1001
tion - 8.3 percent of males and 7.1 percent of females - to be disabled in 1988. 9 In a
study using data from the National Health Interview Survey (NHIS), Chirikos (1989)
estimated that 12.8 percent of the working-age population had some activity limitation
in 1988. If the disabled are restricted to those with a "major" limitation, the percent-
age falls to 9.3. Disney and Webb (1990) studied the prevalence of disability among
the older working-age population in the United Kingdom in the mid-1980s, and found
16 percent of men aged 60-64, 9 percent of men aged 55-59, and 5 percent of women
aged 55-59 to be disabled.
The prevalence of working-age disability clearly depends on the stringency of the
definition that is used as the basis of the estimate. The elasticity of the disability preva-
lence estimate with respect to the disability criterion is shown by Haveman et al. (1995)
using Survey of Income and Program Participation (SIPP) data for 1984. Using self-
reports of poor or fair health as a criterion, 11.8 percent of the working-age population
was classified as disabled. The prevalence of disability when the presence of one or
more limitations of activities of daily living (ADLs) is the criterion is 14.9 percent; a
narrower criterion of two or more limitation of ADLs yields an estimate of 6.9 per-
cent. Combining the fair/poor health and the one or more limitation of ADLs classifies
19.6 percent of the working-age population as disabled.1 0 Similarly, SIPP-based esti-
mates for 1994-1995 based on the presence of any of a wide range of functional limi-
tations or disabilities that include but are not limited to work find more than 20 percent
of the working-age population to be disabled, though the use of a more stringent def-
inition (severely disabled) suggests that approximately 10 percent of the working age
population is disabled [McNeil (1997)].11
For the United States, most of the recent time-series studies that maintain a con-
stant definition of disability have found a slight decrease in the prevalence of disability
over time. For example, using the two-part criterion and CPS data, Haveman and Wolfe
(1990) find the prevalence of disability to have fallen from 10-11 percent in the period
9 Those classified as disabled owing to work limitations are those who report being unable to work or unable
to work full time, full year because of the presence of limiting health conditions. The disabled identified by
the program participation criterion are those who receive benefits from the Social Security Disability Insur-
ance program, the disability component of the Supplemental Security Income program, Railroad Retirement
program benefits for the disabled, and Workers Compensation. The program participation criterion tends to
expand the number of people classified as having disabilities, while the work limitation criterion is more lim-
ited than that applied by other researchers relying on CPS data. These studies include all those that respond
positively to a question involving the presence of a health problem or disability which prevents work or which
limits the kind or amount of work, plus those who report receiving SSI or Medicare. See US Bureau of the
Census (1989).
10 While these different definitions do not always identify the same individuals, the overlap is substantial as
can be seen from the proportions noted.
11 The limitations include: a work disability; a functional limitation in seeing, hearing, speaking, lifting,
climbing stairs, or walking; a limitation in activities of daily living that include bathing, eating, toileting,
getting around inside the home, getting in or out of bed or a chair; or a limitation in instrumental activities
of daily living that include going outside the home, keeping track of money and bills, preparing meals, doing
light housework, or using the telephone; or a mental or emotional disability.
1968-1980 to about 8 percent by 1988.12 In contrast, disability prevalence in European

countries appears to have increased, at least through the late 1980s. For example, in the
United Kingdom disability prevalence for older men increased from 14 percent in 1979
to 16 percent in 1984; comparable numbers for men and women aged 55-59 also show
an increase over this time period. 13
An alternative measure of the prevalence of working-age disability across countries
can be obtained from the rates of disability transfer receipt. 14 Aarts and de Jong (1996)
present data on recipiency rates for five OECD countries and three age groups for the
years 1970 to 1994 (see Table 1). Several interesting patterns are shown: (1) the rate
of disability transfer recipiency has increased in all of the countries;' 5 (2) across the
countries, prevalence rates increase with age, the rate for those aged 60-64 being at
least four times the rates for those aged 45-59; (3) a clear ranking exists across the
countries in the rates of disability transfer benefit receipt, with the Netherlands having
about twice the rate of the next highest country in 1990 (the latest date for which data
are available for all of the countries).
2.3. The demographic composition of the disabled
While the following overview of the characteristics of the working-age disabled popu-
lation relies on United States data, the patterns that we show are present in other indus-
trialized nations as well.
Consider first the age, race, and education patterns among the approximately 9.5 per-
cent of the working-age population that is identified as disabled by the self-reported
presence of a work limitation in 1993 CPS data.16 Whereas about 15 percent of all peo-
ple aged 25-64 have less than a high school degree, more than 40 percent of the disabled
population are high school dropouts. The disabled population is also disproportionately
older; 42 percent of the working-age disabled are over age 45, compared to 32 percent
12 See Haveman and Wolfe (1990), Burkhauser, Haveman, and Wolfe (1993), and Haveman et al. (1995).
A recent report from the US Bureau of the Census (1989) estimates a similar downward trend over the 1981
to 1988 period, though at a somewhat higher prevalence rate. Chirikos (1989) estimates that the disabled
population fell from 13.6 percent of the population in 1973 to 12.8 percent in 1988. Burkhauser and Daly
(1996) note that cross-sectional estimates of the disabled population oversample "long-stayers" (those whose
disabilities are permanent and, for many, have occurred long ago). From this they conclude that cross-sectional
snapshots of the work, earnings, and demographic characteristics of the disabled population cannot distinguish
those whose disabilities are caused by onset late in life from those disabled from childhood, and hence may
convey a misleading picture of the opportunities for intervention, or the impact that "becoming disabled" has
on relative well-being.
13 These estimates are from Disney and Webb (1990), who used the Family Expenditure Survey for 1979
and 1984.
14 As opposed to survey-based measures, this rate reflects both differential prevalence of health-related prob-
lems and the accessibility of public transfer benefits.
15 In Germany, however, the rate appears to have peaked in the mid-1980s.
16 The following tabulations are from the March 1993 Current Population Survey, and are taken from unpub-
lished reports by the US Bureau of the Census. See Mashaw et al. (1996).
Table 1
Disability transfer recipients per 1,000 active labor force participants, by age, in
five OECD countries
1970 1975 1980 1985 1990 1995
Age 15-44
Germany 7 6 7 8 5 6
Netherlands 17 32 57 58 62 57
Sweden 18 20 19 20 21 32
United Kingdom 8 9 11 20 23
United States 11 17 16 20 23 39
Age 45-59
Germany 75 64 84 103 75 87
Netherlands 113 179 294 305 339 271
Sweden 66 95 99 108 116 151
United Kingdom 48 46 51 97 119
United States 33 68 83 71 72 103
Age 60-64
Germany 419 688 1,348 1,291 1,109 890
Netherlands 299 437 1,033 1,283 1,987 1,872
Sweden 229 382 382 512 577 716
United Kingdom 219 195 209 357 413
United States 154 265 285 254 250 314
Source: Aarts and de Jong (1996) and Aarts, Burkhauser, and de Jong (1997).
of all working-age people. The rate of disability for African Americans is about 50 per-
cent greater than that for both whites and Hispanics; for those with severe disabilities,
the rate for African Americans is nearly twice that for others.
The prevalence of disability among the working-age population is likely to be affected
by a variety of factors, some demographic and others economic. For example, during
periods of high unemployment older workers or workers in redundant industries may be
viewed as unable to obtain employment and hence disabled. As a result, changes over
time in both the social and demographic structure of the population and the performance
of the economy are likely to be related to intertemporal changes in the composition of
the disabled population and overall disability rates.
Using a definition of disability that reflects both health limits on work and the receipt
of disability benefits,17 Wolfe and Haveman (1990) first identified the working-age dis-
abled population, and then statistically related a number of socioeconomic characteris-
tics to the probability of being classified as disabled. Their results for males and females
are shown in Table 2, for both 1973 and 1984. The estimates shown are partial deriva-
tives from logit regressions with the dependent variable taking a value of one if the
individual is disabled and zero if not. Consistent with the above cross-tabulated results,
17 See note 12, above.

1004 R. aveman and B. Wolfe
Table 2
Partial derivatives from logit regression: determinants of disability status among
working-age men and women, 1973 and 1984
Men Women
1973 1984 1973 1984
Race (white = 1) -6.61* 0.66 -5.20* -3.35*

Age 0.35 0.40* 0.24 0.35*
Age spline 35 0.04 -0.41* 0.16 -0.06
Age spline 54 -0.04 0.50 -0.68* -0.68*
Years of education -1.28 -1.90* -2.04* -1.45*
Never married 5.67* 6.67* 14.46* 9.02*
Widow(er) 1.81 8.60* 17.46* 10.83*
Veteran status 9.10* 4.77* NA NA
Presence of children NA NA -0.47* -0.80*
Source: Wolfe and Haveman (1990).

Note: Insignificant and quantitatively small effects from education squared and
unemployment rate in the state are not reported. Data are from the March Current
Population Survey for 1974 and 1985.
* Significant at 0.05 level.
these results indicate that, ceteris paribus: (1) nonwhites have a higher probability of
being disabled than do whites, (2) older people have a higher probability of being dis-
abled than do younger people (though the pattern for those 54 years old and older differs
between the genders), (3) those with more schooling are less likely to be disabled, and
(4) being unmarried, a veteran (for men), and having no children in the household (for
women) increases the probability of disability status. 18
Over time, the marginal impact of some of these determinants has changed signifi-
cantly. For men, age has become a more important determinant of being disabled. This
is consistent with the substantial increase in early retirement among older men during
recent decades, which choice is often justified by reference to health problems. The op-
posite pattern holds for older women, and it too is consistent with changes in female
labor force participation patterns. Throughout the period, years of education is nega-
tively associated with being disabled among both men and women.
The logit regressions were also used to calculate trends in disability status for eight
prototypical individuals who generally have low economic and labor force status. These
predicted probabilities of being disabled (expressed as percentages) are shown in Ta-
ble 3 for 1973 and 1984. Several patterns are noteworthy. First, age is a very strong
factor both over time and across groups. For all four older prototypical individuals,
the disability prevalence rate in 1984 exceeded 25 percent. Second, those prototypical
individuals with tenuous ties to the labor market show far larger levels of disability
18 While the signs on the unemployment rate variable (not shown) suggest that the absence of good employ-
ment prospects increases the probability of being classified as disabled, they are generally not significant.
Ch. 18: The Economics of Disability and Disability Policy 1005
Table 3
Estimated probabilities of being disabled, prototypical individuals, 1973 and 1984
1973 1984
Totalfemale percentage 9.3% 8.6%

1. Young unmarried black mother 23.1 13.4
2. Middle-aged black wife 9.0 7.8
3. Older white widow 33.7 25.2
4. Older black female 48.0 42.4
Total male percentage
5. Young black male 16.8 10.1
6. Middle-aged veteran 20.4 13.3
7. Older white widower 33.4 32.6
8. Older black male 37.7 26.2
Source: Wolfe and Haveman (1990).

Notes:
1. 21 years old, 10 years of schooling, two children.
2. 40 years old, 12 years of schooling, no children.
3. 60 years old, 12 years of schooling, no children.
4. 60 years old, 10 years of schooling, never married, no children.
5. 21 years old, 10 years of schooling, never married, nonveteran, 20 percent unem-
ployment rate.
6. 40 years old, 12 years of schooling, married.
7. 60 years old, 12 years of schooling, veteran.
8. 60 years old, 9 years of schooling, nonveteran, never married, 20 percent unem-
ployment rate.
prevalence than the total male or female populations of which they are a part. However,
the decreases in prevalence for the younger individuals are noteworthy. This change is
consistent with the reorientation of public disability transfer programs during the pre-
1980 period, which became more lenient toward older relative to younger populations
who were viewed as more employable. 19
The pattern among the disabled in the Netherlands is consistent with this description.
For example, among males and females aged 30-64 in 1988, the disabled tend to be
older than the nondisabled, the mean age being about 50 years. More are unmarried
than among the general population of this age group, and the average education level is
lower. 20 However, compared with the United States and other industrialized countries,
the average age of the disabled in the Netherlands is considerably younger. A logit
estimate [van Soest, Fontein, and Euwals (1996)] suggests a larger probability of being
disabled among males if they are divorced or never married, a lower probability of
19 However, since 1980, there has been relatively rapid growth in the number and proportion of younger
working-age people receiving disability benefits (see below).
20 For more detail on this see van Soest, Fontein, and Euwals (1996). Their estimates are based on analysis
of the Dutch Socio-Economic Panel, a rotating panel which was initiated in 1984.
being disabled if there are children present in the household, and some evidence that
those with low earnings capacity are more likely to be either disabled or unemployed,
suggesting substitution between unemployment and disability.
2.4. Disease patternsunderlying disability
The factors determining the health status of the population, and hence the level of dis-
ability among working-age people, are numerous and interact in complex ways. They
include the levels of economic development and medical care technology, nutrition and
life style, and the access to and utilization of medical care. Clearly, as medical care
advances and access to care increases, certain diseases can be controlled. However,
stresses, pollution, and older age tend to offset some of these gains by introducing other
illnesses and health limitations. The underlying health and mental health conditions that
lead to impairments and hence disability tend to differ somewhat across nations and by
level of industrialization.
Table 4 presents estimates of the leading causes of "years lived with disability" world-
wide, and for developed regions. There is substantial variation in the leading causes
of disease across countries. The table illustrates this variation between the pattern for
developed countries and the world. Among the industrialized nations, depression, acci-
dents, alcoholism, dementia and arthritis dominate as the primary health conditions that
lead to disability. While there is a good deal of overlap in the leading causes between
the industrialized and less developed nations, anemia, falls, and more severe categories
of mental illness appear to play a relatively larger role as leading causes of ill health and
disablement in the nonindustrialized nations.
2.5. Employment patterns among the working-age disabledpopulation
An important characteristic of the disabled population, by any definition, is its diversity.

This applies to observed employment patterns as well. For example, in 1992 working-
age men with disabilities in the United States indicated a remarkably high rate of em-
ployment. Table 5 indicates the labor force participation rates of all working-age men
and men with disabilities over the 1970-1992 period. It shows that although work par-
ticipation of men with disabilities in 1970 was substantial (in excess of 70 percent for
men aged 54 or less), these rates were lower than those for all men. However, in the sub-
sequent two decades, the falloff in work participation was much larger for those with
disabilities than for the remainder of the working-age population. By 1992, somewhat
less than 60 percent of men with disabilities aged 54 or less were participants compared
to about 85 percent of nondisabled men in this age range.
The pattern for women is quite different from that for men, as Table 6 shows. Irre-
spective of age group, the labor force participation rates of women with disabilities rose
over the 1970-1992 period. For some of the age groups, the increases were only slightly
Table 4
Leading causes of years lived with disability, 1990
All countries Developed countries

Total years lived Percentage of Percentage of
with disability total years, ten total years, ten
(in millions) leading causes leading causes
All causes 472.7 75.4

Unipolar major depression 50.8 10.7 13.0
Iron deficiency anemia 22 4.7
Falls 22 4.6
Alcohol use 15.8 3.3 8.1
Chronic obstructive
pulmonary disease 14.7 3.1
Bipolar disorder 14.1 3.0 3.3
Congenital anomalies 13.5 2.9
Osteoarthritis 13.3 2.8 6.2
Schizophrenia 12.1 2.6 4.0
Obsessive compulsive disorder 10.2 2.2 2.8
Cerebrovascular disease 3.1
Road traffic accidents 2.7
Dementia and related disorders 4.3
Diabetes mellitus 2.6
Source: Murray and Lopez (1996, Table 4.13).

Note: Years lived with disabilities are a component of Disability-Adjusted Life Years or DALYs.
Table 5
Labor force participation rates of US working-age males, with and
without disabilities
All men Men with disabilities

1970 1992 1970 1992
Age 18-44 86 85 72 60
Age 45-54 91 85 70 57
Age 55-64 79 65 52 33
Source: Yelin and Cisternas (1996).
Table 6
Labor force participation rates of US working-age females, with and
without disabilities
All women Women with disabilities

1970 1992 1970 1992
Age 18-44 48 69 35 50
Age 45-54 52 72 35 45
Age 55-64 43 47 25 26
Source: Yelin and Cistemas (1996).

less than those for all women. For the youngest group the increase in participation for
women with disabilities was equal to that for all women. 21
In addition to somewhat lower labor force participation rates, the unemployment rate
for workers with disabilities is substantially higher than that for workers without im-
pairments. This pattern reflects the higher probability that less productive workers, in-
cluding workers who are perceived as less productive due to a visible disability, may
be at greater risk of losing their jobs in a recession. These workers are also more likely
to experience lower earnings in a recession, leading to withdrawal from the work force
if disability benefits provide a floor on the reservation wage of disabled workers. 22 In
1993, the unemployment rate for workers without disabilities was 7.3 percent, but stood
at more than 16 percent for workers with disabilities [Mashaw et al. (1996)].
In sum, then, the tie of persons with disabilities to the labor market is weaker than
that of individuals without disabilities. However, substantial proportions of both men
and women with disabilities work, and the overall social and economic trends that affect
the nondisabled population also affect those of working age with disabilities.
In many European countries, disability and nonemployment appear to be at least as
closely related as in the United States. Workers who find they are likely to become
unemployed - for example, those with low earnings capacity - are likely to apply for
and be awarded disability benefits. In many European nations, disability benefits exceed
unemployment benefits, and have a longer duration, leading to low employment rates
among the disabled,
2.6. The economic well-being of the disabled
The socioeconomic and demographic characteristics of the disabled suggest that, as a

group, they are unlikely to fare well in terms of labor market earnings. Their attachment
to the labor market is weaker than that of the nondisabled, and they are likely to work
fewer hours and earn less per hour because of their handicaps. Moreover, their other
human capital characteristics such as age, education, and race suggest relatively low
earnings. These characteristics form a recipe for high poverty rates. Indeed, in spite
of sizable programs supporting the incomes of disabled people, the poverty rate of
working-age people with disabilities in the United States (about 30 percent in the early
1990s) is about three times that of the working-age population without disabilities. 2 3
Burkhauser, Haveman, and Wolfe (1993) provide evidence on the level of labor mar-
ket earnings of disabled men relative to a counterpart group without disabilities. Table 7
21 Among single mothers the impact of disability on labor force participation appears to be substantial. Wolfe
and Hill (1995) estimate that a single mother who is disabled (defined as having difficulty performing one or
more ADLs or who reports poor or fair health) is more than one-third less likely to work than a single mother
who is not disabled; a single mother with a disabled child is about 30 percent less likely to work than other
single mothers.
22 We thank Richard Disney for this point.
23 The US Census classifies as disabled those workers who report a health condition that limits the amount
of work.
Table 7
Economic well-being of men with disabilities relative to men without disabilities, 1967-1987
(in 1987 dollars, in thousands)
1967 1972 1975 1979 1981 1987
Real wage earnings

Amount $15.0 $18.8 $15.0 $13.3 $10.8 $11.3
Relative to those without
disabilities 0.66 0.74 0.66 0.58 0.51 0.49
Real transfer income
Amount $2.7 $3.5 $6.5 $6.0 $5.1 $6.4
disabilities 3.7 3.2 3.2 3.8 3.3 4.1
Real family income
Amount $25.5 $31.7 $30.2 $28.3 $23.7 $28.7
disabilities 0.74 0.80 0.80 0.73 0.66 0.75
Source: Burkhauser, Haveman, and Wolfe (1993).
is taken from their study, and reports the differences between the two groups. During
the period when general earnings trends were positive - 1967 to 1972 - the earnings of
disabled men rose absolutely and relative to those of nondisabled men. By 1972, dis-
abled men earned nearly 75 percent of nondisabled male earnings. However, after the
oil crises of the early 1970s, and continuing through to the mid 1980s, the earnings of
disabled men fell dramatically. From 1972 to 1987, average earnings of disabled males
decreased from nearly $19,000 to somewhat more than $11,000, and from about three-
fourths of the earnings of the nondisabled to about one-half. Although the absolute level
of earnings of the disabled began increasing toward the end of the 1980s, they continued
to erode relative to those of the nondisabled. 24
Assessing the overall levels of well-being, however, requires accounting for the im-
pact of public transfers as well as labor earnings. As Table 6 shows, real per-worker
transfers more than doubled over this period, offsetting to some degree the erosion of
earnings. As a result of the effect of transfers and the incomes of others in the household
- primarily spouses2 5 - the total family income of disabled men remained relatively
24 Haveman and Wolfe (1990) report that in 1984 the average wage rate of a group of men with disabilities
(including those who receive disability transfer payments) was about 54 percent of that of men without work
limitations. Baldwin and Johnson (1994) report an average wage rate in the late 1980s of from 70 to 90 percent
of the nondisabled male wage rate, but their sample excluded the most disabled males. Baldwin, Zeager, and
Flacco (1994) indicated substantial differences between men and women in the disabled-nondisabled wage
ratio.
25 Burkhauser, Haveman, and Wolfe (1993) document the substantial shift in the responsibility for family
support from male heads to wives after 1980, a pattern that is much stronger for families of disabled men than
for the families of nondisabled men.
stable over the entire two-decade period, at about three-fourths of that of nondisabled
males.
The breakdown of these earnings and family income patterns by race and educa-
tion are disturbing. The earnings and incomes of nonwhite and low education men
with disabilities experienced far more erosion over the 1967-1987 period relative to
those of their counterparts without disabilities than did the earnings and incomes of dis-
abled males who were white and had more education. Burkhauser, Haveman, and Wolfe
(1993) conclude: "Overall, the families of doubly disabled workers - those who are dis-
abled and either are nonwhite or have low educations - became increasingly separated
economically from both the families of better-educated men with disabilities and from
families of men without disabilities" (p. 261).26 Daly (1994) charts the labor earnings
of African-American men who are disabled and not disabled over the 1970s and 1980s.
She finds strong cyclical patterns with lower disabled/nondisabled ratios in times of
recession than in times of expansion.
As a result of these patterns, the proportion of the US male poverty population that
was accounted for by men with disabilities increased from about 14 to nearly 17 percent
over the decade of the 1980s. The risk of a disabled man of working age being poor by
the end of the 1980s was about 170 percent of that experienced by a nondisabled man
of working age. This relative risk factor increased by about one-fourth during the 1980s
[Burkhauser, Haveman, and Wolfe (1993)].
The disabled in many European countries are much better off economically than those
in the United States. The combination of substantially higher earnings replacement rates
in the income support programs of European countries (up to 80 percent), in combi-
nation with extensive in-kind benefits, (including both health insurance and housing
assistance) explains this result.27
2.7. The economic costs of disability
The existence of a health-limited working-age population is a drain on the productivity

of the nation's economy and a constraint on its growth. It is important to have a sense
of both the level and the trend of the costs imposed on the economy over time owing to
the presence of these limits on work.
Starting in the early 1970s, several studies have been conducted of the costs to the
United States of ill health and disability. 28 The earliest of these studies focused on the
26 A recent study of university graduates who received rehabilitation services suggests little penalty in terms
of salary reduction for this group compared with the broader universe of persons with disability. See Hen-
dricks, Schiro-Geist, and Broadbent (1997).
27 A recent study notes that Austria has the most generous pension system: the disabled who contributed to
the system receive 80 percent of the average of their highest earnings over the prior 15 years. Thus it is not
surprising to learn that in Austria, nearly half of all men and 25 percent of all women retire on a disability
pension. Pension expenditures account for 15 percent of GDP [Koch and Thimann (1997)].
28 See Cooper and Rice (1976), Rice, Hodgson, and Kopstein (1985), and Rice, Kelman, and Miller (1991).
Hartunian, Smart, and Thompson (1981) and Manning, Keeler, Newhouse et al. (1991) present estimates of
the economic costs of specific diseases and poor health habits.
loss of actual earnings (or the value of lost household services), and estimated that the
costs of health-related work limitations ranged from about 1.7 to 3.5 percent of GNP.
A more recent study [Haveman, Wolfe, Buron, and Hill (1995)] focused on the loss
of earnings capabilities that are attributable to these health limitations. While the earlier
measures of the loss of actual earnings are dependent on individual preferences for work
and leisure, the measure based on lost earnings capabilities is not. In this measure, the
effect of disability on both the wage earned and the hours per year worked are accounted
for in determining how much of an individual's potential earnings are lost because of
the disabling condition. Using the definition of disability based on both self-reports of a
work limitation and/or participation in a disability-related benefit program, the authors
estimate a loss of earnings capability of $131 billion in 1973 (5.3 percent of aggregate
earnings capabilities) and of $128 billion in 1988 [4.5 percent of aggregate earnings
capabilities (in 1988 dollars)]. 2 9
Over time, the loss of potential earnings for the average person classified as disabled
increased from 53 percent of potential earnings to 62 percent, implying that those clas-
sified as disabled are faring more poorly in the labor market in the late 1980s than in the
early 1970s. This finding has important implications for both the operation of the labor
market for workers with low skills, and for the potential demands by disabled people for
public income support. Across demographic groups, the loss of earnings capabilities as
a proportion of potential capability was greatest for females, nonwhites, older workers,
and those with low levels of schooling.30
A recent study [Krueger and Kruse (1995)] that focused on the disabled whose im-
pairment resulted from spinal cord injury found that those with computer skills did
not face an earnings loss after the onset of disability, in comparison to those without
such skills, who experienced substantial disability-related earnings losses. The study
suggests that the increasing use of computers may open opportunities for employment
29 The constancy of the aggregate loss figure results from the offsetting effects of an increase in the number
of working-age people and a decrease in the proportion of them classified as disabled. The proportion of the
working-age population with limitations was estimated to be 9.8 million in 1973 and 7.6 million in 1988.
30 The authors [Haveman et al. (1995)] also demonstrated the dependence of the loss of potential earnings on
the definition of the disabled population. Using Survey of Income and Program Participation data and defining
the disabled population as those with two or more ADLs, the size of the disabled working-age population is
close to that using the dual criterion and the CPS, and the estimate of the aggregate loss of earnings capabilities
is very similar. However, when the definition of disability is expanded to include those with lesser health
constraints, the magnitude of the estimated loss of earnings capabilities expands correspondingly. Defining
all those reporting "fair or poor health" results in a disabled population of 11.8 percent of those of working age
and a loss of aggregate earnings capabilities of $192 billion (or 7.2 percent of the total). Admitting as disabled
those with one or more limitations on activities of daily living yields a disabled population of 14.9 percent of
those of working-age, and a loss of aggregate potential earnings capabilities of $227 billion (or 8.5 percent
of total potential capability). Finally, classifying anyone who reports either poor or fair health or as having
one or more limitations on activities of daily living expands the size of the disabled working-age population
to nearly 20 percent of the population, and a loss of earnings capabilities of $285 billion (or 10.6 percent of
aggregate potential earnings capability). Chirikos (1989) and Chirikos and Nestel (1985) have also provided
estimates of the mean loss of earnings of people with health limitations.
for those with disabilities and hence reduce future losses in earnings attributable to the
onset of disabling conditions.
3. Disability policy toward working-age people: the case of the United

States
Disability policy in industrialized countries has three main goals: easing the burden
of impairments and the reduction in earnings capacity, preventing health impairments
and/or adapting jobs so that persons with physical and mental impairments can be gain-
fully employed, and restoring earnings capacity and the ability to undertake other tasks.
Lying behind these goals is an insurance motive - the reduction of the risk of loss of in-
come and the costs of medical care associated with a long term or permanent reduction
in health.3 1 The first of these objectives is the cornerstone of policy in most Western
nations, and receives the overwhelming share of resources.
Policy in the United States exemplifies this pattern. While public policy programs tar-
geted on working-age people with disabilities take several forms in the United States,
income support programs occupy the central position. The largest of these is the Social
Security Disability Insurance (SSDI) program, a central component of the nation's so-
cial insurance system. It is supplemented by the Supplemental Security Income (SSI)
program, which provides support for working-age individuals with disabilities who are
either not covered by SSDI or whose benefit levels from SSDI are very low. 3 2 Income
support is also provided to workers who have experienced injuries on the job through
the state-based Workers' Compensation system. All of these income support programs
encounter the crucial issue of how to simultaneously ensure that disabled people have
sufficient income and maintain incentive for them to continue to work or to return to
work. 33
An emphasis on maintaining, or encouraging a return to, work is central to other
policy measures that can be viewed as complements to these income support policies.
31 Concern with the potential moral hazard of public disability insurance tends to limit the earnings replace-
ment component of social insurance programs for persons with disabilities.
32 The federally based SSI program replaced a state-based system of aid (Aid to Blind and Disabled Persons)
in 1974. It is administered by the federal Social Security Administration, and has more of the characteristics
of an entitlement program than other United States welfare programs.
33 While nearly all industrialized countries have both a social insurance-based disability pension system and
a basic welfare-type program that provides benefits to some disabled people, the three-pronged United States
arrangement of social insurance based on prior work effort, compensation for work-related injuries, and public
assistance is unique. In most countries, disability has been viewed as a distinguishingcategory in the sense that
providing benefits to this category of persons is approved of by the public, based on the belief that the inability
to work is not the disabled person's fault, and that work disincentives implicit in income provision to disabled
people need be of less concern than income support to other needy groups [Diller (1996)]. Nevertheless, all
countries have struggled with the difficulties of defining the eligible population and have experienced periodic
efforts to constrain access to benefits in order to reduce work disincentives and public costs.
These include the Vocational Rehabilitation program and a variety of other public job
training programs undertaken by both national and state or local governments. Programs
directed at prevention tend to receive only limited resources or attention. These include
regulation and enforcement activities that attempt to reduce the riskiness of jobs (such
as the activities of the Occupational Safety and Health Administration, OSHA, in the
US Department of Labor), financial incentives through experience rating of programs
such as the Workers' Compensation program, and private prevention activities on the
part of employees and employers. Finally, because health concerns are dominant among
people with disabilities, the provision of health care coverage is an important issue for
both those who work and those who do not.
Appendix A outlines the principal characteristics of the primary public programs
that provide income support to disabled people in the United States. The following
sections summarize the full range of public programs targeted at disabled workers in
the United States, emphasizing issues of program coverage, size, and growth, and the
characteristics of those who are served by them.
3.1. The Social Security DisabilityIncome (SSDI) program
The SSDI program has grown rapidly since about 1960, but in distinct spurts. From 1960
to 1980, the number of disabled insured beneficiaries rose from 0.46 million to nearly
2.9 million, while benefit expenditures on disabled workers plus dependents rose from
$0.57 billion to nearly $16 billion, in current dollars. From 1980 to 1985, the number
of disabled beneficiaries declined by 7.2 percent while expenditures remained relatively
constant as efforts to reduce the disability rolls were pursued by the Reagan adminis-
tration. Since 1985, the number of recipients as well as expenditures have again grown
rapidly, by 57.5 percent, reaching nearly 4.2 million persons, and from $19 billion to
about $40 billion in 1995. This represents a 50 percent increase in real expenditures
over this period. From 1970 to 1978 the number of SSDI recipients rose from 27 to
about 40 per 1000 persons aged 15-64, remained at that level until about 1990, and
then increased to 62 by 1994.
Given the downward trend in the prevalence of reported disability among the
working-age population, this growth is surprising. Several factors have been identified,
however, including: (1) the implementation of less restrictive standards for admitting
persons with mental impairments to the rolls, 34 (2) reduction in the number of case re-
views to verify presence of disabling conditions, (3) accession to the rolls of younger
people with longer life expectancies, 3 5 (4) reduction in the number of people leaving the
34 From 1986 to 1994, the number of SSDI recipients with mental impairments increased from about 600,000
to over 1 million, from 22 percent to 29 percent of SSDI rolls. These individuals tend to be substantially
younger than the average SSDI recipient. See US General Accounting Office (1995).
35 Rupp and Scott (1996) demonstrate how the shift over the last two decades toward a lower average age
of program recipients has contributed to both the growth in SSDI rolls and to the longer duration of average
1014 R. Haveman andB. Wolfe
Table 8
Demographic composition of SSDI beneficiaries and new SSDI awardees, 1975,
1985 and 1995
1975 1985 1995
New awardees
Number (000s) 592 377.4 645.8
Age
Less than 35 years 11.0% 16.8% 13.3%
35-54 years 36.0 40.7 48.7
55-61 years 53.0 42.6 37.9
Gender
Female 32% 33% 41.4%
Male 68 67 58.4
Diagnostic group
Infectious 1% 1% 6%
Neoplasms 10 15 16
Endocrine 3 5 5
Mental disorders 11 18 22
Nervous system 7 8 8
Circulatory 32 19 14
Respiratory 7 5 5
Digestive 3 2 2
Musculoskeletal 17 13 12
Other 9 15 9
Current recipients
Number of disabled workers (000) 2489 2656.5 4185
Number of dependents (000) 1863 1251 1672
Source: US House of Representatives (1996, Tables 1-27 and 1-28).
rolls, (5) increases in general life expectancy due to more advanced medical technology,
and (6) changes in macroeconomic (unemployment) conditions. 36
Table 8 describes demographic and diagnostic characteristics of SSDI new awardees
and recipients over time. Consistent with the pattern for new awardees noted above,
benefit receipt of those currently on the rolls. They attribute a part of this shift to the bulge in population in
the 35-50 age category associated with the postwar baby boom, and suggest that future smaller cohorts in this
age range will mitigate this trend, but only slowly. For example, in 1993 the share of new awardees who were
younger than 35 (and, hence, not a part of the baby boom hump) was nearly 20 percent, which is still high
relative to earlier periods. (For example, from 1970 to 1986 the proportion of new awardees aged less than 35
ranged from 12 percent to 18 percent.)
36 Rupp and Stapleton (1995) have presented a thorough pooled cross-sectional, time-series analysis of the
growth in applications for SSDI benefits and benefit awards using state-based data on these variables for the
1980s and early 1990s. They found strong effects of macroeconomic performance variables and of efforts by
states to shift costs to the federal government. Demographic factors regarding the population age distribution,
family structure, and the AIDS epidemic were also found to have influenced the growth of program rolls
during the 1980s and early 1990s.
the average age of beneficiaries has declined over time. In 1960, the average age of an
SSDI worker was 57.2 years; by 1995, that had fallen to 49.8 years.3 7 As seen in the
pattern of new awards, the diagnostic pattern among recipients has changed over time
as well. The proportion of new awardees with circulatory system diagnosis has declined
significantly, the proportions with musculoskeletal, digestive, and respiratory diagnoses
have declined somewhat, while the proportion of those awarded benefits because of
infective problems (including AIDS), cancer-related problems, and mental disorders
has increased. This shift is consistent with the shift to younger ages of beneficiaries. 38
The amount of work by SSDI recipients is small. Among a set of new recipients in
the early 1980s, only about 10 percent had any work experience over the subsequent
ten years. Less than 3 percent of these new recipients left the rolls because of increased
earnings over that decade; only 5 percent attempted trial work that did not lead to de-
parture from the rolls [see Muller (1992)].
3.2. The Supplemental Security Income (SSI) program
The SSI program provides income support benefits to aged, blind, and disabled individ-
uals who are poor (see Appendix A).
As Table 9 indicates, the total number of SSI recipients under the age of 65 rose from
about 1.75 million in 1975 to nearly 4.3 million in 1995. Of these more than four million
recipients, 3.25 million receive only SSI benefits, while more than one million receive
both SSI and SSDI. Among those older than 18, 3.4 million receive SSI, and nearly
one million children receive SSI benefits. The number of disabled recipients who are
children has risen rapidly, from about 100,000 in 1975 to more than 900,000 in 1995. 39
Most of that increase is recent (since 1991), and reflects the revised eligibility criteria
for children as a result of the Supreme Court's Sullivan v. Zebley decision. 4 0
As Table 9 shows, among the adult recipients, more than 60 percent are eligible on
the basis of either mental retardation or some other mental disorder. Relative to the
37 Persons younger than 50 were added to the eligible population only in 1960.
38 This shift may also reduce the average monthly benefit of new beneficiaries to the extent that younger
persons are at a lower point of their earnings profile and to the extent that they gradually withdraw from the
labor market in response to a slow onset of illness [see Haveman et al. (forthcoming b)].
39 See the discussion of disability policy toward children, in Section 5.
40 The Zebley decision of 1990 ruled that the standards being applied to the determination of eligibility
among children was more restrictive than the standards applied to adults, whereas the Social Security Act
required that the standards be comparable. In response the Social Security Administration expanded eligibility
criteria for children. Children (those less than 18 years, or under 22 if a full-time student) are eligible for SSI
benefits if they are unmarried and meet the SSI disability or blindness, citizenship/residency, and income and
resources criteria. In December 1995, almost 63 percent of the SSI children were 12 years old or less, and
an estimated 20 percent of the children were under age 6. About 60 percent of the child recipients are boys,
and nearly one-half are nonwhite. In terms of diagnosis, about two-fifths are eligible because of a diagnosis
of mental retardation and another 22 percent owing to other mental disorders. See US National Commission
on Childhood Disability (1995).
Table 9
Demographic composition of federal SSI beneficiaries (disabled) less than 65
years of age
1975 1986 1995
Current recipients (000s) 1,754 2,713 4,310

Age
Less than 18 years (000) 104 231 898
18-21 years 90 138 219
22-64 years 1,559 1,787 3,193
Gender
Female 54.3%
Male 45.7
Race
White 52.8%
Black 31.3
Other 12.3
Diagnostic group 1995 Aged 18-64 Younger than 18

Infectious 1.7% 0.3%
Neoplasms 1.5 1.5
Endocrine 4.1 1.1
Mental retardation 28.2 39
Other mental disorders 31.2 22
Nervous system 10.1 12
Circulatory 5.0 0.6
Respiratory 2.6 2.5
Musculoskeletal 7.1 1.2
Other 8.4 19.6
Source: US House of Representatives (1996, Tables 4-12, 4-13, and 4-15).
general population, a far higher proportion of recipients of SSI are black (31.3 percent
of the SSI population, but 11.5 percent of the general population in 1995), and slightly
more are women. 41 Nearly all of the growth in the number of recipients since 1991 is
attributable to adults with mental impairments, children, and noncitizens. These groups
grew at an annual average rate of 11.0, 16.4, and 15.5 percent, respectively, from 1986
through 1993, compared with just under 5 percent for all SSI recipients [US General
Accounting Office (1995)]. The GAO report found that recent SSI awardees tend to
be younger, stay on SSI longer, receive larger benefits, and depend more on SSI as a
primary source of income than those awarded benefits earlier in the program's life.
41 As of January 1, 1997, about 35,000 persons previously eligible for SSI benefits solely because they were
disabled due to substance abuse are no longer eligible.
Ch. 18: The Economics of Disability andDisability Policy 1017
3.3. The Workers' Compensationprogram andjob-relatedhandicaps
The Workers' Compensation (WC) program is a mandatory state-based program which

provides annuity or lump-sum benefit awards to temporary, permanent partial, and total
disablement injuries that arise out of or occurred during the course of employment. It
is a true insurance program, financed by individual employers whose premia (or self-
insurance costs) are related to their own job-related injury experiences. 42 Like SSDI and
SSI, WC benefits are untaxed at any level of government, and include medical care cov-
erage for the impairment leading to the claim. 4 3 In 1995, WC programs covered some
113 million workers, a covered payroll of $2.5 trillion, and paid out nearly $44 billion
in benefits to more than 1.5 million families (including about $1.4 billion under fed-
eral Black Lung benefit programs) [US House of Representatives, Committee on Ways
and Means (1996, pp. 949, 950)]. In the early 1990s about 14 percent of recipients of
WC also received benefits from SSDI or SSI while a much smaller proportion (about
3 percent) received mean tested cash benefits targeted on persons of very low income
[Burkhauser and Wittenberg (1996)].
3.4. Accident and Injury Preventionprograms
As in other industrialized nations, the United States undertakes extensive regulatory

efforts designed to reduce the frequency and severity of accidents and injury at the
workplace. These programs come under the jurisdiction of the Occupational Safety and
Health Administration (OSHA) of the US Department of Labor, and focus both on pre-
vention in terms of setting minimum standards to protect workers and a financial penalty
on firms who have high rates of workplace injury. Little is known about their effective-
ness.4 4
42 Employers can self-insure if they are large, or can purchase coverage individually or in groups from private
insurers (or in some states, the state in competition with private insurers). In all cases, the premia paid are
related to the claims cost experience of the employer (policy holder) and to the industry of the employer, al-
though the basis for experience rating varies substantially across the various insurance categories [see Worrall
and Butler (1986)]. These programs vary across states in the US with insurance provided by private insurance
companies, public (state) funds and self-insured firms. As of the mid 1990s, private insurance firms accounted
for about 55 percent of the market; self insurers for about 22 percent.
43 Medical care coverage for the work-related impairment is virtually unlimited in the WC program, and as
of 1993, over 40 percent of all benefits paid were for medical care.
44 Gray and Jones (1991) attempt to study the impact of such programs for the US manufacturing sector.
Lanoie and Str6Liski (1995) have studied a similar program in the province of Quebec (Canada) over the
1983-1990 period. They suggest that the financial incentives of a workers' compensation program that is
experience rated has some impact on the prevention of accidents, as does the use of inspections in risky
sectors. The overall impact is described as "minor", however. See Butler (1995) for a review of the incentives
for safety in worker's compensation programs.
3.5. Vocational Training, Rehabilitation,and Return-to-Workprograms
Public provision of rehabilitation and training to disabled workers - programs consistent

with the objective of recovery of earnings capacity - are also common components
of disability policy in industrialized nations. These programs are designed to increase
productivity and the return to work of those experiencing disabling conditions, and are
intended to offset the underprovision of rehabilitation and training services by firms and
individuals that results from the absence of experience rating and the presence of moral
hazard.45 If the programs are successful, disabled people will work and earn more and
will tend to rely less on public income support, hence avoiding some of the adverse net
welfare effects accruing from reduced labor supply described above. Estimating the net
welfare effect of these efforts requires application of standard benefit-cost techniques to
potential increases in work and productivity (and the associated reductions in leisure)
and the associated labor and capital costs required of the programs.
In the United States, the Vocational Rehabilitation (VR) program is comprised of
state-operated programs providing rehabilitation services to people with disabilities.
Current spending on the VR program is in excess of $2 billion, about 80 percent of
which comes from federal government sources, and equals less than 4 percent of total
federal spending on cash transfers for working-age people with disabilities, down from
6 percent in 1975.46
The VR programs offer a wide range of services to about one million people
with disabilities each year, including diagnosis, assessment, medical treatment, edu-
cation/training, counseling and job placement.4 7 Potential clients are referred to the
program by health care providers and representatives of benefit programs. After assess-
ment, the VR program either accepts or rejects the client, who in turn can either choose
to participate in the program or not. The services provided vary widely across clients,
and are tailored to meet the work or living requirements of the client. 48
The numerous efforts to evaluate the program, often using benefit-cost techniques,
have indicated modest employment and earnings impacts. 4 9 Earnings gains from clients
who have been placed in jobs are estimated to be from $1,000 to $2,000 per year. This
compares to an average cost of about $10,500 per SSDI participant whose VR claim has
been approved.
45 The issues here are similar to the provision of training in firms more generally. Firms that provide training
may lose the most successful "students" to other firms.
46 See the discussion of SSDI efforts to link income support with rehabilitation, described in the Appendix.
47 Barnow (1996) reports that less than one-half of VR clients are given any type of training or education
services.
48 Rehabilitation programs may also provide for corrective devices such as wheelchairs and accommoda-
tions for the workplace, including specialized computers or transportation devices that allow an individual to
commute from home to work.
49 References to these early studies are in Dean and Dolan (1991), who provide their own more positive
evaluation. See also US General Accounting Office (1993).
In addition to the VR program, a number of other programs provide work-related

training and education services to people with disabilities. These include the Job Train-
ing Partnership Act Title II programs (about 10 percent of the adults served and nearly
20 percent of the youths served are people with disabilities), state-based Vocational Ed-
ucation programs (nearly 13 percent of whose clients are people with disabilities), and
the Employment Service (less than 3 percent of whose clients are people with disabili-
ties).5 0
In contrast to the United States, both Germany and Sweden have sizable vocational
rehabilitation programs. The program in Sweden includes the provision of benefits to
workers in rehabilitation that are significantly higher than those paid to disabled workers
who are not employed; in this case workers in rehabilitation receive 90 percent of their
lost earnings. The training is conducted by labor market institutes, which also train per-
sons who are not disabled. Increasing resources have been shifted toward rehabilitation
and away from cash benefits. In Germany, large amounts are also spent on rehabilita-
tion, and employers are subsidized to adapt jobs for the handicapped. New programs
in the United Kingdom are also designed to increase employment. The rehabilitation
program in the United Kingdom is designed to encourage volunteer and part-time work,
allowing a disabled individual to continue to receive benefits while they undertake such
activities.
3.6. Health InsuranceAspects of Disabilityprograms
In most industrialized countries, the health care needs of persons with disabilities tend to
be covered by some form of universal health insurance. In the United States, however,
universal health care coverage does not exist, and most coverage is employer-based.
As a result, persons who are no longer employed, such as disabled workers, would
be required to pay the premiums in order to maintain or obtain coverage. Moreover,
because this private sector coverage is experience-rated, firms will be reluctant to offer
or to continue employment of a worker with an impairment that is likely to lead to
sizable medical care costs.
For these reasons, a special public program fills this gap for disabled persons in the
United States. Recipients of benefits from the primary cash-transfer disability programs
(SSDI, SSI, and WC) receive health insurance coverage as part of the benefit package. 5 1
In some cases, these medical benefits are of substantial value and influence the desir-
ability of establishing eligibility and remaining eligible for disability-related benefits.
50 These statistics are from Barnow (1996) who reviews these non-VR programs providing employment-
related services to people with disabilities, emphasizing the lack of integration among them and inconsisten-
cies in the definitions of disability that are used.
51 SSDI recipients are covered by Medicare after 24 months; SSI recipients are immediately covered by
Medicaid, and the WC programs pay directly for medical care related to the occupationally based injury. Both
Medicare and Medicaid are publicly supported entitlement programs that provide health insurance. Medicare
is a national health insurance program for the aged and certain disabled persons. Medicaid provides coverage
under a federal-state program for low-income persons who are disabled (and certain other groups).
1020 R. Haveman and B. WolJ
Expenditures for medical care associated with disability program benefit receipt are
very large and increasing. About 36 percent of the sum of cash plus health care ex-
penditures on SSDI recipients, $20 billion (in 1994), is for Medicare expenditures. For
SSI blind and disabled recipients, the situation is even more extreme: more than two-
thirds of the sum of cash plus health care expenditures, or $43 billion (in 1994), is
accounted for by Medicaid expenditures (authors' calculations). For the WC programs,
medical and hospitalization costs are about $18 billion per year, and account for more
than 40 percent of total WC expenditures. In total, more than $80 billion per year of pub-
lic health care expenditures are devoted to recipients of public cash transfer programs
targeted on disabled working-age people. 52
The level of medical care expenditures on behalf of disability transfer program re-
cipients has grown to these large sums largely because of growth in the proportion of
recipients who are using public medical care benefits. For example, the proportion of
SSDI recipients using Medicare has grown from less than 50 percent in 1975 to more
than 75 percent in 1994. While the average annual Medicare expenditure per SSDI re-
cipient was $1,548 in 1975, by 1994 it had grown to more than $6,000. 5 3 Similarly,
the average annual Medicaid expenditure per SSI recipient grew from $800 in 1972 to
$7,735 in 1994. Although only 16 percent of Medicaid recipients are SSI beneficiaries,
more than 40 percent of Medicaid "vendor costs" were devoted to this population.
4. The economics of policies toward working-age people with disabilities
Disability policy measures create incentives for a variety of individual behaviors, in-
cluding work, benefit receipt, fertility, and the structure of living arrangements. They
also affect the structure of risks and the organization of markets in a number of areas,
especially those providing insurance, medical care, and health maintenance services.
Because such measures are publicly financed, the size and composition of public ex-
penditures and taxes are also affected. As a result of these incentives, market impacts,
and fiscal allocations, the distribution of income and well-being among households re-
flects the effect of disability policy. In this section, we identify these impacts and discuss
the research that has been done on them.
4.1. Disabilitypolicy and its economic effects: a roadmap
Understanding the economic effects of disability policy requires that a counterfactual to

the world with these policy measures in effect be stipulated and evaluated. 54 Setting out
52 See US House of Representatives, Committee on Ways and Means (1996).

53 This level of per capita expenditure is 30 percent higher than for the elderly Medicare population. Statistics
cited are from US House of Representatives, Committee on Ways and Means, various years.
54 Danzon (1993) has discussed the need for identifying the appropriate counterfactual in assessing the effects
of a particular constellation of interventions.
this benchmark is difficult. One possible alternative would be a full information, first
best optimum, but this option seems both unrealistic and infeasible. 5 5 A more realistic
alternative is a world in which private insurance would exist without public interven-
tions, but with actual amounts of private information (and hence adverse selection and
moral hazard).56 To the extent that public interventions simply duplicate the private pro-
visions in such a world, there would be no net effects of the public measures. However,
there are several reasons why the structure of public measures will differ substantially
from those in this counterfactual, including distributional objectives and adverse selec-
tion and moral hazard effects. Hence, disability policy will have behavioral effects with
welfare implications.
Disability policy encompasses measures that are designed to provide income support
to people with disabilities, to enable disabled workers to return to economically produc-
tive and normal lives, and to regulate wage rates paid to disabled workers, the provision
of accommodations, and prevention of discrimination in hiring. These policies also es-
tablish a screen - and screening mechanism - to determine who is, and who is not,
eligible for disability-based benefits. 5 7
Consider, first, income support in the form of social disability insurance, or cash
transfers to people with disabilities. The provision of disability-related income support
will affect the amount and kind of work activities in which people engage, that is, it
creates a moral hazard. By providing support to individuals conditional on their having
reduced earnings capacity, and in fact earning little, cash and in-kind benefit measures
have an income effect which encourages the substitution of nonwork for work. Simi-
larly, because disability income support tends to be withdrawn as earnings and other
income rise, incentives for reduced work hours by disabled workers, conditional on
working, are created. Finally, to the extent that public disability insurance is financed
by distorting payroll taxes, taxpayers in general will reduce their work effort, and these
changes must be taken into account as well. However, the net welfare effect of these
productivity changes must also account for the benefits in the form of increased leisure
time accruing to those induced to reduce work effort. Because the private counterfac-
tual to public provision will also have adverse work incentives, it is the response to the
55 Information asymmetries faced by potential private insurers including adverse selection and moral hazard
generally preclude the existence of a private market that would insure against the loss of earnings.
56 Aarts and de Jong (1998) develop and estimate a model of the private demand for disability insurance
which covers earnings losses through age 65. The model is based on a simple utility maximization framework
and shows that risk-averse workers would buy full coverage of expected earnings loss if charged an actuarially
fair premium. Comparing this to a social insurance scheme based on the Dutch system, they find that about
50 percent of the population would leave a social insurance scheme that charges an actuarially fair but flat rate
if permitted to do so. Those that leave are at lower risk (adverse selection); the result is that the contribution
rate must increase by about 50 percent for the fifty percent that remain. Moral hazard is not taken into account
in their model or simulations.
57 Diamond and Sheshinki (1995) discuss optimal program design including screening processes. They argue
that to the extent that a screen does in fact include those with significant disabilities and exclude those without
such disabilities, it may be efficient. Unfortunately, the cost of the screen is not included in the calculation.
net increment in disincentives that must be considered. Given imperfect information (as
reflected in the difficulty in accurately measuring disability) and moral hazard (such as
incentives to claim disability among those with limited labor market success), there is a
presumption that, especially for low-wage workers, public benefits will exceed private
provision and net disincentives will be associated with the public program.
The public provision of disability insurance is likely to have a number of other effects
on well-being. Consider, first, some additional costs associated with the public provision
of disability insurance in addition to these work-effort related effects. To the extent
that the financing of public disability benefits fails to provide experience-based rates to
private employers (relative to those provided by the counterfactual), the incentives for
preventing the onset of disabling conditions among workers, or for engaging in post-
disability rehabilitationwill be reduced. Finally, to the extent that the applicationof
eligibility criteriawill be less effective and more permissive than in the counterfactual,
58
the public program will attain supra-optimal size, and additional costs.
A variety of benefits can also be associated with public disability insurance provision.
First, assuming the public insurance program is mandatory, the costs associated with ad-
verse selection risks that both limit private insurance coverage and increase its premia
can be avoided, and these are risk-reduction benefits accruing to all covered citizens
to be attributed to the intervention. Moreover, because public insurance benefits are in-
dexed while private benefits are not, additional "indexation" benefits are to be assigned
to public disability income support [see Bernheim (1987)]. An additional source of ben-
efit may come from the equity gains assigned by society to the intragenerational transfer
59
of income from high lifetime income people to those with lower expected incomes.
In addition to policies providing income support to disabled people, there are policies
designed to enable disabled workers to return to work. These efforts, and the assessment
of their effects, have been described above.
Finally, policy measures are undertaken with the aim of attaining equal wages for
disabled and nondisabled workers (with comparable productivities), mandating cost-
effective accommodation investments, and prohibiting discriminatory hiring policies.
58 This is related to the increased incentive that individuals face to "fake" disability (or to incur disability)
when the provision of public insurance affords a higher level of consumption when disabled than when not,
relative to the counterfactual. A series of recent analyses have addressed the issue of the administrative costs
involved when imperfect disability evaluation by those responsible for "tagging" some applicants for benefits
as eligible and others ineligible. Because tagging is imperfect, and involves both type I and type II errors,
an agency problem exists between the government and those responsible for eligibility determination, who
possess private information regarding their decisions. These analyses attempt to identify an optimal set of
arrangements involving benefit levels and administration costs associated with work requirements and direct
monitoring of those responsible for eligibility determination under a variety of conditions regarding the pres-
ence of alternative income transfer programs, public objectives, and access to information. See Diamond and
Sheshinski (1995), Parsons (1996), and Boadway, Marceau, and Sato (1997).
59 This discussion neglects the potential intergenerational effects of public disability insurance and the dis-
tortions that accompany pay-as-you-go financing of social insurance programs, both of which are likely to
be negative. The unresolved question of the effects of pay-as-you-go financing of social insurance on aggre-
gate national saving is debated in Feldstein (1982) and Leimer and Lesnoy (1982). It is also part of recent
discussions of reform of social insurance in the US.
Each of these regulatory efforts will have its own particular costs and benefits of the
sorts described above, and these will depend upon assumptions regarding information
imperfections, the ability of firms to offer long-term contracts, the elasticities of the
labor supply response to net wages, the productivity of nonmarket time, the abilities
of firms to monitor disability status and productivity, and the investment costs of the
resources devoted to these activities. 6 0
4.2. The labor supply effects of disabilitybenefits
When jobs are scarce or pay is low or when disability-related cash transfers are raised or
tied benefits-in-kind are increased, there is a clear incentive for individuals to apply for
disability-based benefits and correspondingly, to reduce their labor force participation.
The stringency of the screen to become eligible for disability-based benefits may also
play a role in an individual's decision to apply for benefits and to reduce his or her labor
force participation. Increased uncertainty of future labor market opportunities may act
in a similar way leading to a decrease in labor force participation.6 1
Accompanying the increase in the value of SSDI benefits and the number of recip-
ients of them over the period from 1965 to the 1980s (described above) was a rapid
decrease in the labor force participation rate of older workers. The growth in the SSDI
rolls from 450,000 to 2.9 million from 1960 to 1980 (and the 50 percent increase in the
wage replacement rate of disabled worker benefits over this period), was accompanied
by a threefold increase, from 4 percent to nearly 12 percent, in the labor force nonpartic-
ipation rate of workers aged 45 to 59. The presumption of a tie between these variables
is obvious.
Numerous researchers have attempted to determine the magnitude of the relationship
between the generosity and accessibility of disability benefits and this increase in the
rolls (and the simultaneous decrease in older male labor force participation). This lit-
erature has employed cross-sectional, and often longitudinal, household survey data to
measure the effect of the level of disability-related benefits on presence on the rolls (or
labor force participation, its virtual complement). Estimation is usually based on a one-
period, static model, in which working-age individuals decide whether or not to work
on the basis of their expectations regarding disability benefits and labor market oppor-
tunities. In the estimation, the expected value of benefits is imputed for those not on the
rolls, and a wide variety of individual attributes (e.g., race, age, education) and regional
characteristics are employed to characterize labor market expectations.
These studies rely on cross-sectional variation in the expected value of disability ben-
efits, and obtain estimates of the effect on program participation (labor force nonpartic-
ipation) of this variable. Table 10 summarizes the principal studies of this relationship.
60 Danzon (1993) presents a discussion of the economic effects of such regulatory measures using a model
of optimal insurance, post-injury accommodations, and labor supply.
61 See Kreider (1998, forthcoming) for a fuller discussion of the role of uncertainty in influencing the choice
to apply for benefits.
1024 R. Havemnan and B. Wolfe
Table 10
Labor force effects of disability-related benefit levels
Study Data set Sample analyzed Results
Leonard (1979) Social Security Survey 1,685 men aged 45-54 Elasticity of beneficiary
of Health and Working in 1972 status = 0.35
Conditions
Parsons (1980a) National Longitudinal 3,219 men aged 48-62 Elasticity of labor force
Survey in 1969 nonparticipation = 0.63
Parsons (1980b) National Longitudinal 4,831 men aged 45-59 Elasticity of labor force
Survey in 1996 nonparticipation = 1.80
Slade (1984) Longitudinal 5,403 men aged 58-63 Elasticity of labor force
Retirement History in 1969 nonparticipation = 0.81
Survey
Haveman and Wolfe Panel Study of Income 741 men aged 45-62 Elasticity of labor force
(1984b) Dynamics in 1978 nonparticipation =
0.06-0.21
Haveman, de Jong, and Panel Study of Income 2,163 men aged 45-62 Elasticity of labor force
Wolfe (1991) Dynamics in 1978 nonparticipation = 0.97
(single female heads) =
0.23 (married women)
Gruber (1996) Canadian Labor Force Men 45-59 1985-89 Elasticity of labor force
Survey nonparticipation =
0.25-0.32
As seen there, the implied elasticities vary widely. When they are used in simulation
analyses of predicted responses to observed benefit changes, the largest parameter es-
timates or elasticities [Parsons (1980a, 1980b), Slade (1984)] attribute virtually all of
the decrease in older male labor force participation rates during the 1970s to changes in
benefit levels. The remaining estimates are able to "explain" a portion of the observed
participation change, but not all of it.62
The estimates from these studies rest on a number of assumptions, and these deter-
mine their relative (and absolute) strengths and weaknesses. All of the studies ignore
both cohort effects (and the larger payout available to workers who take advantage of
62 Bound (1989) and Leonard (1986) review and critique several of these studies. An important issue in inter-
preting the findings of these studies is the positive and redistributive relationship between expected disability
transfers and past earnings history. Finding that workers with higher replacement rates are more likely to be
labor force nonparticipants may only reflect the fact that such low-wage workers may have a greater desire to
cease working.
benefit availability early in the life of these programs) and life-cycle effects, such as
the loss of future human capital or pension accrual associated with the disability benefit
recipiency (labor force nonparticipation) option. Like all estimates of the determinants
of dichotomous individual choice behavior, researchers are able to observe utilities or
incomes in only one of the states being evaluated; utility or income in the unobserved
state must be estimated, and the estimate is sensitive to the procedures used. For exam-
ple, while the true value of unobserved disability benefits for those working depends
upon individual wage history, family structure, and the idiosyncracies of the benefit
structure, the estimates employed in the studies fail to distinguish one or more of these
determinants, sometimes in persistent ways leading to biased estimates. 63
In much the same vein, the replacement rate (or the separate values of expected dis-
ability transfers and labor market incomes) used may fail to reflect: (1) the value of
Medicare coverage if a disability transfer recipient, (2) the value of fringe benefits if a
labor force participant, (3) the substitute relationship between expected public disabil-
ity transfers and private insurance coverage, and (4) the ability to combine Workers'
Compensation and Supplemental Security Income benefits with Social Security Dis-
ability Insurance benefits. If these neglected values are correlated in particular ways
with the program or labor market values used in estimation, the coefficient on the ef-
fects of expected disability transfers on nonparticipation may be biased either upward
or downward.
In addition to opportunities available in the labor market and as a disability recipient,
individual health status is an important determinant of the choice between these options.
The measures of health status used in these studies are primarily self-reported indica-
tors of limitations on the ability to work. This indicator is a noisy proxy of true health
status; more seriously, it may be either endogenous to the choice itself or insufficiently
accurate to identify those individuals with disabling conditions so severe as to eliminate
the choice of the work option (an attrition bias issue). 6 4
More recent estimates of the effect of available disability benefits on the work effort
of older men have adopted different approaches. Bound (1989) uses the labor force par-
ticipation rates of rejected applicants for disability insurance as an upper bound for the
rates of recipients were they not on the rolls, and concludes from this that growth in the
SSDI program can explain no more than 40 percent of the increase in the older male
nonparticipation rate. As he recognizes, this procedure rests on the dubious assumption
that the withdrawal from the labor force required for applying for SSDI benefits has
63 See, for example, the criticism of Parsons' estimates by Haveman and Wolfe (1984a, 1984b).
64 Danzon (1993) discusses these assumptions and procedures and attempts to identify the likely direction
by which they may bias the coefficient(s) describing the response of individuals to the generosity and ac-
cessibility of public disability benefits. Dwyer and Mitchell (1999) explore whether self-rated health status
measures are endogenous in the decision to leave the labor force. They also explore the use of a wide variety
of alternative measures, including one based on mortality risk of an individual's diagnosis and two based on
measures of activity limitations.
no effects on subsequent work effort of these rejected applicants. 65 Halpern and Haus-
man (1986), Parsons (1991b), and Gruber and Kubik (1997) measure the response of
older workers to over-time or across-state denial rates in the SSDI program, treating
the uncertainty of benefits reflected in these rates as a policy tool designed to reduce
moral hazard in the decision to leave the work force. Kreider (1998) adds a measure of
prior earnings volatility to his analysis of the same issue. All of them find a sizable rela-
tionship between denial rates and either nonparticipation rates or applications to SSDI
benefits, although the strength of the relationship varies across studies 66 Gruber (1996)
translates these rejection rates into equivalent benefit generosity levels, and concludes
that the response to changes in denial rates implies an elasticity of nonparticipation
with respect to benefit generosity of from 0.12 to 0.17, which is at the low end of the
estimates in Table 10.
Finally, Gruber (1996) uses a large shift in disability benefit generosity in the non-
Quebec (Canada) disability benefit program relative to the otherwise equivalent Que-
bec disability benefit program to evaluate the labor supply response to benefit levels.
He concludes that the elasticity of labor force nonparticipation with respect to disabil-
ity benefit levels is from 0.25 to 0.32, which estimates derive from both difference-in-
difference and parameterized model estimates. This estimate is also at the low end of
those presented in Table 10, but it too rests on several assumptions, in particular that the
estimating procedure has effectively controlled for differences between Quebec and the
remainder of Canada in any labor market or other policy variables that could influence
the work decisions of older men.
A related issue is the role of disability related transfers and other programs on en-
couraging or discouraging return-to-work of beneficiaries. The rate of return to leaving
beneficiary status appears to be quite low in all countries with disability transfer sys-
tems. Certain factors may lead to higher rates of return, including: (1) an official benefi-
ciary review system in which disability determination is reviewed at regular intervals for
some beneficiaries or those who might be expected to recover based upon the original
diagnosis, (2) a reduction in the marginal tax rates associated with a return to work (in-
cluding continuation of disability-related medical insurance, provision of work-related
cash payments or subsidies, and a promise of immediate recertification of eligibility if
the condition worsens), and (3) provision of the services of vocational rehabilitation and
employment programs.
There is some time series evidence from the US indicating that disability review poli-
cies play a role in encouraging a return to the work force - or at least termination of ben-
efit receipt. Dykacz and Hennessy (1989) estimate that on average 11 percent of a given
cohort of beneficiaries eventually returns to work, with the rate varying inversely with
age at which eligibility is determined. Two experiments, the Transitional Employment
65 See Parsons (1991a) and Bound (1991) for extensive discussion of the validity of this assumption.
66 An exception to this are the results for women estimated by Riphahn and Kreider (1997). They find the
response of women to a reduction in the probability of acceptance to be less than a quarter of men's response.
However, there is some question as to whether they have appropriately specified the opportunity set of women.
Training Demonstration and Project Network, which provided extensive rehabilitation

and employment services, found that only very small percentages of beneficiaries are
likely to return to work even with such services. The experience of the private sector
is more encouraging in terms of the probability of return to the work force. This may
however reflect selectivity in who is covered, and the availability of generous benefits.
On the basis of all of this evidence, and in spite of the wide variance of elasticity
estimates, it seems clear that older workers do respond to the incentives to cease working
that are provided by disability-related income transfer programs. However, the response
elasticities reported by the later, and more reliable, studies are rather modest. Applied to
the large benefit increases recorded over the 1960 to 1980 period, the choices made in
response to expected benefits can explain only about 10 to 20 percent of the increase in
nonparticipation recorded over this period [see Gruber (1996) and Bound and Waidman
(1992)].
The structure of WC programs also contain a variety of incentives, many of which
may generate responses that increase the costs of the program. For example, the pro-
grams have a waiting period before which benefits can be claimed; hence, workers have
incentive to extend the duration of nonwork spells beyond the waiting period in order
to qualify for indemnity benefits. Moreover, if the injury is sufficiently serious in terms
of work days lost, benefits covering the waiting period will be paid, and again there is
incentive to prolong the duration of nonwork. While seriously injured WC recipients
are eligible for vocational rehabilitation services, the effect of successful rehabilitation
is to reduce the probability of becoming classified in a higher-benefit, permanent partial
disability category; there is incentive either to forgo the use of rehabilitation services
or to fail to take advantage of their potential effectiveness. As with SSDI and SSI, the
level of benefits alone creates incentives to either reduce injury-prevention actions or to
promote claims filing. Finally, the large number of definitional uncertainties and possi-
bilities for conflict in the determination of eligibility, classification of eligible disabled
workers into the various categories, and possible criteria for determining benefit levels
that inhere in the WC program encourage the substantial amount of litigation and legal
costs that surround the program. 6 7
The behavioral responses to many of these incentives have been studied, especially
the effects of the WC benefit levels on injury rates and claims frequency. The consistent
finding from these studies is that both of these variables are positively and strongly
related to the level of available WC benefits. Although the degree of responsiveness
estimated by these studies is large, the estimated elasticities have a wide range. Far
less research has been done on the effect of: (1) experience rating on employers' safety
records or investments in work place safety, (2) benefit opportunities associated with
67 Worrall and Butler (1986) describe the situation as follows: "When we introduce uncertainty, lack of
perfect information, insurers' incentives and the possibility of profit, monitoring costs, conflicting medical
testimony, injured and angry workers, regulators, political power blocs, alternative recovery mechanisms, and
a workers' compensation bar (to name a few), we have the workers' compensation system". They describe
the system as being "rife with the fertilizer to grow litigation" (p. 112).
1028 R. Haveman andB. Wolfe
a delay in the return to work on the duration of work loss, or (3) program structure
and benefit levels on the level and effectiveness of the use of vocational rehabilitation
services. An assessment of the research that has been done on these issues concludes
that, while these incentives for inefficient use of the program are substantial, it seems
unlikely that the responses of firms or workers are as large or pervasive as the relatively
large response of injury rates and claims frequency [see Worrall and Butler (1986),
Butler (1995)].
A related issue is whether or not the employer provides some accommodation to the
employee's limitation. Such accommodation would permit an employee to stay working
for the employer, though perhaps in a position that is less demanding and less rewarding
monetarily. Empirical research suggests that such accommodations do play a role; for
example, Burkhauser, Butler, and Kim (1995) find that workers who report such accom-
modation were about twice as likely to be working for their employer after the onset of
impairment, than workers who did not report such accommodation. However, this may
reflect favorable selection; that is, employers may provide accommodation for workers
where such accommodation is likely to pay off - where either it is inexpensive or the
worker is difficult to replace.
4.3. Cash benefits, economic well-being, andpoverty reduction
While the demographic and health characteristics of the disabled working-age popula-
tion would seem to make them likely candidates for being classified as poor, numerous
and sizable programs providing income support tend to offset the low earnings capacity
of people with disabilities. Moreover, because poverty is measured on a household ba-
sis, rather than on an individual basis, many disabled people escape the risks of poverty
status by having a high-earning spouse or other relative in their family.
A rough picture of the effectiveness of the cash transfer programs in offsetting the
lack of earnings capability of people with disabilities is provided in Table 11. A single
person with disabilities and with no earnings or other income would be assured of cash
income plus food stamps equal to nearly 90 percent of the poverty line in 1996. This
percentage has been increasing slightly since the late 1980s, but is below the 91.4 per-
cent figure attained in 1975. Eligible couples receiving SSI and other benefits fare better
Table 11
Cash benefits for the disabled (in 1996 dollars and as percentage of the poverty line)
Single individual Married couple
Poverty line $7,309 $9,221

SSI benefits (maximum) 5,643 (77.2%) 8,460 (91.7%)
SSI benefits plus SSDI 5,876 (80.4%) 8,700 (94.3%)
SSI benefits plus SSDI plus food stamps 6,380 (87.4%) 9,540 (103.4%)
Source: US House of Representatives (1996, Tables 4-9, 4-10).
Table 12
Antipoverty effectiveness of cash transfers to the disabled
Men Women
1982 1991 1982 1991
Percentage in poverty with

no cash transfers 60 62 63 72
Percentage in poverty after
SSDI and SSI receipt 21 17 25 25
Percentage of pretransfer poor
removed from poverty by SSDI and SSI 65 73 60 65
Source: Haveman et al. (forthcoming a, forthcoming b).
relative to the poverty line than do single individuals. Receipt of the full complement of
benefits would raise the couple above the official poverty line.
These estimates suggest that the simultaneous receipt of benefits is important in se-
curing levels of well-being above the poverty line for people with disabilities, and raises
the question of the extent of multiple benefit recipiency. Some evidence on this is pro-
vided in an analysis of the respondents to the Social Security Survey of Disability and
Work in 1978. Nearly 40 percent of all of the WC recipients also received benefits from
another income support program, most often SSDI. Of this 38 percent, nearly 40 percent
received benefits from two or more income support programs other than WC. 6 8
In two recent studies, Haveman et al. (forthcoming a, forthcoming b) employed data
from the 1982 Social Security New Beneficiary Survey to measure the effectiveness of
public cash transfers in removing disabled people from poverty. Table 12 summarizes
their findings concerning disabled men and women who were awarded SSDI and SSI
benefits in 1982, and who were reinterviewed in 1991.
These estimates indicate that, for this group of severely disabled individuals - those
judged to be unable to undertake any gainful activity - poverty rates (excluding all in-
kind benefits) ranged from 17 to 25 percent. These are to be compared to the national
poverty rate of 14.2 and 15 percent respectively for the years for which the tabulations
were made. For this group of very low-earnings-capacity people, cash benefits from
the two social security programs raised about two-thirds of the families from below to
above the official poverty line.6 9 Hence, while people with disabilities have a somewhat
68 This is taken from Worrall and Butler (1986). The programs other than WC from which benefits could be
received include SSDI, SSI, Veterans' benefits, public assistance, unemployment insurance, civil service, and
Aid to Families with Dependent Children.
69 This general result is consistent with an estimate of the risk of being in poverty of men with disabilities
(using the two-part definition of disability described above and Current Population Survey data) relative to
nondisabled men. In 1983, a disabled working-age man had a risk of being in poverty, after accounting for
all of the income sources in his family, that was 137 percent of that of a nondisabled man; by 1987, this
percentage had increased to 171 percent. See Burkhauser, Haveman, and Wolfe (1993).
1030 R. Havenian and B. Wolfe
higher than average chance of being in poverty than do those without disabilities, the
system of cash transfers appears to be generally effective in raising well-being to close-
to-poverty-line levels for most severely disabled people.
4.4. Training, rehabilitation,and the return to work
A current priority in most industrialized countries is to reduce the proportion of the

working-age population who are no longer working owing to an impairment. This is
particularly the case when considering younger persons who have a substantial number
70
of remaining years of potentially productive employment. As noted above, the prob-
71
lem is that prior vocational efforts have been largely unsuccessful. Continuing reviews
of eligibility for income transfer programs have been one approach to encouraging de-
partures from the programs, but these have met with a good deal of resistance as well as
little success in removing persons from these cash transfer programs.
Recent discussions suggest a multifaceted approach, combining a variety of finan-
cial incentives with improved opportunities to acquire medical care insurance. First, an
earned income tax credit would encourage disabled persons to work or return to work.
The core idea is to make work more financially rewarding. Second, persons would be
given a voucher to be used for training that they believe will be most advantageous to
them. This would encourage the setting up of new training operations as an expanded
market for such training is introduced. Third, the payment made to the training agency
accepting the voucher could be based on success. In one particular approach, recom-
mended by Monroe Berkowitz (1997), and adopted by the Disability Policy Panel of
the National Academy of Social Insurance, the payment would be a proportion of saved
social insurance payments as individuals trained by the enterprise no longer received
(full) benefits. Fourth, public medical care insurance would be made available to those
with partial or full disabilities, perhaps on a sliding fee basis.
70 In the United States there is rising interest in this issue, as both SSDI and SSI disability rolls increasingly
are composed of a significant number of younger individuals. A recent governmental report [US General
Accounting Office (1997)] states that "among working-age SSDI and SSI beneficiaries, one out of three is
under the age of 40". In addition, in 1993, 35 percent of the 84,000 SSDI beneficiaries who responded to an
Social Security Administration questionnaire expressed an interest in receiving rehabilitation or other services
that could help them return to work, an indication of motivation. Moreover, a substantial portion - almost one
in two - of a cohort of SSDI beneficiaries had a high school degree or some years of education beyond high
school.
71 For example, the primary training agencies in the United States are state vocational rehabilitation agencies.
The US General Accounting Office (1993) evaluated the results of these programs by examining the employ-
ment status of clients (including SSDI beneficiaries) over an 8-year period following receipt of services. They
found that among the most successful - those who had been placed in employment for at least 60 days -
earnings returned to near the pre-VR program level after 8 years. There was also evidence of selection: clients
who had been successfully rehabilitated had better work and earnings histories than clients who had dropped
out of the program. Clients who had received many but not all of the services that rehabilitated clients had
received did no better in later employment and earnings than VR dropouts who had received no services after
an initial VR evaluation.
A demonstration project, Project Network, tested a program in which the training

programs are rewarded based on their success (a variation of point three, above).72 Pre-
liminary results suggest but limited success; less than 3 percent of those enrolled in
the treatment group successfully acquired jobs. Even this proportion may be an over-
estimate: evaluation of an Ontario (Canada) program indicated that 40 percent of those
who returned to work left employment within a few years because of longer-term health
problems, and another 21 percent had to take time from work owing to their disabling
health condition. This combination of low initial success and subsequent failure makes
it difficult to see how such programs could be justified on benefit-cost grounds. It also
means that entrepreneurs may not establish a variety of training programs given the
low expected payout associated with the poor employment prospects of persons with
disabilities. 73
4.5. Public health care and in-kind benefits
As we have described, demands on the health care system by severely disabled individ-
uals are substantially greater than those of the nondisabled working-age population, and
account for a sizable proportion of the total public costs devoted to these people. Other
evidence supports this pattern. When limited to the working-aged population, people
whose disabilities are sufficiently severe as to prohibit carrying out major life activities
generate as much as six times the hospital costs and three times the physician visits as
those with no disabilities. 74
However, other data suggest that some people with disabilities do not have medical
care coverage. Clearly, Medicare, Medicaid, and Workers' Compensation do not cover
all working-age people who are captured in many definitions of disability. A 1993 report
found that nearly 16 percent of the working-age population with an activity-limiting dis-
ability, or 2.1 million people, lacked health care coverage. 7 5 Moreover, underinsurance
appears to be a common and severe problem for many of the disabled. Those with pri-
vate health insurance often find that they have coverage for hospitalization and provider
costs for acute care, but are not covered for items related to long-term needs such as
medical equipment or supplies, rehabilitation, or personal assistance services.
72 Project Network is a US Social Security Administration demonstration project designed to test alternative
approaches to assisting people with disabilities in finding and maintaining employment. The demonstration,
which began in June 1992 and ended in March 1995, operated in eight sites throughout the nation. The
project solicited voluntary participants from Disability Insurance and Supplemental Security Income for the
blind and disabled, tested four case management and referral models, and randomly assigned participants
either to treatment or control groups.
73 William Johnson (1997) suggests that only about 1 of 100 would succeed in a long-term position. One can
hardly imagine entrepreneurs being attracted to operate facilities with such low and uncertain returns.
74 See DeJong, Batavia, and Griss (1989). In addition, data from the Health Insurance Association of America
indicates that the 3 percent of the population with the most serious severe or chronic disabilities or illnesses
account for over one-half of the nation's total health care costs. It should be noted that these estimates are
dominated by people who are beyond working age, many of whom are in nursing home care.
75 See Watson (1993). Nearly 15 percent of the nondisabled population in 1984 had no health insurance.
The enacted expansion of Social Security and SSI to permit persons who return to
work to remain eligible for publicly provided health insurance is a recognition of the
incentive that the United States health insurance arrangements create to remain an SSDI
or SSI recipient once on the rolls. This is especially likely to be the case as the cost of
medical care and hence the value of Medicaid increases. 76 Decoupling these benefits
reduces the incentive to remain out of the labor force and eligible for cash benefits in
order to retain public health insurance coverage. 77
Finally, the relationship between public income support and health care coverage in
the United States is illustrated by what has come to be called the Monday effect. The
basic idea of the Monday effect is that among persons with no or limited health insur-
ance coverage, there will be an attempt to claim that impairments that occurred during
nonwork, or weekend, hours in fact happened while at work (in which case medical
care costs will be covered the Workers' Compensation program). Recent evidence from
a single state suggests that the Monday effect exists, but that it is hard to detect and may
be quite small [Card and McCall (1996)].
4.6. Antidiscriminationlegislation
Among many disabled working-age people (and their advocates), much of the relatively
poor labor market (and economic) position of people with disabilities is attributable to
discrimination by employers in both hiring and in wages. This judgment lay behind
the political efforts that resulted in the passage of the Americans with Disabilities Act
(ADA) of 1990. The ADA sought to reduce this discrimination by both mandating al-
terations in work environments so that work limitations do not prevent persons with
disabilities from working, and by providing legal relief of the sort given to racial and
gender minorities in cases of perceived violations of civil rights guarantees of equal
employment opportunities.
While discrimination by employers against persons with disabilities may explain
some of the employment and wage gaps between them, research evidence is quite uncer-
tain regarding the extent to which this factor, as opposed to differences in human capital
and other characteristics, is responsible. 78 If prejudice, rather than true human capital
differences, lies at the heart of observed wage and employment differences between
76 Yelowitz (1998) estimates that approximately 20 percent of the growth in SSI participants over the 1987-
1993 period can be explained by the growth in average Medicaid expenditures per capita.
77 For example, the 1619b provision allows persons once eligible for SSI due to disability (or blindness) to
continue their eligibility for Medicaid past the SSI disregard level subject to four conditions and to regain
eligibility should their attempt at work fail. The recommendation by the National Commission on Childhood
Disability to allow children who were covered by SSI to remain eligible for Medicaid as long as their disability
continued is also a recognition that the coupling of Medicaid with SSI provides an incentive for children to
remain on SSI once eligible.
78 Although some evidence suggests that employer discrimination may play an important role in explaining
the disabled-nondisabled wage gap for men, attributing all unexplained variance to this factor is not reli-
able when extensive unobserved human capital and other differences exist between disabled and nondis-
disabled and nondisabled people, the effect of the ADA in reducing these differences
may be small. Indeed, because of the large informational requirements imposed on em-
ployers who hire people with disabilities, ADA could actually reduce the employment
opportunities for such workers [Baldwin (1997)]. Preliminary evidence suggests that
the ADA has increased the costs of hiring persons with disabilities and that on net, may
have led to a decrease in employment of disabled men and younger disabled women
[Acemoglu and Angrist (1998)].
5. Children's disability and public policy
As indicated in Section 4, most public policy and economic research concerned with
disability has focused on the working-age population. Since the end of the 1980s, how-
ever, increased research and policy attention has been turned to the problem of children
with disabilities. Much of this concern developed because of the rapid increase in the
number of children added to the SSI rolls over the past decade and the large and growing
transfer and medical care costs associated with this growth. The growth in the preva-
lence of special education for children with disabilities in public schools also carried
important cost consequences that have aroused concern.
5.1. SSI benefitsfor children
From the start of the SSI program in 1974 until 1989, the number of children younger
than 18 receiving benefits never exceeded 300,000. Over the subsequent five years the
rolls nearly tripled (to more than 900,000), and program costs allocated to children grew
from about $1 billion to nearly $5 billion. The number of children receiving SSI benefits
grew from 3 percent of the total SSI recipient population in 1975 to more than 14 per-
cent by 1995. About one-third of the children on the SSI rolls are physically impaired,
42 percent are mentally retarded, and about 28 percent have mental or emotional impair-
ments. Those children with impairments have been the subject of the greatest concern,
growing out of claims that eligibility may have been established through either fraud or
"parental coaching" of children to feign impairment when none is present.79
Children who receive SSI benefits are also eligible for Medicaid 80 benefits, including
children in families with income above the cutoff level for SSI (if such children enter an
abled people. The evidence on the role played by discrimination in accounting for wage and employ-
ment gaps between disabled and nondisabled women is even less clear. See Baldwin and Johnson (1994,
1995) and Salkever and Domino (1997).
79 Many of the child recipients in this category have gained access to the rolls through diagnosis in the
Individualized Functional Assessment (IFA) process (see below), being diagnosed in particular as having
Attention Deficit Disorder or other learning disorders.
80 However, as with SSI for adults, states may exclude certain children from the Medicaid program. In these
states, recipients are allowed to deduct medical expenses in eligibility determination, a process known as
spend-down.
institution and remain there for 30 days). 81 The average annual Medicaid expenditure
per child SSI recipient was $6,400 in 1993, but has fallen to about $5,000 since then.
It is the value of the combination of this medical care benefit and the cash benefit that
is seen by some as promoting unwarranted application for benefits, and perhaps the
feigning of symptoms of impairment. This incentive, of course, is the greatest in those
states with the lowest welfare benefits. 82
Concern with the large number of children receiving SSI benefits, and the growth in
children's recipients and expenditures on them, has led to legislative proposals designed
to reduce the potential for abuse, and hence stem the increase in the number of children
awardees. There is substantial debate regarding the cause of the rapid increase in chil-
dren recipients. Some analysts see the increase as a one-time-only phenomenon, due to
the change in eligibility criteria following the Zebley decision (see note 40, above). Oth-
ers concluded that the increase in children's application and award rates would continue
in line with the growth in the early 1990s. (For example, a report by the US General Ac-
counting Office (1995) suggests that if recent trends had been allowed to continue, the
number of children who are SSI recipients would grow to 1.9 million by the year 2000.)
The report of the US National Commission on Childhood Disability (1995) concludes
that four factors have accounted for the rapid increase in child recipients over the 1989-
1994 period: the Zebley decision, a revision of the listings of mental impairments (which
made more children eligible for benefits), increased outreach efforts to increase aware-
ness of the program, and a significant increase in the proportion of children living in
families with income below the poverty line.
The revision in the mental impairments listings was not independent of the Zebley
decision. Prior to this decision, children could only be found eligible for SSI benefits
if they were determined to have one (or more) of a qualifying set of medical impair-
ments which needed to be comparable in terms of severity to one that would prevent an
adult from engaging in substantial gainful activity. Following the Zebley decision, this
standard was reinterpreted to be an impairment that substantially reduced the child's
81 Recognizing the incentive for unwarranted institutionalization created by this provision, 1982 federal leg-
islation permitted states to allow children from such above-cutoff families to have Medicaid coverage if that
child would be eligible for Medicaid if living in an institution, and the cost of home care does not exceed
the cost of an institution (alternatively, states could provide long-term care at home under the home and
community-based services waiver). This provision is known as the "Katie Beckett provision" from the legal
decision that prompted the legislation.
82 Subject to eligibility determination, SSI pays a disabled child or adult a monthly cash benefit. Under
current law [section 402 (a) (24) of the Social Security Act], the income received by a person on SSI is not
considered as available for others in the unit. Hence, in the case of a family eligible for AFDC and containing
an individual receiving SSI, the person receiving SSI is not included in the calculation of eligibility or the
monthly benefit to be paid to the family unit. For a family eligible for both programs, the combination of SSI
and AFDC for the family will be significantly greater than AFDC alone, even accounting for the marginal
AFDC payment for the child with a disability. And the incentive to get on SSI may be enhanced by the time
limits welfare established in the 1996-US welfare reform legislation. SSI recipiency requires reevaluation, but
has no time limit.
ability to function in an age-appropriate and effective manner. Individualized Func-

tional Assessment (IFA) and a "functional equals" criterion were added in response to
the Zebley decision. The IFA process seeks to detect whether a child behaves in an age-
inappropriate manner, and a finding that this is the case may lead to approval of an SSI
application. By 1995, about 225,000 of the 890,000 children on SSI had qualified under
an IFA.
New legislation applying to child applicants for SSI benefits was passed as part of the
1996 welfare reform legislation. That law terminates the Individualized Functional As-
sessment provision. 83 Although this new approach eliminates the provision of benefits
to about 235,000 children annually who would have qualified under the IFA process,
the number of children receiving SSI is still predicted to grow modestly, from about
995,000 to 1,089,000 between 1996 and 2002. 8 4
5.2. Special education in the public schools
The large and growing number of children enrolled in "special education" programs in
the public schools, and the associated costs of this program, also prompted concern. 85
Beginning with Mills v. Board of Education (348F. Supp. 866, in 1972), followed and
enhanced by the 1975 Education for All Handicapped Children Act (now known as
the Individuals with Disabilities Education Act, or IDEA), states have been required
to provide appropriate public education, known as special education, to all school-age
children with disabilities. 8 6
83 Under the new law, the Personal Responsibility and Work Opportunity Reconciliation Act (Public Law
104-193) signed into law on August 22, 1996, the definition of eligibility changes to "An individual under
the age of 18 is considered disabled under SSI if the child has a medically determinable physical or mental
impairment, which results in marked and severe functional limitations, and which can be expected to result in
death or which has lasted or can be expected to last for a continuous period of not less than 12 months".
84 Many children who have qualified under the IFA process have disabilities that will make them eligible
under the medical standards. Hence the exact number of children who will be affected by the elimination of
one eligibility criterion is not clear.
85 In addition to concerns with the size and costs of special education programs, questions regarding their
purposes and effectiveness were also raised: Are children who are screened into these programs better off
with the additional resources invested in them? Are there negative consequences to children of spending
time in special education? Is there a stigma associated with the program that has negative consequences?
Are children in these programs given less motivation, or is their peer group less challenging? Are "problem
children" being siphoned off into such programs only to relieve teachers in traditional situations from having
to deal with them?
86 According to official procedures, teachers refer students who they suspect may have a disability for an
evaluation which is carried out by a group of specialists matched on the basis of the suspected disability. [Ho-
cutt (1996) estimates that 3-5 percent of the school-age population is referred for such evaluation each year;
more than 90 percent are tested, and nearly three-quarters of those tested are placed in special education.] This
group then determines whether or not the child has a disability making them eligible for special education. If
the child is so diagnosed, the child is entitled to services under IDEA. Having been so diagnosed, a child is
provided with an individualized education program. About ninety-five percent of special education students
remain in public schools, where they spend an average of 30 percent of their time in some special education
service.
Special education is now a major program, accounting for more than $32 billion in
annual expenditures and providing services to some 4.5 million public school children,
or more than 10 percent of public school enrollment. 8 7 According to Richard Rothstein
(1995), from 1967 to 1991 spending on special education increased nearly fivefold, ris-
ing from less than 4 percent of school district expenditures to 17 percent. Lankford and
Wyckoff (1996), in a study of education resources in New York, found that spending on
special education accounted for a significant share of the increase in total expenditures
over a fourteen-year period, 1979-1980 to 1993-1994: in New York City it accounted
for more than 50 percent of real increased per pupil spending from 1980 to 1989 and
more than 40 percent in the 1989-1990 to 1993-1994 period. The number of children
being served has been relatively stable since 1980, but a major shift in their composition
has occurred. While less than a third were children with learning disabilities in 1980,
children with this diagnosis accounted for more than half of all disabled special educa-
tion children by 1993 [US Department of Education (1994)]. More than 5 percent of all
public school children are currently diagnosed as having a learning disability.
It is toward this "learning disabilities" component of the program that most of the
concern has been directed. While it is relatively easy to diagnose a child with significant
physical disabilities or severe retardation, diagnosing children with a learning disability
is far more difficult. 8 8 Perhaps more important is the incentive structure of the program,
which enables school districts both to increase their total financial resources and to
enhance their performance evaluation by shifting students from regular classrooms to
those designated for special education. 89
Unfortunately, few data are available, and hence few studies have been conducted,
on the effectiveness of special education programs.9 0 Some studies have assessed the
performance of special education children when they are placed in a regular classroom.
These studies suggest that little accommodation is made for these students, and what-
ever differentiation takes place favors students who either do well or pay attention;
in other words, not those needing special education. Other studies concern the prac-
tices in special education classrooms as compared to regular classrooms; they are based
largely on interviews with students and assessments of teaching strategies. 91 Most such
87 As of 1993, 8 percent of these children were aged 3-5 years, and about 5 percent were aged 18-21.
88 Children designated as having learning disabilities may exhibit a significant discrepancy between expected
performance and actual performance, or difficulty in paying attention, reasoning and/or organization.
89 Costs (for example, for a learning aide) dedicated to a student who has been designated a "special educa-
tion" student tend to be covered by federal and state dollars rather than school district or local dollars. The
same service provided to a regular student would have to be covered by the district's own budget. Hence, there
is incentive to the school district to increase the number of students officially diagnosed as having a disability.
Moreover, if schools or school districts are rewarded on the basis of academic success, there is an additional
incentive to reallocate students who are performing poorly from the regular classroom to special education
status, where they are excluded from formal achievement measures.
90 Hocutt (1996) reviews this limited literature.
91 See, for example, Ysseldyke, Christenson, Thurlow. and Bakewell (1988), and Fuchs, Fuchs, and Bishop
(1992).
studies find differences between the two types of classrooms, but the degree and di-
mensions of differentiation noted are inconsistent across the studies. A third strand of
research addresses outcomes of special education. Most of these studies have severe
methodological problems, ranging from small sample sizes to the lack of an appropriate
92
counterfactual.
Hocutt (1996) concludes that special education programs have a positive impact on
elementary-school-age students in terms of self-perception and reading ability. 93 A set
of related studies of "Effective Schools" suggests that placing students with special ed-
ucation needs in regular classrooms in such schools does have a positive impact (p. 92).
The combination of limited evidence on effectiveness of these programs and increas-
ing pressure on resources for education are likely to lead to increased efforts to reduce
the growth of (and perhaps to decrease) the real resources devoted to, special education
in the future. What this will mean for children whose need for special education is based
on disabilities other than learning disabilities is not clear. Those worst off may be best
served if the program is narrowly targeted.
6. Disability policy across OECD countries
As we have emphasized, public policies toward working-age people with disabilities

have three objectives: the provision of income support (largely through cash assistance),
rehabilitation, and the prevention of physical and mental impairments. While all indus-
trialized countries address each of these objectives, the relative emphasis on them varies
widely across countries and across time. Today there is a move toward convergence and
toward less generous policies toward persons with disabilities [Aarts, Burkhauser, and
de Jong (1997)].
It is difficult to characterize the widely varying emphases or to understand the basis
for them. To some extent, differences in income levels or economic performance play
a role. Higher-income countries, those with low unemployment, or those with high ra-
tios of working to dependent people may tolerate higher income support benefits, with
their accompanying labor supply disincentives. Countries also differ in the extent to
which they seek to tailor their policies among various population groups or individuals
92 Virtually no use is made of controlled experiments or random assignment; hence, the "selection" of stu-
dents into special education (and assignment of particular services for specific children) makes a comparison
with other children without these unique characteristics impossible without random assignment. Moreover,
placement in special education may be confounded with other school experiences such as grade retention and
school mobility; none of the evaluation studies controls for these factors. Further complicating the matter is
the failure of most districts to keep records on achievement of students in special education, at least in terms
of standardized achievement tests, making comparisons using more than one school district difficult. Finally,
the full impact of special education programs may not occur for a considerable period of time, but there are
few longitudinal studies of the consequences of special education placement.
93 These studies were conducted on very small numbers of students; for example, 11 poor readers and 21 stu-
dents with learning disabilities (Hocutt, p. 88).
with different levels of impairment. For example, some countries provide much stronger
encouragement for rehabilitation of younger than older workers, while accepting gen-
erous transfer benefits for older working-age disabled people. Finally, countries differ
substantially in the extent to which implicit social contracts or preferences persist which
guarantee generous income minima as a matter of right. 94
Aarts and de Jong (1996) and Aarts, Burkhauser, and de Jong (1996) characterize
the programs of five OECD countries (Germany, the Netherlands, Sweden, the United
Kingdom, and the United States), and contrast their programs in terms of "freedom
from want" and "freedom from idleness". 95 Countries that are characterized by the first
label tend to emphasize income transfers as the main instrument of disability policy;
those that are characterized by the second emphasize rehabilitation and the prevention
of impairments. Of the five countries, the Netherlands emphasizes income support (free-
dom from want) to the greatest extent, and does so essentially for all disabled persons.
Sweden and Germany separate the disabled by age, offering income support for those
60 and over but placing far greater weight on rehabilitation and the prevention of de-
pendency among younger disabled people. The United Kingdom spends very little on
rehabilitation, but also has relatively low income support expenditures, largely because
of lower benefit levels. 96 Most European countries have a compensation program for
those injured at work that is quite similar to the Workers' Compensation program in the
United States.
Across the European countries, there is some evidence that disability income sup-
port policy has been used for purposes of accommodating increases in unemployment
resulting from either macroeconomic or structural changes (e.g., declining regions or
industries, or increases in the supply of younger or female workers), especially during
the 1970s. These studies suggest that disability policy has been used to hide unem-
ployment and has been attractive to workers because of the more generous and less
94 For example, many countries of continental Europe base their program on the tradition of a responsibility
of the state to protect residents from poverty and to work toward social solidarity. Bismarck invented Worker's
Compensation using private insurance models in the 1870s. Winston Churchill and Franklin Roosevelt estab-
lished the tradition of protection of those who could not provide for themselves in the Atlantic charter of 1942.
Beveridge elaborated this for the UK. Across the European members of OECD there tends to be a relatively
high minimum wage, flat-rate means-tested transfers for all citizens, and universal health insurance. This
combination protects people with disabilities from severe poverty, and guarantees medical care coverage.
95 An earlier analysis of cross-national disability policies and their impacts is Haveman, Halberstadt, and
Burkhauser (1984).
96 According to Aarts and de Jong (1996), the Netherlands spends the greatest proportion of its GDP on
programs for the disabled (5.2 percent). Very little is spent on rehabilitation, but 12 percent of total disability
policy expenditures are in support of sheltered workshops or the creation of jobs for the disabled. In Sweden,
on the other hand, about 19 percent of its total disability policy expenditures (slightly more than 4 percent
of GDP) are on vocational rehabilitation and public sector jobs. Germany allocates about 2 percent of its GDP
to disability policy, of which about 10 percent is in support of vocational rehabilitation efforts and sheltered
workshops. The United Kingdom allocates about 2 percent of GDP to disability policy, and virtually none of
it is in support of rehabilitation or employment. The German program may be the most work oriented, in that
it has mandatory rehabilitation and a quota system that requires employers to offer jobs to the disabled. The
majority of working aged persons with disabilities work full-time.
stigmatizing nature of disability, relative to unemployment, benefits.97 Given this it is

perhaps not surprising that the few existing studies of the impact of disability benefits
on labor force participation of older workers find a sizable impact due to generosity of
the program. Nevertheless, Riphahn (1996), using German panel data, finds that dis-
ability benefits have only a small impact on the likelihood of seeking disability benefits;
she states that "even cutting benefits in half would not reduce the predicted cumula-
tive probability of disability retirement by age 62 by more than 5 percentage points,
from 38.2 to 33 percent" (p. 20).
In some cases, eligibility for support has been explicitly linked to unemployment, es-
pecially for older workers. According to Aarts and de Jong (1992), until the late-1980s,
disability law in the Netherlands required the eligibility process to take labor market
opportunities into account with the result that about 40 percent of beneficiaries could
be termed socially disabled or unemployed. In Germany if either an individual who is
disabled cannot find a job, or the Public Employment service is unable to place the
individual in a job that has pay and prestige equal to that of the job held prior to dis-
ability, the individual must be granted full disability benefits within a year. In Sweden,
econometric evidence suggests that among younger persons, especially women, there
is a significant increase in disability claims during periods of poorer labor market per-
formance [Rebick (1994, p. 211)]. Sweden has used disability benefits as an alternative
to less generous unemployment benefits through about 1992; recent Swedish unem-
ployment rates suggesting a sharp increase in unemployment indicate that this policy
was largely abandoned due to the high costs incurred [Aarts, Burkhauser, and de Jong
(1997)]. 9 8
Replacement rates tend to be more generous in European countries than they are in
the United States. Blondal and Pearson (1995) calculated replacement rates for a proto-
type individual who became entitled to disability benefits at age 40 having worked since
age 18. As of 1993, they calculated that full disability benefits for a single average earner
would provide a replacement rate of 76 percent in the Netherlands, 79 percent in Swe-
den; 56 percent in Germany, compared to a 24 percent rate in the US. These differences
are likely to be somewhat smaller now as these countries have attempted to scale back
their disability programs. For example, the Netherlands, has established a new formula
which cuts benefits for younger and higher paid workers [Aarts, Burkhauser, and de
Jong (1997)].
In our discussion of US policy, we emphasized children's participation in disability
programs. This experience is unique among the industrialized countries, in part because
97 For example, until 1987 the Netherlands explicitly recognized the difficulty that even partially impaired
workers have in finding employment. As a result, even those with a small loss in their earnings capacity
were often awarded full benefits. Aarts and de Jong (1992) estimate that perhaps as much as 40 percent of
those on the disability benefit rolls in Holland were long-term unemployed or those with perhaps only a
15 percent reduction in earnings capacity. Sweden also used disability policy to deal with problems of long-
term unemployment, although changes in the law in 1992 reduced this practice.
98 The Swedish unemployment rates went from 1.5 to 7.6 percent over the 1990-1995 interval [Aarts,
Burkhauser, and de Jong (1997, Table 3)].
of the absence in the United States of child allowances, which provide cash assistance
to all or to lower-income families with children, and hence reduce the incentives for
securing support for children through programs reserved for persons with disabilities.
Among OECD countries there are two primary approaches to determining eligibility for
special benefits to families with a child who is disabled: a list of impairments and an
evaluation of functioning deficits together with an assessment of special needs related
to the deficit(s). Payments tend to be made either through additions to or extensions of
the child allowance, or through the existing program for adults with disabilities. The
child allowance benefits take the form of an additional amount for families with a dis-
abled child who meets the eligibility requirements (Australia, Ireland, Luxembourg, and
New Zealand), or a tiered benefit tied to required care or degree of disability (United
Kingdom, Finland, Germany, Belgium, France and Sweden) [Zeitzer (1995)1. 9 9
7. Unresolved issues and research needs
All industrialized countries continue to struggle with their policies toward working-age
people with disabilities. The conflict at the simplest level is between the two fundamen-
tal objectives of disability policy: freedom from want and freedom from idleness. 0l°0°
But that is too simple. Disability policies seek simultaneously to provide a safety net
for those who are truly disabled, to encourage work among those with impairments
who are capable of work, and to reduce the probability of disablement.l 0 ' If disability
were unidimensional, objective, and defined in terms of ability to work, achieving these
three objectives simultaneously might be possible. But with a more subjective and mul-
tidimensional definition, achieving these joint goals is very difficult. This assessment
suggests several areas in which further research is necessary.
* Is it possible for disability determination to be based on a combination of impairments
and the status of job opportunities in such a way that when either medical technology
improves or job prospects change, persons can automatically be reclassified? And, by
developing reliable assessments of earnings capacities and making this information
known to prospective employers, can they be encouraged to modify jobs and to hire
workers with disabilities, especially in situations of tight labor markets?
99 Zeitzer (1995) argues that the approach of most of the developed countries appears to focus on the individ-
ual needs of a child who is disabled with the goal of helping the child "reach his or her potential", taking into
account the situation of the family. There is no evidence of the effects of alternative systems on the functioning
of the child with a disability, either as a child or as an adult.
100 For example, profit maximization and being internationally competitive are difficult to combine with
expensive redesign of jobs to facilitate employment of persons with disabilities.
101 And, in periods of significant unemployment, there is yet another objective: disability policies may offer
a way to open up jobs for the able-bodied young while providing an attractive option for many older persons
with some health limitation. In this instance, disability policy is really an unemployment-reduction or job-
creation policy for young workers.
* Is it possible to identify those impairments where improvement is possible and hence

target disability review and retraining on persons with such impairments?
* Can incentives or financial subsidies be designed to encourage employers to modify
jobs with a view to hiring primarily younger persons with impairments?
* Can greater use of temporary benefits be employed to improve the incentives in in-
come support programs for persons with disabilities?
* Can effective incentives for accident and injury prevention be designed by, say, mod-
ifying the experience rating of programs such as workers' compensation?
* Can successful training and rehabilitation programs in private firms be identified and
publicized so as to encourage imitation?
* Does the introduction of computers open up possibilities to effectively redesign jobs
so as to allow persons with disabilities to work? And if so, how can we design
computer-oriented training programs that provide necessary knowledge to persons
with disabilities?
* How do we deal with a lack of jobs in certain locales and the need for greater flex-
ibility among certain disabled regarding hours worked and the need for time off for
treatment and maintenance activities?
* For younger persons with impairments, is there a gain in terms of future employment
to providing more education, including a (or an additional) college degree? If so,
could such a policy pass a benefit-cost test? And could it be designed to be narrowly
and appropriately targeted?
* Would an earnings tax credit for the disabled be effective in increasing employment
and decreasing the use of cash transfers? If so, what is the best design?
* Can cost-effective policies be designed that are flexible enough to provide benefits
when jobs are not available, but not when jobs are available?
Turning to issues related to children with disabilities, reliable information regarding
the increased resources associated with caring for the child, the opportunity cost to the
parent, and how best to provide for these needs would seem to be of high priority. Indus-
trialized countries with a children's allowance and universal medical care would appear
to have in place some of the elements designed to meet the additional needs of raising
a child with a disability. The opportunity cost and additional costs of modifications to a
home or car also are important components on which research is needed.
As many of these points suggest, the all-or-nothing nature of many disability pro-
grams may be the source of some of the difficulty in meeting the diverse objectives
which we have assigned to disability policy. While this situation may be most serious
in the United States (affecting all of its cash transfer programs, except Workers' Com-
pensation), it applies to other nations as well. However, establishing a scale of degrees
of disablement and eligibility for benefits of various magnitudes and periods requires
more knowledge of reasonable expectations and possible job modifications (and their
costs) than is currently available.
We simply do not know how to design successful sets of programs to simultaneously
achieve all of the three objectives that we have set for disability policy - income main-
tenance, rehabilitation, and prevention. Heretofore, most research has been focused on
defining the disabled, evaluating a single experimental program, analyzing the adequacy
of specific benefit schedules, or studying the incentive effects of a single program or re-
lated set of programs. Perhaps we might begin by considering a set of policies that could
provide incentives for work but also reduce significant income uncertainty due to a dis-
ability. In addition, an important need is for the design of an effective set of incentives to
encourage effective prevention activities. Experimenting with alternative arrangements
for policy improvements, and then assessing the benefits and costs of each, would seem
to be of high priority.
Unfortunately, at least in the near future, it seems clear that all industrialized countries
are likely to continue to face dilemmas with their policies toward the disabled. Design-
ing an effective system to combine freedom from want for the disabled with freedom
from idleness is a most difficult objective to secure.
Appendix A. United States income support programs for working-age

disabled people
A. 1. The Social Security Disability(SSDI) program
The SSDI program is the major source of income support for disabled people and pro-
vides social security coverage to about 95 percent of the US workforce. It pays out more
than $40 billion in benefits each year, and has more than five million beneficiaries. SSDI
was added to the nation's social security system (the Old-Age, Survivors, and Disability
Insurance program, OASDI) in 1957, and provides partial earnings replacement bene-
fits to those working-age people with severe, long-term handicaps that eliminate the
possibility of engaging in "any substantial gainful employment". Funds for support of
the program come from the Federal Insurance Contributions Act (FICA) which imposes
payroll taxes on both employees and employers.
The law covers working-age individuals who (1) by virtue of a "medically deter-
minable physical or mental impairment" that is either expected to last for one year or
result in death, are unable to engage in "any substantial gainful activity" (this SGA con-
straint means being unable to earn $500 per month),10 2 (2) have sufficient quarters of
employment in jobs covered by the social insurance system,103 and (3) are not working
(or working less than the SGA constraint). Determination of eligibility for benefits is
made by employees of the Social Security Administration in state and regional offices. If
the application is denied, an appeals process can be pursued, with ultimate adjudication
by an administrative law judge.
102 Deductions for impairment-related work expenses (IRWE) are taken into account in determining if earn-
ings exceed the SGA limit.
103 Twenty quarters of coverage out of the 40 quarters preceding application for benefits are required, except
for younger workers.
The SSDI benefit is related to the worker's predisability earnings record, and aver-
ages more than $700 per month. The benefit schedule is redistributive, and the effective
replacement rate (benefits/predisability earnings) ranges from nearly 80 percent for low
earners (average earnings of, say, $500 per month) to less than 30 percent for high
earners (say, $4500 per month). The replacement rate is 42 percent for workers with
average earnings. 10 4 In addition, Medicare benefits are provided to all SSDI recipients
after 24 months on the rolls. The average value of these health benefits (measured on
an expenditure basis) is about $500 per month per recipient, nearly $6,000 per year per
recipient. 105
As in the disability pension programs of most countries, the work incentives in the
SSDI program are complex, and can only be understood by considering various situa-
tions that benefit applicants can experience. Rather distinct work disincentive patterns
are observed for the application/waiting period, the period of benefit recipiency, the
grace period following completion of a trial work period, and an extended period of
eligibility. 106 In addition, various of the provisions create incentives for entry into the
program.
Since the very origin of the SSDI program, there have been efforts to link the pro-
vision of income support benefits with the encouragement of rehabilitation. Several
104 These values are for 1996 and are based on calculations by Koitz (updated by Cornelius) (1996, p. 9).
105 The Canadian Disability Program is similar in structure to the SSDI program in the United States [Maki
(1993)].
106 The following discussion draws from Hoynes and Moffitt (1996). First, during the application/waiting
period, earnings cannot exceed the SGA amount. This limit imposes a constraint on work effort until the
$500 amount, at which point there is an effective notch. Both above and below the $500 notch point, the
marginal tax rate on earnings reflects only non-SSDI tax/transfer provisions. This incentive structure also
applies during the period of benefit recipiency, except that a trial work period (TWP) provision exists to
encourage work effort. If a recipient is in a TWP, nine months of employment over a 60-month period are
permitted with no reduction in the SSDI benefit payment, regardless of how much is earned. Hence, the
marginal tax rate during the TWP is also zero. However, the ability to both earn and receive benefits during
this period creates an income effect that would tend to reduce work effort. Introduction of the TWP provision
by itself creates an incentive for beneficiaries to work some more while receiving benefits, but discourages
an exit from the program for those who can work at jobs paying in excess of the SGA. The provision, by
increasing the perceived generosity of the program, would also tend to encourage entry into the program,
resulting in increased caseloads and reduced work.
If the beneficiary completes the TWP and is found to be able to continue in SGA employment, a three-
month grace period is granted, and benefits stop thereafter. Beneficiaries continuing to work in the grace
period enter a 36-month extended period of eligibility (EPE); during this period, full benefits are paid if
earnings fall below SGA, but fall to zero if earnings exceed SGA. However, reapplication is not necessary
if SGA earnings cannot be sustained. Hence, during this period too, the marginal tax rate is zero at earnings
levels below $500 per month, but becomes greater than 100 percent at the notch created at that level; the
marginal tax rate again falls to zero at earnings levels above $500. This pattern gives incentive for maintaining
earnings below the SGA level, but discourages work at jobs paying more than $500. Medicare benefits are
provided 39 months after the recipient enters the EPE, and for three months past the end of the EPE period if
the worker succeeds in leaving the rolls. The insurance-against-failure provided by the EPE tends to increase
work effort by recipients during the EPE period, but like the TWP may stimulate entry into the program, with
the concurrent reduction in work effort.
approaches have been followed. The most prominent effort is known as the Benefi-
ciary Rehabilitation Program (BRP), which was in effect from 1965 (nine years after
the beginning of the SSDI program) until 1981. The BRP program allocated 1 percent
(increasing to 1.25 percent in 1973 and 1.5 percent in 1974) of total disability insurance
payments from the trust fund financing the program to the Vocational Rehabilitation
(VR) program on a prospective basis, and without state matching, for the purpose of
providing rehabilitation services to SSDI recipients (and after 1974, to SSI recipients).
By the end of the 1970s, the average cost per rehabilitation was about $8,000 (in current
dollars), and a series of academic and governmental studies concluded that the program
was, at best, returning $1 in benefit savings for every $1 that was spent on services, and
the BRP was generally viewed as a failed effort. 107
In 1981, the BRP program was terminated, the funds allocated to the VR program
fell precipitously, and a new system in which payments were retrospective, and reim-
bursement paid only for SSDI cases who were removed from the rolls, was put into
place. However, until the present time the number of SSDI cases terminated from the
rolls because of VR services remains at about 6,000 per year (less than 0.2 percent
of the total number of SSDI recipients), about 60 percent of the recovery claims for
which VR programs have sought reimbursement. VR programs have had to seek other
sources of revenue, and because of the high rate at which their claims for SSDI reim-
bursement are rejected tend to shy away from accepting SSDI recipients as clients. 108
A new program, "Ticket to Independence", has been proposed to replace the current
VR programs. It would use private and public providers of rehabilitation, and pay the
provider according to SSA savings based on successful return to employment.
A.2. The Supplemental Security Income (SSI) program
The SSI program, established in 1972 and begun in 1974, provides income support
benefits to aged, blind, and disabled individuals who are poor. To be eligible for benefits,
disabled individuals must meet the same "gainful employment" disability criteria as for
the SSDI program, must have income and assets below eligibility requirements (e.g.,
$670 per month of "countable" income per couple, and $3,000 of assets, excluding
home and automobile in 1996), and not work or work less than the SGA level. The
107 By the end of the BRP program, the annual number of SSDI recipients terminated from the rolls because
of VR services was about 7,000, equal to about 0.3 percent of total recipients. The reasons for the disap-
pointing effectiveness of the program are not hard to find. The program served only people who had applied
and passed the waiting period for SSDI benefits, and who were receiving benefit payments; moreover, long
delays in gaining beneficiary status (from about 6 months to more than 1.5 years if appeals are made) are
not uncommon, and during this time applicants are focused on their impairment, rather than on investment
in productivity-restoring activities. Berkowitz and Dean (1996) and Hunt, Habeck, Owens, and Vandergoot
(1996) discuss the SSDI approach to encouraging return to work and the reasons for its modest success.
108 Berkowitz and Dean (1996) state: "The link between VR and [SSDI] continues to resemble a long funnel
into which... [cases are poured]... to have only a few trickle out the other end. The disparities between the
ends of the funnel are staggering" (p. 240).
calculation of "countable income" 10 9 yields a marginal tax rate of 50 percent above the
deducted amount.
Benefits in the program equal $446 per person ($669 per couple) less countable in-
come. 1 0 The majority of states supplement the federal benefit for at least some partici-
pants. And, in most states, all SSI recipients are automatically granted Medicaid health
benefits.' 1 Only about 6 percent of working-age SSI recipients work at all.
In order to encourage SSI recipients to work, Section 1619 was passed by Congress
in 1986. Under this provision, blind and disabled recipients with substantial earnings
can continue to be eligible for Medicaid and also receive cash benefits. 112
A.3. The Workers' Compensationprogram andjob-relatedhandicaps
The Workers' Compensation (WC) program is the nation's oldest social insurance pro-
gram, having been first introduced in 1908. While WC laws are state-based, all reflect
the same underlying principle - employers should assume the costs of occupational
disabilities without regard to fault. 113 By 1949, all states had WC laws.
Unlike SSDI and SSI, WC (1) is a state-based program with mandatory coverage
in all states, (2) involves disability determinations that permit recognition of temporary
and permanent partial as well as total disablement, (3) requires that the impairment arise
out of or occurs during the course of employment, (4) can pay out benefits in annuity
or lump sum fashion, with some benefits being indemnity awards according to fixed
schedules, and (5) is a true insurance program, being financed by individual employers
whose premia (or self-insurance costs) are related to their ownjob-related injury experi-
ences. 14 Like SSDI and SSI, WC benefits are untaxed at any level of government, and
109 Countable income equals income from all sources minus all disability-related work expenditures, plus $20
of unearned income, $65 of earned income, and 0.5 of remaining earnings.
110 Average payments were $383 per month for an individual and $624 for a couple in December 1995.
11 A limited number of states use the "209(b)" option, which enables states to impose Medicaid eligibility
criteria which are more restrictive than SSI criteria, so long as the criteria chosen are not more restrictive
than the state's approved Medicaid state plan in January 1972. Eleven states used this option in 1996, which
account for about 19 percent of the SSI recipient population nationwide.
112 Recipients can receive cash benefits (reduced according to a sliding scale) up to a "break-even point"
of $1,025 per month in 1996 in states without supplementation. Even with earnings beyond the break-even
level, recipients can continue Medicaid coverage as long as they continue to have an impairment, meet the
non-earnings requirements for SSI eligibility, be "seriously inhibited from continuing to work" if Medicaid
was terminated, and cannot afford a reasonable substitute for Medicaid.
113 Prior to the enactment of WC laws, workers were only protected in cases in which employer negligence
could be proven as the cause of injury or death.
11
4 Employers can self-insure if they are large, or can purchase coverage individually or in groups from
private insurers (or in some states, the state in competition with private insurers). In all cases, the premia
paid are related to the claims cost experience of the employer (policy holder) and to the industry of the
employer, although the basis for experience rating varies substantially across the various insurance categories.
See Worrall and Butler (1986).
Table A.1
Growth of Disability and Workers' Compensation programs in the United States since 1960
Growth of recipiency of disability programs (in thousands)
Year SSI disabled program SSDI program

Nonelderly Disabled Spouses Children
<18 18-21 22-64
1960 455 77 155

1965 988 194 558
1975 104 90 1,559 2,489 453 1,411
1983 191 122 1,517 2,569 308 936
1986 231 138 1,787 2,727 291 965
1989 256 139 2,091 2,895
1995 898 219 3,193 4,185 264 1,409
Growth of Workers' Compensation program
Year Covered workers Benefits paid Share medical Cost of program

(millions) total (millions) as %covered payroll
1960 44.9 $1,295 33.6% 0.93%

1965 50.8 1,814 33.1 1.0
1975 67.2 6,598 30.8 1.32
1983 78.0 17,575 32.3 1.67
1986 86.0 24,613 35.1 1.99
1989 93.7 34,316 39.1 2.27
1991 93.6 42,169 39.9 2.40
1993 96.1 42,925 40.8 2.30
Source: Green Book 1996 Table 4-12 (SSI), Social Security Bulletin Table 1-24, Annual Statisti-
cal Supplement, 1995 Tables 5.A4, 9.B1.
include medical care coverage for the impairment leading to the claim. 115 Appendix
Table A. 1 compares growth in the disability and WC programs.
WC programs provide both cash and medical benefits to persons with job-related
disabilities, and survivors' benefits to dependents of those whose death resulted from
a work-related accident or illness.1 6 Cash compensation for lost earnings is usually
based on a percentage of weekly earnings at the time of injury or death. While most
states set these at about two-thirds of gross (pretax) lost earnings, others set them as
a percentage of lost "spendable" (after-tax) earnings, usually replacing 75 or 80 per-
115 Medical cost coverage for the work-related impairment is virtually unlimited in the WC program; as of
1993, over 40 percent of all benefits paid were for medical care.
116 About 5-8 percent of cash benefits paid by WC programs are awarded to survivors.
cent. 117 Most workers' compensation benefits are paid by insurance companies through
policies purchased by private employers that are tied to the benefits required by the state
law covering the employer.
References
Aarts, L., and P. de Jong (1992), Economic Aspects of Disability Behavior (North-Holland, Amsterdam).
Aarts, L., and P. de Jong (1996), "European experiences with disability policy", in: J. Mashaw, V. Reno,
R. Burkhauser and M. Berkowitz, eds., Disability, Work, and Cash Benefits (Upjohn Institute for Employ-
ment Research, Kalamazoo, MI).
Aarts, L., and P. de Jong (1998), "The cost of coercion: an empirical study of the willingness to pay for dis-
ability insurance", Paper presented at the Fifth International Research Seminar on Issues in Social Security,
Sjudarhojden, Sweden.
Aarts, L., R. Burkhauser and P. de Jong, eds. (1996), Curing the Dutch Disease: An International Perspective
on Disability Policy Reform (Avebury, Aldershot, England).
Aarts, L., R. Burkhauser and P. de Jong (1997), "Convergence: a comparison of European and United States
disability policy", mimeo (Leiden, The Netherlands).
Acemoglu, D., and J. Angrist (1998), "Consequences of employment protection? The case of the Americans
with Disabilities Act", NBER Working Paper 6670.
Baldwin, M.L. (1997), "Can the ADA achieve its employment goals?", in: W.G. Johnson, ed., The Americans
with Disabilities Act, Special Issue of The Annals of the American Academy of Political and Social Science
549:37-52.
Baldwin, M.L., and W.G. Johnson (1994), "Labor market discrimination against men with disabilities", Jour-
nal of Human Resources 29(1):1-19.
Baldwin, M.L., and W.G. Johnson (1995), "Labor market discrimination against women with disabilities",
Industrial Relations 34:568-572.
Baldwin, M.L., L.A. Zeager and P.R. Flacco (1994), "Gender differences in wage losses from impairments",
Journal of Human Resources 29(3):865-887.
Barnow, B. (1996), "Policies for people with disabilities in US Employment and Training Programs", in:
J. Mashaw, V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability, Work, and Cash Benefits (Upjohn
Institute for Employment Research, Kalamazoo, MI).
Berkowitz, E., and D. Dean (1996), "Lessons from the Vocational Rehabilitation/Social Security Adminis-
tration Experience", in: J. Mashaw, V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability, Work, and
Cash Benefits (Upjohn Institute for Employment Research, Kalamazoo, MI).
Berkowitz, M. (1997), "Linking beneficiaries with return-to work services", in: V. Reno, J. Mashaw and
B. Gradison, eds., Disability: Challenges for Social Insurance, Health Care Financing and Labor Market
Policy (National Academy of Social Insurance, Washington, DC).
Berkowitz, M., and A. Hill, eds. (1986), Disability and the Labor Market (Industrial and Labor Relations
Press, Ithaca, NY).
Berkowitz, M., W.G. Johnson and E.H. Murphy (1976), Public Policy toward Disability (Praeger, New York).
Bernheim, B.D. (1987), "The economic effects of social security", Journal of Public Economics 33:273-304.
Blondal, S., and M. Pearson (1995), "Unemployment and other non-employment benefits", Oxford Review
of Economic Policy 11(1): 136-169.
117 Permanently disabled workers receiving WC benefits also may be eligible for SSDI benefits. However,
they are limited to a maximum replacement rate of the higher value of 80 percent of former earnings or
80 percent of the total family SSDI benefit.
1048 R. Haveman and B. Wolft
Boadway, R., N. Marceau and M. Sato (1997), "An agency model of welfare and disability assistance", Work-
ing Paper 46 (Universit6 du Quebec Montreal, Center for Research on Economic Fluctuations and Em-
ployment).
Bound, J. (1989), "The health and earnings of rejected disability insurance applicants", American Economic
Review 79(3):482-503.
Bound, J. (1991), "The health and earnings of rejected disability insurance applicants: reply", American
Economic Review 81(5):1427-1434.
Bound, J., and T. Waidman (1992), "Disability transfers, self-reported health, and the labor force attachment
of older men: evidence from the historical record", Quarterly Journal of Economics 107:1393-1420.
Burkhauser, R., and M. Daly (1996), "Employment and economic well-being following the onset of a disabil-
ity: the role for public policy", in: J. Mashaw, V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability,
Work, and Cash Benefits (Upjohn Institute for Employment Research, Kalamazoo, MI).
Burkhauser, R., R. Haveman and B. Wolfe (1993), "How people with disabilities fare when public policies
change", Journal of Policy Analysis and Management 12(2):251-269.
Burkhauser, R., J.S. Butler and Y.W. Kim (1995), "The importance of employer accommodation on the job
duration of workers with disabilities: a hazard model approach", Labour Economics 3(1):1-22.
Burkhauser, R., and D. Wittenburg (1996), "How current disability transfer policies discourage work: analysis
from the 1990 SIPP", Journal of Vocational Rehabilitation 7(August):9-27.
Butler, R. (1995), "Safety incentives in workers' compensation", in: J. Burton and T. Schmidle, eds., 1995
Workers' Compensation Year Book, vol. 1 (Horsham, PA) 82-91.
Card, D., and B. McCall (1996), "Is workers' compensation covering uninsured medical costs? Evidence from
the 'Monday effect' ", Industrial and Labor Relations Review 49(4):690-706.
Chirikos, T. (1989), "Aggregate economic losses from disability in the United States: a preliminary assay",
Milbank Quarterly 67(Suppl. 2, Part 1):59-91.
Chirikos, T., and G. Nestel (1985), "Further evidence on the economic effects of poor health", Review of
Economics and Statistics 67(1):61-69.
Cooper, B.S., and D. Rice (1976), "The economic cost of illness revisited", Social Security Bulletin 39(2):21-
36.
Daly, M. (1994), "The economic well-being of men with disabilities: a dynamic cross-national view", Ph.D.
Dissertation (Syracuse University).
Danzon, P.M. (1993), "The economic implications of public disability insurance in the United States", Journal
of Labor Economics 11(1, Part 2):S 170-S200.
Dean, D.H., and R.C. Dolan (1991), "Assessing the role of vocational rehabilitation in disability policy",
Journal of Policy Analysis and Management 10(4):568-587.
DeJong, G., A.I. Batavia and R. Griss (1989), "America's neglected health minority: working-age persons
with disabilities", Milbank Quarterly 67(Suppl. 2, Part 2):311-351.
Diamond, P., and E. Sheshinski (1995), "Economic aspects of disability benefits", Journal of Public Eco-
nomics 57:1-23.
Diller, M. (1996), "Entitlement and exclusion: the role of disability in the social welfare system", UCLA Law
Review 44(2):361-465.
Disney, R., and S. Webb (1990), "Why social security expenditure in the 1980s has risen faster than expected:
the role of unemployment", Fiscal Studies 11(1, February):1-20.
Disney, R., and S. Webb (1991), "Why are there so many long term sick in Britain?", Economic Journal
101:252-262.
Dwyer, D., and 0. Mitchell (1999), "Health problems as determinants of retirement: are self-rated measures
endogenous?", JHE 18:173-193.
Dykacz, J., and J. Hennessey (1989), "Postrecovery experience of disabled worker beneficiaries", Social Se-
curity Bulletin 52(9):42-66.
Feldstein, M. (1982), "Social security and private saving: reply", Journal of Political Economy 90:630-642.
Fuchs, L.S., D. Fuchs and N. Bishop (1992), "Teacher planning for students with learning disabilities: differ-
ences between general and special educators", Learning Disabilities: Research and Practice 7:120-128.
Gray, W., and C.A. Jones (1991), "Are OSHA health inspections effective? A longitudinal study in the man-
ufacturing sector", Review of Economics and Statistics 71(3):504-508.
Gruber, J. (1996), "Disability insurance benefits and labor supply", Working Paper No. 5866 (National Bureau
of Economic Research, Cambridge, MA).
Gruber, J., and J. Kubik (1997), "Disability insurance rejection rates and the labor supply of older workers",
Journal of Public Economics 64(1):1-23.
Halpern, J., and J. Hausman (1986), "Choice under uncertainty: a model of applications for the Social Security
Disability Insurance program", Journal of Public Economics 31:131-161.
Hartunian, N., C. Smart and M. Thompson (1981), The Incidence and Economic Costs of Major Health
Impairments (Lexington Books, Lexington, MA).
Haveman, R., V. Halberstadt and R. Burkhauser (1984), Public Policy toward Disabled Workers: Cross-
National Analyses of Economic Impacts (Cornell University Press, Ithaca, NY).
Haveman, R., and B. Wolfe (1984a), "The decline in male labor force participation: comment", Journal of
Political Economy 92(3):532-541.
Haveman, R., and B. Wolfe (1984b), "Disability transfers and early retirement: a causal relationship?", Journal
of Public Economics 24:47-66.
Haveman, R., and B. Wolfe (1990), "The economic well-being of the disabled, 1962-1984", Journal of Human
Resources 25(1):32-55.
Haveman, R., B. Wolfe, L. Buron and S. Hill (1995), "The loss of earnings capability from disability/health
limitations: toward a new social indicator", Review of Income and Wealth 41(3):289-308.
Haveman, R., P. de Jong and B. Wolfe (1991), "Disability transfers and the work decision of older men",
Quarterly Journal of Economics 106(5):939-949.
Haveman, R., K.C. Holden, B. Wolfe, P. Smith and K. Wilson (forthcoming a), "The changing economic
status of disabled women, 1982-1991: trends and their determinants", in: D. Salkever and A. Sorkin, eds.,
Essays in The Economics of Disability (JAI Press).
Haveman, R., K.C. Holden, B. Wolfe, P. Smith and K. Wilson (forthcoming b), "The economic status of
younger and older disabled men: trends and their determinants, 1982-1991", Empirical Economics.
Hendricks, W., C.A. Schrio-Geist and E. Broadbent (1997), "Long-term disabilities and college education",
Industrial Relations 36(1):46-60.
Hocutt, A. (1996), "Effectiveness of special education: is placement the critical factor?", The Future of Chil-
dren 6(1):77-102.
Hoynes, H., and R. Moffitt (1996), "The effectiveness of financial work incentives in social security disabil-
ity insurance and supplemental security income: lessons from other transfer programs", in: J. Mashaw,
V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability, Work, and Cash Benefits (Upjohn Institute for
Employment Research, Kalamazoo, MI).
Hunt, H.A., R.V. Habeck, P. Owens and D. Vandergoot (1996), "Disability and work: lessons from the private
sector", in: J. Mashaw, V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability, Work, and Cash Benefits
(Upjohn Institute for Employment Research, Kalamazoo, MI).
Johnson, W.G., ed. (1997), "The Americans with Disabilities Act: social contract or special privilege?", An-
nals of the American Academy of Political and Social Science 549.
Koch, M., and C. Thimann (1997), "From generosity to sustainability: The Austrian pension system and
options for its reform", Working Paper WP/97/10 (International Monetary Fund, Washington, DC).
Koitz, D. (1996), "Social security: brief facts and statistics", Congressional Research Service Report No. 94-
27 EPW (Washington, DC).
Kreider, B. (1998), "Workers' applications to social insurance programs when earnings and eligibility are
uncertain", Journal of Labor Economics 16(4):848-877.
Kreider, B. (forthcoming), "Disability insurance: applications, awards, and lifetime opportunity costs", Jour-
nal of Labor Economics.
Krueger, A., and D. Kruse (1995), "Labor market effects of spinal cord injuries in the dawn of the computer
age", Working Paper No. 5302 (National Bureau of Economics Research, Cambridge, MA).
1050 R. Havenman and B. Wolfe
Lankford, H., and J. Wyckoff (1996), "The allocation of resources in New York State school districts: 1979-
80 to 1993-94", in: R. Berne, ed., Cost Effectiveness in Education (New York State Board of Regents,
Albany, NY).
Lanoie, P., and D. Str6Liski (1995), "L'Impact de la rglementation en matire de santd et scurit6 du travail
sur le risque d'accidents an Quebec: des nouveaux rsultats", Centre Interuniversitaire de Recherche en
Analyse des Organisations, Quebec.
Leimer, D., and S. Lesnoy (1982), "Social security and private savings: new time-series evidence", Journal of
Leonard, J.S. (1979), "The Social Security Disability Program and labor force participation", Working Paper
No. 392 (National Bureau of Economic Research, Cambridge, MA).
Leonard, J.S. (1986), "Labor supply incentives and disincentives for disabled persons", in: M. Berkowitz and
A. Hill, eds., Disability and the Labor Market (Industrial and Labor Relations Press, Ithaca, NY).
Maki, D.R. (1993), "The economic implications of disability insurance in Canada", Journal of Labor Eco-
nomics 11(l):S148-S169.
Manning, W., E.B. Keeler, J.P. Newhouse et al. (1991), The Costs of Poor Health Habits (Harvard University
Press, Cambridge, MA).
Mashaw, J., V. Reno, R. Burkhauser and M. Berkowitz, eds. (1996), Disability, Work, and Cash Benefits
(Upjohn Institute for Employment Research, Kalamazoo, MI).
McNeil, J.M. (1993), "Americans with disabilities: 1991-1992", Current Population Reports, Household Eco-
nomic Studies, Series P70-33 (US Bureau of the Census).
McNeil, J.M. (1997), Americans with disabilities: 1994-95, Current Population Reports, Household Eco-
nomic Studies, Series P70-61 (US Bureau of the Census).
Muller, L.S. (1992), "Disability beneficiaries who work and their experience under program Work Incentives",
Social Security Bulletin 55:2-19.
Murray, C., and A. Lopez, eds. (1996), The Global Burden of Disease (World Health Organization, Harvard
University Press, Cambridge, MA).
Nagi, S.Z. (1969), Disability and Rehabilitation (Ohio State University Press, Columbus, OH).
Parsons, D. (1980a), "The decline in male labor force participation", Journal of Political Economy 88:117-
134.
Parsons, D. (1980b), "Racial trends in male labor force participation", American Economic Review 70:911-
920.
Parsons, D. (1991a), "The health and earnings of rejected disability insurance applicants: comment", Ameri-
can Economic Review 81:1419-1426.
Parsons, D. (1991b), "Self-screening in targeted public transfer programs", Journal of Political Economy
99:859-876.
Parsons, D. (1996), "Imperfect 'tagging' in social insurance programs", Journal of Public Economics 62:183-
207.
Rebick, M. (1994), "Social security and older workers' labor market responsiveness: the United States, Japan
and Sweden", in: R. Blank, ed., Social Protection versus Economic Flexibility: Is There a Trade-Off?
(University of Chicago Press, Chicago).
Reno, V., J. Mashaw and B. Gradison, eds. (1997), Disability: Challenges for Social Insurance, Health Care
Financing and Labor Market Policy (National Academy of Social Insurance, Washington, DC).
Rice, D., T.A. Hodgson and A.N. Kopstein (1985), "The economic costs of illness: a replication and update",
Health Care Financing Review 6(1):61-80.
Rice, D., S. Kelman and L.S. Miller (1991), "Estimates of economic costs of alcohol and drug abuse and
mental illness, 1985 and 1988", Public Health Reports 106(3):280-292.
Riphahn, R. (1996), "Disability retirement among German men in the 1980s", mimeo (University of Munich,
Germany).
Riphahn, R., and B. Kreider (1997), "Applications to the US disability programme: a semiparametric ap-
proach", Discussion Paper 1559 (Centre for Economic Policy Research, London).
Rothstein, R. (1995), Where's the Money Gone? (Economic Policy Institute, Washington, DC).
Rupp, K., and C. Scott (1996), "Trends in the characteristics of DI and SSI disability awardees and duration
of program participation", Social Security Bulletin 59(1):3-21.
Rupp, K., and D. Stapleton (1995), "Determinants of the Growth in the Social Security Administration's
disability Programs - an overview", Social Security Bulletin 58(4):43-70.
Salkever, D., and M. Domino (1997), "Within group 'structural' tests of labor-market discrimination: a study
of persons with serious disabilities", NBER Working Paper 5931.
Slade, F.P. (1984), "Older men, disability insurance and incentive to work", Industrial Relations 23(2):260-
277.
Soest, A. van, P. Fontein and R. Euwals (1996), "Earnings capacity and labour market participation", mimeo
(Tilburg University, The Netherlands).
US Bureau of the Census (1989), "Labor force and other characteristics of persons with work disabilities,
1981 to 1988", Current Population Reports, Special Series No. 160.
US Department of Education, Office of Special Education Programs (1994), Sixteenth Annual Report to
Congress (Government Printing Office, Washington, DC).
US General Accounting Office (1993), Vocational Rehabilitation: Evidence for Federal Programs' Effective-
ness Is Mixed, GAO/HEHE 93-19 (Government Printing Office, Washington, DC).
US General Accounting Office (1995), Supplemental Security Income: Growth and Changes in Recipient
Population Call for Re-Examining Program, GAO/HEHS 95-137 (Government Printing Office, Washing-
ton, DC).
US General Accounting Office (1997), Social Security: Disability Programs Lag in Promoting Return to
Work, GAO/HEHS-97-46, March 17 (Government Printing Office, Washington, DC).
US House of Representatives, Committee on Ways and Means (various years), Green Book (Government
Printing Office, Washington, DC).
US National Commission on Childhood Disability (1995), "Supplemental security income for children with
disabilities", Report to Congress (Washington, DC).
Watson, S. (1993), "An alliance at risk: the disability movement and health care reform", The American
Prospect 12:60-67.
Wolfe, B.L., and R. Haveman (1990), "Trends in the prevalence of work disability from 1962 to 1984, and
their correlates", Milbank Quarterly 68(1):53-80.
Wolfe, B., and S. Hill (1995), "The effect of health on the work effort of single mothers", Journal of Human
Resources 30(1):42-62.
World Health Organization (1980), International Classification of Impairments, Disabilities and Handicaps
(World Health Organization, Geneva).
Worrall, J.D., and R.J. Butler (1986), "Some lessons from the Workers' Compensation Program", in:
M. Berkowitz and A. Hill, eds., Disability and the Labor Market (Industrial and Labor Relations Press,
Ithaca, NY).
Yelin, E., and M. Cisternas (1996), "The contemporary labor market and the employment prospects of persons
with disabilities", in: J. Mashaw, V. Reno, R. Burkhauser and M. Berkowitz, eds., Disability, Work, and
Cash Benefits (Upjohn Institute for Employment Research, Kalamazoo, MI).
Yelowitz, A. (1998), "Why did the SSI-disabled program grow so much? Disentangling the effect of Medi-
caid", JHE 12:321-349.
Ysseldyke, J.E., S.L. Christenson, M.L. Thurlow and D. Bakewell (1988), "Are different kinds of instructional
tasks used by different categories of students in different settings?", School Psychology Review 81:305-
311.
Zeitzer, I. (1995), "Social insurance provisions for children with disabilities in selected industrialized coun-
tries", Social Security Bulletin 58(3):32-48.
Chapter 19
CHILD HEALTH IN DEVELOPED COUNTRIES*
JANET CURRIE
Dept. of Economics, UCLA
Contents
Abstract 1054
Keywords 1054
1. Introduction 1055
2. A simple economic model of the demand for child health inputs 1056
3. Evidence regarding the utility of the simple model 1059
3.1. The Law of Demand 1059
3.2. Health as an investment 1062
4. The importance of failures in the market for child health 1063
4.1. Imperfect information 1063
4.2. Externalities 1065
5. The role of government in the market for child health: the provision of insur-
ance coverage 1067
5.1. Institutional background 1067
5.2. Insurance and children's health 1068
5.3. The efficiency of utilization of care 1071
5.4. Crowdout 1074
5.5. Insurance coverage and inequities in children's utilization of care 1075
5.6. Summary 1076
6. The role of government in the market for child health: other types of interven-
tions 1076
6.1. Direct provision of services 1077
6.2. Public health programs 1078
6.2.1. Education and informational campaigns 1078
6.2.2. Product safety and other government mandates 1080
6.2.3. Home visiting 1081
6.2.4. Incentives and/or sanctions 1082
6.2.5. Community-based programs 1082
*The author thanks the Canadian Institute for Advanced Research, and the National Institutes of Health for
financial support, and the editors for helpful comments. All opinions are those of the author and are not
necessarily supported by any funding agency.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P.Newhouse

O 2000 Elsevier Science B. V All rights reserved
1054 J Currie
6.2.6. Summary 1082

7. Directions for future research 1083
References 1084
Abstract
This chapter provides an overview of the literature on child health in developed coun-
tries. I first lay out a simple economic model of the demand for child health inputs, and
discuss whether the evidence is consistent with that model. Next, two main causes of
market failure in the market for child health inputs - lack of information and external-
ities - are analyzed. These failures may provide an economic rationale for government
intervention in the market for health care. Much of the literature on child health has
focused on one such intervention, the provision of public health insurance. However,
the utilization of health care is only one input into the production of child health, and
it is far from the most important input for most children. Hence, the last section of this
chapter offers a brief review of what we know about the effects of government interven-
tions designed to address other threats to child health. The chapter concludes with some
opinions about useful direction for future research and data collection efforts.
Keywords
children, infants, health
JEL classification: 112, I118, I11

Ch. 19: Child Health in Developed Countries 1055
1. Introduction
What can economics bring to the study of child health? Some possible answers are: the
recognition that people respond to economic incentives; the idea that expenditures on
child health can be regarded as investments by families and societies; the concept of
market failures and the resulting theoretical justification of government intervention in
the market; and the idea that cost-effectiveness is a criterion for making social choices.
While the relevant economic theory can be applied in any setting, this review focuses
on child health in developed countries. The primary reason is that there are several
excellent recent surveys dealing with child health in developing countries [cf. Behrman
and Deolalikar (1988) and Strauss and Thomas (1998)]. In contrast, recent changes in
the health insurance coverage of low-income children in the United States have sparked
renewed interest in research on children's health in rich countries, and this research has
not yet been summarized. Much of this work relies on changes in US institutions to
identify econometrically the effects of interest. While the institutions themselves may
be of limited interest to non-Americans, the broader conclusions that can be drawn
about the determinants of child health are of general interest, as are the identification
strategies that have been employed. A second body of recent literature focusing on child
health in developed countries deals with the relationship between child health and adult
health and well-being, and with the overall relationship between inequalities in health
and inequalities in economic status.
Although many of the relevant theoretical and empirical issues are the same, there
are also important differences in the way child health has been examined in rich and
poor countries. First, studies of children in developing countries often look at objective
measures of child health such as nutrient intakes, height, weight, or arm circumference.
In contrast, studies of children in richer countries often focus on the utilization of care
and maternal reports of child health. Reliance on measures of this type creates special
problems. For example, both utilization of care and maternal awareness of some condi-
tions rise with maternal education, so that while objective health status can be expected
to improve with maternal education, reported health may actually decline [cf. Dow et
al. (1997)].
Second, the responsiveness of child health to changes in the level of various health
"inputs" can be expected to be highly non-linear. For example, at very low levels, small
increases in caloric intake will save lives and prevent disease, while at higher levels,
small increases in the number of calories consumed might lead to weight gain but pose
no immediate threat to health. Similarly, children without any doctor visits can be pre-
sumed to be going without preventive and palliative care, but it cannot be assumed that
increasing the number of doctor visits from, for example, 3 to 4 per year will have any
measurable effect on health. Since children in richer countries generally consume health
care at a higher level than children in poorer countries, we might expect the size of any
measured effects of changes in health inputs to be smaller.
This chapter first lays out a simple model of demand for child health inputs (in Sec-
tion 2) and then discusses evidence consistent with that model (in Section 3). Section 4
1056 J Currie
discusses two main causes of market failure in the market for child health inputs - lack
of information and externalities. These market failures provide an economic rationale
for government intervention in the market for health care. Section 5 discusses one of the
most common forms of intervention in the health care market, the provision of public
health insurance.
However, the utilization of health care is only one input into the production of child
health, and there is overwhelming evidence that in most cases it is far from the most
important input. Improvements in standards of living, advances in knowledge about
disease and hygiene, and public health measures such as improved sanitation and the
provision of clean drinking water have done more to improve child health in the past
150 years than even the most spectacular advances in personal medical care [Preston
(1977), Mokyr (1998)].
Accidents and violence are the major killers of young children in wealthy countries
after the first year of life [MacArthur and Pless (1990)]. Accidents are often viewed as
unavoidable and violence is seen as a problem for the criminal justice system rather than
a public health problem [cf. Rosenberg et al. (1990)]. Yet variation in rates of death from
these causes across countries suggest that many deaths could be prevented. For example,
in the US there are approximately 10 deaths per 100,000 per year from external causes
compared to 4.3 deaths per 100,000 in the Netherlands [Williams and Miller (1992)].1
Moreover, substance abuse and poor eating habits threaten children from conception
and beyond.
Hence, Section 6 moves beyond the narrow focus on insurance coverage and the
utilization of care that has occupied so much of the literature and offers a brief review
of what we know about government interventions designed to address other threats to
child health. One of the main conclusions of this review is that economists interested in
child health should pay more attention to these public health issues. Overall conclusions
are presented in Section 7.
2. A simple economic model of the demand for child health inputs 2
In the standard economic model of the determinants of child health, parents are assumed
to maximize an intertemporal utility function such as:
Et(l/l1 + o)tUt + B(AT+i), (1)

t=l
Some of the difference is accounted for by a much higher rate of deaths in motor vehicle accidents in the
US (5 per 100,000 in the US compared to 2.5 per 100,000 in the Netherlands). Some of this difference in
turn may be due to differences in patterns of motor vehicle use, rather than to differences in required safety
equipment or driver education for example.
2 The model in this section is drawn with small alterations from Blau (1996).
where a is the discount rate, B is a bequest function, A denotes assets, and Ut is given
by:
Ut = U(Qt, Ct, Lt; Xt, ul, £lt), (2)
where Q is the stock of child health, C is consumption of other goods, L is leisure, X

is a vector of exogenous taste shifters, u 1 is a vector of permanent individual specific
taste shifters, and E denotes a shock to preferences. Utility is maximized subject to the
following set of constraints:
Qt = Q(Qt-l, Gt, Vt; Zt, u2, 2t), (3)

Ct = Yt - PgtGt - (At+ - At), (4)
Yt = It+ wtHt + rAt, (5)
Lt + Vt + Ht = 1, (6)
where G and V are material and time inputs into health production, Z is a vector of ex-
ogenous productivity shifters, u2 are permanent individual specific productivity shifters,
82t is a productivity shock, Y is total income, P represents prices, I is unearned income,
w is the wage, H is hours of paid work, r is the interest rate, and endowments of health
and assets, Q and Ao are assumed to be given. Equation (3) can be interpreted as a
"production function" for child health which describes the way that inputs are can be
converted into health.
Health inputs are valued by consumers not for their own sake, but because they affect
child health, which in turn has a direct effect on parental utility. Non-market time is an
input into both health production and the production of other valued non-market goods
(i.e. leisure activities). This model is dynamic in the sense that the stock of child health
today depends on past investments in health, and on the rate of depreciation of health
capital (which is one of the elements of u2).
The model can be solved to yield Frisch demand functions for Ct, Ht, Gt, and Vt of
the following form:
Ct , Ht, Gt, and Vt = F(iXt, Xt, Zt, wt, Pt, Mt, r, ar, ul, u,2 £Et, 2t), (7)
where t is the marginal utility of wealth and Mt is a vector of moments of the distri-
bution of {Xk, Zk, Wk, Pk, lk, E2k}, and k = t + 1, ... T.
Substituting the solutions for G and V into (3) yields a Frisch demand function for
Qt that is conditional on Qt-l:
Qt = Q*(Qt-1, t, Xt, Zt, wt, Pt, Mt, r, oa, ul, u2, it, 2t). (8)
Repeated substitution of lagged versions of (8) into (9), yields a Frisch demand for Qt
that is not conditional on Qt- l:
Qt = Q** (QO , Jt, , t, Zt, wt, Pt, Mt, r, a, ul, u2, Elt, 2t), (9)
1058 J. Currie
where Jt = Xk, Zk, Wk, Pk, Elk, 2k}, and k = ,...,t - 1.

Finally, substituting in the determinants of Xt yields a Marshallian demand function
for Qt:
Qt =Q***(Qo, A, Jt, XtZ,

Zt, t, It, Pt, Mt*, r, , u, U2, lt, 2), (10)
where Mt* is a vector of moments of the distribution of {Xk, Zk, Wk, Ik, Pk, Elk, E2k},
and k = t + 1, ... , T and Jt*= Xk, Zk, wk, Ik, Pk, Elk, 2k},and k = 1, ... , t - 1.
This model can be considerably simplified by assuming that the elements of M* are
functions of current and past realizations of the exogenous variables. In panel data, we
can also control for Qo, Ao, r, a, ul, u2 by including a child-specific fixed effect, t to
yield the following reduced form demand for health:
Qt = Q(J* t, Zt, wt, It, Pt, 1-,El2t), ()
Alternatively, we could start with (8), allow the elements of Mt to be functions of current
and past realizations of the exogenous variable, and capture At, r, a, u l, u2, with either
a child-specific or family specific fixed effect to arrive at:
Qt = Q"(Qt-1 J, Xt, Zt,

zt, Wt P, , eI, 82t). (12)
This simple model yields several insights. First, consumption of child health inputs will
fall as the price of these inputs rises. This observation can be used to lend structure to
discussions of "access" problems, for example. Large distances to providers, waiting
times, transportation problems and so on, can all be interpreted as things that raise the
"price" of health inputs and thus lower their utilization. Second, child health this period
depends on child health last period, which in turn reflects past health inputs. Thus, the
choice of inputs today will affect health not only today, but also in future. It is this
feature of the model that leads us to view health as an investment. Third, in this model
health care is not itself an outcome that provides utility. It is merely one of many inputs
into the production of healthy children. This perspective suggests that the literature
should place more focus on health outcomes, and on the relationship between inputs
and outputs rather than viewing the provision of health inputs as an end in itself.
Drawing out the relationship between health inputs and outcomes is however, easier
said than done. The fact that we do not know the form of the health production function
is one reason that it is so difficult to estimate a structural model of the type laid out
above. A second problem is that even if we knew all of the inputs, we would be hard
pressed to come up with adequate measures of most of them. Other problematic issues
include the endogeneity of some of the factors that affect household budget constraints
(such as marital status and number of children), and the correct modeling of bargaining
within the households (since mothers and fathers may place different weights on child
health).
3. Evidence regarding the utility of the simple model
3.1. The Law of Demand
There is a great deal of evidence that the "Law of Demand" holds when it comes to the
health care utilization of children: Higher consumer prices mean less care. Perhaps the
most compelling evidence regarding the relationship between utilization of care and the
price of medical care per se comes from the Rand Health Insurance Experiment.3 This
experiment, conducted from the mid 1970s to the early 1980s, covered approximately
2,000 non-elderly families. Families were assigned to one of 14 insurance plans which
varied the prices of available services: Some families had access to free care, while
others faced varying degrees of cost sharing. 4
The results of the experiment suggested that the price elasticity is non-linear in in-
come: There was no significant response to cost sharing among non-poor children, while
for poor children, there were significantly fewer claims for several diagnoses under cost
sharing. The reduction in claims for preventive services (vision examinations and gen-
eral medical exams) and for illnesses that may have deleterious long-term effects if
untreated (ear infections, acute upper respiratory infection, acute bronchitis, and var-
ious injuries) was particularly noteworthy. Cost sharing was also found to reduce the
probability of immunization: 59% of children in the free care program were fully im-
munized by 18 months, compared to 49% in the cost-sharing program. What may be
most remarkable about these immunization figures however, is how low they are even
for the free care group. They provide strong evidence that monetary costs are not the
only barrier to obtaining recommended preventive care.
In a further exploration of these results, the experimenters divided diagnoses into
categories ranging from those for which care was "rarely effective" to those for which
care was "highly effective". They found that cost sharing reduced the propensity to seek
"rarely effective" care for all children. However, among poor children, the probability
of making at least one claim for "highly effective" outpatient care was 56% of the
probability among those with free care. The equivalent figure was 85% for non-poor
children.
Additional evidence on the relationship between the consumer price of care and uti-
lization comes from non-experimental comparisons of the Medicaid and non-Medicaid
populations. Since 1965, Medicaid has been the main system of public health insurance
for low income women and children in the United States. It differs from many sources
3 Some of the results of the RAND Health Insurance Experiment have recently been replicated in a non-
experimental setting by Matthew Eichner (1997). Eichner takes advantage of the fact that in families where
one member has an accident that results in the insurance deductible being reached, other family members face
much lower costs of medical care for the rest of the year. He finds an estimated elasticity of demand of -0.32,
somewhat larger than those obtained from the RAND experiment.
4 For further information about the design of the experiment, see Newhouse and the Insurance Experiment
Group (1993).
1060 J. Currie
of private insurance in that there are no copayments or deductibles, and all preventive
services are covered. Thus, the model would lead one to predict that children on Medi-
caid were at least as likely to receive care as those with private insurance. The evidence
on this point is somewhat mixed however. Himmelstein et al. (1995) find that children
on Medicaid are similar to the uninsured in terms of their probability of being inade-
quately immunized. On the other hand, St. Peter et al. (1992) report that poor children
on Medicaid were more likely than other poor children to have a usual source of care
and to receive routine care within appropriate time intervals.
These mixed results may be due in part to the fact that children on Medicaid are likely
to differ from other children in unobserved ways. Currie and Thomas (1995) use data
from the National Longitudinal Survey of Youth (NLSY) to follow the same children
over time. They show that becoming covered by Medicaid increases the probability of
receiving sick visits to the doctor as much as private health insurance, but that it has a
greatereffect on well visits. This is consistent with the theory since many private health
insurance plans pay for sick visits but not for well-child visits.
A second complication is that the fees that providers receive for treating Medicaid
patients are often as much as 50% lower than those they would charge privately in-
sured patients. Thus, Medicaid patients may experience significant difficulties finding
physicians who are willing to treat them [cf. Currie, Gruber, and Fischer (1995)]. One
quarter of US doctors report that they do not accept Medicaid patients in their prac-
tice [Rowland and Salganicoff (1994)]. Among pediatricians, 23% do not participate in
Medicaid and 40% limit the number of Medicaid patients in their practices [Yudkowsky,
Cartland, and Flint (1990)]. Perhaps as a result, children on Medicaid are less likely to
be seen by private physicians, and more likely to be seen in hospital emergency rooms
and outpatient clinics. Decker (1992) finds that private doctors who do treat Medicaid
patients typically spend less time per patient than other doctors. Zambrana et al. (1994),
Riportella-Muller et al. (1996), and Brown (1989) all cite unfriendly or unsympathetic
providers as a barrier to care. These differences suggest that measurement of the ef-
fects of price changes may be complicated by the fact that poor parents may "purchase"
health care of different quality than wealthier parents.
Barriers to care such as transportation problems, long waiting times, language or
cultural problems, and lack of child care for other children, are often cited as particu-
larly salient for low income parents [cf. Zambrana et al. (1994), Riportella-Muller et al.
(1996)]. As discussed above, these barriers can be regarded as increasing the effective
"price" of medical care, and they may obscure the relationship between measured price
and quantity demanded if they are ignored. We observe large differences in the utiliza-
tion of care by socio-economic status even in countries with universal health insurance
coverage. This suggests that non-monetary barriers to care may be extremely important.
Yet virtually no research has been conducted examining the relevance of these factors
in a multivariate context. Currie and Reagan (1998) provides one exception. They show
that among central-city black children, the probability of having had a routine doctor's
checkup in the past year decreases with distance to hospital, suggesting that this group
is particularly reliant on hospitals for primary care.
In principle, further information about the effects of price of care on utilization could
be obtained by studying the growth of "managed care" in the United States. Both the
privately and publicly insured are increasingly enrolled in some form of managed care.
Much of the growth in managed care among Medicaid patients has resulted from the
liberalization of federal rules governing the Medicaid program in 1993. These changes
permitted state governments to conduct trials in which some federal regulations cov-
ering Medicaid (such as regulations that ensure freedom of choice of providers) are
waived, in return for expansions in the proportion of the population that is covered. Be-
tween 1993 and 1994, the fraction of the Medicaid population enrolled in managed care
grew from 14 to 23% of the caseload. Currently, over half of the Medicaid enrollees in
California and New York are in some form of managed care, and most of those enrolled
in managed care are low income women and children, rather than elderly or disabled
Medicaid recipients.
In many managed care systems, services are only paid for if they are approved by a
"gatekeeper" physician. But since all approved services are paid for in full (or with only
minimal co-payments), the plan has an incentive to promote the utilization of preventive
care while limiting or eliminating unnecessary use of more specialized care. However,
the evidence to date is mixed. A 1995 review of over 130 studies of Medicaid managed
care by the Kaiser Foundation [Rowland et al. (1995)] found that it reduced utilization
of emergency room and specialist care, but did not lead to any consistent changes in
the overall number of doctor visits. Similarly, access to preventive care did not seem to
consistently rise or fall, and remained lower for the Medicaid population than for the
non-poor population.
Several US General Accounting Office reports (1990a, 1995b) have criticized quality
control standards in Medicaid managed care plans. For example, the 1995 study found
that California had no way of verifying whether children in these programs had actu-
ally received necessary preventive services or not. Problems with quality control may be
particularly acute in the Medicaid population for several reasons. First, it is possible that
"churning" of the Medicaid caseload reduces a provider's incentive to control long-term
costs by promoting preventive care. In the past Medicaid coverage was related to receipt
of cash welfare. Since most welfare spells are short, most new Medicaid enrollees could
be expected to be on the program for relatively short periods. Short, Cantor, and Monheit
(1988) found that fewer than half of Medicaid enrollees were covered for a continuous
32 month period. I am not aware of comparable figures for privately insured children,
but since the average worker spends more than 20 years at his or her longest job, there
is reason to believe that children with private, employer-provided coverage might en-
joy a more stable source of insurance coverage. Nor is it clear how recent expansions
of Medicaid to persons of higher income and welfare reform may have affected stabil-
ity of coverage among the Medicaid-eligible. However, there is reason to suspect that
these changes may have increased churning since women on welfare may find it easier
to document their eligibility than women who are not. The evidence from states that
have started to enforce time limits suggests that many women who remain eligible for
Medicaid have lost their benefits due to such administrative problems [Rubin (1997)].
1062 J. Curie
3.2. Health as an investment
There is growing evidence that child health affects adult health, although the extent to
which children can recover from some insults to their health (for example, those caused
by under-nutrition) early in life is controversial. Indeed, some researchers believe that
health in the womb affects adult health. David Barker and his colleagues have linked a
number of adult disorders, including heart disease, to under-nutrition of the mother dur-
ing critical gestational periods [cf. Barker and Martyn (1992)]. Animal studies suggest
that stressful events or positive stimulation during "critical periods" can have long-term
effects on brain chemistry and on the endocrine system with resulting effects on health
[Hertzman and Wiens (1996)]. Since health affects wages and labor force participa-
tion [cf. Currie and Madrian (1999)], poor health in childhood can lower future utility
directly by affecting future health, and indirectly by lowering future wages and partici-
pation.
Child health may also be linked to adult labor market success, through its effects
on schooling [cf. Grossman (1975), Perri (1984), Wolfe (1985), Wadsworth (1986) and
Grossman (2000)]. Some of this effect may be direct, through lost school days, while
some of it may be indirect, through effects on cognition.
Many studies find positive effects of anthropometric measures of health such as birth-
weight, weight, height, head circumference, and absence of abnormalities on the cog-
nitive development (measured using test scores) of children of various ages. 5 For ex-
ample, Broman, Nichols and Kennedy (1975) look at 4 year olds; Edwards and Gross-
man (1979) examine white children 6 to 11 years old, Shakotko et al. (1981) look at
teenagers, while Wilson et al. (1986) examine children over 12 years old. Chaikind and
Corman (1991) and Rosenzweig and Wolpin (1994) look at the effects of birth weight
on later cognitive achievement. Outside of economics, the volume of literature on this
subject is even greater [cf. Brooks-Gunn et al. (1993), McCormick (1992)].
Kaestner and Corman (1995) examine tests of cognition and find positive effects of
birthweight, and negative effects of stunted growth (e.g., weight or height less than the
10th or 25th percentiles) in models estimated using cross-sectional data. These effects
largely disappear when child fixed effects are added to the model; however, given the
large amounts of measurement error in the test scores this result is perhaps to be ex-
pected. Alternatively, Kaestner and Corman suggest that their results may be weaker
than those of Rosenzweig and Wolpin (who use the same NLSY data) because Rosen-
zweig and Wolpin focus on a subsample of more disadvantaged children. That is, the
5 Birthweight is the single most important indicator of infant health since children of low birthweight (birth
weight less than 2500 grams) experience post-neonatal mortality rates 10 to 15 times those found among
infants of normal birthweight [US Office of Technology Assessment (1987)]. Weight can respond quickly to
changes in child health, and so is often regarded as a shorter-term measure of well-being than height, which
responds more slowly. Anthropometric measures like these reflect not only the effects of under-nutrition, but
also the effects of illness, since frequent illness interferes with growth. See Martorell and Habicht (1986) for
more discussion of the interpretation of various anthropometric measures.
ill effects of poor health on cognition may be greater for more disadvantaged children
than for children who are better off. All of these studies focus on US children. Koren-
man, Miller, and Sjaastad (1995) also find positive relationships between anthropomet-
ric measures and various measures of cognition in the NLSY data, and note that the
effects are particularly large for short-term memory.
The Kaestner and Corman paper also examines the effects of several maternally re-
ported measures of child health including the presence of a serious illness and the num-
ber of illnesses requiring medical attention. The former has a negative effect on cogni-
tive outcome measures in cross-sectional models, whereas the latter has no consistent
effect. Maternal reports are subject to a number of sources of bias. First, more edu-
cated and higher income mothers are likely to take their children to the doctor more
frequently, even though we would expect their children to be in better health. Second,
greater contact with health care providers may result in a higher probability of some
diagnoses (such as asthma). Third, mothers who report themselves to be in poor health,
may be more likely to report poor health in their children, whether or not the children
are actually sick [cf. Strauss and Thomas (1996)]. It is possible though that a dichoto-
mous measure of whether or not there is a serious illness is more accurately reported
than the objective number of illnesses.
This discussion suggests that investments in child health do pay off in the form of bet-
ter adult health, and superior educational outcomes. It also highlights important mea-
surement issues that are ubiquitous in this literature. Results may depend on whether
maternal reports of health status (e.g., good, fair, poor) or more intrinsically objective
measures such as height and weight are used. And even when more objective measures
are chosen, it is important to determine whether or not they are directly observed or
based on maternal reports. Strauss and Thomas (1996) give an interesting example from
the NLSY. In some cases, maternal reports of heights and weights were used rather than
direct measures. Maternal reports were more likely to be used for more educated moth-
ers. But since children grow, maternal reports tend to understate child height. Thus,
there is a spurious negative relationship between maternal education and child height
and weight in these data.
4. The importance of failures in the market for child health
While the simple economic model of household decision making is useful for under-
standing many aspects of the demand for health care, the health care market is a great
deal more complex than the market for oranges, for example. Important differences
arise because of market failures, especially imperfect information and the presence of
externalities.
4.1. Imperfect information
There is a strong presumption in the medical literature that many parents lack appro-
priate information about the care their children need. It has been shown, for example,
1064 J. Currie
that the advice received during prenatal care visits can have a significant impact on the
incidence of low birth weight [Kogan et al. (1994)]. And home visiting programs, which
generally stress parent education as well as the monitoring of at-risk children, have been
shown to improve child health along a number of dimensions including reductions in
the risk of injury and improvements in cognition [cf. Gomby et al. (1993)]. The evi-
dence regarding under-immunization of preschool children in the United States is also
suggestive of a lack of information: Most under-immunized children visited their health
care providers 6 to 8 times during a given year [US General Accounting Office (1995a)]
but did not receive recommended shots, perhaps because parents did not realize that
they should receive them. Miller et al. (1994) present more direct evidence that chil-
dren of parents who did not know the recommended age for Measles-Mumps-Rubella
vaccination were more likely to have delayed immunizations, although the direction of
causality is not clear. Finally, there is a good deal of evidence that the probability that a
child receives recommended preventive medical and dental care increases with maternal
education, regardless of family income and family structure [Zill (1996)]. Thus, lack of
knowledge on the part of parents is one explanation for under-utilization of preventive
care.
A key insight of models of physician-induced demand such as those of Pauly (1980)
and Dranove (1988) is that asymmetric information can also lead to excessive con-
sumption of medical services if physicians take advantage of their superior information
to "sell" services that patients do not need. In Dranove's (1988) model, this capacity to
sell services is limited by the fact that patients will not always buy them - doctors who
over-prescribe treatments may develop a reputation that causes patients to view their
recommendations with skepticism. In the model, the extent of demand inducement is
shown to be higher for services that have high clinical value when they do happen to be
warranted. For example, antibiotic treatments are very effective for bacterial infections
but not at all effective against viruses, and so the model might predict that antibiotics
would be over-prescribed. Demand inducement also rises with the supply of physicians.
Inducement falls with the price of the service and with increases in the patient's diag-
nostic skills, which in turn may be correlated with the patient's level of education and
income.
This model suggests that there may be considerable scope for inducement in the mar-
ket for children's health care: Most services required by children are inexpensive, and
services such as antibiotic treatments for ear infection may have high clinical value
when they are warranted. Moreover new parents, in particular, may lack diagnostic skill.
There is evidence that medically inappropriate pediatric hospitalizations are sensitive to
insurance status - one study found that among the uninsured, 14% of pediatric hospital
days were inappropriate compared to 22% of hospital days among children with insur-
ance [Kemper (1988)]. The results of the RAND Health Insurance Experiment are also
broadly consistent with some of these predictions: Cost sharing reduced demand more
among the poor than among the rich, and the demand for less effective services fell by
more than the demand for highly effective services. However, these differences were
not statistically significant among children.
Models of induced demand are premised on the idea that doctors have better infor-
mation than parents. But the fact that it is difficult for doctors to assess adequately the
risk status of patients without agreeing to see them is a form of asymmetric information
that may also contribute to inadequate health care for children. One possible reason is
that doctors may use public insurance as a marker for "high risk". Patients with public
insurance are generally poor and more likely to suffer from other factors that place them
at risk. If doctors use insurance status to screen out potential patients, then they are in
effect practicing a form of statistical discrimination because they cannot observe actual
risk status without treating the patient.
This observation suggests that even if there were no difference in fees between pub-
lic and private insurers, higher costs of treating high-risk patients might still lead to a
"two-tier" system in which some doctors opted out of the public program, or otherwise
acted to limit their involvement with low-income patients. Physicians have been found
to spend more time with patients who are "well-groomed" [Hooper et al. (1982)], and
poverty and minority status have been found to have effects on children's access to care
that are independent of insurance status [Newacheck et al. (1996)]. Travis (1996) finds
that increases in Medicaid fees paid to physicians in Washington state increased utiliza-
tion of prenatal care for whites and Hispanics but not for blacks. While discrimination
is not the only explanation for this pattern of results, it is certainly consistent with them.
Seminal papers in the asymmetric information literature focus on lack of informa-
tion among insurers [cf. Akerlof (1970)]. If insurers cannot distinguish high-risk from
low-risk individuals, then at any given price it will be the higher risk individuals who
purchase insurance. Insurers will be forced to keep raising the price of insurance until
the market fails and insurance becomes unavailable. The health insurance market seems
to function in such a manner in the United States - families that do not receive public
insurance, and are unable to purchase insurance in large pools created by employers,
often remain uninsured because the price of insurance is unaffordably high.
A final possibility is that appropriate information about medical care is simply lacking
on all sides. For example, a recent survey of US pediatricians found that one quarter did
not understand that breast milk is the best food for infants, and 40 to 50% did not know
that supplementary feedings in the first weeks of life can lead to breast-feeding failure
[Freed et al. (1995)].
4.2. Externalities
The most obvious externalities are related to the transmission of disease - children who
receive inadequate care and become ill may pose a serious threat to others. Parents who
do not take account of costs imposed by these externalities may not provide the optimal
level of care for their children [cf. Philipson (1996)]. It is also possible that parents, who
must trade off child welfare against other goods, will not choose the level of investment
in health care that the child would choose if he or she were able to make his or her own
decision. This too can be regarded as an externality. Finally, as long as society feels an
obligation to provide emergency care for children as well as services for children with
1066 J. Currie
special needs, then inadequate prenatal care or primary care of children will impose
fiscal externalities on all taxpayers.
There is a good deal of evidence that children in the United States have not been
receiving optimal levels of at least some forms of preventive care. During 1989 and
1990 the United States experienced a measles epidemic among preschoolers that led to
43,000 cases and more than 100 preventable deaths. Since measles is the most conta-
gious of the vaccine-preventable diseases, this increase may signal that many children
are also vulnerable to pertussis, polio, mumps and rubella [The National Vaccine Ad-
visory Committee (1991)]. Indeed, the incidence of pertussis (whooping cough) also
rose in the late 1980s and early 1990s [US Centers for Disease Control and Prevention
(1992)].
Low birthweight, smoking while pregnant, drug-exposure in utero, lead poisoning,
and anemia are all preventable, and can lead to cognitive deficits that impair children's
school performance, increasing costs for special education and possibly for social pro-
grams down the road [Chaikind and Corman (1991), Olds et al. (1994), Hoyme et al.
(1990), Parker (1989), Needleman et al. (1996)]. For example, Chaikind and Corman
find that in addition to any extra medical costs incurred by children of low birthweight,
these children are 50% more likely to be enrolled in special education classes. They
calculate that these extra enrollments raised special education costs by $371 million
dollars in 1989. The risk of abnormal neurodevelopmental handicaps is 3 times higher
among low birthweight infants (those less than 2,500 grams) and 10 times higher among
infants of very low birthweight (those less than 1,500 grams). The rate of impairment
approaches 25% among very low birthweight infants, and includes survivors with men-
tal retardation, cerebral palsy, blindness, and deafness.
Low birthweight and exposure to illicit drugs are also linked to higher neonatal med-
ical costs. One hospital reported a median cost of $5,500 1990 dollars per drug-exposed
infant compared to $1,400 for an infant who had not been exposed. These infants have
longer hospital stays, are more likely to end up in intensive care, and between 25 and
60% of them will require foster care [US General Accounting Office (1990b)]. Lead
poisoning has been linked to delinquency in children and teenagers and to adult crim-
inality, which clearly imposes externalities on those outside the affected individual's
family [Needleman et al. (1996)].
Preventable hospitalizations and inappropriate use of expensive emergency room care
also impose high costs on society. Casanova and Starfield (1995) find that the rate of
pediatric hospitalizations for ambulatory care sensitive conditions is lower in Spain, a
country with universal health insurance, than in the United States. They attribute this
finding to greater utilization of primary preventive care and to the availability of out-
patient alternatives to hospitalization in Spain. Lower socio-economic status (SES), re-
liance on neighborhood clinics for well-child care, and living in an area with relatively
few primary care physicians are all correlated with a reliance on emergency rooms for
routine sick care [Halfon et al. (1996)]. To the extent that parents are unable to pay for
care in hospitals and emergency rooms, hospitals will pass on the costs to governments
and to the privately insured. Uncompensated care is a significant expense for Ameri-
can hospitals: For example, Long, Marquis, and Harrison (1994) found that in 1989,
bad debt and charity care accounted for about 9% of the costs incurred by hospitals for
maternal and infant care.
5. The role of government in the market for child health: the provision of
insurance coverage
These market failures create a case for government intervention to assure that all chil-
dren get necessary preventive care. A good deal of research has been devoted to eval-
uating the sometimes unintended consequences of these interventions. Much of this
research focuses on recent changes to Medicaid, the main US program providing in-
surance to low-income children. Because health insurance coverage for low-income
children has been phased in gradually in the US, it is possible to use differences across
states and over time to analyze the effects of providing insurance. In contrast, most Eu-
ropean countries offer the analyst a single before/after comparison so that it is difficult
to control for other confounding factors. Moreover, attempts to measure the effects of
health insurance by comparing countries with different systems are suspect because of
the great differences in other factors across countries.
While there are also differences between American states, it seems reasonable to
assume that these differences are smaller than those that exist between countries. Thus,
one can view the recent changes in the United States as a unique opportunity to address
some general questions about the effects of extending public health insurance to low-
income children. This "experiment" can be used to address several questions. First, can
it be shown that public insurance has a positive effect on the utilization of care and on
children's health? Second, how efficiently is care provided under a means-tested public
health insurance system? And third, to what extent do differences in access to care by
socio-economic status stem from differences in the availability of health insurance?
5.1. Institutionalbackground
The United States does not have universal health insurance coverage, but does have
public insurance programs which cover the elderly, the disabled, and some women and
children in poor families. Medicaid is the government program that covers the latter
group. Medicaid was implemented in the late 1960s and early 1970s and was phased
in at different rates across the states. From its inception until the mid-1980s, Medicaid
coverage was tied to the receipt of cash welfare benefits. Income thresholds for welfare
varied widely across states, but in general only female-headed households were eligible
for benefits. Hence, as many as 30% of poor children lacked health insurance coverage
[Bloom (1990)]. A good deal of research has established that uninsured children have
lower utilization rates, a less efficient distribution of utilization across sites of care, and
worse health outcomes [cf. Kasper (1986), Short and Lefkowitz (1992)].
1068 J. Currie
In response to this lack of coverage and to rates of mortality and morbidity among
US children that were higher than in many other developed countries, the US govern-
ment began expanding Medicaid eligibility to previously uncovered groups of pregnant
women and children in the mid-1980s. By July 1, 1991, states were required to cover all
children born after Sept. 30, 1983 whose family incomes were below the poverty line.
Currie and Gruber (1996a) estimate that these Medicaid expansions roughly doubled
eligibility for Medicaid coverage among women of child bearing age from 15 to 35%,
while Currie and Gruber (1996b) find that eligibility among children increased from
15 to 30%. It may surprise readers who think of the US medical insurance system as
primarily private to learn that approximately 40% of births were paid for by Medicaid
in 1995 [National Governor's Association (1997a)].
Typically, states were first given the option of extending coverage to specific groups,
and then required to do so. The important point is that since states took up these options
at different rates, and programs varied tremendously in terms of generosity to begin
with, there has been a great deal of variation across states in both the income thresholds
and the age limits governing Medicaid eligibility. 6 This variation in eligibility across
states, years, and child age groups, can be used to identify the effects of eligibility for
public insurance among the poor and near-poor children who became eligible.
5.2. Insuranceand children'shealth
The least controversial measure of health is mortality. Mortality is affected both by

underlying health status and by medical care received, and it is not possible to separate
out these factors using mortality data alone. In the case of infants, it is possible to
proxy underlying health status using birth weight. For example, Horbar et al. (1993)
found that in a sample of very low birth weight children weighing between 601 and
1300 grams at birth, each increase in birth weight of 100 grams was associated with
a decrease of approximately 10% in the probability of death, other things being equal.
In contrast, infant mortality rates reflect not only the health of the fetus but also the
effect of any interventions that occur during or after birth. New technologies have had
dramatic effects on the survival rate of low birth weight infants. Buehler et al. (1985)
report that improvements in birth weight-specific mortality rates accounted for 91% of
the overall decline in neonatal mortality in the US between 1960 and 1980.
Though it is effective in saving lives, high-tech neo-natal care is very expensive 7 and
surviving low birth weight babies are at increased risk of handicaps such as cerebral
6 This point can be illustrated with reference to 11 of the largest states. As of January 1988, 8 of these
states had taken advantage of the option to extend Medicaid coverage to previously ineligible children. By
December 1989, all 11 of them had done so. However there was still a great deal of variation across states
in the generosity of the program. For example, Pennsylvania had extended coverage to children up to 6 years
old in families with incomes below 100% of the poverty line, while New York only covered children up to
one year old, but extended coverage to infants in families with incomes up to 185% of the poverty line.
7 For example, in 1984, the cost of caring for a surviving baby less than 2500 grams in the US was $9,712
compared to $678 for a heavier infant [Office of Technology Assessment (1987)].
palsy of significant degree, major seizure disorders, blindness, deafness, and learning
disabilities [McCormick et al. (1992)]. Hence, for both humanitarian and economic
reasons, the best way to reduce infant mortality is to reduce the incidence of low birth
weight and pre-term delivery through appropriate use of preventive prenatal care.
Accordingly, Currie and Gruber (1996a) focus on the recent extensions of Medicaid
eligibility in the US to pregnant women and infants and ask whether these extensions
reduced the incidence of low birth weight in addition to reducing the infant mortality
rate. They use state rules to calculate the fraction of 15 to 44 year old women in the
March Current Population Survey (CPS)8 who would be eligible for Medicaid coverage
in the event of pregnancy in each state and year from 1979 to 1990. They then estimate
models in which the fraction of low birth weight infants in the state, and the state infant
mortality rates are functions of the fraction of women who are potentially eligible. State
and year dummies are included in the models in order to control for any state or year
specific determinants of mortality.
A possible drawback to this strategy, is that the since Medicaid is means-tested, the
actual fraction eligible for Medicaid may depend on business cycle effects, or on omit-
ted variables specific to states and years. It is possible to construct an eligibility measure
that reflects only variations in state rules by using a nationally representative sample and
calculating the fraction of women in this sample who would be eligible for Medicaid
in each state and year. This "simulated eligibility" measure will be exogenous as long
as state rules can themselves be treated as exogenous variables. The plausibility of this
assumption is bolstered in this case by the fact that much of the change in legislation
(though not all) was in response to federal mandates, and thus not directly under the
control of state legislators. An additional advantage of this procedure is that sampling
variation due to the fact that there are small cell sizes in some states and years is elimi-
nated.
Use of this simulated measure suggests that the estimated effect of Medicaid eligi-
bility is indeed biased towards zero when actual eligibility is used. Hence, the fraction
eligible is positively correlated with omitted characteristics of states (such as poverty)
that are themselves correlated with poor birth outcomes. Estimates using the simulated
eligibility measure suggest that the observed 20 percentage point increase in eligibility
over the 1980s reduced the incidence of low birth weight by 2% and the incidence of
infant mortality by 8.5%. Cole (1994) reports similar results regarding the incidence of
low birthweight using county-level data.
This finding supports an earlier study by Hanratty (1996) which showed that the
introduction of universal health insurance in Canada was associated with a decrease in
the infant mortality rate. 9 As in the US, health insurance was adopted by the Canadian
8 The CPS is used to generate unemployment rate figures and many other government statistics. About
100,000 individuals are surveyed each month, but the nature of the questionnaire changes from month to
month and the most detailed demographic information is asked in March.
9 In 1966, the federal government passed the Medical Care Act which offered federal funding to all provinces
that introduced comprehensive universal coverage. Between July 1, 1968 and April 1, 1972, all the Canadian
provinces took up this offer.
1070 J. Currie
provinces at different rates. She also asked whether the introduction of national health
insurance increased birthweights, with ambiguous results.
It is more difficult to document a relationship between public insurance coverage and
health among older children. Currie and Gruber (1996b) explore this issue using data
about children less than 15 years of age from the pooled 1984 to 1992 waves of the
0
US National Health Interview Surveys (NHIS). By combining household information
with information about state laws and the child's age, it is possible impute Medicaid
eligibility for each child. Perhaps the most useful measure of utilization available is
whether or not the child has been to the doctor in the past 12 months. Since every child
should receive at least one visit a year, this measure is a relatively clean indicator of
utilization that is not confounded by morbidity. On the other hand, it is not clear that
the productivity of most visits (in terms of improvements in child health) is very high.
Perhaps surprisingly, Medicaid-eligible children are as likely to have at least some
visits in the year as other children. However, they are much more likely to be reported
by their parents to be in poor health. Hence, conditional on need, it appears that Medi-
caid eligible children may be less likely to receive care. The Medicaid children are also
younger on average, which is to be expected given the way the expansions have been
phased in covering younger children first - and one might expect younger children to
have more doctor visits, other things being equal. Medicaid-eligible children are dis-
advantaged in most observable respects and it is also likely that they differ from other
children in ways that are not observed, and that may be correlated with utilization of
care and health status. Hence, simple comparisons of the effects of Medicaid eligibility
on the variables of interest are likely to be subject to omitted variable bias.
In order to address this problem, Currie and Gruber (1996b) instrument individual
eligibility using the simulated fraction of children in the same state, year, and age group
who are eligible for Medicaid. Dummy variables for state, year, and each year of age, as
well as interactions between age and state and age and year are included in both the first
and second stage regressions. These variables control for characteristics of states, years,
and ages that could be correlated with the outcome variables, as well as for omitted
variables such as nation-wide technology changes or state programs affecting specific
age groups. Using this instrumental variables procedure they find that Medicaid eligi-
bility increases the utilization of care - the probability that a child did not receive a
doctor's visit in the past year falls 10 percentage points from a baseline level of 19 per-
cent. That is, becoming eligible for Medicaid is estimated to reduce the probability of
going without a doctor's visit by more than half. In contrast, the OLS estimate is only
2.5 percentage points suggesting that these estimates are indeed biased towards zero by
omitted variables.
to The sample is limited to children less than 15 years of age to avoid complications due to "children having
children" - under the Medicaid expansions teenage girls could become eligible for Medicaid coverage inthe
event of pregnancy, even ifthey were not previously eligible.
11On the other hand, number of doctor visits would not be a good measure of health, since conditional on
measures of medical need, lower SES individuals get fewer visits [Aday (1975)].
Currie and Gruber (1996b) also estimate models in which aggregate state-level child
mortality rates depend on the fraction of children eligible in each state, year, and age
group.12 Using this objective measure of child health they find that the 15% increase
in the fraction of children eligible for public insurance between 1984 and 1992 was
associated with 0.2 percentage point decline in child mortality, which translates into a
5.1% decrease in child mortality. Moreover, this difference is statistically significant for
deaths from internal causes such as disease (which one might expect to be affected by
medical intervention), but not for external causes such as accidents, homicides, etc.
In summary, the evidence in this section suggests that extending public insurance to
pregnant women and children is associated with significant decreases in infant and child
mortality, and with increases in the use of preventive care among children.
5.3. The efficiency of utilization of care
A second set of questions that can be posed is whether these improvements in health
were "purchased" in the most efficient way possible? If it is true that "an ounce of
prevention is worth a pound of cure" then we would hope that public insurance increased
the use of preventive care as well as care for sick children. More generally, one might
want to step back a bit further and ask whether spending money on providing medical
care (of whatever variety) for children is the most cost-effective way to improve their
health?
Cole (1994) tackles the question of whether insurance promotes preventive or pal-
liative care directly by examining the effect of the Medicaid expansions on the use of
prenatal care. She follows a two-step procedure. First, she regresses individual-level in-
formation about birth weight and prenatal care from American Vital Statistics records
(1983 to 1990) on maternal characteristics. Then, she aggregates the residuals up to
the county level, and regresses these measures on the fraction of Medicaid eligibles
in the county and on other county characteristics. This procedure "nets out" variations
between counties in the characteristics of their residents. In a sample of high poverty
counties, she finds that an increase in the fraction eligible increased the use of prenatal
care: The 20% increase in eligibility that took place over the 1980s is estimated to have
reduced the fraction of women who delayed obtaining prenatal care beyond the first
trimester from 22% to 17%.
She also aggregates the individual-level data by state, race, age group, and marital sta-
tus, and alternatively by state, race, age group, and education group. These regressions
show that among teenaged mothers and unmarried mothers the Medicaid expansions
significantly increased the utilization of prenatal care and reduced the incidence of low
birth weight and preterm births. Among these "at-risk" populations, a 20 percentage
point increase in Medicaid eligibility was associated with a 25% decline in the number
of women delaying prenatal care, and decreases in the incidence of low birth weight
and preterm births of approximately 4% and 8%, respectively.
12 Data are available for two age groups: children 1 to 4 and children 5 to 14.
1072 J Currie
On the other hand, Currie and Gruber (1996a) find evidence that some groups of
newly eligible pregnant women were not drawn into prenatal care. In particular, they
distinguish between expansions that were targeted to very low income women who had
been ineligible for Medicaid because of family structure (e.g., because they were mar-
ried), and broader expansions that made women eligible by raising income thresholds.
They find that only the more narrowly targeted expansions improved child outcomes al-
though both types of expansions raised Medicaid costs. The targeted expansions raised
expenditures for both outpatient and inpatient hospital services, while the broad expan-
sions increased only expenditures on hospitals.
Why were the broader expansions to women of higher income less effective and what
conclusions can be drawn from this about the efficacy of means-tested public health
insurance programs? It appears that the broader expansions failed to increase Medicaid
coverage.13 That is, women who became eligible for Medicaid as a result of higher
income thresholds did not take up their new benefits. On the other hand, poorer women
did take up their benefits at about the expected rate. 14 These results hold even when the
sample is restricted to women without private insurance coverage.
The problem may be due to the fact that women who became eligible as a result of
higher income thresholds were unfamiliar with the social insurance system, and did not
realize that they were eligible for Medicaid coverage. This interpretation is supported by
case studies. For example, Piper et al. (1990) found that a 1985 expansion of eligibility
in Tennessee increased Medicaid enrollments, but that most of the increase took place
within thirty days prior to delivery.
Clearly, this represents a very inefficient pattern of usage. The women did not take
advantage of the insurance to obtain prenatal care, but appear to have been enrolled
when they arrived at the hospital to deliver. This pattern is consistent with increases in
costs only for hospital services. The fact that coverage of pregnant women went up at
all probably reflects the incentives that hospitals have to recoup the costs of treating
otherwise uninsured women from the Medicaid program.1 5
Note that this example suggests that eligibility may have very different effects than
coverage. Eligibility may affect the provider's incentives to provide certain services to
women in labor or to children seeking attention in hospital clinics and emergency rooms.
13 Individuals were asked whether they were covered by Medicaid at any point in the past year. Currie and
Gruber use each woman's socio-economic characteristics to calculate whether or not she would have been
eligible for Medicaid in the event of pregnancy, under either the targeted or the broad expansions. Coverage
is then regressed on eligibility. The models control for demographic characteristics and income as well as for
state and year. There are approximately 456,000 observations. Their results regarding low take-up of means
tested social insurance programs are consistent with those of Blank and Ruggles (1996) and Blank and Card
(1991).
14 They estimate that for every 100 women made potentially eligible under the targeted expansions, 9.6 took
up their benefits, which represents almost full takeup given current fertility rates.
15 Hospitals that accept any federal funds are not permitted to turn away a woman who is already in labor
[US Office of Technology Assessment (1987)] and maternity care is the single largest component of uncom-
pensated care for US hospitals [Saywell et al. (1989)].
Coverage also reflects the parent's decision to seek care. This distinction between the
effects of eligibility and coverage does not arise in a system with universal health insur-
ance coverage. But it is an important feature of the US system, and one that has been
largely ignored to date.
In a related study using Vital Statistics data on every birth in the US over the 1987 to
1992 period, Currie and Gruber (1997) find that women who became eligible for insur-
ance coverage of their pregnancies were more likely to receive a number of obstetrical
procedures. There was however, little evidence of associated changes in the health of
newborns except among women whose closest hospital had a Neonatal Intensive Care
Unit. These results suggest that a considerable amount of the procedure use that is in-
duced by increases in the generosity of public insurance coverage (e.g., increases in the
use of Cesarean sections) may have only marginal health benefits.
While the studies discussed above focus on infants, Baker and Royalty (1996) use
data from a longitudinal survey of California physicians observed in 1987 and 1991 and
find that expansions of Medicaid eligibility to previously uninsured women and chil-
dren increased the utilization of care provided by public clinics and hospitals but had
little effect on visits to office based physicians. On the other hand, using the methods
described above, Currie and Gruber (1996b) find some evidence that the eligibility ex-
pansions increased the number of visits to doctor's offices but did not affect the number
of visits to other sites of care. In principle, one would like to shift visits from emergency
rooms and hospital clinics to office-based physicians because in addition to possibly re-
ducing program costs, such a shift would likely improve the continuity of care. But the
question of whether or not such a shift has occurred remains open given the conflicting
evidence and the limited amount of research to date.
Taken together, these studies suggest that while the expansion of insurance coverage
to the low income population in the United States did improve infant and child health,
it may not have done so in the most efficient way possible. They point to one important
benefit of universal programs - if everyone is covered, people will not fail to seek care
solely because they are unaware of their entitlements. In a system in which only some
children are eligible, outreach may be very important. Several states have accompanied
changes to their Medicaid programs with such programs. To the extent that people are
aware of their eligibility but still fail to seek care, the elimination of other administrative
barriers to becoming covered may also be warranted. In the same way that evaluations
of the Medicaid eligibility expansions have contributed to our knowledge of the benefits
and limitations of insurance coverage, evaluations of outreach programs and efforts to
simplify administrative procedures could shed more light on the importance of other
non-price barriers to care. 16
16 See US General Accounting Office (1994) for a description of some of the problems facing people wishing
to establish Medicaid eligibility.
1074 J. Currie
5.4. Crowdout
In 1993, 67% of US children were covered by private health insurance provided pri-
marily by their parent's employers, 20% were covered by Medicaid, and 13.5% were
uninsured [US General Accounting Office (1995c)]. A good deal of recent research fo-
cuses on the issue of whether public insurance tends to "crowd out" private insurance.
The figures on insurance coverage for children are extremely suggestive: Despite the
dramatic increases in public insurance coverage discussed above, the fraction of chil-
dren without insurance coverage has stayed remarkably constant in recent years because
private health insurance coverage has fallen by the same amount that public insurance
coverage has risen [US General Accounting Office (1995c)]. However, private health
insurance coverage has also been falling among groups that one would not expect to be
affected by the Medicaid expansions, such as single men. Thus, it is not obvious to what
extent the relationship between increases in public insurance and decreases in private
insurance is causal.
Estimates of the extent of crowdout are sensitive to the methods used to control for
possibly pre-existing trends in the provision of private health insurance coverage. At the
high end of the spectrum of estimates, Cutler and Gruber (1996, 1997) estimate that for
every two people covered by the Medicaid expansions, one person lost private health
insurance. However, some of these people (such as household heads who decided they
would no longer purchase health insurance once their children became eligible) were
not themselves eligible for Medicaid - so not all of the people crowded out ended up
getting insurance at public expense. They calculate that in fact about 40% of those
crowded out ended up on Medicaid. Other observers have posed the question somewhat
differently, and come up with corrTespondingly different estimates. For example, Dubay
and Kenney (1997) find that about 22% of the increase in Medicaid coverage came from
people who used to be privately insured. Since not everyone who became eligible for
Medicaid did so as a result of the expansions, this number is necessarily smaller than
Cutler and Gruber's estimate. Finally, one might ask what share of the overall decline
in private insurance coverage is a result of the Medicaid expansions. The answer to this
question is about 15%, which suggests that a great deal of research remains to be done
on the causes of this decline.
One issue obscured by the focus on crowdout is the fact that Medicaid insurance cov-
erage may be better than what is privately available to many families. For example, as
discussed above, many private policies do not cover routine pediatric preventive care
such as immunizations. Hence, the substitution of Medicaid for private insurance cov-
erage may improve children's health care, and this improvement should be valued when
the costs and benefits of the expansions are weighed.
From a societal point of view, it does not matter whether private or public insurers
pay for health care, except in so far as taxation creates a dead-weight loss, but policy
makers reluctant to raise (or eager to cut) taxes are deeply concerned about this issue.
If public insurance does consistently crowd out private insurance to a significant extent,
then it may prove impossible to have the two systems side by side in equilibrium.
5.5. Insurancecoverage and inequities in children'sutilization of care
The third question posed above was the extent to which disparities in the utilization of
care can be attributed to lack of insurance coverage. In the United States, race is often
used as a proxy for socio-economic status. For example, Pui et al. (1995) reports that
higher mortality rates among black pediatric cancer patients reflect inferior care, but it
is not clear to what extent this is due to differential insurance coverage. It is possible
that insured children from low SES families will receive less care than other children
because their parents are not informed about the value of medical care, or because they
face other barriers such as transportation or child care costs.
In addition, children insured by the Medicaid program may lack access to care be-
cause many providers do not accept Medicaid payments, or limit the amount of time
that they spend with Medicaid patients [Sloan, Mitchell, and Cromwell (1978), Decker
(1992)]. Doctors have many reasons for refusing to participate in the Medicaid program,
ranging from low reimbursement rates relative to those paid by private insurance, to red
tape, to fears that high-risk patients may sue. As discussed above, some providers may
prefer to avoid poor, high-risk patients even in countries with universal health insur-
ance. It has been argued that in the US, problems of access are particularly acute among
black children because of residential segregation [Fossett and Peterson (1989)]. On the
other hand, urban teaching hospitals tend to be located in inner-city areas with high
concentrations of African-American residents - hence children in these locations may
have access to the best quality hospital care even if they find it difficult to gain access to
office-based physicians for primary care.
Currie and Thomas (1995) ask whether Medicaid coverage has differential effects on
whites and blacks. As discussed above, they use data from the National Longitudinal
Survey of Youth (NLSY), which has followed children of the initial respondents since
1986. The longitudinal nature of the data allows them to include a fixed effect for each
child which controls for any unobserved, constant, characteristics of the home envi-
ronment and of the child that might be correlated with Medicaid status. Their estimates
indicate that both private insurance coverage and Medicaid coverage are associated with
a higher number of visits for illness among white children, while for black children, in-
surance coverage has no significant effect on the number of sick child visits. A possible
explanation for this result is lack of power due to a relatively small sample of black
children. However, as discussed above, the study did find a significant effect of insur-
ance coverage on the utilization of preventive care. Thus, it is unlikely that low power
accounts for the lack of a significant impact on sick visits among blacks.
The study does not resolve the question of whether these racial disparities can be
attributed to a shortage of providers, other barriers to care, or differing attitudes to-
wards care. The influential Black report in Great Britain concluded that class-based
disparities in mortality actually increased following the introduction of national health
insurance and that other policies might be more effective in promoting equality in health
outcomes [Townsend, Davidson, and Whithead (1988)]. As discussed below, many Eu-
ropean countries have acted to remove other potential barriers by sending public health
1076 J. Currie
nurses directly to children's homes, requiring parents to bring children to designated

centers for care, or rewarding parents who bring their children in for care and penaliz-
ing those who do not. It would be interesting to conduct rigorous evaluations of these
programs.
5.6. Summary
The American example suggests that means-tested public insurance programs improve
the utilization of preventive care and child health, but that when introduced alone, they
may not result in a very efficient allocation of medical resources. In particular, there is
some evidence that the eligible under-utilize preventive care which may lead in some
cases to over-utilization of sick care. Similarly, there is a great deal of evidence that the
poor are more likely to seek care from expensive sources such as emergency rooms and
hospitals rather than from doctors in private practice.
To the extent that universal eligibility and aggressive public health programs address
these problems, they may result in a more efficient allocation of medical resources.
A second conclusion is that minority children appear to benefit less from public insur-
ance coverage than other children. It is possible that research in Europe could shed light
on whether this finding is peculiar to the American system or reflects a more widespread
form of "discrimination" against lower SES individuals.
6. The role of government in the market for child health: other types of
interventions
Despite the increasing availability of sophisticated sick care for children, child mortal-
ity rates are higher in the US than in Europe, Japan, or Canada for all age groups, and
especially among 1 to 4 year olds and 15 to 19 year olds. Some of this difference may
reflect under-utilization of basic preventive care: Immunization rates for US preschool-
ers have been lower than in many European countries (though they are improving), and
US children are less likely than their counterparts in these other countries to have a
regular source of medical care [Williams and Miller (1992)]. But since, for example,
very few American children die of communicable diseases, we must look to other dif-
ferences between the US and other countries if we wish to explain American "excess
child mortality".
Some of the mortality gap may reflect the effects of a broad array of public health
measures that are in place in other countries. However, even their advocates admit that
the health policies in place in most developed countries have not been subjected to any
systematic evaluation. For example, Williams and Miller (1992, p. 991) state that "One
of the most impressive aspects of health policy implementation among the European
study countries is the realization that the programs were put in place not because of
extensive documentation on cost effectiveness, but out of a value system that cherishes
equity in health care". Thus evaluating the costs and benefits (or even the efficacy!) of
the various aspects of the policy environment in other countries remains an important
focus for future research.
Once the question of child health is framed as a public health issue, it becomes an ex-
tremely broad topic indeed. The following discussion provides only the briefest possible
overview of some of the many policies that fall under this rubric. 17
6.1. Directprovision of services
Economic theory suggests that the direct provision of in-kind benefits is generally in-
efficient unless services are limited to those that are deemed to be strictly necessary
for every child. Otherwise, some children will receive services that they do not really
need at public expense. Yet, despite this caveat, most developed countries outside of
the United States have chosen to engage in the direct provision of preventive services
for children. For example, Japan has a network of public health centers that provide a
wide range of pediatric preventive services. Sweden has a system of child health centers
throughout the country that provide free immunizations and screenings, while in Ger-
many, infants and preschool children are entitled to a pre-determined number of free
comprehensive examinations [Chaulk (1994)]. In France, a health care team including
a physician, nurse, child psychologist and social worker monitors children beginning in
their preschool years [Bergmann (1996)]. These countries typically have much higher
rates of compliance with recommended schedules of preventive care than are typical in
the United States.
One of the keys to making direct provision work seems to be setting up a tracking
system to ensure that eligible children are reached. In the United States, many children
who regularly see health care providers are not immunized; in fact these "missed oppor-
tunities" are considered a major cause of under-immunization [cf. Holt et al. (1996)].
England, France, Germany, Japan, and the Netherlands all have tracking systems that
notify health authorities of a birth and initiate computer tracking of preventive care
[General Accounting Office (1993)]. In the Netherlands, for example, computerized
punch cards indicate to a central office that a scheduled visit has been completed and
that the appropriate care has been received [Williams and Miller (1992)]. Studies in
the US have shown that tracking pregnant women and high risk newborns and assisting
them in coordinating their care can be effective in improving birth outcomes among
poor women [Buescher et al. (1991), Myerberg et al. (1995)]. Implementing a track-
ing system for high risk children has also improved immunization rates in Washington
state [National Governor's Association (1997b)]. The effectiveness of tracking may re-
flect the importance of information failures in this market - parents who do not know
the recommended schedules and doctors who do not know what care the patient has
received may both contribute to having children "fall through the cracks" in the system.
17 These issues are discussed further in Currie (1998).

1078 J. Currie
6.2. Public health programs
The discussion thus far has focused on the provision of medical services for children.
Yet, as discussed in the introduction, many of the gravest threats to child health are not
medical in nature. Even in rich countries, there may remain considerable scope for the
improvement of child health through public health measures aimed at preventing acci-
dents, curbing violence, reducing substance abuse, and encouraging healthy lifestyles.
Public health programs can take various forms ranging from the provision of informa-
tion via educational campaigns, to mandating the use of safety equipment and sanction-
ing undesirable behaviors. To the extent that social ills are due to imperfect information
about, for example, the importance of good nutrition or seatbelts, economic theory sug-
gests that provision of the requisite information should be among the most efficient
forms of government intervention in the market for child health. Hence, it would be
especially interesting to evaluate the extent to which public education (broadly defined)
can contribute to improved child health. The rest of this section discusses a variety of
public health initiatives along with the available evidence regarding their effectiveness.
6.2.1. Education and informationalcampaigns
There is little doubt that some efforts to educate parents and children about good health
practices pay off. For example, several states have implemented a program called "Baby
Your Baby" which consists of a massive advertising campaign for prenatal care along
with a toll free number people can call to get information about prenatal care available
in their area. In Nevada, the infant mortality rate declined from 9 deaths per 1,000
before implementation of the program to 5.7 deaths per 1,000 after the introduction
of this program - and it cost only $468,000 in fiscal year 1997 [National Governor's
Association (1997b)].
Perhaps the most startling recent example of a successful public education campaign
is the decline in the infant mortality rate from 8 deaths per thousand live births to 7
deaths per 1000 between 1994 and 1996. This improvement has been attributed to a
concerted effort by hospitals and other health care providers to get people to put babies
to sleep on their sides or backs rather than on their fronts. Recent research has shown
that sleeping on the back reduces the risk of Sudden Infant Death Syndrome (SIDS), a
leading baby killer. All of the two-year decline in the infant mortality rate came from
a reduction in SIDS deaths. Given the approximately 4 million births per year that take
place in the United States, this decline works out to 4000 fewer deaths per year [National
Center for Health Statistics (1997)].
The extent to which smoking can be influenced by informational campaigns is an
extremely important issue given the major public health threat posed by tobacco. It is
estimated that 1 in 6 children may take up smoking and that one quarter of them will die
as a result [US DHHS (1991)]. Moreover, 80% of smokers take up the habit before age
18 [US CDC (1998)]. Smoking during pregnancy has been identified as "the single most
powerful determinant of poor fetal growth in the developed world" [National Center on
Addiction and Substance Abuse (1996, p. 1)]; it has been shown to double the risk that
a baby will be born underweight.
Despite considerable controversy, it is clear that tobacco use is influenced by adver-
tising in mass media. For example, in the early 1990s, the state of California conducted
an unprecedented multi-million dollar campaign against smoking. Between April 1990
and June 1993 it spent 26 million dollars, more than any other state government had
ever spent. Hu, Sung, and Keeler (1995) found that the program had a large negative
effect on cigarette consumption - in fact, this program has been credited with reducing
smoking in California up to 40%.
However, it has been countered by increased advertising by tobacco companies,
which has had the predicted neutralizing effect. Tobacco companies spend 14 mil-
lion dollars a day to advertise their products in the United States [Clary (1998)]. This
problem was addressed in the recent tobacco settlement in Minnesota which severely
restricted tobacco advertising by banning advertisements on billboards, bus shelters,
taxis, and buses, and by prohibiting marketing or promotions aimed at children [Wein-
stein (1998)].
A cautionary note comes from Arizona's recent anti-tobacco education campaign,
which apparently backfired. The theme that smoking was "not for kids" appears to have
only made it more attractive to teens [Clary (1998)]. Similarly, programs that group
deviant youths together for counselling can actually increase drug use, since peer effects
may outweigh any positive effects of the programs [Botvin (1997)].
In fact, interventions that rely on individual counselling or education efforts often
show disappointing results. For example, Kendrick et al. (1995) report that clinics that
integrated smoking cessation advice into regular prenatal care achieved lower reported
levels of smoking but that quit rates verified through urine tests were no different in
treatment and control clinics. Perhaps unsurprisingly, a meta-analysis of controlled trials
of smoking cessation interventions found that the intensity of the intervention (e.g.,
number of counselling sessions) was critical [Kottke et al. (1988)]. This finding suggests
that it is not enough to tell people that smoking is harmful. They also need information
about how to quit and reassurance that back-sliding can be overcome.
Interventions designed to prevent teen pregnancy or to reduce violent behavior have
generally taken the form of providing either group or individual counselling and educa-
tion. The birth rate for 15 to 19 year old girls in the United States was 54.7 per 1,000
in 1995. This rate can be compared to 6 per 1,000 in the Netherlands, 9 per 1,000 in
Denmark, and 32 per 1,000 in Great Britain. Teen pregnancy is thought to have impor-
tant negative social and health consequences both for mothers and their children [US
DHHS (1991)]. In his review of the literature on teen pregnancy prevention programs,
Douglas Kirby (1997) concludes that some educational programs may have been effec-
tive in persuading teens to delay intercourse and/or use condoms, especially in areas
in which AIDS was perceived as a salient threat. But he finds no studies which have
demonstrated effects on pregnancy rates. On the other hand, the teen birth rate has been
declining since the early 1990s and this decline has been driven primarily by the in-
creased use of contraception (especially condoms) rather than by decreases in sexual
1080 J. Currie
activity or increases in abortion rates [Child Trends (1997)]. It is possible that this de-
cline represents a cumulative effect of education gleaned from many sources.
Educational interventions aimed at reducing violent behavior typically seek to teach
children to identify feelings and act on them in an appropriate way. Many of these
programs also emphasize remedial education in an effort to alleviate frustration due to
academic failure. While some programs of this type have shown promising results with
very young children [cf. Tremblay et al. (1997)], there is little evidence that programs
aimed at seriously aggressive adolescents are effective in modifying behavior [Tolan
and Guerra (1994)].
Thus, the limited amount of information that is available suggests that as economic
theory would suggest, education can be a potentially very cost effective method of im-
proving public health in some cases. However, some educational strategies are more
successful than others, and certain forms of "education" may actually do more harm
than good!
6.2.2. Productsafety and other government mandates
Tengs et al. (1995) review several government interventions aimed at protecting chil-
dren. They estimate that child restraint systems in cars cost $73,000 per year of life
saved, or almost $5.5 million for a child with a 75 year life expectancy. Comparable
figures for flammability standards for children's sleepwear can be as high as $1 billion,
while various school bus safety measures started at $10 million. In contrast, figures
presented in Currie and Gruber (1996a) suggest that Medicaid expansions which were
targeted to low income women cost approximately $1 million per life saved. While
the exact cost figures are of course sensitive to the assumptions used to obtain them,
this comparison suggests that safety mandates are not always particularly cost effective
public health measures.
Viscusi (1995) argues that this is because mandates are often put into place with-
out any cost/benefit analysis, and because they often incorporate a seemingly irrational
"status-quo bias". The later refers to people's willingness to bear existing "natural" risks
(e.g., consume naturally occurring chemicals in food) coupled with their unwillingness
to bear additional smaller "synthetic" risks (e.g., consume chemical additives to food).
On the other hand, it may not be irrational for people to be skeptical of pronouncements
that synthetic chemicals are harmless given previous examples of "harmless" chemicals
which turned out to be toxic, carcinogenic, and/or environmentally disastrous. Thus, the
status-quo bias may be a reasonable response to imperfect information in the market for
risk.
Williams and Kotch (1990, p. 108) examine "excess mortality" among children in the
United States and note that deaths from motor vehicle accidents, firearms, fires, poison-
ings, and drownings are all greater in the US than in other developed countries. They
call for the extensive use of government mandates to limit these risks, concluding that
"limiting exposure [to risk] through regulation of handguns, greater use of public trans-
portation, and affordable and accessible day care are among the measures that should
be implemented". Obviously, mandates in areas such as public transportation or gun

control involve issues far beyond the scope of this survey. However, this argument does
highlight the message that many public policies outside the health care arena affect child
health, and that these effects ought to be included in the cost/benefit calculus when such
policies are considered.
6.2.3. Home visiting
Canada, Germany, France, Sweden, Japan and Great Britain all have home visiting pro-
grams for high risk pregnant women, for post-partum care, and for "at risk" children.
For example, Great Britain uses public health nurses to provide preventive health infor-
mation and examinations in the homes of all families with newborns. All six countries
also have lower child and infant mortality rates though it is not clear that these desirable
results can be attributed only to home visiting programs [Chaulk (1994)]. Home visit-
ing programs have had a long history in the US, beginning in the 1850s with volunteer
"Friendly Visitors" programs in Philadelphia and other cities. There are now thousands
of US programs which use home visitors to deliver a variety of services to families with
young children.
Although few cost/benefit studies of home visiting have been conducted, there is
some evidence that home visiting programs targeted to "at risk" families with young
children can reduce the incidence of serious and costly problems such as low birth-
weight, child abuse and neglect, and educational failure. For example, pregnant teens
in a rural South Carolina county with a "Resource Mothers" program for pregnant ado-
lescents had half the number of low birthweight babies as similar teens had in a com-
parison county without such a program [Heins et al. (1987)]. Mayer (1988) found that
low-income Michigan mothers who were visited had better birth outcomes both when
compared to their previous pregnancies and when compared to a control group with
similar demographic characteristics. Larson (1980) reports that children randomly as-
signed to a home visiting program had fewer accidents in their first year and higher
immunization rates. In Florida, low birthweight children who were randomly assigned
to receive home visits for 2 years were less likely to need additional developmental ser-
vices later on [Resnick et al. (1987)]. Randomized trials have also shown that children
who experience "failure to thrive", benefit from home visiting programs [Black et al.
(1995)], and that home visited children "at risk" are less likely to be subjected to abuse
or neglect [Olds et al. (1995)].
The Olds et al. study is notable because it did undertake a cost/benefit evaluation
which was based on a randomized control/treatment design [Olds et al. (1993)]. The
results suggested that government savings from welfare (Aid to Families with Depen-
dent Children and Food Stamps), Medicaid, and Child Protective Services slightly out-
weighed the costs of providing the home visiting program. This study focused on rural
women in upstate New York, but positive results have also been achieved in a similar
intervention with a population of urban black women in Memphis Tennessee [Olds et
al. (1997)].
1082 J. Currie
However, many home visiting programs fail to reach their goals Olds and Kitz-
man (1993)]. Success appears to depend on a number of elements such as clarity of
goals, comprehensiveness of services, guidance and supervision of staff, and targeting
to at-risk families - European-style programs that provide home visits to all children
irrespective of risk have not been shown to be cost-effective, though of course this does
not mean that they are not.
6.2.4. Incentives and/orsanctions
Germany and Japan offer financial incentives to seek early prenatal care, while France
penalizes women who do not obtain care by reducing the child allowances they are enti-
tled to [Bergmann (1996)]. In Britain, it is the physicians who get bonuses for achieving
high immunization rates [Chaulk (1994)]. While these policies have not been subjected
to rigorous evaluation, economists in any case would expect parental behavior to be
sensitive to direct financial incentives.
6.2.5. Community-basedprograms
Community-based programs can take an almost infinite variety of forms. The distin-
guishing characteristic is that they focus on the neighborhood or community rather than
on changing individual behavior. For example, the "Safe Kids/Healthy Neighborhoods
Injury Prevention Program" in central Harlem worked to renovate playgrounds, and to
set up safe supervised activities. The project also had an educational component, as
injury and violence prevention education was introduced in local schools. Targeted in-
juries (vehicle injuries, outdoor falls, assaults, and handgun injuries) fell over the study
period while non-targeted injuries (poisonings, ingestions, bums, etc.) did not. Similar
community based approaches in Massachusetts and Sweden have also shown positive
effects [Davidson (1994)].
Some public health professionals are also starting to look at crime as a health issue
[cf. Earls and Reiss (1994)]. Their emphasis on uncovering risk factors for criminal
behavior complements economic analyses of criminality in that it focuses on preference
formation and constraints facing individuals, rather than on incentives to commit crime
given the preferences and constraints that are in place at any given point in time.
6.2.6. Summary
There are many types of initiatives that have been proven to be effective in improving
children's health, at least in some settings. What is lacking is any systematic evaluation
of these initiatives that would rank programs in terms of cost effectiveness. One factor
complicating such an evaluation is that programs aimed at improving child health are
often implemented as part of a package of measures. Moreover, many policies which are
not primarily aimed at improving child health may nevertheless have significant effects
on health.
Indeed, many child advocates feel strongly that child health would be most effectively
improved by a combination of measures that included attention to economic and social
deprivation as well as health [cf. Kliegman (1995)]. It seems clear, for example, that
the fact that black infants are twice as likely to die as white infants in the United States
is to some extent socially determined, because this differential is observed even among
college educated women, and because one does not observe the same divergence in
countries such as Cuba [Schoendorf et al. (1992), Hogue and Hargraves (1993)].
7. Directions for future research
One of the most important factors limiting research on determinants of child health
in more developed countries is a lack of objective health data. Much of the discussion
above has focused on infant and child mortality, because it is virtually the only objective
measure that is widely available. But mortality is an extreme outcome, and may not be
closely related to other indicators of child health. Other relatively objective indicators
that might be measurable in large scale surveys include whether or not children are
anemic, stunted, over or under-weight, whether they have uncorrected vision or hearing
impairments, and whether they are experiencing developmental delays.
Much of the research reviewed here focuses on the utilization of care. Unfortunately,
it is usually difficult to link utilization with health outcomes, rendering it impossible to
know whether utilization of care should be considered to be primarily a consumption
good rather than an investment good. In addition, utilization data are generally self-
reported which raises the possibility of systematic measurement error. For example,
less educated parents may be less likely to report some conditions, such as asthma, if
they know less about the symptoms of those conditions.
Another difficulty is that administrative data sets created to record information about
such things as birthweight and the incidence of communicable diseases or accidents,
are often lacking important information about children's income or family backgrounds.
Moreover, it is often impossible to link individual-level data about utilization to "supply-
side" data that measures access to care (since the geographic identifiers are either un-
available, or available at only a very aggregate level).
Relatively simple improvements in data collection efforts which would have a poten-
tially large payoff in terms of research include: the development of protocols allowing
the release of geographical identifiers that would allow the merger of "demand" and
"supply-side" data sets; the addition of measures of family income and background to
administrative data sets; and the collection of more objective measures of child health.
Assume for the moment however, that we had data linking measures of child health,
family background, utilization of care, and access to care. What are the most interesting
outstanding questions that could be addressed? One of the most pressing questions for
health policy is identifying interventions in child health that are good investments. For
example, clinical studies have demonstrated that prenatal care for high-risk women can
have a significant effect on birth outcomes [cf. Institute of Medicine (1985)]. On the
1084 J. Currie
other hand, economic studies of the effects of prenatal care in large populations tend
to find small or insignificant effects [cf. Rosenzweig and Schultz (1982, 1983, 1988),
Frank et al. (1991), Corman et al. (1987), Grossman and Joyce (1990)]. These results are
not necessarily inconsistent because the benefits could well be large for "at risk" women
and small or non-existent for other women. Also, while it is likely to be important for
all women to get some prenatal care (so that the high-risk women can be identified
and treated), it may not be useful for all women to receive all of the visits currently
recommended.
A second line of research would explore the links between socio-economic status and
health. We know that higher incomes and education are associated with better health,
but we know little about the reasons why. Also, we know that there is correlation in
income across generations, and hence there must be correlation in health status, but we
do not know what the causal linkages are, or what interventions might be effective in
breaking these linkages among low-income people.
Third, the relationship between utilization of care and the supply of services could
be examined. Lack of access to care is often cited as a problem when utilization of
care is below recommended levels, but it is difficult to demonstrate a causal link be-
tween utilization and access when other factors, such as socio-economic status or lack
of appropriate information, might be implicated.
Finally, this review has focused a good deal of attention on a range of public health
measures aimed at improving child health. It would be extremely useful to have more
systematic analysis of these programs, particularly studies aimed at identifying the com-
ponents of successful programs and the costs and benefits of implementing them.
References
Aday, L.A. (1975), "Economic and noneconomic barriers to the use of needed medical services," Medical
Care 18(6):447-456.
Akerlof, G. (1970), "The market for lemons: qualitative uncertainty and the market mechanism", Quarterly
Journal of Economics 84:488-500.
Baker, L., and A. Royalty (1996), "Medicaid policy, physician behavior, and health care for the low-income
population", xerox (Department of Economics, Stanford University) December.
Barker, D.J., and C. Martyn (1992), "The maternal and fetal origins of cardiovascular disease", Journal of
Epidemiology and Community Health 46(8).
Behrman, J., and A. Deolalikar (1988), "Health and nutrition", in: H. Chenery and T.N. Srinivasan, eds.,
Handbook of Development Economics (North-Holland, New York).
Bergmann, B. (1996), Saving Our Children from Poverty: What the United States Can Learn from France
(Russell Sage, New York).
Black, M., H. Dubowitz, J. Hutcheson, J. Berenson-Howard and R. Starr (1995), "A randomized clinical trial
of home intervention for children with failure to thrive", Pediatrics 95(6):807-814.
Blank, R., and D. Card (1991), "Recent trends in insured and uninsured employment: is there an explana-
tion?", Quarterly Journal of Economics 106:1157-1190.
Blank, R., and P. Ruggles (1996), "When do women use AFDC and food stamps?: The dynamics of eligibility
vs. participation", Journal of Human Resources 31(1):57-89.
Blau, D. (1996), "The effect of income on child development", xerox (Department of Economics, University
of North Carolina at Chapel Hill) June.
Bloom, B. (1990), "Health insurance and medical care", Advance Data from Vital and Health Statistics of the
National Center for Health Statistics, #188 (NCHS Public Health Service, Washington, DC) October 1.
Botvin, G. (1997), "Substance abuse prevention through life skills training", in: R. de V. Peters and R. McMa-
hon, eds., Preventing Childhood Disorders, Substance Abuse, and Delinquency (Sage Publications, Lon-
don).
Broman, S., P. Nichols and W. Kennedy (1975), Preschool IQ: Prenatal and Early Developmental Correlates
(Lawrence Erlbaum Associates, Hillsdale, NJ).
Brooks-Gunn, J., F.R. Liaw, P.K. Klebanos and D. Spiker (1993), "Enhancing the development of low-
birthweight, premature infants: changes in cognition and behavior over the first three years", Child De-
velopment 64:736-753.
Brown, S. (1989), "Drawing women into prenatal care", Family Planning Perspectives 21(2):73-80.
Buehler, J.W., J. Kleinman and C. Hogue (1985), "Birth weight-specific infant mortality, US, 1960 and 1980",
Public Health Reports 102(2):3564-3567.
Buescher, P.A., et al. (1991), "An evaluation of the impact of maternity care coordination on Medicaid birth
outcomes in North Carolina", American Journal of Public Health 81:1625.
Casanova, C., and B. Starfield (1995), "Hospitalizations of children and access to primary care: A cross-
national comparison", International Journal of Health Services 25(2):283-294.
Chaikind, S., and H. Corman (1991), "The impact of low birthweight on special education costs", Journal of
Health Economics 10:291-311.
Chaulk, P. (1994), "Preventive health care in six countries: models for reform?", Health Care Financing Re-
view 15(4):7-19.
Child Trends, Inc. (1997), 1997 Facts at a Glance (Child Trends, Washington, DC).
Clary, M. (1998), "Teen-driven ad campaign puts heat on big tobacco", Los Angeles Times (May 8):A1.
Cole, N. (1994), "Increasing access to health care: the effects of Medicaid expansions for pregnant women",
xerox (Abt Associates) June (Cambridge, MA).
Corman, H., T. Joyce and M. Grossman (1987), "Birth outcome production functions in the United States",
Journal of Human Resources 22(Summer):339-360.
Currie, J. (1998), "Caring for kids: public policy and child health", manuscript (UCLA, Department of Eco-
nomics).
Currie, J. (1999), "Do children of immigrants make differential use of public health insurance?", in: G. Borjas,
ed., Issues in the Economics of Immigration (University of Chicago Press for National Bureau of Economic
Research, Chicago).
Currie, J., and B. Madrian (1999), "Health, health insurance and the labor market", in: D. Card and O. Ashen-
felter, eds., Handbook of Labor Economics (North-Holland, New York).
Currie, J., and D. Thomas (1995), "Medical care for children: public insurance, private insurance and racial
differences in utilization", Journal of Human Resources 30(Winter): 135-162.
Currie, J., and J. Gruber (1996a), "Saving babies: the efficacy and cost of recent expansions of Medicaid
eligibility for pregnant women", Journal of Political Economy (December).
Currie, J., and J. Gruber (1996b), "Health insurance eligibility and child health: lessons from recent expan-
sions of the Medicaid program", Quarterly Journal of Economics 111 (May):431-466.
Currie, J., and J. Gruber (1997), "The technology of birth: health insurance, medical interventions, and infant
health", Working Paper #5985 (National Bureau of Economics Research, Cambridge, MA) April.
Currie, J., J. Gruber and M. Fischer (1995), "Physician payments and infant health: effects of increases in
Medicaid reimbursements", American Economic Review 85(2): 106-111.
Currie, J., and P. Reagan (1998), "Distance to hospital and children's access to care: is being closer better, and
for whom?", xerox (UCLA, Department of Economics) September.
Cutler, D., and J. Gruber (1996), "Does public insurance crowdout private insurance?", Quarterly Journal of
Economics lll(May):391-430.
Cutler, D., and J. Gruber (1997), "Medicaid and private insurance: evidence and implications", Health Affairs
(January/February): 194-200.
1086 J. Currie
Davidson, L.L., et al. (1994), "The impact of the Safe Kids/Healthy Neighborhoods Injury Prevention Pro-
gram in Harlem, 1988 through 1991", American Journal of Public Health 84(4):580-586.
Decker, S. (1992), "The effect of physician reimbursement levels on the primary care of Medicaid patients",
xerox (New York University, School of Public Service).
Dow, W., P. Gertler, R. Schoeni, J. Strauss and D. Thomas (1997), "Health care prices, health, and labor
outcomes: experimental evidence", xerox (RAND) March.
Dranove, D. (1988), "Demand inducement and the physician/patient relationship", Economic Inquiry 26:281-
299.
Dubay, L., and G. Kenney (1997), "Did Medicaid expansions for pregnant women crowd out private cover-
age", Health Affairs (January/February): 185-193.
Earls, F., and A. Reiss (1994), Breaking the Cycle: Predicting and Preventing Crime (US Department of
Justice, Washington, DC).
The Economist (1995), "Womb for Improvement", November 18, 89-90.
Edwards, L., and M. Grossman (1979), "The relationship between children's health and intellectual devel-
opment", in: S.J. Mushkin and D.W. Dunlop, eds., Health: What is it Worth? (Pergamon Press, Elmsford)
273-314.
Eiclmner, M. (1997), "Incentives, price expectations and Medical expenditures: an analysis of claims under
employer-provided health insurance", xerox (Columbia University Graduate School of Business) April 12.
Fossett, J., and J. Peterson (1989), "Physician supply and medicaid participation: the causes of market failure",
Medical Care 27:386-396.
Frank, R., D. Strobino, D. Salkever and C. Jackson (1991), "Updated estimates of the impact of prenatal
care on birthweight outcomes by race", Working Paper #3624 (National Bureau of Economic Research,
Cambridge, MA) February.
Freed, G., S. Clark, J. Lohr and J. Sorenson (1995), "Pediatrician involvement in breast-feeding promotion:
a national study of residents and practitioners" 96(3):490-494.
Gomby, D., C. Larson, E. Lewit and R. Behrman (1993), "Home visiting: analysis and recommendations",
The Future of Children 3(9):6-22.
Grossman, M. (1975), "The correlation between health and schooling", in: N.E. Terleckyj, ed., Household
Production and Consumption (Columbia University Press, New York) 147-211.
Grossman, M. (2000), "The human capital model", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of
Health Economics (Elsevier, Amsterdam) Chapter 7.
Grossman, M. and T. Joyce (1990), "Unobservables, pregnancy resolutions and birth weight production func-
tions in New York City", Journal of Political Economy 98(5):983-1007.
Halfon, N., P. Newacheck, D. Wood and R. Peter (1996), "Routine emergency department use for sick care
by children in the United States", Pediatrics 98(1):28-34.
Hanratty, M. (1996), "Canadian national health insurance and infant health", American Economic Review
86(1):276-284.
Heins, H., et al. (1987), "Social support in improving perinatal outcome: The Resource Mothers Program",
Obstetrics and Gynecology 70(2).
Hertzman, C., and M. Wiens (1996), "Child development and long-term outcomes: a population health per-
spective and summary of successful interventions", Social Science and Medicine 43(7): 1083-1095.
Himmelstein, D., and S. Woolhandler (1995), "Care denied: US residents who are unable to obtain needed
medical services", American Journal of Public Health 85(3):341-344.
Hogue, C.J., and M.A. Hargraves (1993), "Class, race, and infant mortality in the United States", American
Journal of Public Health 83:9-12.
Holt, E., et al. (1996), "The contribution of missed opportunities to childhood underimmunization in Balti-
more", Pediatrics 97(4):474-480.
Hooper, E., L. Comstock, J. Goodwin and J. Goodwin (1982), "Patient characteristics that influence physician
behavior", Medical Care 20(6):631-638.
Horbar, J., E. Wright, L. Onstad et al. (1993), "Decreasing mortality associated with the introduction of
surfactant therapy: an observational study of neonates weighing 601 to 1300 grams at birth", Pediatrics
92(2): 191-196.
Hoyme, E., K. Jones, S. Dixon, T. Jewett, J. Hanson, L. Robinson, M.E. Msall and J. Allanson (1990), "Pre-
natal cocaine exposure and fetal vascular disruption", Pediatrics 85(5):743-747.
Hu, T.W., H.Y. Sung and T.E. Keeler (1995), 'The state anti-smoking campaign and the industry response: the
effects of advertising on cigarette consumption in California", American Economic Review 85(2):85-90.
Institute of Medicine (1985), Preventing Low Birthweight (National Academy Press, Washington, DC).
Kaestner, R., and H. Corman (1995), "The impact of child health and family inputs on child cognitive devel-
opment", Working Paper #5257 (National Bureau of Economic Research, Cambridge, MA) September.
Kahn, K., et al. (1994), "Health care for black and poor hospitalized Medicare patients", Journal of the Amer-
ican Medical Association 271(15):1169-1174.
Kasper, J. (1986), "Health status and utilization: differences by Medicaid coverage and income", Health Care
Financing Review 7:1-17.
Kemper, K. (1988), "Medically inappropriate hospital use in a pediatric population", The New England Jour-
nal of Medicine 318(16):1033-1037.
Kendrick, J.S., et al. (1995), "Integrating smoking cessation into routine public prenatal care: The Smoking
Cessation Pregnancy Project", American Journal of Public Health 85(2):217-222.
Kirby, D. (1997), No Easy Answers: Research Findings on Programs to Reduce Teen Pregnancy (National
Campaign to Prevent Teen Pregnancy, Washington, DC).
Kliegman, R. (1995), "Neonatal technology, perinatal survival, social consequences, and the perinatal para-
dox", American Journal of Public Health 85(7):909-912.
Kogan, M., G. Alexander, M. Kotelchuck and D. Nagey (1994), "Relation of the content of prenatal care to
the risk of low birth weight", Journal of the American Medical Association 271(1):1340-1345.
Korenman, S., J. Miller and J. Sjaastad (1995), "Long-term poverty and child development in the United
States: results from the NLSY", Children and Youth Services Review 17(1/2): 127-155.
Kottke, T., R. Battista, G. DeFriese and M. Brekke (1988), "Attributes of successful smoking cessation inter-
ventions in medical practice", Journal of the American Medical Association 259(19):2883-2889.
Larson, C. (1980), "Efficacy of prenatal and postpartum home visits on child health and development", Pedi-
atrics 66(2).
Long, S., M.S. Marquis and E. Harrison (1994), "The costs and financing of perinatal care in the United
States", American Journal of Public Health 84(9):1473-1478.
MacArthur, C., and I. Pless (1990), "Unintentional injury mortality in Canada", Pediatrics 86(6):s1073-1077.
Martorell, R., and J.P. Habicht (1986), "Growth in early childhood in developing countries", in: F Faulkner
and J.M. Tanner, eds., Human Growth: A Comprehensive Treatise (Plenum Press, New York).
Mayer, J. (1988), "Evaluation of maternal and child health community nursing services: application of two
quasi-experimental designs", in: Health Action Papers, Vol. 2.
McCormick, M., et al. (1992), "The health and developmental status of very-low-birthweight children at
school age", Journal of the American Medical Association 267.
Miller, L.A., R. Hoffman, A. Baron, W. Marine and P. Melinkovich (1994), "Risk factors for delayed immu-
nization against measles, mumps, and rubella in Colorado two year olds", Pediatrics 94(2):213-219.
Mokyr, J. (1998), "Why was there more work for mother? Knowledge and household behavior, 1870-1945",
xerox (Northwestern University, Department of Economics) October.
Moore, K., B.W. Sugland, C. Blumenthal, D. Glei and N. Snyder (1995), Adolescent Pregnancy Prevention
Programs: Interventions and Evaluations (Child Trends, Washington, DC).
Myerberg, D.Z., R. Carpenter, C. Myerberg, C. Britton, C. Bailey and B. Fink (1995), "Reducing postneonatal
mortality in West Virginia: a statewide intervention program targeting risk identified at and after birth",
American Journal of Public Health 85(5):631-637.
National Center for Addiction and Substance Abuse (1996), Fetal Tobacco Syndrome (NCASA, Columbia
University, New York).
1088 J. Currie
National Governor's Association (1997a), State Medicaid Coverage of Pregnant Women and Children (NGA,
Washington, DC) September 30.
National Governor's Association (1997b), Promising Practices to Improve Results for Young Children (NGA,
Washington, DC).
National Vaccine Advisory Cormnittee (1991), "The measles epidemic: the problems, barriers, and recom-
mendations", Journal of the American Medical Association 226(11):1547-1552.
Needleman, H.L., J.A. Riess, M.J. Tobin, G.E. Biesecker and J.B. Greenhouse (1996), "Bone lead levels and
delinquent behavior", Journal of the American Medical Association 275(5):333-369.
Newacheck, P., D. Hughes and J. Stoddard (1996), "Children's access to primary care: differences by race,
income, and insurance status", Pediatrics 97(1) 26-32.
Olds, D., C. Henderson and R. Tatelbaum (1994), "Intellectual impairment in children of women who smoke
cigarettes during pregnancy", Pediatrics 93(2).
Olds, D., C. Henderson, C. Phelps, H. Kitzman and C. Hanks (1993), "Effect of prenatal and infancy nurse
home visitation on government spending", Medical Care 31(2):155-174.
Olds, D., C. Henderson, H. Kitzman and R. Cole (1995), "Effects of prenatal and infancy nurse home visita-
tion on surveillance of child maltreatment", Pediatrics 95(3):365-372.
Olds, D., and H. Kitzman (1993), "Review of research on home visiting for pregnant women and parents of
young children", The Future of Children 3(3):53-92.
Olds, D., et al. (1997), "Long term effects of home visitation on maternal life course and child abuse and
neglect", Journal of the American Medical Association 278(8):637-643.
Parker, L. (1989), The Relationship Between Nutrition and Learning (National Education Association, Wash-
ington, DC).
Pauly, M. (1980), Doctors and Their Workshops (University of Chicago Press, Chicago).
Perri, T.J. (1984), "Health status and schooling decisions of young men", Economics of Education Review
3:207-213.
Philipson, T. (1996), "Private vaccination and public health: an empirical examination for US measles", Jour-
nal of Human Resources 31(3):611-630.
Piper, J.M., et al. (1990), "Effects of Medicaid eligibility expansion on prenatal care and pregnancy outcomes
in Tennessee", Journal of the American Medical Association 264:2219-2223.
Preston, S. (1977), "Mortality trends", Annual Review of Sociology 3:163-178.
Pui, C.H., et al. (1995), "Outcome of treatment for childhood cancer in black as compared with white chil-
dren", Journal of the American Medical Association 22(Feb.):633-637.
Riportella-Muller, R., M. Selby-Harrington, L. Richardson, P. Donat, K. Luchok and D. Quade
(1996), "Barriers to the use of preventive health care services for children", Public Health Reports
11 l(Januayly/February):71-77.
Resnick, M.B., et al. (1987), "Developmental intervention for low birth weight infants: improved early devel-
opmental outcomes", Pediatrics 80(1).
Rosenberg, M., J. Rodriguez and T. Chorba (1990), "Childhood injuries: where we are", Pediatrics
86(6):s1084-1090.
Rosenzweig, M., and T.P. Schultz (1982), "The behavior of mothers as inputs to child health: the determi-
nants of birth weight, gestation, and rate of fetal growth", in: V. Fuchs, ed., Economic Aspects of Health
Rosenzweig, M., and T.P. Schultz (1983), "Estimating a household production function: heterogeneity, the
demand for health inputs, and their effects on birth weight", Journal of Political Economy 91:723-746.
Rosenzweig, M., and T.P. Schultz (1988), "The stability of household production technology: a replication",
Journal of Human Resources 23:535-549.
Rosenzweig, M., and K. Wolpin (1994), "Are there increasing returns to the intergenerational production
of human capital? Maternal schooling and child intellectual development", Journal of Human Resources
29(2):670-693.
Rowland, D., and A. Salganicoff (1994), "Commentary: lessons from Medicaid - improving access to office-
based physician care for the low-income population", American Journal of Public Health 84(4):550-552.
Rowland, D., S. Rosenbaum, L. Simon and E. Chait (1995), Medicaid and Managed Care: Lessons from the
Literature (The Henry J. Kaiser Family Foundation, Washington, DC).
Rubin, A. (1997), "Poor children falling out of Medicaid's Safety Net", Los Angeles Times, November 18,
Al.
Saywell, R.M., T.W. Zollinger, D.K. Chu, C.A. Macbeth et al. (1989), "Hospital and patient characteristics of
uncompensated hospital care: policy implications", Journal of Health Politics, Policy, and Law 14:287-307.
Schoendorf, K., C. Hogue, J. Kleinman and D. Rowley (1992), "Mortality among infants of Black as com-
pared with White college-educated parents", New England Journal of Medicine 376(23):1522-1992.
Shakotko, R., L. Edwards and M. Grossman (1981), "An exploration of the dynamic relationship between
health and cognitive development in adolescence" in: J. van der Gaag and M. Perlman, eds., Health, Eco-
nomics, and Health Economics (New York, North-Holland) 305-326.
Short, P., J. Cantor and A. Monheit (1988), "The dynamics of Medicaid enrollment", Inquiry 25:504-516.
Short, P., and D. Lefkowitz (1992), "Encouraging preventive services for low-income children: the effect of
expanding Medicaid", Medical Care 30:76-80.
Sloan, F., J. Mitchell and J. Cromwell (1978), "Physician participation in state Medicaid programs", Journal
of Human Resources 13:211-245.
St. Peter, R., P. Newacheck and N. Halfon (1992), "Access to care for poor children: separate and unequal?",
Journal of the American Medical Association 26(20):2760-2764.
Strauss, J., and D. Thomas (1996), "Measurement and mismeasurement of social indicators", American Eco-
nomic Review 86(2):30-34.
Strauss, J., and D. Thomas (1998), "Health, nutrition, and economic development", Journal of Economic
Tengs, TO., et al. (1995), "Five-hundred life-saving interventions and their cost-effectiveness", Risk Analysis
15:369-390.
Tolan, P., and N. Guerra (1994), What Works in Reducing Adolescent Violence: An Empirical Review of the
Field (Center for the Study and Prevention of Violence, Boulder, CO).
Townsend, P., N. Davidson and M. Whitehead (1988), Inequalities in Health (Penguin, London).
Travis, K. (1996), "Physician payment and the effect on access to prenatal care for Medicaid patients", xerox
(Department of Economics, University of Nevada, Las Vegas, NV).
Tremblay, R., L. Masse, L. Pagani and F. Vitaro (1997), "From childhood physical aggression to adolescent
maladjustment", in: R. deV. Peters and R. McMahon, eds., Preventing Childhood Disorders, Substance
Abuse, and Delinquency (Sage Publications, London).
US Centers for Disease Control and Prevention (1992), "Pertussis surveillance - United States, 1989-1991",
Morbidity and Mortality Weekly Report SS8(41):11-19.
US Centers for Disease Control and Prevention (1998), "Tobacco use among high school students -US 1997",
Morbidity and Mortality Weekly Report 47(12):229-233.
US Department of Health and Human Services (1991), "Healthy people 2000: National health promotion and
disease prevention objectives", PHS 91-50212 (US DHHS, Washington, DC).
US General Accounting Office (1990a), "Medicaid: Oversight of health maintenance organizations in the
Chicago area", GAO/HRD 90-81 (US Government Printing Office, Washington, DC).
US General Accounting Office (1990b), "Drug exposed infants: a generation at risk", GAO/HRD-90-138
(Government Printing Office, Washington, DC) June.
US General Accounting Office (1993), "Preventive health care for children: experience from selected foreign
countries", GAO/HRD-93-62 (US Government Printing Office, Washington, DC) August.
US General Accounting Office (1994), "Health care reform: potential difficulties in determining eligibility
for low-income people", GAO/HEHS-94-176 (Government Printing Office, Washington, DC).
US General Accounting Office (1995a), "Vaccines for children: Barriers to immunization", GAO/T-PEMD-
95-21 (Government Printing Office, Washington, DC) May 4.
1090 J. Currie
US General Accounting Office (1995b), "Medicaid managed care: more competition and oversight would
improve California's expansion plan", GAO/HEHS-95-87 (Government Printing Office, Washington, DC).
US General Accounting Office (1995c), "Health insurance for children: many remain uninsured despite med-
icaid expansion", GAO/HEHS-95-175 (Government Printing Office, Washington, DC).
US Office of Technology Assessment (1987), "Healthy children: investing in the future", OTA-H-344 (Gov-
ernment Printing Office, Washington, DC).
US National Center for Health Statistics (1997), "Births and deaths US: 1996" (Centers for Disease Control,
Washington, DC) www.cdc.gov/nchswww/nchshome.htm
Viscusi, K. (1995), "Carcinogen regulation: risk characteristics and the synthetic risk bias", American Eco-
nomics Review 85(2):50-54.
Wadsworth, M.E. (1986), "Serious illness in childhood and its association with later-life achievement", in:
R. WiLkinson, ed., Class and Health (Tavistock, London).
Weinstein, H. (1998), "Big tobacco settles Minnesota lawsuit for $6.6 billion", Los Angeles Times, May 9,
Al.
Williams, B., and A. Miller (1992), "Preventive health care for young children: findings from a 10-country
study and directions for United States Policy", Pediatrics 89(5):S983-998.
Williams, B., and J. Kotch (1990), "Excess injury mortality among children in the United States: comparison
of recent international statistics", Pediatrics 86(6):s1067-1073.
Wilson, D.B., et al. (1986), "Growth and intellectual development", Pediatrics 78:646-650.
Wolfe, B. (1985), "The influence of health on school outcomes: a multivariate approach", Medical Care
23:1127-1138.
Yudkowsky, B., J. Cartland and S. Flint (1990), "Pediatrician participation in Medicaid: 1978-1989", Pedi-
atrics 85(4):567-577.
Zambrana, R., K. Ell, C. Dorrington, L. Wachsman and D. Hodge (1994), "The relationship between psy-
chosocial status of immigrant Latino mothers and use of emergency pediatric services", Health and Social
Work 18(2):93-102.
Zill, N. (1996), "Parental schooling and children's health", Public Health Reports 111(Jan-
nuary/February):34-43.
Chapter20
THE INDUSTRIAL ORGANIZATION OF HEALTH CARE

MARKETS*
DAVID DRANOVE and MARK A. SATTERTHWAITE

Kellogg Graduate School of Management, Northwestern University, Evanston, IL 60208
Contents
Abstract 1094
2. Agency and the physician-patient relationship 1098
3. Monopolistic competition in health care markets 1102
3.1. Demand under monopolistic competition 1104
3.2. Information and equilibrium price and quality 1107
4. Regime I: Independent physicians and hospital cost reimbursement 1111
4.1. Price competition among physicians 1112
4.2. Quality competition and the medical arms race 1114
5. Regime II: Hospital PPS and regulated physicians 1117
5.1. Quality choice under prospective payment 1118
5.2. Utilization review 1121
5.3. Dynamic cost-shifting 1123
6. Regime III: Managed care and contracted physicians 1125
6.1. Payer-driven competition 1126
6.2. Multinomial choice models 1131
6.3. Measuring quality 1133
7. Conclusions 1134
References 1135
*Jim Malcomson, Joe Newhouse, Tom McGuire, and Mike Scherer each generously provided detailed com-
ments and saved us from many errors of both commission and omission. All remaining errors are our respon-
sibility alone.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P Newhouse
O 2000 Elsevier Science B. V All rights reserved
1094 D. Dranove and M.A. Satterthwaite
Abstract
Health care markets fail to satisfy many requirements for perfect competition, includ-
ing large numbers of consumers and firms, zero search costs, and marketability of all
goods and services. Over time, health care markets have evolved to overcome the re-
sulting inefficiencies. We combine the theory of agency with a model of monopolistic
competition to explore three regimes of health care organization that were dominant at
different points in time: (1) independent physicians and cost-based reimbursement for
hospitals; (2) regulation; and (3) managed care. Each regime represents, for its time,
a sensible response to market failure. Each regime has predictable consequences for
prices, costs, and quality. We examine the theoretical arguments and review the empir-
ical evidence about each regime. A consistent message emerges: Providers respond to
economic incentives in a manner consistent with theory.
JEL classification:D43, D83, Ill, L13, L15

Ch. 20: The IndustrialOrganizationof Health Care Markets 1095
1. Introduction
It has been said that writing about history is about making choices. The same may be
said for writing this chapter. Industrial organization focuses on individual, imperfectly
competitive markets and seeks to understand the behavior of the firms that compose
them and the resulting efficiency, i.e., performance, of those markets. This is clearly a
very broad focus - the Handbook of Industrial Organization contains over 1500 pages
and virtually every topic in that two volume set can be reevaluated within the health
care marketplace. Forced to make choices, we have elected to focus on those areas of
the industrial organization of health care that are either unusually important in health
care or instructive for economists whose interests extend beyond health care. We pay
little attention to topics that, while potentially important, have received little attention
by researchers, either in the form of novel theories or important empirical studies. Thus,
we devote considerable attention to topics such as agency, decision making under uncer-
tainty, and nonprice competition while touching lightly at most on topics such as entry,
exit, and product differentiation that are unstudied despite their importance.
Health care markets fail to satisfy the substantial list of requirements that must be
met to be classified as perfectly competitive: large numbers of consumers and firms,
free entry and exit, marketability of all goods and services including risk, symmetric
information with zero search costs, and no increasing returns, externalities, or collu-
sion. While health care markets satisfy none of these requirements fully, they fail the
requirements of symmetric information, zero search costs, and the marketability of all
products most dramatically.
Asymmetric information. Each consumer's demand for care depends both on his in-
dividual health state and how his physician perceives and interprets that health state. As
Starr (1982, pp. 134-142) remarks, one of the great transformations of medical prac-
tice to occur in the early twentieth century was for the physician to gain independent
access to the health status of each of her patients through diagnostic tools and tests. 1
Consequently the physician may appropriately prescribe expensive, inconvenient, and
even dangerous interventions for a patient who, based on personal perception, feels per-
fectly healthy. A patient's demand for care is not independent of his physician's medi-
cal knowledge and, often, neither the patient nor a third party can directly evaluate the
usefulness of her advice. Consequently a physician may abuse the asymmetry of infor-
mation by "inducing" patients to systematically over- or underconsume services [Fuchs
(1978)].
Costly search. A patient necessarily chooses an individual physician (or at most a
small team) for his care. Each physician has different experience, knowledge, skills, and
personal manner. Given a patient with a serious illness and any two physicians randomly
chosen from the appropriate specialty, then almost certainly one physician or the other
1 For ease of exposition, we use the male pronoun to refer to a patient or consumer and the female pronoun
to refer to a physician or other health care provider.
is a better match for that patient, often by a wide margin, in terms of the expected utility
of the health outcome. In the absence of reliable, detailed sources of information about
individual providers, the consumer can not shop among physicians with confidence that
he will select the one who offers the most preferred combination of expected outcome
and price without incurring prohibitive search costs. This has the direct effect of reduc-
ing average quality, but it also has the indirect effect of failing to give physicians and
other providers clear signals as to what patients value. These unclear signals may, for
example, cause physicians systematically to underproduce clinical quality, overproduce
patient convenience, and ignore price competitiveness.
Marketability of risk. Illness is a random event that every person faces and that may
threaten his life style, his ability to hold a job, and his existence. The expected effi-
cacy of treatment is often sufficient to make the expected benefits from the purchase
of health care exceed its price, by a huge amount, even when that price is high relative
to the individual's liquidity position. As a consequence, an individual who falls ill is
at tremendous risk for losing much of his wealth. This creates a demand for insurance
unlike that for virtually all other goods and services. 2 In principle riskiness is no barrier
to market competitiveness, for every risk is a product and markets can be established
for each possible risk allowing every consumer to obtain full insurance. The informa-
tional difficulties of health care, however, make this full marketability impossible. Full
efficiency in the health insurance market would require that all insurance be in the form
of lump sum indemnities for each possible adverse health incident. Insurance contracts
can neither be written nor administered at that level of complexity. Instead health in-
surance generally pays charges or costs, perhaps with a deductible or coinsurance, and
does not impose on the consumer the full marginal cost of the last unit of care he con-
sumes as efficiency requires. Moral hazard is therefore possible with its consequent
overconsumption of health care services [Pauly (1968). 3
No other market of substantial importance violates these three requirements of per-
fect competition so radically. This justifies the often made claim that the health care
market is "different" than other markets and implies that any evaluation of its perfor-
mance must be based on models that explicitly take into account its deviations from the
competitive market's prerequisites. The model of perfect competition can still serve as
the benchmark of optimal performance, but generally it can not be used to illuminate
the health care market's specific functioning.
The design of the paper is straightforward. In the next two sections we set up our
review by introducing two analytical frameworks that enable us to unify (largely) the
2 Another example is homeowner's insurance.

3 Perhaps even more significant than the nonmarketability of the financial risks that the threat of illness im-
poses on consumer is the nonmarketability of the personal risk associated with illness. Diagnosis is frequently
imprecise and there is no fully predictable relationship between treatment and outcome. In other words, con-
sumers face uninsurable risk for inappropriate care. Even more serious, illness threatens each consumer's
ability to partake in life fully, or even to partake in life at all. Insurance is completely unable to spread this
personal risk, for how can insurance make a person whole when they learn that illness will shortly deprive
them of existence.
Ch. 20: The Industrial Organization of Health Care Markets 1097
discussion of individual papers. The first framework, agency, discussed in Section 2, is

the idea that physicians, by acting faithfully in their patients' interests, can overcome
many, or even most, of the problems that the market's failure to meet competitive pre-
requisites may produce. As Arrow (1971, p. 184) writes in a seminal article,
"I propose the view here that, when the market fails to achieve an optimal state,
society will, to some extent at least, recognize the gap, and nonmarket social
institutions will arise attempting to bridge it. Certainly this process is not nec-
essarily conscious; nor is it uniformly successful in approaching more closely
to optimality when the entire range of consequences is considered. It has al-
ways been a favorite activity of economists to point out that actions which on
their face achieve a desirable goal may have less obvious consequences.... But
it is contended here that the special structural characteristics of the medical-care
market are largely attempts to overcome the lack of optimality due to the non-
marketability of the bearing of suitable risks and the imperfect marketability of
information. These compensatory institutional changes, with some reinforcement
from the usual profit motives, largely explain the observed noncompetitive behav-
ior of the medical-care market, behavior which, in itself, interferes with optimal-
ity. The social adjustment towards optimality thus puts obstacles in its own path."
The second analytical tool is a model of equilibrium in a market in which consumers
can only imperfectly observe the product attributes that are important to them. This
model, which Dranove and Satterthwaite (1992) introduced, allows us to discuss in a
consistent manner the range of informational difficulties that the literature considers.
Given these preliminaries, we then follow the program that is implicit in Arrow's
remarks. Broadly speaking, over the last thirty years there have been three main insti-
tutional structures that the United States has employed to deal with the imperfections
of the health care marketplace: first, independent physicians supported by nonprofit
hospitals financed through cost reimbursement; second, prospective reimbursement of
hospitals using diagnosis related groups (DRGs) coupled with increasing price regula-
tion of physician fees; and third, managed care. The three regimes overlap. Physician
independence and hospital cost reimbursement persist even as physicians and hospitals
simultaneously adapt to the regulatory reforms of Medicare and the competitive pres-
sure of managed care. Most of the studies that fall under the industrial organization
of health care rubric can be characterized as attempts to better understand how well
these institutional regimes overcome the market's basic difficulties. Therefore, since
these studies typically focus on one of the regimes, Sections 3, 4, and 5, respectively,
discuss the literature concerning the performance of markets with cost reimbursement,
prospective payment, and managed care. Section 6 concludes.
For each institutional regime the papers we discuss divide into two main types. 4 The
first consists of theoretical papers that posit a specific stylized description of the insti-
4 This division is somewhat artificial because many of the most important papers contribute importantly to
both our theoretical understanding and our empirical knowledge.
tutional arrangements within the market, model the behavior of providers, and trace out
how the market's imperfections affect its performance. The second consists of empiri-
cal papers that systematically examine data from the market for evidence of particular
behaviors and performance difficulties suggested by the theory. Most commonly the
benchmark against which performance is compared comes from the competitive model:
consumers should be insured against the financial risks illnesses engender, the quality of
care provided should be that which consumers would demand if they had full informa-
tion and if they had perfect indemnity insurance, and prices should be just high enough
to elicit over the long run a sufficient number of providers to meet consumer demand.
A potential limitation of our discussion is that it focuses almost exclusively on the
United States' experience. This reflects the fact that most of the empirical literature that
is relevant to this chapter uses American data. The paper's overall theme, however, is
more universal. Every developed country is wrestling with how to organize its health
services to bring its citizens the fruits of modern medical care at an affordable cost. The
applicability to other venues of the ideas presented here is illustrated by recent work on
the 1989 reforms to the United Kingdom's National Health Service (NHS). Chalkley
and Malcomson (1996) review incentives these reforms give hospitals to compete on
price and quality. Using a sample of the prices that NHS hospitals posted over the 1991-
1995 period, Propper (1998) and Propper, Wilson, and S6derlund (1998) provides weak
evidence that the new market pressures have affected their pricing behavior towards both
local Health Authorities and General Practice Fundholders. The economic literature - as
opposed to the institutional and political literatures - that each of these papers references
and builds on is largely the US-focused literature that we examine in our discussion of
managed care.
2. Agency and the physician-patient relationship
The central role of the physician is a dominant organizational feature of the health care
market. Even today, with the great growth of managed care providers, most services are
ordered by a physician acting as the agent of the patient. The reason for this, in Arrow's
analysis, lies in the fact that physicians primarily sell information and information has
a peculiar nature relative to the usual commodities people trade. When a consumer
consults a physician it is normally to resolve uncertainty: what is the diagnosis and what
therapy should he purchase? The physician responds with information that reduces the
patient's uncertainty. The full cost of this information is very expensive - witness the
time and money costs of medical training - and consequently it is inefficient for patients
to make their own diagnoses and therapeutic choices (even if this were psychologically
feasible). The market niche that physicians fill is in response to this fixed cost and
the efficiency of spreading it across many consumers. But the difficulty with selling
information is that its value "is frequently not known in any meaningful sense to the
buyer; if, indeed, he knew enough to measure the value of the information, he would
know the information itself." [Arrow (1971, p. 183)].
Arrow then observes that, since consumers can neither possess adequate knowledge
to make purchase decisions nor rely on contracts to assure efficient delivery of health
care service, they rely on the professional standards of physicians to make appropri-
ate decisions on their behalf. Physicians, in turn, are bound by ethical constraints,
drummed into them throughout their training, not to exploit their superior knowledge.
Certification of medical professionals provides further assurance to patients. The physi-
cian/patient relationship, provided the patient is trusting and the physician is trustwor-
thy, cuts through the difficulties of selling information and offers a substantial welfare
gain relative to a system of either self-diagnosis and treatment on the part of consumers
or fully self-interested behavior on the part of physicians.
A well functioning agency relationship between the patient and physician also ame-
liorates the difficulties associated with insurance. Health insurance traditionally has re-
imbursed providers for costs or charges within limits such as "usual and customary"
fees. This raises the possibility of moral hazard: the patient pays less than the marginal
cost of that care - perhaps even pays nothing - and therefore overconsumes care [Pauly
(1968)]. Arrow (1971, p. 203) observes that the physician, if she is a faithful agent to
both her patient and to society, can prevent overconsumption. As a perfect agent, she
knows the probability distribution of both the cost and outcome of treatment, she knows
the values her patient places on different outcomes, and she follows the decision rule of
only prescribing care for which the marginal benefit exceeds the marginal cost.
In the third of a century since Arrow wrote, medical knowledge has increased explo-
sively with the consequence that the prevalence and importance of physician specializa-
tion has increased enormously. Each consumer, particularly as he grows older and health
problems accumulate, now typically has a set of doctors, each of whom treats a subset
of his complaints. This need to assemble and marshal a set of doctors complicates the
consumer's decision problem and multiplies the usefulness of a knowledgeable agent.
The concept of the primary care physician as chief agent and coordinator has been the
nonmarket, professional response to this increasing complexity. Two difficulties, how-
ever, prevent this from being a full solution to the consumer's informational problem.
First, the ability of primary care physicians to act as a good agent varies. Just as the
consumer is better off if his primary care physician helps him choose his specialists, he
would be even better off if he had an information resource to aid in selecting his primary
care physician. In other words, the consumer could use a "super" agent (some informa-
tional resource knowledgeable about primary care physicians) to choose his "manage-
rial" agent (the primary care physician) to choose his "medical" agent (the specialist).
Second, no matter how good the primary care physician that the consumer selects, she
has far from perfect information about the range, the qualities, and the prices of available
health care providers. Though an individual physician forms judgements of the quality
of other providers with whom she interacts, this circle of colleagues is inevitably small
and her judgements may be dependent on anecdotes and interpersonal relationships.
Because of the inability of consumers and physicians to measure quality systematically,
organizations that can develop and disseminate accurate quality information about in-
dividual providers may add substantial value to both consumers and their primary care
physicians. Private consulting firms began providing quality audits for hospitals in the
1980s. The Joint Commission for Accreditation of Healthcare Organizations began its
quality audits of hospitals in the mid-1990s. The National Committee for Quality As-
surance (NCQA) began auditing the quality of managed care organizations in the mid-
1990s. 5 With the exception of the publication by several states of surgical mortality
rates, we are unaware of similar broad-based auditing of physician quality.
The agency relationship in healthcare has at least two levels. At the top level are the
physicians who take responsibility for guiding a patient's care. The physicians, how-
ever, have relationships with a set of second level agents: the hospital most importantly,
but also labs, physical therapy centers, outpatient radiology centers, etc. These agents
accept orders from the physicians as to what therapies or tests should be carried out,
but then exercise their own professional skills in independently providing the required
services. These specialized second level agents exist for two reasons. First, given the
small size of most physician groups, the organization serving as the second level agent
can only obtain minimum economic scale if it serves several physician groups. The hos-
pital is the most dramatic illustration of this. Second, organizationally and financially
separating the second level agent from the primary agent preserves the integrity of the
primary agent's relation with the patient. To the extent that the physician only sells the
patient information about his diagnosis and treatments appropriate to that diagnosis and
refers the patient to independent second level agents for the actual therapy, then the pri-
mary care physician does not have a financial incentive to be a less than perfect agent
and to prescribe unnecessary treatment. It is for this second reason that Medicare for-
bids payments by hospitals and other second level agents to physicians in exchange for
referrals.
Jeffrey Harris (1977) argues that these agency relationships between the patient and
physician and, at the second level, between physician and hospital explain the traditional
"dual" hierarchy of hospitals. Hospitals are generally organized so that the hospital
administrator has responsibility for most resource allocation decisions, but defers to the
medical staff and its chief as to how to deploy the resources in the service of particular
patients. As Harris puts it, "the patient buys a promise from the doctor to be fixed up.
The hospital in turn (that is, the administration part) supplies the necessary inputs to
the doctor.... There is a strong ethical presumption that the doctor be left alone to do
whatever is necessary for the patient's well being". 6
Harris points out that uncertainty and technological complexity make it difficult to
specify norms for producing hospital services. Complex problems frequently arise, hos-
pital resources often grow scarce, and inappropriate decisions about what care to pro-
vide and to whom could have dire consequences. Harris argues that, under these condi-
tions, it is more efficient to assign final decision-making authority to physicians rather
than the administrator. Whether it is due to their superior knowledge (vis-a-vis admin-
istrators) or their fealty to patients that causes them to give high weight to outcomes
5 For a discussion of the origins and goals of NCQA, see Corrigan (1995).
6 Harris (1977), pp. 473-474.
relative to cost, physicians are less likely to make choices that would have disastrous
outcomes. Moreover, a price mechanism would be impossible to implement, due to the
complexity of the resource allocation problem and the frequent need for immediate de-
cisions.
Harris' analysis is a nice example of a more general argument about residual rights
of control that Grossman and Hart (1986) and Hart, Shleifer, and Vishny (1997) have
made. These papers argue that if the performance of agents can not be pinned down
contractually because of the practical impossibility of writing complete contracts for
the services that they provide, then residual rights should rest with those agents whose
decisions most heavily influence outcomes. In this case, uncertainty and technological
complexity favor leaving clinical decision-making authority in the hands of physicians,
as it is their decisions that most affect outcomes, while leaving to the hospital's admin-
istrators the managerial and logistical issues of seeing that the necessary resources are
available when needed. This is a neat division of labor that preserves the physician's
role as the patient's agent intact. It also allows hospital administrators to be assigned
strong incentives to minimize cost subject to the constraint of the physicians' clinical
decisions. 7
The focus that Arrow and Harris place on physician agency may seem inappropriate
in an era of managed care in which the physician no longer has unfettered authority
in committing resources to a patient. As we discuss below, yet a third type of agent,
the payer, employs utilization review to monitor and limit to an appropriate level the
resources that a physician orders for an individual patient. But the message of their work
is only partly about the special role physicians with their informational resources play
and the consequent development of a strong patient-physician agency relationship. It is
equally about the responsiveness of the institutions that dominate health care markets
and the problems created by risk, moral hazard high information costs, inducement,
and the need to spread fixed costs over a sufficient number of patients. It is certainly
plausible that at one point in time unfettered physician agency might have represented
the best feasible solution. It is equally plausible that as technology, costs, and societal
expectations change, modified institutional structures may become more attractive.
7 This argument also relates closely to an important paper of Holmstom and Milgrom (1991) in the principal-
agent literature. Consider a product whose production process requires completion of two tasks and suppose
that it is easy to measure the quality with which the first task is performed, but it is difficult to measure the
quality with which the second task is performed. They show that the two tasks should not be assigned to the
same agent. If they are assigned to the same agent and the agent is given strong performance incentives based
on the measurement of the first task, then the agent will tend to neglect the quality of the second task. If the
agent is not given strong performance incentives, then the agent will tend to shirk the first task. The tasks for
which hospital administrators are responsible are much more measurable than the tasks for which physicians
are responsible; therefore, Holmstrom and Milgrom's analysis prescribes that the tasks be assigned to distinct
agents who face very different incentive schemes.
3. Monopolistic competition in health care markets
While agency defines the vertical relations among the panoply of different health
providers, competition has shaped their horizontal relations in each of the three insti-
tutional regimes that we use to subdivide our discussion. In the first regime, which is
characterized by cost-reimbursed hospitals and independent physicians, hospitals com-
pete on the basis of quality attributes for physicians and physicians compete on the basis
of price and quality attributes for patients. In the second regime, which is characterized
by prospectively reimbursed hospitals and (incompletely) price regulated physicians,
cost constrained hospitals and price constrained physicians compete on the basis of
nonprice attributes. In managed care, the third regime, payers compete on the basis
of price, administrative convenience, and the attractiveness of their provider networks
while hospitals and physicians compete on the basis of price and quality attributes, both
for inclusion into provider networks and for individual patients. Despite the distinctive-
ness of these three regimes, the competition among providers to which each gives rise
can be understood through a common model: monopolistic competition with explicit
representation of the quality attributes that the competitors possess and, within limits,
manipulate.
A consumer who has a specific complaint generally seeks care from a member of
the class of providers that is appropriate for that complaint and limits his considera-
tion to providers within his own metropolitan area. Consequently health care consists
of many separate geographic markets that are further segmented by type of provider.
A given market is at (or tending towards) a short run monopolistically competitive equi-
librium if each provider sells a differentiated product, if there are a sufficient number of
providers that each provider in selecting its own action takes the actions of the compet-
ing providers as given, and if providers are maximizing given the actions of the other
providers. If, in addition, no substantial barriers to entry or exit exist, then the market
8
tends towards a long run monopolistically competitive equilibrium.
Monopolistic competition describes the market for most health care services tolera-
bly well. Health care providers sell a differentiated product: patients idiosyncratically
prefer one provider to another based on real or perceived differences in ability and the
idiosyncratic match. As a consequence, a provider can increase price (or decrease some
quality attribute that consumers value) without losing all its patients to other providers.
Most health care providers compete in market segments that have a substantial number
of other providers whose services are good, but not perfect substitutes for their own
services.9 The competition is therefore diffuse: for a specific provider no one or two
8 Eaton and Lipsey (1989) in their lucid discussion of models of product differentiation review models of
monopolistic competition. The original statement of the model, and still a valuable reference, is Chamberlin
(1933).
9 A small city may be able to support only one or two hospitals. Hospital services there are either monop-
olistic or oligopolistic. But this exception is of limited importance, for most Americans live in metropolitan
areas of at least 500,000 people that are large enough to support enough hospitals, physicians, and managed
care plans to limit oligopolistic interactions and outcomes.
Ch. 20: The Industrial Organizationof Health Care Markets 1103
competitors are central enough that is fruitful for it to think how the competitor will
react to a change in price or service attributes. Instead providers tend to take their com-
petitors' strategies as given and to choose their own actions as best responses to the
actions of competing providers.l°
Entry and exit are relatively easy for individual health providers such as physicians
and managed care plans because each metropolitan area is a distinct market. Conse-
quently if a shortage of physicians exists in a quickly growing metropolitan area (as
indicated by higher than normal real incomes), then physicians will tend to relocate
from other areas into this more promising area. While there are sunk opportunity costs
in establishing a practice (including idle production time), these are minimal for mid-
career physicians and non-existent for physicians finishing their training. Entry and exit
is harder, however, for hospitals. With respect to entry, regulatory barriers may exist
and incumbent hospitals may expand either at their current locations or at new locations
in order to meet growing market demand. Nevertheless consumers' preferences for ge-
ographical convenience and prejudices against massive, impersonal institutions along
with limited scale economies have ensured that each large metropolitan area in the US
is served by a set of independent hospitals that is too large to form a tight oligopoly. 1l, 12
As alternatives to exit, hospitals have proved effective at downsizing and, in order to
share fixed costs across several locations, merging. 13 Moreover the nonprofit legal sta-
tus of many hospitals means that the management of a marginal hospital is not under the
same pressure to reallocate its invested capital to a higher value use. As a consequence
some degree of stickiness may exist in hospitals' adjustment to a new long run equilib-
rium after an inward shift of demand. 14 That some segments of the health care industry
are sticky, and only uncertainly driven to long run equilibrium, is not a problem for our
analysis because, except where otherwise noted, we rely on the short run equilibrium
properties only.
10 Dynamic oligopolistic interactions such as tit-for-tat pricing normally occur among a small, closed group
of firms that mutually regard themselves as the only competitors within a defined market segment. The spatial
character of competition among health care providers does not create neatly closed groups among whom such
interactions are likely to develop, but rather overlapping sets of competitors among whom noncooperative
coordination is difficult.
11 Metropolitan areas with a population above one million usually have ten or more hospitals. For example,
Milwaukee, with a metropolitan area population of 1.4 million, has over 15 community hospitals.
12 The purpose of this paper is not to speculate about the future evolution of health care in the US, but it is
worth remarking that one of the unintended consequences of managed care may be a strong incentive for the
buyers of services - insurance companies, etc. - to consolidate in order to gain scale and scope economies
and market power. To the extent this process continues our characterization of the market as monopolistically
competitive may become outdated and issues of monopsony power [see Pauly (1998)] and dynamic oligopoly
behavior may become central.
13 These mergers have attracted intense antitrust scrutiny. See Gaynor and Vogt's (2000) chapter in this
Handbook.
14 An action has option value if it can be delayed, if uncertainty about future returns exist, and if waiting will
resolve some of the uncertainty. Closing a hospital is an irreversible act that generally can be delayed in the
hope that the uncertain competitive situation will improve. The exit decision's option value creates stickiness
in the adjustment process towards long run equilibrium. See Dixit (1992).
3.1. Demand under monopolistic competition
The heart of monopolistic competition theory is the demand structure in which each
provider as a result of both its differentiation and consumers' poor information faces a
downward sloping demand curve. In fact, with respect to health care, our intuition is
that consumer demand is very unresponsive to an individual provider's choice of price
and, consequently, steeply downward sloping. This intuition derives from three facts:
(A) Consumers often consider the purchase of health care services to be a necessity.
Therefore their quantity purchased is not responsive to the price that they must pay.
(B) Consumers in choosing among several substitutable health care providers must con-
sider many nonprice attributes. They are loyal to individual providers with whom
they have had good experiences and, unless they are dissatisfied with some other
aspect of the providers care, they are unlikely to switch to gain a small price reduc-
tion.
(C) Because of insurance, consumers do not bear the full burden of the price differential
between a high priced provider and another provider. This reduces the importance
of price relative to other determinants of provider choice.
These three facts, however, do not imply that the demand an individual provider faces
is not elastic with respect to its price and quality attributes. This is fortunate, for if an
individual provider's demand were not elastic with respect to its own price, then the
provider's optimal price would be unbounded and market forces could not be depended
on to limit providers' incomes.
To see why this is so we must review Chamberlin's model of demand for a mo-
nopolistically competitive firm. Each provider i chooses its price pi and the levels of
two quality attributes, xi and yi, to maximize its objective function (whether it be net
income, subjective utility, or social welfare) subject to the constraint of its demand func-
tion
qi = Qi (pi, xi, yi; p-i, x-i, y-i, 1), (3.1)
where qi is its expected quantity demanded, P-i, x-i, and y-i are the vectors of the
other providers' prices and quality levels that shift i's demand curve, and I is the level
of insurance coverage in the market. Figure 3.1 replicates this standard diagram of de-
mand. The curve dd is firm i's demand; it is traced out as firm i varies its price Pi,
all other firms keep price fixed at price po, and all quality attributes remain unchanged.
Given that each provider regards its competitors' choices of prices and quality attributes
as fixed, this is the provider's demand curve in the sense that it is the curve that con-
strains its choice of price. By contrast the curve DD traces out the quantities that are
demanded from i if as it varies its price pi, the other providers vary their prices in lock-
step with firm i (i.e., p = P2 = ... = i = ... ), and all quality attributes xi, yi, x-i,
and y-i are constant. Necessarily the DD curve intersects the dd curve at the price po.
To establish notation, let q/Q and /,d be the price elasticities of the DD and dd demand
Pi
L
Po
/7.
r YiI
Figure 3.1. Curve dd is firm i's demand as it varies its price pi and other firms keep their prices at P0. Curve
DD is firm i's demand as it varies its price Pi in concert with the other firms.
curves, respectively. Also let /q and yq, respectively, be the quality elasticities of qi
with respect to provider i's quality decisions xi and yi.
The DD curve may be expected to have the same price elasticity of demand as does
the industry's aggregate demand curve because, with the other firms following firm
i's price in lockstep, provider i neither gains nor loses market share as it varies its
price pi. For adults seeking outpatient care, Newhouse et al. (1993, p. 121) estimate
that aggregate industry demand's price curve's elasticity IQ is -0.31, which is quite
inelastic and is in accordance with fact (A). The dd curve evidently must be at least
somewhat more elastic than this because if provider i reduces its price below the price
po that its competitors are charging, then some patients will switch to it and increase its
market share.
Somewhat more elastic, however, is not sufficient if health care is monopolistically
competitive: if a provider is maximizing profits, then the price it sets must be in the
elastic region of its dd demand curve. Thus, necessarily q < -1. But while it is easy to
argue that the dd curve is at least a bit more elastic than the DD curve, fact (B) suggests
that it can not be much more elastic because for most patients price is secondary to
quality. This intuition, however, is misleading as Satterthwaite (1979) shows. The idea
is that each physician has a panel of consumers, each of whom consider her to be his
doctor. Each month, for a variety of reasons, a small proportion, say 2.0%, of her panel
of patients switches to other physicians. Nevertheless, in equilibrium, her panel does not
shrink because an equal number of patients switch into her panel from other physicians.
Now suppose that she increases her price 10% and that this price increase causes the
proportion of her patients who drop off her panel each month to increase from 2.0%
to 2.5%. This small increase in her switching rate is sufficient to make her long run
demand price elastic: q < -2.5. Thus loyalty, as measured by low switching rates, is
consistent with the substantial price elasticities of demand that individual physicians
must face if the theory of monopolistic competition is to be applicable.
This is seen as follows. Suppose there are M physicians and N consumers in the
community. Focus on physician i and let Ni be the number of consumers in her panel.
The number of patient visits per month she receives is qi = v(pi)Ni, where Pi is the
fee she charges, vi (Pi), the visit rate, is the expected proportion of patients in her panel
who each month require care, and Ni is implicitly a function of both Pi and P-i. Let
the visit rate's price elasticity be p; it is approximately equal to the elasticity of the
DD curve because the visit rate reflects only patients' decisions to consume less care as
price increases, not their propensity to switch physicians.
Given that v only has the elasticity of the DD curve (approximately -0.3), most of
the elasticity of the dd curve must come from the response of Ni to changes in pi. To
see how this can be note that if i's panel is in equilibrium, then it is neither growing
nor shrinking, which is to say that the number of patients joining the panel must be
offsetting the number leaving. Equilibrium for the physician i's panel therefore requires
M
-s(pi, p-i)Ni +a(pi,P-i) E s(pj,p_j)Nj=0, (3.2)
j=l1jAi
where:
s(pi, p i) is the switching rate: the proportion of physician i's panel that
switches out each month when Pi is i's price and p-i is the vector of
competing physicians' prices,
a (pi, p-i) is the acquisition rate: the proportion of patients who are switch-
ing from other physicians and who join her panel, and
YEj=ljoi s(pj, pj)Nj is the total number of consumers that are leaving
the panels of the other M - 1 physicians and that physician i's panel
can potentially acquire.
Each consumer is a member of only one physician's panel at a time; therefore,
YjM=1Nj = N. If the market is in equilibrium, then Equation (3.2) holds for each of
the M physicians. Together the resulting M + 1 equations can be solved to show that,
if there a reasonably large number of physicians [so that (M - )/M 1] and if each
physician has demand that is symmetric to the demand of every other physician in the
market, the price elasticity of each physician's demand curve is
q t pq+ - + rSi (3.3)
where rip is the cross elasticity of physician j's switching rate s (pj, p_j) with respect
to i's price Pi and tp, ip, and rip are, respectively, the elasticity of physician i's visit
rate, acquisition rate, and switching rate with respect to her own price pi.
Two points regarding formula (3.3) should be noted. First, it shows that, as asserted
above, a change in the switch rate from 2.0% to 2.5% per month in response to a 10%
increase in price implies long run demand that is price elastic. To be specific, suppose
s (40, 40) = 0.02, i.e., if physician i and other internists charge $40 for a standard office
visit charges, then 2% of her patient panel switches to another physician each month. If
she increases her charges 10%, then her switch rate increase 25% to 0.025, which im-
plies that her switching rate's price elasticity rp is 2.5. Equation (3.3) therefore implies
that her price elasticity, r/q is less than -2.5.15 Second, Equation (3.3) generalizes to
any other variable that physician i controls. For example, if one dimension of quality is
the length, x, of a standard office visit, then rq = r + r x --rrs - x , where is the
elasticity of her visit rate with respect her visit length xi, etc.
3.2. Information and equilibriumprice and quality

q
The price and quality elasticities of demand, r/q, r/x , and ri, together with marginal
costs, determine the equilibrium levels of price and quality that a physician selects.
Equilibrium price is guaranteed to be above the benchmark value of marginal cost be-
cause each provider's downward sloping demand curve gives it market power. The is-
sue of whether quality competition can lead to excessively high, or possibly excessively
low, levels of quality is considered by Dranove and Satterthwaite (1992). Their model
of monopolistic competition suggests that when purchasers are relatively price insen-
sitive, but are quality sensitive, then the equilibrium level of quality may be excessive.
On the other hand, if purchasers are price sensitive, but not quality sensitive, then the
equilibrium level of quality may be too low.
The key actors in their model are the purchasers whose observations of providers'
price and quality levels are noisy. Each purchaser optimally searches for the provider
that best meets his preferences, correcting in a Bayesian manner for the noise he knows
contaminates his price and quality observations. The effect of his correction is to cause
him to weight less heavily in his decision those attributes whose signals are noisy than
their importance in his utility function would otherwise prescribe. Noise in an attribute
therefore reduces the responsiveness of demand to changes in that attribute, i.e., the
elasticity of demand with respect to an attribute decreases in absolute value - becomes
less elastic - as the noisiness of the consumer's observations of it increase.
Though Dranove and Satterthwaite's model is complex because of their explicit rep-
resentation of the noisy observations, most of the key insights can be obtained by ex-
amining a model of price and quality choice by a single provider. Suppressing explicit
15 This is moderately elastic demand, yet physician i's switching rate is very small and the increase in her
switching rate that her price increase causes is even smaller. As a result she undoubtedly has difficulty per-
ceiving how much, if at all, her price increase changes her switching rate. This suggests that physicians (and
other health care providers) do not find the profit maximizing price through a process of explicit optimization,
but find it as a result of an evolutionary process in which individual physicians imitate the pricing decisions
of other physicians who seem to be earning a higher net income. To our knowledge, however, no one has
explicitly applied the ideas of evolutionary game theory to health care markets.
1108o D. Dranove and iM.A. Satterthwaite
consideration of competing providers does little harm because the primary determinant
of providers' equilibrium behaviors are the elasticities of demand with respect to the
price and nonprice service attributes that they face. Consider, to be concrete, a provider
who chooses the profit maximizing levels of three attributes: price (p), clinical qual-
ity (x), and patient amenities (y), where the subscript i is suppressed. Her demand is
q(p, x, y); it is decreasing in p and increasing in x and y. For simplicity, suppose that
total costs are
C(q, x, y) = q(a + bx + cy) + F, (3.4)
where a, b, and c are positive constants, (a + bx + cy) is the constant marginal cost of
producing an additional unit of care, and F is fixed costs. If she maximizes profits, then
her objective function is
n(p, x,y) = pq(p, x, y)- C[q(p, x, y), x, y]

= q(p,x, y)(p - a - bx - cy)- F. (3.5)
Taking first order conditions, solving for the optimal p in terms of cost parameters a,
b, and c and the three demand elasticities qp, 7xq, and Biy, and solving for the optimal
levels of x and y in terms of p, the cost parameters, and the demand elasticities results
in the formulas]6 :
p, =ar_ (a + bx* + cy*)lqp
q + . q
x*= b(l+. aq P* 1 eqq ) b " q (3.6)
X*= a _ (3.7)
q
ar17 q Y (3.8)
)
c(l + rqP ++r + ry -c qp
P
These formula capture the equilibrium effects of how changes in the market's infor-
mation structure influence the equilibrium choices of the providers. There is a second
effect, the matching effect, that identifies the welfare gains that consumers realize from
being able, with better information, to identify better providers who offer attractive
price/quality bundles. Numerical experiments, however, suggest that this effect is less
important quantitatively than the equilibrium effect.
16 These formulas assume that providers are profit maximizers. Many health care providers are nonprofits,
including most hospitals. However, most nonprofits must respond to market forces or face bankruptcy. More-
over, many nonprofits may elect to maximize profits in some sectors (e.g., privately insured patients) so as to
undertake unprofitable activities in other sectors (e.g., provide uncompensated care). If we were to modify the
model to account for some eleemosynary objectives, our main qualitative insights would remain.
The implications of these formulas are more easily understood if further symmetry is
imposed on the model's parameters. Suppose clinical quality (x) and patient amenities
(y) have identical marginal costs of unity (a = b = 1) and suppose further that they
appear symmetrically in consumers' utility functions. These two assumptions together
imply that if no information problems existed in the market, then consumers' demand
for the two attributes is symmetric and, at equilibrium, ix = 1q, x* = y*, and x* and
y* are optimal in a market without insurance. 1 7 With this as a baseline the implications
of formulas (3.6)-(3.8) can be outlined:
* Equilibrium price p* is greater than the marginal cost of production a + bx + cy even
in the perfect information case. This is due to the fact that each provider faces down-
ward sloping demand as a result of its intrinsic differentiation. If information with
respect to price is imperfect in the sense that consumers can not observe it perfectly,
then providers' demand is distorted towards inelasticity, causing equilibrium price
and equilibrium price-cost margins to increase. The quite inelastic overall demand
for healthcare implies that the increased price may not have a serious welfare loss
in the sense of restricting consumers' purchases of care. Excess price-cost margins,
however, do cause two potentially serious welfare losses. First, as discussed, below,
they may induce providers to overproduce the nonprice attributes x and y. Second,
in the short run they cause profits to be earned in the industry that, in the long run,
attract entry. Entry does not necessarily drive down prices; instead it may dissipate
these profits - in fact, may waste them - because each new firm's fixed costs F are
a direct deduction from total industry profits. To the extent that insurance further re-
duces the elasticity of demand facing sellers, we would expect price-cost margins
to be even higher when there is insurance, therefore magnifying these two welfare
losses.
* Information problems can cause the equilibrium levels of quality x* and y* to be
too high. This can occur if quality attributes remain perfectly observable, but price
is only observable with noise. This distorts the price elasticity irq towards zero (i.e.,
demand becomes less elastic with respect to price) and causes equilibrium price to
rise. 18 This increased price-cost margin gives each provider an incentive to increase
its quality levels beyond the first-best levels. 19
* Information problems can cause the equilibrium levels of quality x* and y* to be too
low. This can occur if relative to the baseline perfect information case price remains
perfectly observable, but the two quality attributes are only observable with noise.
This distorts the quality elasticities /x7and irq towards zero and causes equilibrium
quality levels x* and y* to fall below the first-best levels.
17 See Section 3 of Dranove and Satterthwaite (1992). Their equation (20) gives the equilibrium level of
quality and their equation (24) gives the first-best level of quality when there is no insurance.
18 In Dranove and Satterthwaite's model changing the observability of price has a small effect on the quality
elasticities, but the effect on the price elasticity dominates.
19 This is the well-known Dorfman-Steiner (1954) effect.
* The distortions on x and y can go in opposite directions. Note that the ratio of their
equilibrium levels is
)C Ccx4_ AX
1-
7- o(3.9)
where the second equality follows from the assumption that b = c. Suppose that con-
sumers have difficulty observing price accurately; this forces the price-cost margin
up and tends to cause quality levels to be superoptimal. Further suppose, as appears
plausible, that consumers have great difficulty observing x (clinical quality), which
distorts ri towards zero, while they have little difficulty observing y (patient ameni-
ties), thus leaving sql essentially unchanged from the baseline case. Inspection of (3.9)
shows that this may reduce x* substantially below its first-best value and inspection
of (3.8) shows that y*, which was pushed above its first-best value as a result of price
being imperfectly observed, is decreased marginally back towards its first-best level.
The point of this exercise is to illustrate how information imperfections may distort the
equilibrium for a given health service. The last example, in which clinical quality may
be seriously underproduced, is meant to show the seriousness of the difficulties with
which the theory is perfectly consistent.
One may object, however, that the example is in fact implausible because profes-
sional ethics constrain physicians and other providers from changing their clinical care
substantially in response to economic incentives. The counter argument to this is to con-
cede that physicians are unlikely to knowingly compromise the quality of care that they
provide. Nevertheless if there are not strong economic incentives to stay current with
advances in clinical practice, then physicians may well not make the investments in con-
tinuing education that are necessary. Thus over a career a physician may act in a manner
consistent with this model not by compromising her standards, but by underinvesting
and becoming obsolete. 2 0 The essence of this information based view of competition in
health care is captured by the managerial maxim, "You get what you measure".
20 Anecdotal support for this possibility is provided by T. Burton in a front page article in the Wall Street
Journal ("United Health Care finds drugs, tests are often underutilized", 8 July 1998, p. Al). Beginning in
1996, United Health Care embarked upon a program of using its claim records to grade the physicians who
treat its enrollees on how well the physicians conform to standard medical practice guidelines. For example, a
well established standard is that heart attack patients should receive beta blockers to prevent a second attack.
United found that only 37 percent of the heart attack patients of its Ohio cardiologists, had been put on
beta blockers within two months of their heart attacks. As Waenard L. Miller, a Texas cardiologist who was
surprised that he did not conform to the standards as well as he would have liked, remarked, "I'm 51 years old
and almost none of these standards were talked about during medical school and training. They are virtually
all subsequent standards."
4. Regime I: Independent physicians and hospital cost reimbursement
By the mid-twentieth century medical knowledge had advanced sufficiently that physi-
cians could do much more than diagnose symptoms, correct a few acute ailments with
surgery, and provide reassurance as disease followed its natural course. With the avail-
ability of new and effective therapies came the growth of hospitals as essential medical
institutions, for without a modem hospital a physician could not employ many of the
new diagnostic techniques and therapeutic regimes. Hospitals, rather than their private
offices, became in Pauly's phrase "doctors' workshops". 21 These exciting and beneficial
changes in technology created two difficulties for the medical care system:
* Health care was now successful enough that consumers began regarding it as a neces-
sity, a necessity whose need is largely unpredictable and occasionally very expensive.
The created the need for health insurance that the nonprofit Blues and the competing
commercial plans filled. In 1965, Medicare was established to ensure coverage for
the elderly. These plans by and large pegged physician payments to observed market
outcomes, i.e., physicians received either their usual and customary fee or the pre-
vailing fee within the community. Hospitals were reimbursed primarily on the basis
of their costs.
* Hospitals needed to be expanded and modernized. Rapidly growing areas of the
country needed new hospitals. Communities were determined to meet this need; to-
gether private philanthropy, local governments, and occasionally profit-making firms
financed and organized an enormous growth of the hospital system.
The adaptation of the health care system to the opportunities created by new technology
fits the agency perspective. Physicians were the trusted agents of patients. The physi-
cians needed their workshops and the consumers needed health insurance so that they
could afford the care that modem medicine made possible. The nonprofit hospitals and
the nonprofit Blues, together with their copycat commercial brethren, responded to this
need in a cooperative and, at least initially, very successful manner.
By the late 1960s the success of the health care enterprise became its problem: it had
become a big, expensive, rapidly growing segment of the economy and questions be-
gan to be raised concerning the efficiency of its operation. Understanding the workings
and efficiency of this set of institutional arrangements is the task of this section. We
focus on two issues that attracted much attention at the time, price competition among
physicians and the "medical arms race" among hospitals. These issues remain important
for two reasons. First, generating effective price competition through managed care (or
some other set of institutional arrangements) is essential if a market oriented health care
system is to achieve anything close to optimality. The discussions concerning physician
pricing under this first regime have relevance to more recent discussions of managed
care pricing. Second, the new institutional structures of regulation and now managed
21 See Pauly and Redisch (1973) and Pauly (1980).

care have been built on top of this regime of independent physicians and cost reim-
bursed hospitals. Aspects of this regime persist to the present, so it is well to understand
it.
4.1. Price competition amongphysicians
The agency perspective, as first explained by Arrow, provides a convincing explanation

of why consumers of health care goods and services do not resemble consumers of more
"traditional" goods and services. In a nutshell, most patients are neither motivated nor
capable of being good shoppers for two principal reasons:
* The vast majority of consumers have health insurance, and this reduces their incen-
tives to seek out the best price.
* While consumers may have good information about providers' amenities and inter-
personal aspects of quality (such as bedside manner), they have little basis for com-
parison shopping on those aspects of quality that affect outcomes. 2 2
Consumers understand both these points and, as a consequence, realize that they are
better off relying on their primary care physicians to act as their agents in the health
care purchase decision. Thus competition among hospitals and among medical spe-
cialists was largely driven by insured, poorly informed patients and their primary care
physicians. Dranove, Shanley, and White (1993) describe this as "patient-driven com-
petition."
There has been little theoretical research directed towards understanding how
providers set prices under patient-driven competition. Certainly, the factors affecting
pricing decisions may differ for various types of providers. While hospitals may have
constrained their pricing decisions on account of their nonprofit status, no coherent the-
ory has been proposed as to why medical specialists such as surgeons, whose services
have for many years been largely covered by insurance, did not increase their fees even
more rapidly than was the case during the period. On the other hand, for some physicians
such as general practitioners, internists, and pediatricians the agency argument against
normal market behavior did not apply as strongly. For these physicians insurance was
much less of a factor than for surgical care and hospital services. Copayments and de-
ductibles represented a larger percentage of the total price of non-hospital physician
services, if they were covered at all, than for hospital and surgical services. Addition-
ally, consumers may have had better information about physicians who provide primary
care than about specialized physicians and hospitals because, if no other reason, con-
sumers were likely to have had more experience buying physician services than hospital
services. Thus, one might expect that the "normal" rules of competition might apply to
physician services outside the hospital.
22 These distinctions are further discussed by Donabedian (1980). Haas-Wilson (1994) reports that consumers
of optometry services were unable to comparison shop on the basis of technical quality (e.g., the accuracy of
the prescription and workmanship of the eyeglasses) and relied instead on interpersonal aspects of quality. As
a result, prices reflected the latter but were uncorrelated with technical quality.
Ch. 20: The Industrial Organization ofHealth Care Markets 1113
While there is little theoretical research about patient-driven competition, there has
been substantial empirical research. An early study by Feldstein (1970) apparently
found an upward sloping demand curve for physician services. He posited that physi-
cians kept prices artificially low so as to create "permanent excess demand," thereby
enabling themselves to only work on the most interesting cases. Feldstein's empirical
methods left much to be desired [see Ramsey (1980)], however, and researchers looked
for other evidence about provider markets.
A number of researchers, most notably Evans (1974) and Fuchs (1978), note a ro-
bust empirical fact: prices for physician services were positively correlated with the per
capita supply of physicians across communities. Fuchs (1978) offered one possible ex-
planation based on demand inducement. As detailed in McGuire's chapter in this Hand-
book, the inducement literature has been plagued by a host of methodological issues, so
that it is difficult to draw any conclusions about the effects of competition.
Another possible explanation, which we take up here, is that there can be an equi-
librium correlation between physician supply and prices without a shift in the market
demand curve. Evans (1974) offers a simple explanation as to how this might occur.
He proposes that physicians have target incomes. If the supply of physicians increases
relative to the demand, then physicians raise their prices to maintain their income tar-
get. Thus, physician prices are directly related to the supply of physicians per capita, as
indicated by the data. The assumed causality is from supply to price.
Satterthwaite (1985) argues the opposite causality. He offers a model in which (i)
physicians maximize income and (ii) the price elasticity of demand facing individual
physicians varies across markets for exogenous reasons. As a consequence, prices vary
across markets. Assuming that physicians make their location choices at least partially
in response to income opportunities, long run equilibrium necessarily implies equivalent
real incomes across communities. 2 3 This, in turn, implies that communities in which
elasticities facing individual physicians are inherently low will have high prices and
more physicians.
Satterthwaite (1979) also offers an explanation as to why elasticities facing individual
physicians may differ across markets. Physicians in short run equilibrium price in the
usual way - setting marginal revenue equal to marginal cost. He argues that the price
sensitivity within a market is a function of the information that consumers have about
alternative providers and that, among other determinants of consumer information lev-
els, increases in the number of providers may reduce their information that is relevant
to shopping for physicians. The reason is that when there are more physicians in a com-
munity, patients are unable to form concrete opinions about their abilities, because it is
difficult to obtain and remember substantial amounts of information about any partic-
ular physician. 24 Thus, increases in physician supply may by itself conceivably drive
prices up.
23 Newhouse, Williams, Bennett, and Schwartz (1982) provide convincing empirical evidence that physi-
cians' location decisions do tend to equalize real incomes across communities.
24 This is analogous to the "too many breakfast cereals on the shelf" concern about product proliferation and
pricing.
While Satterthwaite's paper is best known for this counterintuitive result, its main
contribution is to argue that information is a critical determinant of the price elasticity
of demand facing individual providers. Pauly and Satterthwaite (1981) provide evidence
to support this argument. They develop a number of variables that plausibly proxy for
the availability of information about physicians, including population density (increases
information), recent move-ins to the area (decreases information), and availability of
public transportation (increases information). Each variable has the expected effect -
factors that increase information are associated with lower prices.
Satterthwaite (1979, 1985) and Pauly and Satterthwaite (1981) are quick to state that
this model of price competition only applies to providers whose patients bear part or all
of the cost of care. Thus in their empirical work, Pauly and Satterthwaite only apply it
to physicians who are involved in primary care, not to surgeons and hospitals. Indeed,
it seems at least plausible that a hospital facing patient-driven competition faced very
inelastic demand with respect to price. If so, it is necessary to ask why such hospitals did
not charge much higher prices than they did. Perhaps, as suggested above, the answer lay
in the fact that the vast majority of hospitals were and continue to be nonprofits. 2 5 How
traditional notions of structure, conduct, and performance, such as the role of market
concentration, affected pricing by nonprofits under patient-driven competition remains
largely unknown.
4.2. Quality competition and the medical arms race
The notion that physician pursuit of self-interest can generate inefficient outcomes mo-
tivates the theory of the medical arms race (MAR). The MAR hypothesis begins by
noting that physicians act as agents for patients seeking hospital care, and influence
their choice of hospital. Hospitals seeking to increase admissions must therefore be
attractive to physicians. Because it is considered unethical to pay physicians for admis-
sions (and it is illegal to do so for Medicare and Medicaid patients), hospitals compete
for admissions by providing services that complement physician work effort, thereby
enhancing physician productivity. The costs of these additional services were passed
through to insurers who, during this first regime, cost reimbursed hospitals. Physicians,
because of their independence, the theory predicts, capture at least some of the gains.
Implicit in this story is that both financial constraints and, in the case of nonprofit hos-
pitals, a sense of public responsibility constrained the willingness of hospitals to meet
physicians' demands.
Economists have conjectured about the mechanism whereby physicians compel hos-
pitals to provide complementary services. For example, Robinson and Luft (1985) state,
"physicians affiliated with one institution may implicitly or explicitly threaten to shift
25 For-profits did enter some markets and prospered until the advent of managed care.
their patients elsewhere if the desired clinical services are not made available". Thus,
cardiologists may threaten to switch hospitals if their present hospital does not offer
cardiac catheterization services. This threat may be made credible if the cardiologists
have privileges to practice in other local hospitals. Or, physicians may exercise their
demand for complementary services by expressing their voice on hospital boards. To
date, however, there have been no formal analyses of these or other conjectures about
how physicians have been able to capitalize on their control of admission decisions.
In the context of our model, the MAR is just an example of nonprice competition in
which the choice of quality exceeds the social optimum. In the MAR, hospitals compete
for physicians and their patients by providing the latest technology, excessive staffing,
and lavish amenities. Enthoven (1978) and others explicitly state that such nonprice
competition drives up costs, and implicitly assume that the final level was above the
social optimum. While the bulk of the systematic evidence (discussed below) suggests
that the MAR led to higher costs, only anecdotal evidence exists that costs exceeded
the social optimum. For example, Finkler (1981) finds that open heart surgery centers
in urban markets - precisely the sort of service one would expect to proliferate under
the MAR - operate at less than minimum efficient scale. Ironically, if there is a posi-
tive relationship between clinical quality and volume at a particularhospital, then this
MAR-induced proliferation of high-tech services may actually lower quality. 26
Evidence on quality competition. The MAR hypothesis may take one of two forms:
(i) increasing the number of sellers in the market increases the elasticity of demand
with respect to quality that each seller faces or (ii) increasing the number of sellers in
the market leads to higher levels of quality. The intuition underlying these statements is
that as the number of hospitals within the market increases, each physician in effect
has more bargaining power and therefore, by playing the hospitals off against each
other, can extract more services and facilities for her patients. More formally, the first
statement may be thought of the "pure" MAR hypothesis. The empirical literature to
date examines the second statement, and typically involves estimation of the reduced
form:
Y = f(H, X), (4.1)
where Y is some measure of cost or service intensity, H is a measure of concentration,

and X is a set of control variables.
Recall from Equations (3.1) and (3.2) that a change in market concentration affects
quality both directly (through the effect on the quality elasticity) and indirectly (through
the effect it has the price elasticity and the price-cost margin). Thus, the existing liter-
ature measures this combined effect. This is not a serious problem for studies of the
MAR during the era of cost-based reimbursement because the price-cost margin was
not necessarily related to market concentration. It is, however, problematic in later years
26 See Luft, Hunt, and Maerki (1987) for evidence on the volume-outcome relationship.
when managed care organizations replaced cost-based reimbursement with negotiated

fees that depended in part on their market power. Simultaneous equation estimation
is necessary to disentangle the conflicting effects of market competitiveness on quality
choice.27 Occasionally researchers recognize the simultaneity problem when discussing
their findings, but often they do not.
Robinson and Luft (1985) provide one of the earliest studies of the MAR. They com-
pute the number of competitors faced by each hospital in the United States, considered
to be all other hospitals within fifteen miles of the hospital in question. Controlling for
a wide range of potential determinants of costs, they found that hospitals with more
competitors had higher costs per admission. Using similar methods, Robinson and a
variety of coauthors show that hospitals in the 1970s and early 1980s that faced more
competitors tended to have higher ratios of employees to patients [Robinson (1988)], of-
fered more angioplasty and coronary bypass facilities [Robinson, Garnick, and McPhee
(1987)] and offered more mammography, emergency, cobalt therapy, heart surgery and
catheterization services [Luft et al. (1986)]. All of these studies lend support to the MAR
hypothesis.
Zwanziger and Melnick (1988) point out that the changing nature of competition
might cause the negative relationship between cost and concentration to erode and even-
tually disappear. Without appealing to a formal model, they argue that hospitals under
pressure from managed care could more effectively attract patients by restraining costs
and securing additional managed care contracts than providing underutilized facilities
for a few demanding physicians. This is the indirect effect that the price elasticity of
demand r/q has on quality choice in Equations (3.7) and (3.8). Zwanziger and Mel-
nick were among the first economic researchers to exploit the well-known financial
and utilization data provided by the California Office of Statewide Health Planning and
Development (OSHPD). They estimate a cost/competition regression similar to that in
Robinson and Luft, with two differences: they focus on California and they explore the
1983 to 1985 time period, which was a period of managed care growth in California.
Zwanziger and Melnick find evidence of an MAR in 1983 that disappeared by 1985,
i.e., the cost/concentration relationship is negative in 1983 and zero in 1985. These
results are surprising inasmuch as managed care growth in California during 1983-1985
was steady but unspectacular. Moreover, many of the costs associated with the MAR,
such as the costs of sophisticated imaging equipment, are fixed in the medium run. If
these costs were present in competitive markets in 1983, then they would still be present
to some extent in 1985. It is possible that the introduction of Medicare's Prospective
Payment System in 1983 placed downward pressure on costs in urban hospitals, and
that this accelerated the erosion of the cost/concentration relationship. While Zwanziger
and Melnick attempt to control for this pressure, they are hampered by data that do not
permit accurate measurement of the profitability of treating Medicare patients.
27 Whitney, Milgrom, Conrad, et al. (1997) use two stage least squares to determine the positive effect of
price on the quality choice of dentists, but do not consider the role that competition may have on the quality
elasticity.
Ch. 20: The Industrial Organization ofHealth Care Markets 1117
Conner, Feldman, and Dowd (1997) use national data from the American Hospital
Association to test the possibility that the MAR disappeared during the 1980s and early
1990s. They find an inverse and statistically significant cost/concentration relationship
in 1986 and a direct and marginally significant one in 1994. They also find that between
1986 and 1994 expenses rose significantly more rapidly in more concentrated markets.
Dranove, Shanley, and Simon (1992) challenge the empirical methods used by other
MAR researchers. They argue that their regressions are misspecified because they fail
to account for the endogeneity of hospital supply and economies of scale in provision of
costly hospital services, including high-tech services. Hospitals may be more plentiful
in markets that have greater demand, thereby inducing a correlation between "competi-
tion" and expenditures. 2 8 At the same time, economies of scale may limit the supply of
high-tech services such as magnetic resonance imaging equipment in smaller markets,
again inducing a correlation between competition and cost. They use ordered probit re-
gressions similar to those in Bresnahan and Reiss (1991) to identify the determinants
of the supply of high tech equipment in sixty market areas in California in 1983. The
market areas they examine include urbanized areas, as well as smaller, isolated, com-
munities. They control not only for a community's own demographics, but also for the
population and distance to nearby communities.
Dranove, Shanley, and Simon find that market competition is a significant determi-
nant of the supply of only two of the 13 service that they examine. Market competition
appears to have only a minimal effect on the overall supply of services. They conclude
that there is scant evidence of an MAR in California in 1983. This finding stands in
contrast to Zwanziger and Melnick's finding that the MAR was still present in 1983 and
did not die out until 1985. The difference may reflect Dranove et al.'s controls for scale
economies and for relational aspects of geographic markets. In particular, Dranove et al.
show that the estimated MAR is substantially larger if one neglects to control for the
population of and distance to surrounding markets.
5. Regime II: Hospital PPS and regulated physicians
By the late 1970s, there was a widespread perception that health care costs were in-
creasing at an unreasonable rate, fueled by moral hazard and the medical arms race. The
political economy of Medicare and Medicaid in particular was devastating to politicians
as their rapid growth eroded their budget flexibility. The societal response consisted of
reforms that did not break the existing agency relationships: physicians remained the
28 Thorpe (1988) points out a related problem, namely, inner-city hospitals may admit more severely ill
patients due to limited access. At the same time, they may hold greater excess capacity due to the higher
percentage of emergency admissions. This could induce a correlation between cost and concentration that is
independent of the MAR. Interestingly, Thorpe attempts to control for these factors but still finds evidence of
the MAR (using 1981 national data from the American Hospital Association).
agent of their patients while hospitals were regarded as essential and expensive tools for
physicians to use for the benefit of their patients.
We look at two reforms in this section: prospective payment and utilization review.
Each can be regarded as a tool to help physicians, other health professionals, and hos-
pitals be better agents for their patients by (i) giving them better information about
what reasonable norms for treatment costs are and (ii) establishing incentives to act on
this information. The work we report on focuses on the incentives that these reforms
gave physicians and hospitals to alter the quality (or the intensity) of the services they
provide and on establishing empirically that these incentives do change behavior. We
also review the literature on cost-shifting, a possible unintended effect of prospective
payment, and other changes that placed constraining financial incentives on hospitals.
Overall, the evidence suggests that the changed incentives do lead to modified ser-
vice intensities. Whether or not the changes in service intensity are large enough to
cause measurable changes in important outcomes such as mortality remains unresolved.
Without evidence on outcomes, it is difficult or impossible to evaluate whether the re-
forms in fact aided or interfered with physicians in their role as patients' agents.
5.1. Quality choice under prospective payment
Medicare's Prospective Payment System (PPS) was established in 1983. Under it Medi-
care began paying a prospectively set fee for each hospital admission. The fee varies
according to the patient's illness and required treatment. That is, in the terminology of
the program, the fee is determined by the diagnosis related group (DRG) into which the
patient fits. The fee for each DRG category approximates the nationwide average cost
of treating patients in that DRG. The straightforward intuition behind PPS is that by re-
placing a fee-for-service based system with a prospective system, hospitals would have
incentives to reduce costs as well to compete for patients indirectly through investment
in facilities desired by physicians. Many researchers have explored the effectiveness of
this reform and the subtle implications it has on the quality choices of hospitals and
physicians.
Analysis of incentives. Shleifer's (1985) model of "yardstick competition" formal-
izes the intuition behind the shift to PPS. Shleifer assumes that all sellers are in unique
markets (one might think of sellers as local monopolists) and - to mirror the PPS -
face prices equal to the average cost across them. Hence, each seller competes against
a cost "yardstick". He shows that sellers in yardstick competition have incentives to se-
lect the most efficient production technology. Dranove (1987) expands Shleifer's model
to permit local market competition. Dranove's model introduces the important assump-
tion that the cost of treating a patient in a given DRG is a function of both the hospi-
tal's efficiency and the severity of the patient's illness. Moreover, the average severity
of illness may vary across hospitals. Hospitals compare the PPS payment with their
own efficiency and their average patient severity, and elect whether to admit patients or
"dump" them onto the local government-owned hospital. Dranove shows that reducing
the within-DRG variation in severity of illness may lower total hospital costs.2 9
Both Shleifer and Dranove assume that quality is fixed. Allen and Gertler (1991) ex-
amine hospital choice of quality when efficiency is fixed, severity of illness varies from
patient to patient, and hospitals may not dump.3 0 Here and elsewhere in this literature,
quality is defined rather loosely as anything that shifts out the demand curve. In addition,
willingness to pay for improved quality is assumed to be higher for more severely ill
patients. Because the DRG payment is independent of severity, hospitals stand to make
money by treating less severely ill patients and lose money by treating more severely ill
patients. Using a partial equilibrium framework, Allen and Gertler show that hospitals
provide more than the first-best level of quality to less severely ill patients and less than
the first-best level of quality to more severely ill patients.
Allen and Gertler also show that it is welfare enhancing for the payer (e.g., Medicare)
to partially "cost-share"; that is, the payer pays hospital i a price Pijk for patient k
classified in DRG j according to a "mixed" reimbursement system:
Pijk = XFj + (1 - X)Cijk,
where Fj is a fixed component, Cijk is hospital i's cost of treating patient k, and is
a weight between zero and one. By partially basing reimbursements on actual costs,
quality is reduced for the least severely ill patients, enhanced for the most severely ill
patients, and social welfare improves accordingly.
There are a number of variants of these models. In Ellis and McGuire (1996) physi-
cians choose an intensity of treatment for hospital inpatients, and hospitals are reim-
bursed a fixed amount per admission. All patients benefit from greater intensity, but the
hospital makes greater profits if physicians select a lower intensity. Ellis and McGuire
assume that physicians balance patient benefits and hospital profits. They show that if
physicians on the margin equate a dollar of patient benefits to a dollar of hospital costs,
then they are "perfect agents" and select the most efficient intensity of treatment. How-
ever, if physicians place more weight on hospital profits than on patient benefits, they
will underprovide intensity. Finally, Ellis and McGuire show that if physicians are not
acting as perfect agents, then a switch to an appropriate mixed reimbursement system
can restore efficient decisions. Pope (1989) reaches similar conclusions and also finds
that as competition between hospitals intensifies, then the optimal level of cost sharing
(i.e., 1 - ) decreases.
29 A number of classification schemes were proposed during the 1980s that would reduce within category
variation in severity. For example, the Prospective Individualized Reimbursement model created fewer cat-
egories than the DRG system, but had only half the level of within-DRG variation in expenses [Johansen
(1986)].
30 Rogerson (1994) examines the interaction between a monopoly payer and monopoly nonprofit hospital.
The hospital, which values quality and quantity, may vary service intensity across disease categories but must
break-even overall. Rogerson finds that the payer will vary the payment so that the hospital earns positive
(negative) profits on services with low (high) intensity elasticities of demand.
Many recent treatments of the same issue, including Ma (1994), Gal-Or (1996), and
Ellis (1998), also support the use of mixed reimbursement systems. For example, Ma
(1994) extends previous work by permitting hospitals to separately choose their efforts
at cost reduction and clinical quality enhancement. He shows in this more general set-
ting that even if providers can refuse to treat expensive patients, the first-best cost and
quality can be reached through a mixed reimbursement system.
The consistent finding from this literature is that mixed reimbursement systems are
necessary to optimally balance cost and quality. Despite this finding, most insurers do
not use mixed reimbursement systems. 31 If the finding is correct, then the competitive-
ness of insurance markets should force insurers to adopt mixed reimbursement. There
may be simple explanations for the absence of mixed reimbursement. For example, per-
haps the efficiency issues presented by Shleifer and Dranove dominate the incentives
to promote quality. Alternatively, the absence of mixed models may result from com-
plex strategic interactions among payers and hospitals. Glazer and McGuire (1994), for
example, present a model in which hospitals can manipulate cost allocations causing
payers who use mixed systems to bear a disproportionate share of cost. In the resulting
equilibrium, every payer may choose a fully prospective payment scheme.
Empirical evidence on PPS's effects. Several studies have documented the effects
on access and quality of PPS and other government payment programs that set fees
prospectively. Anecdotal reports at the time of the reforms strongly indicated that in-
patient lengths of stay for Medicare patients dropped after the imposition of PPS - this
led to the well-known expression "quicker but sicker". The conclusion of the systematic
studies has been to confirm this conventional wisdom: the changed incentives do affect
access and quality as well as cost.
Langa and Sussman (1993) study the effects of reductions in Medicaid payments in
California on the provision of coronary revascularization. They find that Medicaid pa-
tients are significantly less likely to receive this costly intervention than are comparably
ill privately insured patients. They attribute the difference to the fact that Medicaid reim-
burses hospitals at a lower rate than do private insurers. Dranove and White (1998) find
that hospitals in California that were heavily dependent on Medicaid payments reduced
service levels to Medicaid patients following reductions in Medicaid reimbursements
in the 1980s. They also find some evidence that these service reductions were felt by
privately insured patients as well.
Ellis and McGuire (1996) examine changes in lengths of stay for psychiatric pa-
tients covered by New Hampshire's Medicaid program. In 1989, that program switched
from cost-based reimbursement to a prospective payment system similar to Medicare's
PPS. Controlling for demographic trends, they find that inpatient lengths of stay for
31 McClellan (1997) observes that the Medicare PPS appears to be mixed, because payments are increased
when certain procedures are performed. But payments do not change when patients receive more diagnostic
testing, nursing visits, etc. This suggests that the mixed payment is not intentionally tailored to match the
incentives suggested by economic theory.
psychiatric patients fell by approximately five days as a result of the move to prospec-
tive payment. They further find that reductions in length of stay were most pronounced
for patients with the longest stays, and that short stay patients actually experienced in-
creases in length of stay. As the former are the least profitable and the latter are the most
profitable, these results are supportive of the quality competition models such as Allen
and Gertler and Ellis and McGuire.
Models of PPS predict that payment reductions should affect admissions as well as
quality. This could be a demand response (fewer patients seek admission because quality
is lower) or a supply response (hospitals turn away unprofitable patients). To test this
prediction, Newhouse (1989) constructs a "PPS pressure index", Iji, which measures
the expected profitability for hospital i of treating a patient in DRG j. He finds that
admissions of patients in DRG j at hospital i are an increasing function of Iji, which
is consistent with PPS models. Although he is unable to sort out all of the supply and
demand side explanations, he does rule out the possibility that hospitals are transferring
potentially unprofitable patients shortly after admission.
Gertler (1989) develops a model of quality choice that is a variant of Spence's (1975)
well-known model of quality choice by a regulated monopolist. In Spence, the marginal
revenue from an increase in quality depends on the price-cost margin and the elasticity
of demand for quality. Gertler's distinctive variant accounts for a peculiarity of the New
York nursing home market for which he has data. During the period he studies (the
1970s), most homes in New York were at capacity and could only admit more private
patients by reducing admissions of Medicaid patients. Gertler's model shows that under
these conditions, as Medicaid payments fall, nursing homes have an incentive to boost
private admissions, in part by boosting quality. He finds strong evidence to support
this prediction. Nursing homes receiving lower reimbursements from Medicaid offered
higher levels of service to private patients. Nyman (1988) adopts a similar approach in
his examination of nursing home expenses in Wisconsin. As in New York, the supply
of beds was constrained and most homes were at capacity. Nyman finds that in those
regions of Wisconsin in which excess demand existed, nursing homes competed less
intensely.
Finally, Cutler (1995) notes that implementation of PPS changed the effective prices
that hospitals received for inpatient care, but that the changes were not uniform across
all hospitals. He computes the change in reimbursements for hospital/DRG pairs and
finds a significant negative correlation between changes in reimbursement levels and
changes in within-hospital mortality, i.e., decreases in reimbursement are associated
with increases in inpatient mortality. However, decreases in reimbursement have no
effect on mortality in the first full year post-discharge, thus clouding his first finding.
5.2. Utilization review
Upon implementing the PPS, Medicare created Peer Review Organizations

(PROs) to assure the appropriateness of care provided to Medicare recipients. To do this,
PROs perform utilization review. During the 1980s, many private insurers also imple-
I 122 D. Dranove and M.A. Satterthwaite
mented UR (often employing agencies established by the same companies performing

UR for the PROs).
UR includes preadmission certification, presurgical second opinion, and case man-
agement for costly illnesses. Insurers contract with UR service agencies to monitor
treatment. Providers and their patients must obtain approval for a proposed treatment
plan (e.g., a proposed surgery) from the UR service agency. If the UR service agency
rejects the plan, it may work with the provider to develop another plan, grant an excep-
tion, or deny coverage.
Utilization review (UR) may usefully be thought of as a modification of physician
agency that is broadly consistent with the informativeness principle, which states that
if the principal can observe a signal that carries information about the appropriateness
of the agent's actions, then the principal should incorporate that information into the -
perhaps implicit - contract he has with the agent. 3 2 In this case, the information that
is being brought to bear on the medical care process comes from "medical outcomes
research." Outcomes research, which measures the effectiveness of different medical
interventions, has a long tradition in medicine. But reductions in computation and com-
munication costs have enabled researchers to increase outcomes research efforts and
combine results from myriad studies so as to establish practice "norms" for treating
specific illnesses. Given the informational costs, it may be impractical for individual
physicians to keep abreast of these norms. It might therefore be appropriate for a "third
party" (the UR service agency) to intervene in the clinical decision making process by
informing physicians of norms and, if necessary, penalizing physicians (and their pa-
tients) who depart from them.
The medical profession has been highly critical of UR and the accompanying in-
trusion into the medical process. Though such third party intervention into a purchase
decision seems unusual, it is actually common in complex business transactions. For ex-
ample, purchasers of businesses often hire lawyers, investment bankers and accountants
to perform due diligence. To take another example, home buyers usually hire inspectors
and lawyers to evaluate the purchase. What distinguishes UR from these other forms
of oversight is that it simultaneously addresses information and moral hazard problems,
rather than just information problems. As such, the role of UR may be analogous to
the role of the supervisor in many employment settings. Unsupervised, employees may
exercise moral hazard by shirking, and may work inefficiently due to prohibitively high
costs of obtaining information about their work requirements. Supervisors both monitor
workers to prevent shirking and teach workers the information they need to perform
their duties properly and efficiently.
UR has been widely adopted by U.S. health care insurers. As of 1995, nearly 90
percent of conventional indemnity insurance plans and nearly 100 percent of all PPOs
required some form of UR [KPMG Peat Marwick (1995)]. Even so, there is only lim-
ited evidence on the effects of UR on costs, and there is no systematic evidence to date
32 Holmstrom (1979, 1982) and Shavell (1979) develop the informativeness principle. For a nice discussion
of it, see Milgrom and Roberts (1992, Ch. 7, especially pp. 219-221).
of UR's effects on outcomes. Feldstein, Wickizer, and Wheeler (1988) and Wickizer,
Wheeler, and Feldstein (1989) examine the effectiveness of UR programs sponsored by
CNA Insurance. Both studies report statistically significant reductions in inpatient uti-
lization of 12-13 percent by employees in plans containing UR. Wickizer Wheeler, and
Feldstein reject the hypothesis that this reduction reflects the self-selection of certain
employers into plans with UR. Scheffler, Sullivan, and Ho (1991) find that UR plans
adopted by Blue Cross during the 1980s reduced inpatient expenses by about 4 percent,
but they do not test for endogeneity of UR adoption. Khandker, Manning, and Ahmed
(1992) find a 5 percent reduction in inpatient expenditures associated with the UR plans
Aetna adopted in the late 1980s. More recently, Wickizer, Lessler, and Travis (1996) find
that psychiatric UR restricted inpatient psychiatric care, primarily by reducing length of
stay.
The studies cited above provide valuable evidence that UR does alter the medical care
process, but leave many questions unanswered. The studies focus on inpatient use, and
therefore fail to measure any accompanying changes in outpatient utilization. Outcomes
are not studied. Lastly, the studies do not determine if reductions in inpatient utilization
result from effective use of outcomes research, as the agency perspective requires, or
merely reflect a capricious curtailment of services.
5.3. Dynamic cost-shifting33
The typical health care provider receives payment from many different payers - Medi-
care, Medicaid, managed care, traditional indemnity insurance, etc. For a given treat-
ment, say a routine appendectomy, a provider may receive grossly different payments
from different payers. That is, providers price discriminate. Given the practical difficulty
of reselling health care services, such price discrimination is not surprising.
When a class of consumers becomes more price sensitive, we expect sellers to lower
their price to them. However, we do not expect sellers to then substantially change the
prices they charge to other classes of consumers. 3 4 Yet many health care analysts claim
that when one purchaser (e.g., Medicare or Medicaid) extracts a price discount from a
health care provider, the provider increases the price it charges to other purchasers. This
is known as "dynamic" cost-shifting, or simply cost-shifting. As a result of cost-shifting,
one purchaser's gain may be another purchaser's loss.
Dranove (1988) offers a model in which nonprofits cost-shift. He posits that non-
profits maximize a utility function that includes profits and output and observes that
33 The term "cost-shifting" has several meanings, including accounting practices used by hospitals to enhance
revenues under cost-based reimbursement. Morrisey (1994) coined the term "dynamic" cost-shifting to refer
specifically to an increase in the prices that providers charge to private payers in response to cutbacks in
government reimbursements.
34 An exception is when sales to the more price sensitive class of consumers change considerably and
marginal cost is sharply increasing or decreasing [Foster (1985)]. We do not believe these conditions apply to
most health care providers.
hospitals would only cost-shift if they were not charging the highest price the market
could bear. Otherwise the price increase would simultaneously lower profits and output,
leaving the firm unambiguously worse off. A utility-maximizing hospital does not set
the profit maximizing price and, if, for example, Medicare cuts its payments, will in-
crease price to others to recover some of the lost profit and restore the balance between
profits and output. He also shows that the magnitude of the cost shift will shrink as
markets become more competitive.
Estimates of the magnitude of cost-shifting. Morrisey (1994) reviews the empirical
evidence on cost-shifting. Although cost-shifting is a dynamic concept, several studies
[e.g., Hadley and Feder (1985), Zuckerman (1987)] measure the magnitude of cost-
shifting using cross-section data. These studies compute each hospital's "need to cost-
shift", which is based on estimated losses from treating Medicaid, Medicare, and unin-
sured patients. Hospitals with greater needs to cost-shift are predicted to increase the
prices they charge to privately insured patients. 35 Neither paper finds any evidence of
cost-shifting - pricing for privately insured patients seems to be largely independent of
the need to cost-shift. Such cross-section analyses are not convincing. To understand
why not, suppose that a hospital's need to cost-shift increases in year 0. In year 1, it
raises its price to privately insured patients. The chosen price balances the hospital's
goals for profits and output, as described above. If the hospital's need to cost-shift re-
mains unchanged for the next few years, then the hospital may not change the price it
charges to private patients. Thus, a cross-section analysis may fail to find a correlation
between the need to cost-shift and changes in prices to privately insured patients. This
does not imply that the hospital in question would not raise its price again if the need to
cost-shift were to increase again.
Dranove (1988) exploits a unique change in Medicaid reimbursements to explicitly
analyze dynamic cost-shifting. The Illinois Medicaid program sharply reduced pay-
ments to hospitals in the early 1980s. Dranove tied these cutbacks to increases in prices
for privately insured patients. He estimated that for every dollar that a hospital lost from
Medicaid, it was able to recover about fifty cents through cost-shifting.
All of the cited studies use data that predates the growth of managed care. As Dra-
nove (1988) and Morrisey (1994) point out, the ability of providers to cost-shift may
be severely limited in a managed care environment. Two recent studies support this
view. Needleman (1994) examines private payer prices and uncompensated care levels
in California hospitals during the period 1986-1990. The two were significantly posi-
tively related in 1986, consistent with cost-shifting, but unrelated in 1990. Dranove and
White (1998) find that California hospitals that were hard hit by cutbacks in Medicaid
payments during the 1980s actually lowered their prices to privately insured patients.
This is consistent with simple models of price discrimination and is inconsistent with
cost-shifting.
35 A widely discussed study by the Congressional Budget Office (1993) uses a similar though more primitive
measure of cost-shifting. As an artifact of the way that cost-shifting is computed, any hospital whose revenues
equaled or exceeded costs is said to be fully cost-shifting. This is not useful for identifying dynamic cost-
shifting.
6. Regime III: Managed care and contracted physicians
Arrow describes insurance and physician agency as a reasonable solution to the patient's
problems of risk-bearing and high information costs. But physician agency implies that
consumers are neither motivated nor capable of shopping on the basis of price except,
perhaps, for their primary care physician, and there only if they face a substantial de-
ductible amount or copayment. By the early 1980s, the resulting patient-driven compe-
tition had permitted providers to charge prices well in excess of marginal costs.3 6 High
margins, coupled with ever rising costs, anecdotes about duplication of facilities and
empty beds, created a prima facie case that the health care system was inefficient. In
his classic critique of the health care system in which he introduces the concept of man-
aged competition, Alain Enthoven (1978, p. 651) criticizes fee-for-service payments to
physicians, and cost reimbursement for hospitals:
"Hospital charges and physician fees rose faster than consumer prices in general.
Health workers pay overshot equality with other industries. There is great ineffi-
ciency (e.g., duplication of costly underutilized facilities)... People might be just
as healthy with half as much hospitalization."
The federal government reacted to this situation by exercising its bargaining power
and imposing the Prospective Payment System on hospitals in 1983 and the Resource
Based Relative Value System on physicians in 1992. Individual private payers, whether
nonprofit or commercial, did not have the same power to impose new arrangements,
so they instead invented new organizational entities to take advantage of the market
opportunity that the excess capacity created. In particular, payers created new entities
called Preferred Provider Organizations (PPOs) that contracted with hospitals to obtain
discounted prices. These hospitals are dubbed the "preferred providers". In return PPOs
designed their benefits to encourage patients to select the preferred providers. Plan de-
sign typically involved differential cost sharing, so that patients who chose the preferred
providers faced lower copayments.
Selective contracting is the cornerstone of PPOs, for only if a provider perceives
the possibility that it might be excluded from the list of preferred providers does it
have an incentive to offer a discount.3 7 Health maintenance organizations (HMOs) also
selectively contract to obtain the best possible prices. By joining PPOs and HMOs,
enrollees effectively delegate the responsibility of price shopping to these payers. The
payers are motivated to shop around because discounts effectively reduce their cost of
doing business on a dollar for dollar basis. They are also more capable shoppers than are
36 Although there are no published estimates of price/marginal cost margins, Friedman and Pauly (1981)
estimate that the hospital's short run average cost is only half of its long run average cost. Most hospitals set
prices slightly above average costs. We interpret these facts as suggesting that prices were more than double
marginal costs.
37 Initially, state insurance regulations prevented insurers from forming PPOs, so early PPOs were often
formed by firms providing administrative services for a self-funded employer health insurance plan.
patients, for they shop on behalf of potentially thousands of patients and can therefore
afford to purchase information systems that permit comparison of complex cost and
quality information.
There is overwhelming evidence that selective contracting enables providers to obtain
lower prices. 3 8 The early success of PPOs suggests that even small purchasers were able
to obtain discounts. However, evidence offered by Staten, Umbeck, and Dunkelberg
(1988) and Melnick et al. (1992) suggests that larger purchasers obtain bigger discounts.
Staten et al. propose that this is due to traditional monopsony power. This does not seem
to be consistent with the fact that many discounting hospitals have substantial excess
capacity and may not be on the upward sloping portion of their marginal cost curve.
Researchers have yet to offer an alternative explanation for this pricing pattern. 3 9
In exchange for delegating shopping responsibility to HMOs and PPOs, patients give
up free choice of provider. The free choice of provider under patient-driven competi-
tion was, to some extent, illusory. Primary care physicians often limited their referrals
to a narrow network of close associates and their hospital. PPOs had, at least in the-
ory, the potential to create referral networks based on value, rather than professional
relationships.
The invention of the PPO takes shopping for price out of the primary care physician's
list of responsibilities to their patients and allows the physician to focus on patients'
medical needs. 40 PPOs inject into the health care system for the first time a player that
has both the incentive and the ability to seek the best value for the services consumers
demand. Dranove, Shanley, and White (1993) call this payer-driven competition. To un-
derstand payer-driven competition the first task is to develop detailed theories of how
it may affect the market equilibrium and then to estimate quantitatively the strengths
of the possible effects. Beyond that, it would be highly desirable to construct structural
models of how payers select which plans to offer, how consumers choose among those
plans, and how comparative information about the attributes of different providers af-
fects these choices. As we report below, substantial progress has been made on the first
set of questions and limited progress has been made on the second set.
6.]. Payer-drivencompetition
A hospital's average price is not a well defined term; most commonly it is stated as
a price per day or price per discharge. Under payer-driven competition its equilibrium
level is given by Equation (1) from our discussion of monopolistic competition. The
price per discharge is therefore a function of four factors:
38 For early evidence, see Rice et al. (1985).

39 Some insurance executives and hospitals managers with whom we have spoken believe that this reflects
risk averse pricing practices by hospitals that may face bankruptcy if they fail to contract with their largest
purchasers.
40 To one degree or another PPOs depend on primary care physicians along with utilization review programs
to control the quantity of specialized services patients use.
* That hospital's cost of providing average quality to a patient with average illness
severity;
* The illness severity of that hospital's patients;
* The intensity and quality of the services relative to the industry average that the hos-
pital provides its patients;
* The hospital's price/cost margin.
In the short run a hospital's quality levels are fixed and, therefore, the first three factors
determine the hospital's marginal cost. The last factor, the hospital's price/cost margin,
is determined by the hospital's perception of its price elasticity of demand. Given that it
negotiates with the PPOs in its market area one at a time and is able to price discriminate
among them, the major determinant of its price elasticity of demand, q with respect to
a particular PPO is the elasticity of the probability that its final offer to the PPO will be
accepted. 41 Thus in the short run a PPO succeeds in driving down price by convincing
each hospital with which it negotiates that there is a significant threat of exclusion from
its networks if a very competitive offer is not submitted. Once a hospital is accepted in
the network, the number of patients it will admit per year will depend on patient (and
their physician agents) perceptions of its fixed quality level relative to the quality at
other accepted hospitals.
In the long run the hospital's quality levels x and y are not fixed; therefore in the
long run the hospital's equilibrium price and the quality levels are jointly determined
by Equations (1-3) with its price elasticity, 1q interacting with its quality elasticities,
ax and 77Y. Suppose that PPOs do succeed in reducing hospital prices in the short run
by increasing each hospital's price elasticity of demand. The likely long run effect is
that hospitals decrease their quality levels because reducing the price-cost margin has a
strong, indirect, negative effect on optimal quality levels. To the extent that quality was
overproduced under patient-driven competition this is not a problem. However, if the
level of quality under patient-driven competition was appropriate, then any reductions in
quality caused by PPOs would be problematic. In principle, PPOs can prevent hospitals
from reducing quality too much by not only shopping among hospitals on the basis of
price, but also shopping on the basis of quality levels. Shopping for quality increases the
quality elasticities that the hospitals face and therefore forces up their optimal quality
choices.
Thus PPOs, at least in the short run, may solve a substantial problem in physician
agency: primary care physicians are not good price shoppers. In the longer run, however,
PPOs may induce a severe new problem: the hospitals and other providers among whom
41 With quality levels fixed, let the probability that the hospital's offer p to the PPO is accepted be p (p) and
let the expected number of discharges that the PPO will provide per year, conditional on the price p being
accepted by the PPO, be d(p). The expected number of discharges as a function of p is therefore p(p)d(p),
which has price elasticity 17P+ rd, where P and rId are the price elasticities of p(p) and d(p), respectively.
One expects that rd is much smaller than P because d(p) increases as p decreases only if, as the hospital
lowers its price, other hospitals are in expectation excluded from the PPO or patients pay a copayment and
seek a hospital with lower out of pocket costs.
the patient's primary care physician is choosing may be of poorer quality than is optimal.
A perfect agent can not overcome a mediocre set of alternatives.
Evidence on the effect of payer-driven competition. It is difficult for the researcher
to disentangle the various possible effects of local competition on prices. For example,
simple scale economies may permit hospitals in more populous (and therefore likely
more competitive) markets to offer higher quality services [Dranove, Shanley, and Si-
mon (1992)]. This higher quality may attract more severely ill patients. In particular,
if severely ill patients are willing to travel to seek out high quality, then the competi-
tion of large metropolitan areas should be associated with higher prices because large
metropolitan areas attract a more difficult case mix. Conversely, hospitals in small mar-
kets may offer lower quality, and end up treating the least severely ill patients. There
is some evidence to support this set of possibilities. Welch, Larson, and Welch (1993)
find that a hospital's "distant" patients (patients who more than fifteen miles from the
hospital) are more costly to treat than local patients. Adams et al. (1991) and Adams
and Wright (1991) find that rural patients with complex cases are more likely to travel
to urban hospitals than are rural patients with relatively simple cases.
Dranove, Simon, and White (1998) argue that selective contracting is more likely to
succeed in markets where there are several providers to play off against each other and
each has excess capacity that they are eager to fill, even at the cost of giving discounts.
They also argue that selective contracting is more likely to thrive in markets where
the cost of obtaining information about and contracting with providers is lower. They
test these hypotheses using data on physician contracting with HMOs and PPOs from
the American Medical Association. Consistent with their hypotheses, they find that the
penetration of insurance that uses selective contracting is higher in markets with more
hospitals, lower occupancy rates, and more physicians organized into groups.
These results, together with the results discussed above concerning patient-driven
competition and the MAR, suggest that testing for the effects of payer-driven compe-
tition fits the traditional price/concentration approach of empirical industrial organiza-
tion. Under patient-driven competition, the relation between price (or price-cost margin)
and concentration should be negative because of the MAR. Under payer-driven compe-
tition the relation between price and concentration should be positive because, as the
market becomes more concentrated, PPOs are less successful in playing hospitals off
against each other.
Noether (1988) is perhaps the first to recast the relationship between price and con-
centration in health care as a traditional industrial organization problem. Noether stud-
ies hospital competition within SMSAs. She uses a national sample of charge data
for Medicare patients throughout the United States and examines average charges per
admission within eleven different disease categories (e.g., diabetes mellitus, cataract
surgery). She also looks at average expenses per admission, although this is not broken
down by disease category due to the way that Medicare reports its data. Her data are
from 1977 and 1978, which is prior to the implementation of the prospective payment
system, i.e., it is during the period of patient-driven competition.
Ch. 20: The IndustrialOrganization of Health Care Markets 1129
While the question she raises and the methods she uses have been influential,
Noether's choice of data and time period is problematic. One reason is that hospital
choice is determined by patients and their physicians - managed care for Medicare pa-
tients was nonexistent in the study period. The price that most patients paid was invari-
ant to their choice of hospital. 42 Furthermore, hospitals were reimbursed by Medicare
on the basis of costs rather than charges. Noether offers no explanation as to how or
why the Medicare "price" would be related to concentration under such circumstances.
Thus it is not surprising that, controlling for a wide range of possible determinants of
price, Noether fails to find a significant relationship between concentration and price.
She does, however, find that concentration is negatively related to expenses. The latter
finding is consistent with the MAR. Noether concludes that competition simultaneously
drives up expenses and curtails margins. These findings echo those of researchers who
focused on the MAR using later data and are thus an interesting piece of evidence con-
cerning the market's behavior relatively early during its period of payer-driven compe-
tition.
Dranove, Shanley, and White (1993) adopt the methodology traditionally employed
in price/concentration studies to examine changes in the price/concentration relation-
ship that occurred in California during the 1980s. 4 3 Using OSHPD data, they construct
a "list price" (i.e., charges) and "transaction" price for a market basket of eight hospital
services including hotel services, radiology, laboratory, and other services. An advan-
tage of this approach is that it eliminates differences in price that result from differences
in service intensity. (Per diem or per admission prices may vary due to pricing strate-
gies or due to the level of service provided to the patient.) Dranove, Shanley and White
regress prices and price/cost margins against a number of key predictors, including the
Herfindahl index of hospital concentration. They find that margins were unrelated to the
Herfindahl in 1983 and significantly positively associated with the Herfindahl in 1988.
Prices and the Herfindahl were significantly negatively related in 1983 and significantly
positively related in 1988. This is consistent with the switch to payer-driven compe-
tition. Gruber (1992) offers further evidence consistent with the switch. He finds that
price growth among California hospitals was largely independent of market structure
for the period 1982-1984, but that growth was slower in more competitive markets dur-
ing the period 1984-1988. Gruber also finds that during the later time period, hospitals
in competitive markets decreased the amount of uncompensated care that they provided.
Lynk (1995) suggests that the effect of concentration on price may be different for
nonprofits. He uses OSHPD data for 1989 and estimates separate effects for nonprofits
and for-profits. He finds that for-profits charge higher prices in concentrated markets but
nonprofits charge lower prices in concentrated markets. A noteworthy feature of Lynk's
study is that, even though he includes a Herfindahl index on the right hand side (and
42 Many patients would reach the deductible ceiling regardless of which hospital they visited. Others had
insurance to supplement Medicare that may have eliminated copayments altogether.
43 For a thorough discussion of methods in price/concentration studies, see Weiss (1989).
it has the expected positive coefficient), his key predictor variable is market share. If
one restricts attention to nonprofits, the coefficient on share is negative and larger in
magnitude than that on the Herfindahl. He interprets this as evidence that nonprofits
have lower prices in more concentrated markets. Most studies of price and concentra-
tion do not include market share on the right hand side because a negative coefficient
may capture unmeasured differences in efficiency or firm strategy. For example, a hos-
pital that aggressively pursues managed care contracts will have a higher market share
and lower average price than would a hospital that prefers to admit mainly indemnity
patients, ceteris paribus. Hence, it is difficult to determine if the negative coefficient on
share reflects true market power effects or unmeasured firm differences. Because the
issues raised by Lynk are of fundamental importance to antitrust (see Gaynor and Vogt
(2000)), they have been reassessed by several other authors. Citing methodological and
data problems with Lynk's work, Dranove and Ludwick (1999) and Keeler, Melnick,
and Zwanziger (1999) have presented evidence that, they argue, refute Lynk's findings.
Lynk and Neumann (1999), however, are unconvinced and stand by Lynk's earlier con-
clusions.
The studies cited above aggregate different prices paid by different payers to obtain
a single average price per hospital. In addition, they do not have transactions prices and
must estimate them from data on charges and discounts. Several studies examine actual
transactions prices paid to providers by specific payers.
Two studies examine the prices negotiated between Blue Cross plans and hospitals.
Staten, Umbeck, and Dunkelberg (1988) compared the prices that hospitals bid to the
new Blue Cross of Indiana PPO with the prices they had previously received from Blue
Cross. They computed the discount - the difference between the "before price" and
the bid - and determined that discounts were larger in more competitive markets. Blue
Cross' share of the hospital's total business was not a significant determinant of the
discount. Melnick et al. (1992) examine the actual per diem prices paid to hospitals by
Blue Cross of California. They define hospital-specific markets based on the catchment
area for each hospital and the number of other hospitals serving patients in that area.
Controlling for a variety of determinants of cost, they find that the per diem price is
lower in more competitive markets. A ten point decrease in the Herfindahl index implies
about a one percent decrease in price. Unlike Staten et al., they find that Blue Cross
obtained larger discounts in those hospitals in which it had a larger share of business.
On the other hand, hospitals that provide a large share of Blue Cross' business in its
market are able to obtain higher prices. Melnick et al. obtain much weaker results when
they define the geographic market to be the county. Counties may be poor boundaries
for defining geographic markets, especially in California, where some counties stretch
for scores of miles and others are small parts of larger metropolitan areas.
Brooks, Dor, and Wong (1997) examine the actual prices that several fee-for-service
payers paid to hospitals for appendectomy services. These data are attractive for several
reasons. The product is homogeneous, quality is a relatively unimportant issue (appen-
dectomies are highly routinized procedures), and payers ultimately care more about the
price per admission than the per diem price because low per diem prices may be offset
Ch. 20: The IndustrialOrganization ofHealth Care Markets 1131
by longer lengths of stay. The authors find that greater hospital competition is associated
with lower prices. Interestingly, they also find that greater HMO penetration is associ-
ated with higher prices. This is consistent with Wong (1995), who posits that managed
care penetration may facilitate price discrimination by hospitals. Hospitals assume that
enrollees in the remaining indemnity insurance plans are less price sensitive and raise
prices.
Although recent empirical research on price and concentration in hospital markets is
in the tradition of mainstream industrial organization economics, pricing of hospital and
other health care services under managed care does not follow mainstream models. In a
"mainstream" market, a consumer does business with the one seller that offers the most
attractive combination of price and nonprice attributes. But because consumers value
choice, managed care purchasers usually contract with many providers to form a "net-
work." This implies that rather than competing to be the most desirable seller, providers
compete, first, to be in the network and, only subsequently, to be the most desirable
seller among those included in the network. Consumers at the stage of selecting a plan
examine the networks of competing managed care plans and select the most desirable
network. If they should need medical care during the ensuing year, only then do they
select one of the included providers.
A key distinguishing feature of this demand structure is that consumers select their
network prior to knowing their precise medical needs. Thus, consumers may value the
option to access a particular provider, even if they never exercise it.44 Competition un-
der option demand may differ from traditional competition, so that sellers may have
differential opportunities to exploit this source of market power. For example, physi-
cians who are closely aligned with a hospital may raise their prices, effectively free
riding on the overall demand for the hospital's services [Dranove and White (1996)]. To
date empirical work has neither considered nor exploited this distinction.
6.2. Multinomial choice models
The widespread availability of individual patient level utilization data has sparked an
interest in using multinomial choice empirical methods to analyze choices among pay-
ment plans and providers. These are important because they have the potential for allow-
ing structural models to be estimated of (i) consumers' choices among alternative PPOs,
HMOs, and other payment plans that their employers offer, (ii) consumers' choices
among the set of providers that the PPO includes as preferred providers, and, at least
as importantly, (iii) employers' choices of which among available plans to offer as op-
tions to their employees. Porrell and Adams (1995) provide an excellent review of the
multinomial choice literature. We will only highlight a few of the main issues.
44 The expression "option demand" is first used by Dranove and White (1996) who study the pricing of
specialists and hospitals when patients do not know their precise medical needs at the time they select their
hospital network.
Multinomial choice empirical methods are based on random utility models of con-
sumer choice. The utility that patient i obtains from purchasing services from provider
j may be written
Uij = j + Bij Xij - y Pj + ij, (6.1)
where aj represents a provider-specific utility shifter, Xij are characteristics of the

provider that may be evaluated idiosyncratically by the patient (e.g., location, presence
of specific services, etc.), Pj is the provider's price, Bij and y are parameters and Eij
is an idiosyncratic noise term. If eij is distributed with the extreme value distribution,
then Equation (5) may be estimated using either multinomial logit, conditional logit, or
a mixed model. 45
Conditional logit models have been used to assess how a variety of factors influence
patient choice of hospital. Several studies confirm that consumers prefer care close to
home. 46 Burns and Wholey (1992) show that the distance from the admitting physician's
office to the hospital also reduces the patient's preference for that hospital. A number of
studies find that patients prefer hospitals with lower mortality rates.4 7 Not surprisingly,
given the prevalence of hospitalization insurance, no study to date has found higher pa-
tient preferences for hospitals with lower prices. Hodgkin (1996) uses New Hampshire
data and finds that a cardiac patient's preference for a hospital increases if that hospital
offers cardiac catheterization services, even if that patient does not require catheteriza-
tion. This may represent either option demand or some aspect of quality omitted from
the estimation.
Studies using multinomial logit methods examine how different patient characteris-
tics affect their preferences. Dranove, White and Wu (1993) determine that a patient's
insurance status and location are more important determinants of hospital choice than is
their diagnostic condition. Kim (1990) and Bronstein and Morrisey (1991) both study
hospital choice for obstetrics care. Kim's paper is distinctive as it studies hospital choice
in South Korea (virtually all other papers examine the US). These authors find that older
women, white women (in the US), and women with more resources are more likely to
bypass their local hospital.
Berry (1994) observes that estimates of price elasticities of demand based on multino-
mial choice methods may be biased due to an omitted variable bias. Specifically, price
may reflect unmeasured aspects of quality. This may be especially true for estimates
of hospital choice, where the most widely available measure of quality - risk-adjusted
mortality - provides at best a noisy estimate of only one aspect of quality. Berry offers a
45 Multinomial logit is indicated if the right hand side variables are characteristics of the consumer (e.g.,
race, insurance type, distance from the provider). Conditional logit is indicated if they are characteristics of
the provider (e.g., mortality rate, price, presence of specialized services). A mixed model is indicated if both
kinds of variables appear on the right hand side.
46 These include Garnick et al. (1989); Luft et al. (1990); Burns and Wholey (1992); and Phibbs et al. (1993).
47 For example, Gamick et al. (1989) and Luft et al. (1990).
technique for obtaining unbiased estimates of price elasticities in a multinomial choice

setting, but the technique requires instrumenting for price. A candidate instrument such
as local wage conditions would not be acceptable because competing hospitals would
face the same local labor market conditions. To date, no researcher has adopted Berry's
methods to assess price and quality elasticities facing health care providers.
6.3. Measuring quality
In a managed care world the measurement of quality across providers, either implicitly
by consumers or explicitly by specialized firms and government agencies, is essential.
In the long run, if consumers do not have precise measures of quality, then the strong
incentives for price competition lead to reduced price-cost margins and therefore give
providers incentives to cut back on quality. In theory, quality can fall below optimal
levels. At the present time, there is widespread concern that consumers lack precise
information about the quality of their providers and that this prerequisite for an effi-
cient market is not being adequately approximated. As explained by McGlynn (1997),
there are many problems in measuring quality of care. Quality is multidimensional and
conveys different meanings to different people. Moreover, scientifically valid data are
difficult to come by and may also be difficult to explain to consumers.
Despite these difficulties, a growing movement to measure the quality of health care
plans exists. The National Consortium for Quality Assurance (NCQA) represents large
employers who are aggressively seeking to measure the quality of managed care plans.
NCQA has designed a managed care "quality report card" - the Health Employer Data
Information Set, or HEDIS. HEDIS includes information on the utilization of many pre-
ventive services, patient access, and patient satisfaction. Many employers use HEDIS
information to help them select from among competing plans. As with hospital mortal-
ity, HEDIS provides at best a noisy measure of quality, but may provide both search and
equilibrium benefits.
Concerns about the validity of quality data may be overstated. For example, consider
rankings of hospital quality such as the one described by Pauly et al. (1996). They ex-
amine a number of dimensions of quality such as mortality and morbidity. They use risk
adjusters to control for severity of illness, but acknowledge that such controls are not
perfect. The resulting rankings may therefore be imprecise, and some hospitals ranked
below the median might offer above average quality. In this way, the rankings may be
unfair to hospitals. But if several independent measures of quality coexist in the mar-
ket for quality indicators and the measures are positively correlated with actual quality,
then the rankings will generate both the search and equilibrium benefits described by
Dranove and Satterthwaite (1992).
Nevertheless, improved quality information may be of limited benefit to consumers
for at least two reasons. First, Hibbard and Jewett (1997) find that consumers do not
utilize health care quality data if they do not understand them. For example, survey re-
spondents indicated that patient ratings of health care plan performance were more in-
formative than quantitative data on utilization rates and outcomes. Chernew and Scan-
lon (1998) examine a large sample of the health plan choices of the employees of a
Fortune 500 firm that gives its employees substantial choice among plans and carefully
designed report cards based on the HEDIS information set. They find that the HEDIS
ratings have only small effects on employees' choices and speculate that one important
reason is that the meaning of the HEDIS ratings are surprisingly ambiguous from a con-
sumer's viewpoint. For example, all else being equal, every consumer wants short waits
at his physician's office. Seldom, however, is all else equal; high quality, capable, and
caring physicians are precisely those physicians who tend to have overcrowded sched-
ules. Therefore it is no more clear that a consumer would seek out a plan that has many
doctors who are always on schedule than he would select a restaurant that always has
reservations available for Saturday night. Mennemeyer, Morrisey, and Howard (1997)
carefully analyze the effect that the Health Care Financing Administration's now dis-
continued program of reporting hospital mortality had on hospital admissions. They
find only trivial effects of this systematic information, while they find that the occa-
sional anecdotal report in the media of a single unexpected death had a large effect on
the number of admissions.
Second, providers and managed care plans may focus their efforts on performing
well on the quality scorecards, with the result that quality on unmeasured dimensions
suffers. When the measured outcomes are narrow, such as number of mammograms
performed per month, providers can devote too much of their scarce resources to scoring
well in these narrow areas, and devote inadequate resources to unmeasured aspects of
performance. This possibility seems less likely to be a problem when the measured
outcomes are broad, such as overall mortality or patient satisfaction.
Taking a broader, historical view of the information problem, consumers were inef-
fective in price shopping among primary care physicians even when insurance coverage
was much less inclusive than today. While they may find shopping for quality to be
more salient than shopping for price, it is also harder. Once, with the invention of the
PPOs and selective contracting, payers began shopping for price, then suddenly the
previously hard to confirm predictions of monopolistic competition theory became non-
controversial. Providers responded vigorously with competitive price cuts. If PPOs be-
gin seriously weighting clinical quality decisions into their choices of which providers
to accept as preferred providers, then this history suggests that providers would respond
to the new incentives, provided they are clear, consistent, and substantial. Quality mea-
surement organizations, if they do come of age, may play a role on the quality front
analogous to the role PPOs play on the cost front by providing incentives strong enough
to change behavior.
7. Conclusions
In this chapter we have tried to give a coherent view of the evolution of the industrial
organization of the health care system and to back this view with a selective review of
the empirical literature on the topic. We have adopted Arrow's perspective that the in-
stitutional peculiarities of health care can often be best understood as attempts to solve
the principal-agent problem that the consumer faces with respect to his health care. In-
adequate technical knowledge forces consumers to rely on health professionals, most
notably physicians, for guidance on what services they should purchase. As medical
care technology has exploded in the latter half of the twentieth century, the complexity
and difficulty of this agency problem has multiplied. In response new organizations and
programs - utilization review organizations, the PPS, PPOs - have arisen to maintain
appropriate economic incentives within this agency problem. The empirical literature
is supportive of this interpretation of the industry's evolution: health care providers do
respond to the incentives that the evolving institutional structure provides them. The
challenge, of course, is to continue to devise new institutions and new configurations
of old institutions to both correct the long run, unintended consequences of earlier suc-
cesses and to cope with the continuing change that biomedical science imposes on the
system of care.
References
Adams, E., et al. (1991), "Predicting hospital choice for rural Medicare beneficiaries: The role of severity of
illness", Health Services Research 26(5):583-612.
Adams, E., and G. Wright (1991), "Hospital choice of Medicare beneficiaries in a rural market: Why not the
closest?", Journal of Rural Health 7(1):134-152.
Allen, R., and P. Gertler (1991), "Regulation and the provision of quality to heterogeneous consumers: The
case of prospective pricing of medical services", Journal of Regulatory Economics 3:361-375.
Arrow, K. (1963), "Uncertainty and the welfare economics of medical care", American Economic Review
53(5):941-973. Reprinted in Arrow (1971).
Arrow, K., ed. (1971), Essays in the Theory of Risk Bearing (Markham Publishing Company, Chicago) 177-
219.
Berry, S. (1994), "Estimating discrete-choice models of product differentiation", RAND Journal of Eco-
nomics 25(2):242-262.
Bresnahan, T., and P. Reiss (1991), "Entry and competition in concentrated markets", Journal of Political
Economy 99:977-1009.
Bronstein, J., and M. Morrisey (1991), "Bypassing rural hospitals for obstetrics care", Medical Care 16:87-
118.
Brooks, J., A. Dor and H. Wong (1997), "Hospital-insurer bargaining. An empirical investigation of appen-
dectomy pricing", Journal of Health Economics 16(4):417-434.
Bums, L., and D. Wholey (1992), "The impact of physician characteristics in conditional choice models for
hospital care", Journal of Health Economics 11:43-62.
Chalkley, M., and J. Malcomson (1996), "Competition in NHS quasi-markets", Oxford Review of Economic
Policy 12(4):89-99.
Chamberlin, E. (1933), The Theory of Monopolistic Competition (Harvard University Press, Cambridge).
Chernew, M., and D. Scanlon (1998), "Health plan report cards and insurance choice", Inquiry 35(1):9-22.
Congressional Budget Office (1993), "Responses to uncompensated care and public program controls on
spending: Do hospitals cost-shift?" (Congressional Budget Office, Washington, DC).
Conner, R., R. Feldman and B. Dowd (1997), "The effects of market concentration and horizontal mergers on
hospital costs and prices", mimeo (University of Minnesota).
Corrigan, J. (1995), "How do purchasers develop and use performance measures?", Medical Care 33(1):JS18-
JS24.
Cutler, D. (1995), "The incidence of adverse medical outcomes under prospective payment", Econometrica
63:29-50.
Dixit, A. (1992), "Investment and hysteresis", Journal of Economic Perspectives 6(Winter):107-132.
Donabedian, A. (1980), Explorations in Quality Assessment and Monitoring (Health Administration Press,
Ann Arbor).
Dorfman, R., and P. Steiner (1954), "Optimal advertising and optimal quality", American Economic Review
44:835-836.
Dranove, D. (1987), "Rate setting by diagnosis related groups and hospital specialization", RAND Journal of
Economics 18(3):417-427.
Dranove, D. (1988), "Pricing by nonprofit institutions: The case of hospital cost-shifting", Journal of Health
Economics 7:47-57.
Dranove, D., and R. Ludwick (1999), "Competition and pricing by nonprofit hospitals: A reassessment of
Lynk's analysis", Journal of Health Economics 18:87-98.
Dranove, D., and M. Satterthwaite (1992), "Monopolistic competition when price and quality are imperfectly
observable", RAND Journal of Economics 23(4):518-534.
Dranove, D., M. Shanley and C. Simon (1992), "Is hospital competition wasteful?", RAND Journal of Eco-
nomics 23(2):247-262.
Dranove, D., M. Shanley and W. White (1993), "Price and concentration in hospital markets: The switch from
patient-driven to payer-driven competition", Journal of Law and Economics 34:179-204.
Dranove, D., C. Simon and W. White (1998), "The determinants of managed care penetration", Journal of
Health Economics 17(6):729-745.
Dranove, D., and W. White (1996), "Option demand and the pricing of specialist services", Journal of Eco-
nomics and Management Strategy 5(2):277-306.
Dranove, D., and W. White (1998), "Medicaid-dependent hospitals and their patients: How have they fared?",
Health Services Research 33:163-185.
Dranove, D., W. White and L. Wu (1993), "Segmentation in local hospital markets", Medical Care 31(1):52-
64.
Eaton, C., and R. Lipsey (1989), "Product differentiation", in: R. Schmalensee and R. Willig, eds., Handbook
of Industrial Organization, Vol. 1 (North-Holland, Amsterdam) 723-770.
Ellis, R. (1998), "Creaming, skimping, and dumping: Provider competition on the intensive and extensive
margins", Journal of Health Economics 17(5):537-555.
Ellis, R., and T. McGuire (1996), "Hospital responses to prospective payment: Moral hazard, selection, and
practice-style effects", Journal of Health Economics 15(3):257-278.
Enthoven, A. (1978), "Consumer choice health plan", New England Journal of Medicine 298(12):650-658.
Evans, R. (1974), "Supplier-induced demand", in: M. Perlman, ed., The Economics of Health and Medical
Care (Macmillan, London) 162-173.
Feldstein, M. (1970), "The rising price of physicians' services", Review of Economics and Statistics
52(2):121-133,
Feldstein, P., T. Wickizer and J. Wheeler (1988), "The effects of utilization review programs on health care
use and expenditures", New England Journal of Medicine 318:1310-1314.
Finkler, S. (1981), "Cost effectiveness of regionalization: Further results for heart surgery", Health Services
Research 16:325-333.
Foster, R. (1985), "Cost-shifting under cost reimbursement and prospective payment", Journal of Health Eco-
nomics 4:261-271.
Friedman, B., and M. Pauly (1981), "Cost functions for a service firm with variable quality and stochastic
demand: The case of hospitals", Review of Economics and Statistics 63(4):620-624.
Fuchs, V. (1978), "The supply of surgeons and the demand for operations", Journal of Human Resources
13(Supplement):35-56.
Gal-Or, E. (1996), "Optimal reimbursement rules and malpractice reform", mimeo (University of Pittsburgh).
Garnick, D., et al. (1989), "The sensitivity of conditional choice models for hospital care to estimation tech-
niques", Journal of Health Economics 8:377-397.
Gaynor, M., and W. Vogt (2000), "Antitrust and competition in health care markets", in: A.J. Culyer and J.P.
Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 27.
Gertler, P. (1989), "Subsidies, quality and the regulation of nursing homes", Journal of Public Economics
38:33-52.
Glazer, J., and T. McGuire (1994), "Payer competition and cost shifting in health care", Journal of Economics
and Management Strategy 3(1):71-92.
Grossman, S., and 0. Hart (1986), "The costs and benefits of ownership: A theory of vertical and lateral
integration", Journal of Political Economy 94(4):691-719.
Gruber, J. (1992), "The effect of price shopping in medical markets: Hospital responses to PPOs in Califor-
nia", Working Paper 4190 (National Bureau of Economic Research).
Haas-Wilson, D. (1994), "The relationships between the dimensions of health care quality and price: The case
of eye care", Medical Care 32(2):175-182.
Hadley, J., and J. Feder (1985), "Hospital cost-shifting and care for the uninsured", Health Affairs 4:67-80.
Harris, J. (1977), "The internal organization of hospitals: Some economic implications", Bell Journal of Eco-
nomics 8:467-482.
Hart, O., A. Shleifer and R. Vishny (1997), "The proper scope for government: Theory and an application to
prisons", Quarterly Journal of Economics 112(November): 1127-1162.
Hibbard, J., and J. Jewett (1997), "Will quality report cards help consumers?", Health Affairs 16(3):218-228.
Hodgkin, D. (1996), "Specialized service offerings and patients' choice of hospital: The case of cardiac
catheterization", Journal of Health Economics 15(3):305-332.
Holmstrom, B. (1979), "Moral hazard and observability", Bell Journal of Economics 10:74-91.
Holmstrom, B. (1982), "Moral hazard in teams", Bell Journal of Economics 13:324-340.
Holmstrom, B., and P. Milgrom (1991), "Multitask principal-agent analyses: Incentive contracts, asset own-
ership, and job design", Journal of Law, Economics, and Organizations 7(Special issue):24-52.
Johansen, S. (1986), "Comparison of two prospective rate-setting models: The DRG and PIR models", Health
Keeler, E., G. Melnick and J. Zwanziger (1999), "The changing effects of competition on non-profit and
for-profit hospital pricing behavior", Journal of Health Economics 18:69-86.
Khandker, R., W. Manning and T. Ahmed (1992), "Utilization review savings at the micro level", Medical
Care 30(11):1043-1052.
Kim, B. (1990), "A logit analysis of hospital choice behavior in Chollabukdo province of Korea", Social
Science and Medicine 30:1119-1129.
Peat Marwick, K.P.M.G. (1995), Survey of Employer Sponsored Health Benefits.
Langa, K., and E. Sussman (1993), "The effect of cost-containment policies on rates of coronary revascular-
ization in California", New England Journal of Medicine 329(24):1784-1789.
Luft, H., et al. (1986), "The role of specialized clinical services in competition among hospitals", Inquiry
23:83-94.
Luft, H., S. Hunt and S. Maerki (1987), "The volume-outcome relationship: Practice-makes-perfect or
selective-referral patterns?", Health Services Research 22:157-182.
Luft, H., et al. (1990), "Does quality influence choice of hospital?", Journal of the American Medical Asso-
ciation 263:2899-2906.
Lynk, W. (1995), "Nonprofit hospital mergers and the exercise of market power", Journal of Law and Eco-
nomics 38:437-461.
Lynk, W., and L. Neumann (1999), "Price and profit", Journal of Health Economics 18:99-116.
Ma, A. (1994), "Health care payment systems: Cost and quality incentives", Journal of Economics and Man-
agement Strategy 3(1):93-112.
McClellan, M. (1997) "Hospital reimbursement incentives: An empirical analysis", Journal of Economics and
Management Strategy 6(1):91-128.
McGlynn, E. (1997), "Six challenges to measuring the quality of care", Health Affairs 16(3):7-21.
Melnick, G., et al. (1992), "The effects of market structure and bargaining position on hospital prices", Journal
1138 D. Dranove and MA. Satterthwaite
Mennemeyer, S., M. Morrisey and L. Howard (1997), "Death and reputation. How consumers acted upon
HCFA mortality information", Inquiry 34(2):117-128.
Milgrom, P., and J. Roberts (1992), Economics, Organization and Management (Prentice-Hall, Englewood
Cliffs, NJ).
Morrisey, M. (1994), Cost-Shifting in Health Care (American Enterprise Institute Press, Washington, DC).
Needleman, J. (1994), "Cost-shifting or cost-cutting: Hospital responses to high uncompensated care", mimeo
(Harvard University).
Newhouse, J. (1989), "Do unprofitable patients face access problems?", Health Care Financing Review
11(2):33-44.
Newhouse, J., and the Insurance Experiment Group (1993), Free for All: Lessons from the RAND Health
Newhouse, J., A. Williams, B. Benmett and W. Schwarz (1982), "Does the geographical distribution of physi-
cians reflect market failure", Bell Journal of Economics 13:493-505.
Noether, M. (1988), "Competition among hospitals", Journal of Health Economics 7:259-284.
Nyman, J. (1988), "The effect of competition on nursing home expenditures under prospective payment",
Pauly, M. (1968), "The economics of moral hazard: comment", American Economic Review 58:531-537.
Pauly, M. (1980), Doctors and their Workshops (University of Chicago Press, Chicago).
Pauly, M. (1998), "Managed care, market power, and monopsony", Health Services Research 33(5, Part
II): 1439-1460.
Pauly, M., and M. Redisch (1973), "The not-for-profit hospital as a physicians cooperative", American Eco-
nomic Review 63:87-99.
Pauly, M., and M. Satterthwaite (1981), "The pricing of primary care physicians' services: A test of the role
of consumer information", Bell Journal of Economics 12:488-506.
Pauly, M., et al. (1996), "Measuring hospital outcomes from a buyer's perspective", mimeo (University of
Pennsylvania).
Phibbs, C., et al. (1993), "Choice of hospital for delivery: A comparison of high risk and low risk women",
Health Services Research 28:201-220.
Pope, G. (1989), "Hospital nonprice competition and Medicare reimbursement policy", Journal of Health
Economics 8(2):147-172.
Porrell, F., and K. Adams (1995), "Hospital choice models: A review and assessment of their utility for policy
impact analysis", Medical Care Research and Review 52(2):158-195.
Propper, C. (1998), "Market structure and prices: The responses of hospitals in the UK National Health
Service to competition", Journal of Public Economics 61:307-335.
Propper, C., D. Wilson and N. Sderlund (1998), "The effects of regulation and competition in the NHS
internal market: The case of general practice fundholder prices", Journal of Health Economics 17:645-
674.
Ramsey, J. (1980), "A reevaluation of supply and demand concepts in physician care", Final Report submitted
to the Department of Health and Human Services, Contract No. HRA 79-0068.
Rice, T., et al. (1985), "The state of PPOs: Results from a national survey", Health Affairs 4:25.
Robinson, J. (1988), "Market structure, employment, and skill mix in the hospital industry", Southern Eco-
nomic Journal 55(2):315-325.
Robinson, J., D. Gamick and S. McPhee (1987), "Market and regulatory influences on the availability of
coronary angioplasty and bypass surgery in U.S. hospitals", New England Journal of Medicine 317:85-90.
Robinson, J., and H. Luft (1985), "The impact of hospital market structure on patient volume, average length
of stay, and the cost of care", Journal of Health Economics 4:333-356.
Rogerson, W. (1994), "Choice of treatment intensities by a nonprofit hospital under prospective payment",
Journal of Economics and Management Strategy 3(1):7-52.
Satterthwaite, M. (1979), "Consumer information, equilibrium industry price, and the number of sellers", Bell
Journal of Economics 10(2):483-502.
Satterthwaite, M. (1985), "Competition and equilibrium as a driving force in the health services sector", in:
R. Inman, ed., Managing the Service Economy (Cambridge University Press, Cambridge).
Scheffler, R., S. Sullivan and T. Ho (1991), "The impact of Blue Cross and Blue Shield utilization management
programs: 1980-1988", Inquiry 28:263-275.
Shavell, S. (1979), "Risk sharing and incentives in the principal and agent relationship", Bell Journal of
Economics 10:55-73.
Shleifer, A. (1985), "A theory of yardstick competition", RAND Journal of Economics 16:319-327.
Spence, M. (1975), "Monopoly, quality and regulation", Bell Journal of Economics 6:407-414.
Starr, P (1982), The Social Transformation of American Medicine (Basic Books, New York).
Staten, M., J. Umbeck and W. Dunkelberg (1988), "Market share/market power revisited: A new test for an
old theory", Journal of Health Economics 7(1):73-83.
Thorpe, K. (1988), "Why are urban costs so high? The relative importance of patient source of admission,
teaching, competition, and case-mix", Health Services Research 22(6):821-836.
Weiss, L. (1989), Price and Concentration (MIT Press, Cambridge).
Welch, H., E. Larson and W. Welch (1993), "Could distance be a proxy for severity of illness? A comparison
of hospital costs in distant and local patients", Health Services Research 28(4):441-458.
Whitney, C., P. Milgrom, D. Conrad et al. (1997), "The relationship between price of services, quality of care,
and patient time costs for general dental practice", Health Services Research 31(6):773-790.
Wickizer, T., J. Wheeler and P. Feldstein (1989), "Does utilization review reduce unnecessary hospital care
and contain costs?", Medical Care 27(6):632-647.
Wickizer, T., D. Lessler and K. Travis (1996), "Controlling inpatient psychiatric utilization through managed
care", American Journal of Psychiatry 153(3):339-345.
Wong, H. (1995), "Price discrimination in the physician services market: The non-competitive effects of
HMO and PPO growth" (Center for Intramural Research, Agency for Health Care Policy and Research,
Rockville, Maryland).
Zwanziger, J., and G. Melnick (1988), "The effects of hospital competition and the Medicare PPS program
on hospital cost behavior in California", Journal of Health Economics 7:301-320.
Zuckerman, S. (1987), "Commercial insurers and all-payer regulation", Journal of Health Economics
6(2): 165-187.
Chapter 21
NOT-FOR-PROFIT OWNERSHIP AND HOSPITAL

BEHAVIOR*
FRANK A. SLOAN
Duke University, Durham, NC, USA
Contents
Abstract 1142
Keywords 1143
2. Legal distinctions among ownership forms 1146
3. Why the not-for-profit form is dominant 1148
3.1. Transactions costs and assignment of ownership rights 1148
3.2. Fiduciary relationships and complex output 1148
3.3. Public goods 1150
3.4. Implicit subsidies 1151
3.5. Explicit subsidies 1151
3.6. Inertia 1151
3.7. Cartel theory 1152
3.8. Profitability and for-profit entry 1152
3.9. Which rationale fits the best? 1153
4. Models of not-for-profit hospital behavior 1153
4.1. Objectives 1153
4.2. Comparative statics analysis 1154
4.3. Hospital as a physicians' cooperative 1154
5. Empirical evidence on the effect of ownership on hospital performance 1155
5.1. Cost 1155
5.2. Profitability 1158
5.3. Hospital pricing patterns: Cost-shifting in the face of demand shifts 1158
5.4. Uncompensated care 1160
5.5. Diffusion of technology 1161
5.6. Quality of care 1161
*Several persons provided useful comments on previous drafts of this chapter: Joseph Newhouse, Edward
Norton, Charles Phelps, Carol Propper, and William Vogt. This review benefited from seminars at the Na-
tional Bureau of Economic Research and a work group organized by Burton Weisbrod. Remaining errors and
omissions are my responsibility.
1142 F.A. Sloan
5.7. Hospital capital funds and investment 1163

5.8. Bottom line 1165
6. Ownership conversions 1166
7. Role of competition 1167
8. Conclusions and future research agenda 1168
References 1170
Abstract
The for-profit hospital is in the minority numerically in all developed countries. Al-
though the for-profits' market share has been quite stable for decades, for-profit chains
have grown in share and influence in the United States. By contrast, for-profit chains
have made few inroads in other countries.
The literature on hospital ownership addresses three fundamental questions. First,
why do private not-for-profit organizations dominate the hospital industry? Second, how
do private not-for-profits differ from for-profits in their behavior? Third, is the private
not-for-profit form more efficient in this industry?
The main difference between for-profit and private not-for-profit organizations is in
the distribution of accounting profit. The latter do not distribute profits to equity hold-
ers and enjoy some competitive advantages, including tax exemptions and the ability to
receive private donations. Various reasons for why the not-for-profit form is dominant
are explored. Reasons involve: transactions costs of various ownership forms; fiduciary
relationships between patients and providers; public goods; implicit and explicit subsi-
dies; inertia; cartelization; and lack of profit opportunities. The review concludes that
there is merit in an number of explanations, but no single explanation works perfectly.
Certainly the transactions cost of various ownership forms must provide a partial expla-
nation for observed patterns of hospital ownership.
There is a rich empirical literature on hospital behavior. This chapter discusses com-
parative evidence by hospital ownership form on hospital cost, profitability, pricing and
cost-shifting, uncompensated care, diffusion of technology, quality of care, and hospital
capital funds and investment.
In recent years, changes in hospital ownership have been common in the United
States. Changes have occurred in all directions - from private not-for-profit to for-profit
and the reverse, for example. Comparing hospital behavior before and after the con-
version, the fact that a hospital converted seems to be more important than the type of
ownership change that occurred.
As competition among hospitals increases, differences in behavior among hospitals
with different ownership forms should narrow. Privately owned hospitals in the U.S.
were more alike than different. Private not-for-profit hospitals will have less latitude
than previously to produce outputs they deem to be socially worthy.
The chapter ends with the author's agenda for future research.
Ch. 21: Not-For-ProfitOwnership and Hospital Behavior 1143
Keywords
nonprofit institutions, public enterprises, boundaries of private and public enterprise,

privatization, contracting out, personal and professional services, analysis of health
care markets, government policy, regulation, public health, public goods
JEL classification: L31, L32, L33, L84, Ill, 118, H41

1144 FA. Sloan
1. Introduction
A firm may be viewed as a nexus of contracts [Jensen and Meckling (1976), Alchian
and Demsetz (1972)]. Some of these contracts are with suppliers of inputs, others with
suppliers of financial capital, and still others with purchasers of the firm's products.
Conceptually, the appropriate decision as to ownership rights should be based on that
combination that minimizes transactions costs between the firm and its various contrac-
tors [see, e.g., Hansmann (1996)].
In this general sense, health care is no different with one notable exception that drives
the way health industries are organized, the nature of product purchasers. The product
purchaser is distinctive in two respects. First, consumers are often not as well informed
about their purchasers as are the suppliers. Second, consumers typically pay much less
than marginal cost for health care services with parties other than direct consumers
footing the rest of the bill. In general, the identity of this larger consumer is often not
well-defined; nor is this larger consumer's willingness to pay well revealed with nearly
the precision as in "normal" markets.
For both reasons, it has been widely believed among experts in the health field that the
for-profit organizational form may not minimize transactions cost in the health care sec-
tor. Because of asymmetric information, suppliers of care solely motivated by profit may
be inclined to sell more care than consumers would demand if they possessed the infor-
mation of the seller. Because of externalities in consumption, including but not limited
to inoculations against contagious diseases [Pauly (1971)] and the value of maintain-
ing excess capacity to accommodate stochastic variation in demand [Joskow (1980)],
willingness to pay of others than direct consumers of care, including the care of those
not able to pay, are part of total benefit. To the extent that there is direct payment for
these community benefits, a profit-seeking organization would provide socially optimal
levels of care distributed appropriately. However, absent such subsidies, care for which
marginal revenue falls short of marginal cost should be undersupplied. In the context of
the hospital, such care includes services to the uninsured, excess capacity such as in the
emergency room, various patient educational and social services, care for particular di-
agnostic groups, such as AIDS patients or drug addicts, teaching and research. As Gray
and Schlesinger (1997) explained, the reasons for eschewing the for-profit ownership
form for hospitals come down to trust and community benefits.
The for-profit hospital is clearly in the minority numerically in all developed coun-
tries. In the United States, such hospitals constituted only 12 percent of all nonfed-
eral short-term general hospitals in 1994 [American Hospital Association (1995)]. By
contrast, 60 percent of hospitals were organized as not-for-profit hospitals with the re-
maining 28 percent of hospitals being operated by governments or special government
authorities.
Although the for-profit share of hospitals has been reasonably stable historically, for-
profit chains have grown both numerically and in influence since they first appeared
in the late 1960s. Their growth has not been steady, but rather has been characterized
by cycles of growth [Gray and Schlesinger (1997)]. The growth of for-profit chains in
Ch. 21: Not-For-Profit Ownership and Hospital Behavior 1145
particular has elicited substantial concern among health care experts that consolidation
of hospitals under the aegis of publicly traded corporations will mean higher priced and
lower quality care, lower rates of production of unprofitable outputs, and less accessi-
ble care to those with low ability to pay. The remaining for-profit hospitals are small
individually-owned enterprises. They command less interest because of their small size,
and they are a vanishing breed.
The percentage of for-profit hospitals in other countries in the 1980s and 1990s was
generally as low as in the United States, but the mix of public versus private not-for-
profit varied considerably. Hoffmeyer and McCarthy (1994) presented detailed accounts
of health care systems in various developed countries, including descriptions of the
hospital sectors in each of the countries they described as of the late 1980s and early
1990s.
In Canada, for example, 98 percent of hospitals were public facilities. Private hos-
pitals only had a two percent share. In France, 65 percent were public, 16 percent pri-
vate not-for-profit, and 19 percent for profit. However, the public facilities tended to be
larger. In Germany, each of the three ownership types had about one-third share each.
However, in terms of beds, public facilities were 51 percent, private not-for-profit 35
percent, and for profit, the remainder - 14 percent. In Germany, proprietary hospitals
tended to be owned by physicians. In recent years, local public authorities have increas-
ingly established private corporations as direct owners of hospitals to allow greater
flexibility in wage setting for managers and some clinicians. In the Netherlands, pri-
vate for-profit hospitals were prohibited by law. Most hospitals (about 60 percent) were
private not-for-profit (religious) hospitals. In Switzerland, 46 percent were owned by
public bodies (e.g., cantons and communes). Of the remainder, 32 percent were pri-
vate not-for-profit, often supported in part with public funds, and 22 percent were for
profit. The for-profit hospitals only cared for privately insured, and for those with in-
surance packages combining social insurance with a substantial private care element. In
the United Kingdom, most National Health Service (NHS) hospitals were managed by
public district health authorities. Since 1991, some of the NHS hospitals have become
NHS trusts which means that boards of trustees are outside of district health authority
control. A small for-profit sector emerged starting in the late 1970s, but by 1990-1991,
the number of beds under private control actually declined. Generalizing across coun-
tries on this topic is difficult in that payment systems vary markedly [Glaser (1987)],
physicians less often practice as independent entrepreneurs in hospitals as they do in the
U.S. [Pauly (1987)], and legal structures differ among countries.
In this chapter, the empirical evidence presently is largely limited to the United States.
For one, the literature is far greater for the United States than it is for other countries.
Also, the volume is written in the English language, and most readers (including this
author) will find English language articles accessible, but will have difficulty with most
other languages. However, as discussed more fully below, some of the results should
generalize to other countries.
As a chapter in a Handbook of Health Economics, this review focuses on economic
literature. In contrast to the health care experts, many economists have been suspicious
!146 F.A. Sloan
of the view that private not-for-profit and public organizations perform better, measured
in terms of transactions cost. For one, with the profit incentive attenuated, profits may
be distributed to managers, physicians, and hospital personnel rather than to the poor
and to provision of unprofitable outputs.
The literature on this subject addresses three fundamental questions. First, why do
private not-for-profit arrangements dominate the hospital industry? Second, how do pri-
vate not-for-profits differ from for-profits in their behavior? Third, is the private not-for-
profit form more efficient in this industry? The first two are positive questions, and the
third is a normative question.
The first two sections address the first question. Section 2 discusses legal differences
among the various ownership forms. Section 3 reviews various explanations for the
dominance of the not-for-profit form advanced by others. The next four sections ad-
dress the second question. Section 4 presents a capsule description of models of hospital
behavior which provide a framework for discussion of empirical evidence that follows.
This type of model underlies most of the empirical analysis on comparative performance
discussed in the sections that follow. Section 5 discusses in order, cost, profitability, pric-
ing and cost-shifting, uncompensated care, diffusion of technology, quality of care, and
capital funds and investment. Section 6 presents empirical evidence on hospital owner-
ship conversions. Which types of hospitals convert and what happens when they do? In
Section 7, I link ownership with competition among hospitals. Section 8 concludes the
chapter and addresses the normative question about relative efficiency of hospital orga-
nizational forms and implications for public policy. More specifically, given all of the
empirical evidence, does the private not-for-profit form fit the hospital industry? How
certain can we be of the answer, and what types of research is needed to provide a more
definitive answer?
2. Legal distinctions among ownership forms
The main difference between for-profit and private not-for-profit organizations lies in
the distribution of accounting profit. The latter do not distribute such profit to individual
equity holders but rather in the form of a dividend to its sponsors or to whomever it des-
ignates. In principle, the community, however this is to be defined, is the equity holder.
In practice, there are no rules requiring such organizations to define their "community".
However, in recent years, conversions from public or private not-for-profit status have
increasingly received regulatory scrutiny. Thus, at least for the first time, the definition
of community ownership is beginning to be operationalized as the public policy debate
about the merits of hospital conversions to for-profit status unfolds. Incorporation laws
do not preclude private not-for-profit organizations from paying economic rents to em-
ployees, managers, or others who may exercise control over them, such as physicians.
There are no statutory limitations on the amount of profit such organizations can earn.
Some states in the United States do not restrict the scope of activities of such organiza-
tions. Other states limit their activities [Hansmann (1980), especially p. 839].
Private not-for-profits enjoy some government-conferred advantages, including ex-

emption from corporate income and property taxes, somewhat better access to tax-
exempt bond financing, and eligibility for private donations. Until the end of the 19th
century, U.S. hospitals were almost exclusively donative institutions. The more affluent
received care in physicians' offices and at home [Starr (1982), Rosenberg (1987)]. As
discussed below, donations to hospitals have eroded as all private hospitals and even
some public hospitals have changed from largely charitable institutions to large com-
mercial enterprises. Further, the corporate tax advantage has decreased, and for-profit
hospitals have access to tax exempt bonds in the form of industrial revenue bonds. These
advantages now amount to no more than 3 percent of revenue and probably less [Becker
and Sloan (1985)]. For-profit hospitals have a substantial advantage in being able to raise
equity capital through sale of stock.
Another important legal distinction concerns election of boards of directors. Since
private not-for-profit organizations do not have shareholders, there must be an alterna-
tive mechanism for selecting the board. Typically, such boards are self perpetuating,
even though charters may limit their latitude in appointing members. Whether this is an
important difference in practice is debatable. Hansmann (1996, p. 239) speculated that
managers of not-for-profit firms may not be that different from those in large publicly-
traded investor-owned firms in which individual shareholders exercise no meaningful
control. The real difference may be that the for-profit's organizational goal is clearer.
Whom the not-for-profit boards serve is an important policy issue, particularly in situ-
ations in which board members stand to gain from a transaction, such as the sale of a
private not-for-profit to a for-profit hospital organization [Gray (1997)]. Lack of precise
definition of organizational goals and of identity of the owners open the door to personal
gain.
Charters of private not-for-profit organizations generally contain clauses expressly
forbidding private inurement on the part of management and boards of directors. Such
provisions severely limit or even rule out explicit incentive compensation plans that are
managers have in the for-profit sector. Since there is no stock, there is no possibility of
offering stock options to managers and key employees in private not-for-profit organi-
zations. For private inurement provisions to be effective, they have to be enforced, and it
is unclear who the enforcers are and how much real oversight there really is. Although it
might seem logical that offering managers assets in the enterprise motivates managers,
this is not always true [De Meza and Lockwood (1998), Hadlock (1998)].
Public hospital organizations are similar to private not-for-profits in several respects.
They too are not taxpaying, and they have about the same access to tax-exempt bonds.
However, any dividends are paid through the sponsoring governmental entity. The entity
monitors output and appoints the board of directors of the enterprise. Often, public hos-
pitals have less autonomy than their private counterparts, in particular regarding com-
pensation and employment practices. Also, in contrast to even private not-for-profits,
public hospitals are generally not in a position to refuse to care for patients, unless they
do not have the facilities to care for them.
1148 EA. Sloan
3. Why the not-for-profit form is dominant
3.1. Transactionscosts and assignmentof ownership rights
As Hansmann (1998) explained, ownership of a firm involves two essential rights: (1)
the right to control the firm and (2) the right to appropriate the firm's net earnings.
The defining characteristic of private not-for-profit firms is that individuals who control
the firm do not have the right to that firm's earnings. In this sense, such firms have no
owners, although, increasingly, the legal interpretation is that the community, however
this is to be defined, is the owner.
The efficient assignment of ownership minimizes the sum of (1) the cost of contract-
ing - transactions cost between the firm and parties that deal with it either as customers
or as input suppliers and (2) the cost of ownership - the cost of monitoring the managers
and of risk bearing.
Hansmann identified several potential inefficiencies in contracting that may influence
choice of appropriate organizational form. To cope with market power that arises in the
case of natural monopolies, it may be desirable for the customers to be the owners, for
example, consumer cooperatives for electric utilities. Contracting may be costly when
the firm possesses better information that the parties with which it deals. To deal with
asymmetric information, it may be desirable for consumers to be the owners, as in mu-
tual life insurance companies. In some sectors, there are substantial transactions costs of
switching after the initial investment has been made. Such transactions costs are often
high in franchising. A response to franchisor opportunism is to make the franchisees,
as a group, the owners of the franchisor. Worker-owned firms may protect personal in-
vestments made by employees in firm- or location-specific capital. Of course, shirking
by employees is also a risk; this consideration may affect ownership rights in situations
in which it is more likely. Suppliers of capital face the risk that the firm will be reckless
with their funds and therefore face the risk of not being repaid; thus, there is a case for
assigning monitoring rights to suppliers of capital.
On costs of ownership, some parties are better risk bearers than others depending on
the ability to diversity away risk. Efficient assignment of ownership rights also depends
on the ability of owners to be informed about the operations of the firm, and ability to
see that owners' decisions are actually carried out.
Hansmann explained why societies choose to assign property rights to ownership and
why such patterns differ among industries. Several arguments for the dominance of the
private not-for-profit form in the U.S. hospital industry deal with transactions cost of
ownership: fiduciary relationships and complex output; implicit subsidies; and explicit
subsidies.
3.2. Fiduciary relationshipsand complex output
Kenneth Arrow (1963) explained the dominance of not-for-profits as a response to un-

certainty and incomplete markets for risk in markets for medical care. Because such or-
ganizations are not pure profit-seekers, they would not fully exploit their market power
vis-h-vis a patient who experienced a major health shock. In terms of the above frame-
work, the Arrow argument suggests that private not-for-profit hospitals dominate be-
cause of lower costs of contracting with consumers.
Weisbrod (1988, Chapter 3) extended this concept to not-for-profits more generally.
There are many classes of services for which the seller knows more than the patient,
client, or donor. For such "customers", evaluating and rewarding performance is dif-
ficult. Often, for such services, the customer may never know for certain what would
have happened if the service had not been performed or if it had been purchased from
another seller. "If, in the case of health care, the patient improved, was it because of, or
in spite of, the care received? Would an elderly, infirm resident of a nursing home have
been better off at another home? The buyer of legal services will never learn whether he
or she would have been better off dealing with another attorney; was a lawsuit won (or
lost) because of or in spite of the attorney? When college education is purchased, would
the student have received a "better" education elsewhere? When a charitable contribu-
tion is made to an organization, what would have happened to the needy if the donation
had not been made?" (p. 46).
When output is difficult for the customer to measure, Weisbrod argued, it may not be
efficient to reward easily monitored forms of behavior, but rather allow for rewarding,
or at least be more neutral to rewarding, less easily monitorable behavior.
Although this type of argument seems compelling, it does not explain why the pri-
vate not-for-profit form would be dominant for some but not all types of health care
providers. Many types of health services, such as nursing homes and physicians' of-
fices, are predominantly organized as for-profit enterprises. Nursing home care does not
generally involve complex technology, but given the physical and mental condition of
most patients, the potential for exploitation of patients exists. Physicians refer patients
to hospitals. If the need for trust applies to hospitals, it applies with greater force to
physicians. As is now much more commonly recognized than before, physicians are
often uncertain ex ante about what the effects of their care is likely to have on their
patients.
Complex output gives rise to asymmetric information [see, e.g., Easley and O'Hara
(1983)]. Not-for-profit organizations may be less prone to dupe consumers. Also, to
the extent that consumers cannot monitor output, the profit-maximizing quality level
may be lower than the level that consumers would demand if they were fully informed.
However, again, physicians typically monitor hospital care, and they mostly work for
for-profit firms.
For many of its activities, the government cannot fully know in advance or describe
or stipulate what it wants in advance. Such limited contracting opportunities have been
depicted by incomplete contracts theory developed by Grossman and Hart (1986), Hart
and Moore (1990) and Hart (1995). These articles focused on the idea that ownership
of assets confers on the owner control and bargaining power in situations in which the
contracts do not completely specify what is to be done. As a result, ownership increases
the owner's incentive to invest in certain ways that may reduce cost or otherwise inno-
vate to boost profitability since the owner has the power to reap the returns. By contrast,
t150 FA. Sloan
in a government enterprise, and to a lesser degree in a private not-for-profit enterprise

as well, the innovator must share the benefits of innovation with others (government or
a not-for-profit stakeholder).
A recent article by Hart and co-authors (1997) analyzed conditions under which a
government should provide a service in-house versus contracting out its provision. In
their analysis, the choice was between public provision and public contracting for pro-
vision of the service by a for-profit firm. Although this chapter emphasizes comparisons
between private not-for-profit and for-profit organizations, there are close parallels be-
tween public and private not-for-profit provision.
In their model, the provider can invest in improving quality of service or reducing
cost. If contracts are incomplete, the private (for-profit) provider has a stronger incentive
to engage in both quality improvement and cost reduction than a government employee
who actually provides the service has. The latter cannot generally fully appropriate the
gains. The private contractor's incentive to engage in cost reduction is typically too
strong because he ignores the adverse effect on noncontractible quality. In general, the
larger are the potential adverse consequences of (noncontractible) cost cutting on (non-
contractible) quality, the stronger is the argument for public (in-house) provision.
Although the authors' primary application was to prisons, they also commented
(among several areas) on implications for hospitals. They argued that the gains from
innovation in hospitals are potentially enormous, but so is the potential harm to qual-
ity from cost cutting. On balance, the ability of patients to choose their hospital (not a
choice possible in all countries) strengthens the case for provision by a private contrac-
tor. They asserted that "a crucial difference is the limited ability of consumers to assess
the quality of health care they receive. Consumers often cannot tell whether hospitals
have failed to provide care to save costs, and hence would not so readily change suppli-
ers in response to poor quality" (p. 1157). They concluded that this may be the reason
that the public sector is an important provider of care in most countries.
3.3. Public goods
Weisbrod (1977) hypothesized that private not-for-profits arise to satisfy demands of

particular groups for the production of public goods. These goods have two character-
istics: (1) nonrivalness in consumption in the sense that consumption by Person A does
not affect consumption by Person B; and (2) nonexcludability in the sense that indi-
viduals who do not pay cannot be excluded from consuming the good or service. In
the context of hospitals, the vast majority of services, for example, a hospital day or a
laboratory test, do not satisfy these criteria. A public good might be the good feeling
from knowing that everyone in the community has access to care. Another example is a
public health service, such as care for AIDS may satisfy the properties of public goods,
both on grounds that minimizing disease spread reduces a negative externality and for
reasons of public sympathy toward AIDS victims.
The importance of the public good argument depends on the extent to which private
not-for-profit or public hospitals provide more public goods than their for-profit coun-
terparts. This is an empirical question which I address in detail below.
3.4. Implicit subsidies
A pure profit-seeking organization determines outputs where marginal revenue equals

marginal cost. For some outputs of benefit to communities, private marginal revenue
may fall short of marginal cost at the private firm's optimum. An organization not fully
motivated by profit may provide such socially beneficial services anyway.
This rationale is also not without its deficiencies. First, implicit cross-subsidies may
not be the most efficient way to achieve socially optimal outcomes. One reason for this
may be that incentives of principals and agents are not aligned. The boards and man-
agers may be imperfect agents for the principal, the community, however this is to be
defined. Second, penalties for failing to provide implicit cross subsidies are quite lim-
ited. There is no market for hospital boards in which communities can replace boards
that do not act in their interest. Governments do not link tax-exempt status to the constel-
lation of outputs provided. In fact, sanctions would have no legal standing since there
are no explicit numerical standards specifying the outputs such organizations should
provide.
3.5. Explicit subsidies
A category of private not-for-profit enterprise is the donative charity. Individuals donate

to such organizations with the intent that these organizations make private transfers to
needy or worthy individuals. In effect, the donors purchase services from charities that
in turn are given to third parties, the ultimate recipients of the gift. Here the rationale in-
volves both costs of contracting and ownership, and the solution is to eliminate property
rights to the organization's residual [Hansmann (1998)].
Even if there were tax advantages for donating to for-profit enterprises, few donors
would probably want to do this since they would not want to take the chance of enriching
the firm's shareholders [see, e.g., Hansmann (1980, 1998)]. Also, some charities receive
donations from thousands of individuals. Organizing such a dispersed group would be
costly.
This also cannot be a major determinant of ownership mix of hospitals, except as a
legacy of a time in which hospitals were truly donative institutions. As explained below,
donations to hospitals, particularly as a percentage of capital funds, have decreased
appreciably. Given the lack of importance of donors to hospitals, it is hard to see them
as a major factor in disciplining hospitals.
3.6. Inertia
Because private not-for-profit organizations, including hospitals with this ownership

form, have no well-defined owners, transactions converting such firms to other forms
require consent of the firms' managers, who may have a personal stake in opposing
such conversions. For this reason, the forces of market selection operate slowly, and the
form remains dominant long after the rationale for the form has disappeared [Hansmann
(1998, p. 741)].
1152 EA. Sloan
3.7. Cartel theory
The next reason for the dominance of the not-for-profit form, if true, would be a form
of market failure. The theory was not justified on the basis of minimizing the total cost
of ownership. Rather, the private not-for-profit hospital is dominant because this in the
doctors' collective financial interest.
In the United States, most physicians who treat patients in hospitals are not employees
of these organizations. Yet through their power to admit, they potentially exercise an
important influence over hospitals [Pauly (1980)]. Pauly and Redisch (1973) and Shalit
(1977) have argued that hospitals are operated in the interest of a physician cartel. The
Pauly-Redisch model is explained in some detail below.
Cartelization as a rationale for the dominance of the private not-for-profit hospital has
some appeal for not-for-profits under local control, but is less applicable to major teach-
ing hospitals and chain for-profit hospitals. For teaching hospitals, decision-making is
shared among the physicians, their departments, the medical school, and the parent uni-
versity. Local members of for-profit chains report to the home office, which in turn bears
responsibility conferred upon it by the firms' owners. Opposition to the growth of the
latter on the part of physicians may reflect potential loss in the ability to cartelize when
the hospital is owned by a for-profit chain.
Some would argue that private not-for-profit hospitals are less efficient than for-
profits. The conceptual possibility of inefficiency stems from a lack of a well-defined
residual claimant, which is an issue for many kinds of not-for-profit organizations. Yet,
except for the problem of monitoring individual physician behavior at large hospitals, an
issue Pauly-Redisch discussed, physician cartel members, as residual claimants, would
not want to tolerate inefficiency since this would reduce their private returns.l Thus,
to the extent that hospitals are parts of cartels, it would seem unlikely that physicians
would squander their income on perquisites for managers or other employees. 2
3.8. Profitability andfor-profitentry
Lakdawalla and Philipson (1998) argued that private not-for-profit firms are dominant
in sectors in which profits are zero or even negative. The willingness to forgo the for-
profit form falls as the opportunity cost of doing so rises. Exogenous changes, including
growth of demand subsidies that increase product price, will tend to induce entry by
for-profit firms. Similarly, technical change that leads to increased profit-making po-
tential causes entry of for-profits and lowers the private not-for-profit firms' share. The
authors argued that although they lack residual claimants, private not-for-profits depend
on donors with nonpecuniary motives that provides some market discipline as does de-
pendence of investors in the for-profit sector. This explanation was not based on trans-
actions cost. In their model, purchasers of the firms' products are indifferent as to the
1 For additional discussion of motives for forming not-for-profit hospitals, see Clark (1980).
2 See Feldstein (1971) for a discussion of the "philanthropic wage hypothesis" as applied to hospitals.
firm's ownership. However, owners may differ, that is for-profit from private not-for-
profit, in that some are more willing to supply more than the profit-maximizing output
level than are others.
Lakdawalla-Philipson's model was not specifically designed for the hospital sector;
but as the largest sector dominated by private not-for-profits in the U.S. and some other
countries, one might expect the model's predictions to apply to hospitals as well. The
theory is consistent with the finding, described below, indicating that the for-profit hos-
pital market share was higher in states in which capital cost recovery was better.
3.9. Which rationalefitsthe best?
Several theories of the lack of dominance of the for-profit organization as a seller of

hospital care have been advanced. None fits perfectly. Yet, for the United States, there
is something to most of the arguments. For most other countries, given differences in
medical staff organization, the cartel story does not seem plausible, certainly to a much
lesser extent than in the United States. Although one can pick at particular details, there
must be something to the ownership costs explanations. Why would countries uniformly
eschew the for-profit form when this is the dominant form in most other sectors? In
many countries, for-profit hospitals by law have a very limited role, and sometimes,
they are not permitted to exist at all.
4. Models of not-for-profit hospital behavior
4.1. Objectives
Ideally, a private not-for-profit hospital would make input and output decisions to max-
imize the welfare of the community subject to the constraints it faces. In practice, none
of the economic models of hospital behavior specifies the maximization problem in
this way. Zweifel and Breyer (1997, Chapter 9)], for example, identified four hospi-
tal stakeholders: hospital-based physicians; hospital employees; owners - perhaps the
community in this context; and nonphysician managers. Each stakeholder has distinct
and sometimes conflicting objectives. Although hospital policies realistically reflect the
results of a game, no one has formalized conflict in this way. In the Zweifel-Breyer
model, the hospital objective function contained quality, profit, and labor slack, and
capital slack as arguments. Profit in this one-period model is valued because, for one,
such funds can be used for hospital investment in quality-enhancing technology. Slack
or inefficiency may yield utility because nonphysician staff value a slower pace of work
or having extra supplies on hand. Slack, however, is difficult to distinguish from qual-
ity and adding arguments to the objective function greatly decreases a model's analytic
tractability.
In an early and oft-cited model of hospital behavior, Newhouse (1970) assumed that
not-for-profit hospitals maximize quantity and quality subject to a constraint developed
1154 EA. Sloan
from the loci of points where demand curves intersect average cost curves (points of
zero profit). Eliminating slack as an argument from the hospital utility function simpli-
fies the model in useful ways.
4.2. Comparative statics analysis
The Newhouse-type model is readily formalized [Sloan and Steinwald (1980b, pp. 20-
33)]. Let U = U(X, Y, r) where X = output, Y = quality, and wT= profit. The hospi-
tal faces a downward-sloping demand curve in its product market, P = P(X, Y; M),
where P = price and M = any exogenous demand determinant. The hospital's total
cost function is C = C(X, Y; N), where C = average cost and N = any exogenous cost
determinant, such as an input price.
Then according to the first-order conditions, the hospital sets quantity and quality,
respectively, where the marginal utilities of quantity and quality equal the marginal
disutilities of lost profit resulting from extending output and quality beyond their profit
maximizing levels. Under various restrictive assumptions, in particular, a nonnegative
Uxy (small income effect), an increase in demand (dM > 0) raises both quantity and
quality. Conversely, for an increase in cost (dN > 0), both quantity and quality fall. If
Y is dropped from the utility function, unambiguous results are obtained without the
necessity of signing Uxy. That is, hospital output rises in response to outward shifts in
demand and falls when input prices rise.
Retaining the simplified utility function and specifying a production function for out-
put X = f(K, L), where K = capital and L = labor, one can analyze effects of not-for-
profitness on input use. The effect of extending output beyond the profit-maximizing
point is to raise input use. Yet the familiar textbook condition for the optimal input use
ratio (capital and labor employed at points where the input price ratio equals the ratio of
the marginal products of the inputs) is the same as for the profit-maximizing hospital.
Only under the restriction that UKL > 0 are dL/dw (for wage) and dK/dr (for capital
price) unambiguously negative. The effects of changes in w and r on L and K are un-
ambiguously negative if one assumes a competitive rather than a monopolistic market
for hospital care [Sloan and Steinwald (1980b, pp. 24-26)].
This type of model has also been applied to study hospital responses to rate, revenue,
and input regulations [see Sloan and Steinwald (1980b), and Zweifel and Breyer (1997,
Chapter 9)]. For example, if the government fixes price at P*, then the only way for a
hospital to secure additional patients is to raise Y. Under the assumption of Uyp, > 0,
an increase in the regulated price raises quality. Under revenue regulation, the govern-
mental authority sets the hospital's budget (R*). If Uxy is nonnegative, raising R* raises
quantity and quality.
4.3. Hospital as a physicians' cooperative
In an alternative formulation, the private not-for-profit hospital sets output so as to max-

imize the income per doctor of the doctors on the hospital's medical staff [Pauly and
Redisch (1973)]. Then X, K, L, and the number of members of the hospital's medical
staff are set to maximize income per medical staff member. In this model, which is in
the tradition of cooperative-collective models developed in other contexts, supply re-
sponses to changes in product and factor market conditions can be in reverse directions.
That is, an increase in demand (M above) could boost price, reduce output, and result in
lower hospital medical staff sizes. The model is silent about quality, community benefit,
prestige and so on. They are not motives of the physician cartel.
So long as physician staff members cooperate to maximize income per doctor, hos-
pitals should be technically efficient. However, as medical staff size increases, individ-
ualistic behavior on the part of doctor cartel members becomes increasingly difficult
for the collective organization to monitor, leading to the possibility of some slack. In
the more traditional models with U depending on X and Y, there is no room for slack
either. However, as Y increases beyond the profit-maximizing level, so does cost.
5. Empirical evidence on the effect of ownership on hospital performance
There is a rich empirical literature on hospital behavior which focuses on private not-for-
profit hospitals, often in comparison to other ownership forms. In this review, I describe
methods and findings on hospital cost, profitability, pricing and cost-shifting, uncom-
pensated care, diffusion of technology, quality of care, and hospital capital funds and
investment.
5.1. Cost
Cost may vary by ownership for several reasons: competitive advantages conferred
by governments; community benefits, teaching, and/or research; slack; quality; and
casemix severity. But mainly, the cost test for hospital ownership differences has been
conducted to determine whether for-profit hospitals are more efficient. 3 To state con-
clusively that for-profit hospitals are more efficient, it is necessary to hold these other
factors, as well as input prices and scale, constant. Even if one could successfully do
this, it would be difficult to distinguish whether cost differentials were due to slack or
quality.
To the extent that the residual claimant is not well-defined, economists would expect
that private not-for-profit and public hospitals would be less efficient. As seen above,
under some circumstances the not-for-profit hospital's residual claimant may not be
well-defined. An allegation against the for-profit hospitals, especially the chain hos-
pitals, advanced by some noneconomists, is that they "push" ancillary services, such
3 Here I follow the tradition of economists in relating cost to performance. Gray (1991, pp. 91-96) used the
concept of "payer efficiency". He attached normative significance to the prices payers pay. According to this
view, extracting more from payers is "bad". From another perspective, extracting maximum reimbursements
from payers is a measure of good performance (presumably as long as these practices as legal).
1156 EA. Sloan
as diagnostic procedures [Relman (1980)]. This phenomenon too is difficult to mea-

sure with acceptable accuracy without conducting detailed reviews of medical records
which has not been done. Higher rates of use of ancillary services do not necessar-
ily imply abuse.4 A related criticism of for-profit hospitals is that they charge higher
prices. This is an efficiency issue to the extent that there are deadweight losses associ-
ated with high taxes or premiums for public or private health insurance. Some studies
have used price-based measures for comparing performance of alternative ownership
forms.
In empirical work, an important methodological issue relates to measurement of cost.
Economic cost includes a return on equity which reflects competitively-determined cost
of capital. In fact, most studies have either used accounting cost which excludes such
returns or a price measure which may include an economic rent. Cost may vary sys-
tematically by ownership because of differences in the tax laws. Tax-paying hospitals
have an incentive to use accelerated depreciation, which others do not. Finally, costs
could be related to style and amenities in care not captured by any explanatory vari-
ables.
Overall, the empirical evidence demonstrates no systematic differences in efficiency
between for-profit and not-for-profit hospitals. Perhaps, until recently, the hospital mar-
ket did not provide sufficient discipline for any hospital type to be efficient. Using ac-
counting cost per case and not controlling for other factors, for-profit hospitals have the
same or slightly higher cost. However, using per cost them or revenue-based measures
per unit of output, for-profit hospitals tend to be more costly [Institute of Medicine
(1986)]. This is attributable in part to the shorter patient length of stay in for-profit
hospitals.
The studies have used three distinct in methodologies: paired comparisons; regression
analysis; and frontier production functions. Results from the paired comparisons run
the gamut from for-profits are more costly [see, e.g., Lewin et al. (981), Pattison and
Katz (1983)] to for-profits and not-for-profits are equally or for-profits are less costly
[e.g., Herzlinger and Krasker (1987), Sloan and Vraciu (1983)J. The largest difference
reported by Lewin and co-authors and others was in the much higher use of ancillary
services by for-profit hospitals. Part of the difference in findings reflects adjustment for
taxes which logically should be counted as a community benefit.
Among the early regression studies, after controlling for a number of other factors in-
cluding casemrix and teaching status, Becker and Sloan (1985) found that free-standing
for-profit hospitals (not part of a chain) were more expensive than other free-standing
ownership types in terms of cost per adjusted patient day (adjusted to convert outpa-
tient visits into inpatient days), but they were least costly in terms of cost per adjusted
admissions (adjusted to convert outpatient visits into inpatient admissions). The chain
for-profits that had been part of a chain for four years or more were six percent higher
on cost per adjusted admission than were free-standing not-for-profit hospitals, but the
4 Unfortunately, in some writing, "use" equals "abuse".

coefficient on the binary variable for this group was statistically insignificant at con-
ventional levels. Granneman et al. (1986), who focused on assessing scale and scope
economies, included binary variables for hospital ownership. They found that hospi-
tals operated by local or state governments had costs about eight percent lower and
for for-profit hospitals had costs that were 15 percent higher than private not-for-profit
hospitals.
Breyer et al. (1988), using a single cross section of 614 German hospitals, regressed
cost per case on ownership, teaching status, various casemix indicators, and measures
of output and capacity. Public hospitals had higher cost than for-profit hospitals. The
difference between private not-for-profit and for-profit hospitals was not statistically
significant at conventional levels, although the sign on the coefficient for private not-
for-profit was negative and larger than its standard error.
Among types of cost, for-profit hospitals tend to use less labor but more nonlabor
inputs [Sloan and Steinwald (1980b)]. There is no difference in wage rates paid by hos-
pitals of different ownership types, holding other factors constant [Sloan and Steinwald
(1980a)].
Linear programming techniques, applied to individual hospital departments and to
hospitals as a whole can be used to estimate a frontier production function. Inefficiency
is then evaluated in terms of a hospital's divergence from hospitals at or near the fron-
tier. Wilson and Jadlow (1982) applied this technique to nuclear medicine departments
of hospitals, finding that government hospitals were less efficient on average and for-
profit hospitals more efficient than the omitted reference group, private not-for-profit
hospitals. Focusing on one department probably reduces output heterogeneity, but these
authors probably included too few covariates, which may be a source of serious omitted-
variables bias.
Using stochastic frontier regression analysis of 219 community hospitals in New
York, Vitaliano and Toren (1996) found no effect of hospital ownership on hospital
efficiency. With a much larger sample and using the same technique, Zuckerman et al.
(1994) found that for-profit hospitals were more likely to be inefficient than private not-
for-profit hospitals. There were no statistically significant differences in inefficiency
between public and private not-for-profits. Finally, Koop and co-authors (1997) used a
Bayesian approach for estimating a stochastic cost frontier production function which
they applied to a panel of U.S. hospitals. They found government-run hospitals to be the
most efficient, followed by not-for-profit, and for-profit hospitals in last place.
In the same issue in which the Zuckerman et al. article appeared, Newhouse (1994),
commenting on articles in the issue that used the frontier technique pointed to a weak-
ness of the methodology, namely what is measured as inefficiency may be an amenity
or unmeasured aspect of quality that consumers value. Are wider seats in first-class
waste? Probably not. Consumers pay more for comfort on airlines because they value
this. But cost per passenger-mile would be higher in first class. Of course, the same
critique applies to any of the cost techniques economists have used since the analysis
cannot capture quality differences of importance to consumers.
1158 EA. Sloan
5.2. Profitability
If the term is taken at face value, private not-for-profit hospitals should earn a zero
profit. But this applies to economic profit, and measurement has been done on account-
ing profit. Private not-for profit hospitals in the United States have consistently earned
accounting profits. Judged in terms of such profit as a percent of revenue, in three years,
1988-1990, such hospitals were even more profitable than for-profit hospitals [U.S.
Medicare Payment Advisory Commission (1998, p. 72)].
Hospital profits tend to be more variable for for-profit hospitals. Hoerger (1991) stud-
ied profit variability. He tested two hypotheses. First, various models of private not-for-
profit hospital behavior specify that utility is maximized subject to a zero profit con-
straint for the hospital. If so, profits of such hospitals should be less variable. Second,
the profits of not-for-profit hospitals should be less responsive to changes in such ex-
ogenous factors as payment levels of government payers. This would occur under con-
ditions of cost-shifting, a phenomenon to be discussed next. He found empirical support
for both hypotheses.
5.3. Hospital pricingpatterns: Cost-shifting in theface of demand shifts
The notion that hospitals increase price to private payers when the government reduces
the price it pays is widely accepted among health care experts outside the economics
profession. The term given this type of behavior is "cost shifting" [Ginsburg and Sloan
(1984), Dranove (1988), Morrisey (1994)]. This type of behavior is to be distinguished
from price discrimination that occurs when markets can be segregated by the seller and
demand elasticities differ across markets.
Conditions for cost-shifting to occur are (1) the hospital has market power in its
product market (private sector) and (2) such power was not fully exploited before the
government reduced the price it paid hospitals. That is, the hospital did not set price
at the profit-maximizing price before the government decreased price. The intuition is
that hospitals raise price toward the profit-maximizing price to make up for the shortfall
caused by the fall in the price paid for hospital care by government. One would expect
nonprofit maximizing behavior from private not-for-profit but not for-profit hospitals.
More formally, following Dranove (1988), let the private not-for-profit hospital's util-
ity function be U = U[ri (Pi, C) + rj (Pj, C), Xi(Pi)Xj(Pj)]. Market i is the private
market in which the hospitals sets price PI. Market j is the public market. Quantity xj in
the public market is exogenous as is Pj. C is cost. The first-order condition requires that
optimal P be set where the marginal utility of profit from a change in price equals the
marginal utility from loss of output to the private market as Pi changes. If the marginal
utility of the private market output is zero, which holds in the profit-maximizing case,
Pi is set to maximize profit. Totally differentiating the first-order condition and impos-
ing restrictions that (1) ri > 0 and (2) the marginal utility of profit does not decrease
with additional private output, then dPi/dPj < 0, which is cost-shifting.
Ch. 21: Not-For-ProfitOwnership and HospitalBehavior 1159
In his empirical analysis, Dranove found that a one-dollar decrease in hospital profits
from government sources per private admission led to fifty-one-cent increase in price
per private admission. This implies that about half of the revenue loss was recovered.
Morrisey's (1994) book on hospital cost shifting is a very comprehensive review of
theoretical arguments and of empirical analysis of the issue up to the date of the book's
publication. As he noted, the empirical evidence in support of cost-shifting is largely
based on data before 1983 when the Medicare Prospective Payment System (PPS) was
enacted. Other evidence he cited is inconsistent with cost shifting. For example, in one
study of Blue Cross, it was found that Blue Cross saved money after Medicare PPS
was implemented, presumably because the Medicare incentive to boost efficiency was
also applied to patients with different payment sources. More recent studies showed
that reductions in payment levels caused some hospitals to reduce or eliminate care
for Medicaid patients. Studies of California hospitals, a leading state in introducing
competition into the hospital sector, found that hospitals lowered prices in response to
growth of managed care. These price cuts were larger when the hospital faced more
competition. In Morrisey's words, "None of these responses supports the idea that cost
shifting works" (p. 85). He reasoned that it far more likely that if the indigent were
to disappear, hospitals would spend their surpluses on other things, such as research,
new technology, nicer offices for physicians and/or hospital administrators. Only if the
hospital "likes" paying patients would prices charged such individuals fall. As he noted,
"this seems like a thin reed on which to hang much public policy" (p. 86).
Other empirical evidence on private not-for-profit hospitals not setting price to max-
imize profits is from Lynk (1995). In this study, Lynk focused on the difference hori-
zontal mergers of hospitals make on price, depending on the ownership of the merging
organizations. When a merger occurs, price may fall i response to efficiency gains or
may rise to the extent that the merged entity is more able (has more market power) and
willing (wants to exploit this added market power) to raise price. Lynk argued that a pri-
vate not-for-profit hospital may be devoted to its product which is hospital services than
a government organization which may seek to earn a profit from health care to subsidize
a nonhealth good. For-profit hospitals would want to set price to maximize the return
to their owners. Empirically, using data from California, Lynk showed that mergers by
private not-for-profit hospitals lowered price on average. By contrast, such mergers had
the opposite effects for mergers involving the other two merger types.
The data in both studies were for the 1980s. To the extent that hospital markets have
become more competitive since then, cost-shifting by private not-for-profit hospitals
should be decreasing because of diminished ability to raise price, even in markets in
which sellers' output is concentrated. This is because group purchasers of hospital care
have become more aggressive. In fact, a study by Keeler et al. (1999), using more recent
data for California (through 1994), found that private not-for-profit hospitals located in
market areas with higher hospital concentration charged higher prices, and the effect of
concentration on price has increased over time. This is the "normal" relationship one
would expect.
1160 EA. Sloan
The growth of managed care has reduced hospitals' ability to set price. Brooks and
co-authors (1997) studied variation in the difference in price obtained and the minimum
price that the hospital would accept to provide a privately-insured episode of care (for
appendectomy). A large difference was interpreted as high bargaining power by hos-
pitals. As one of many findings, the authors reported "rather paradoxically, for-profit
hospitals had significantly less bargaining power than either public or voluntary hospi-
tals" (p. 431). I am puzzled as well, but mention results such as this as a challenge for
future research. One possibility is that lower markups at for-profits may reflect more
elastic demand for their product.
5.4. Uncompensated care
One public policy concern is that an increase in competition will place the poor and the
uninsured at an even greater disadvantage than previously [see, e.g., Mann et al. (1995)].
However, differences between private not-for-profit and for-profit hospitals in provision
of uncompensated care were very minor, even before the 1990s [Sloan et al. (1986b)].
Uncompensated care is typically defined as the sum of charity care, services for which
the hospital did not expect to be compensated at the time of the patient stay, and bad
debt, services for which the hospital anticipated compensation but did not receive it.
Uncompensated care data for 1994 showed that, for private not-for-profits, uncom-
pensated care was 4.5 percent of revenue. The corresponding percentage for for-profits
was 4.0 percent [U.S. Prospective Payment Assessment Commission (1996, p. 84)].
Differences in provision of uncompensated care by ownership for psychiatric hospitals
appear to be much larger than this [Schlesinger et al. (1997)].
There is evidence that ownership mix in a community affects hospitals' provision
of uncompensated care. In recent years, hospitals in markets with a higher for-profit
market share provided less uncompensated care and were more likely to adopt policies
to discourage admission of uninsured patients [Frank et al. (1990)].
Morrisey et al. (1996) compared uncompensated care provided by private not-for-
profit hospitals in California in 1988 and 1991 to the tax subsidies such hospitals re-
ceived. For all but 20 percent of the hospitals, uncompensated care exceeded the tax
subsidies. Yet this comparison is imperfect since the potential community benefit in-
volves more than provision of uncompensated care. Several recent unpublished studies
have concluded that for-profit hospitals provide less uncompensated care, but at least
some of these studies were conducted to support advocacy positions of private not-
for-profits [Kuttner (1996a)]. Young and co-authors (1997) examined changes in the
provision of uncompensated care by California hospitals that converted from private
not-for-profit to for-profit status and found that conversion did not uniformly reduce
provision of such care. With a national sample of data from 1981, Norton and Staiger
(1994) found that when a for-profit and a private not-for-profit hospital are located in the
same area, they serve an equivalent number of uninsured persons. However, for-profit
hospitals more frequently locate in areas where better-insured persons live.
Frank and Salkever (1991) modeled provision of uncompensated care by private not-
for-profit hospitals. They considered two alternative models: (1) a pure altruism model;
and (2) an impure altruism model. In the first, hospitals have profit and unmet need in
the community as arguments in the hospital's utility function. In this model, increases in
provision of uncompensated care by other hospitals in the market "crowd out" provision
of such care by the hospital in question (except if the income or endowment effect is very
strong). In the alternative, the hospital competes for public goodwill by providing free
care. Here crowding out is less likely since hospitals compete for patients who do not
pay for their reputational value. Using data from Maryland hospitals for 1980-1984, the
authors obtained mixed results for both pure and impure altruism models. Earlier work
by Thorpe and Phelps (1991) found some evidence of crowding out. More specifically,
private not-for-profit hospitals provided less uncompensated care when public hospitals
were present in their markets.
In Gruber's (1994) model, the private not-for-profit hospital maximizes utility which
depends on profit and uncompensated care subject to a zero profit constraint. Under
some restrictive assumptions, his comparative statics analysis showed that uncompen-
sated care should fall as price consumption increases. Greater competition will affect the
hospital by increasing the elasticity of its demand curve. Price competition in California
increased markedly after 1982. Gruber's empirical analysis of California data for 1984-
1988 showed relative decreases in hospital profitability in those areas with the least
output concentration, that is, in the most competitive hospital markets. Likewise, and
perhaps as a result, hospitals in such markets showed the greatest decreases in amounts
of uncompensated care provided. Gruber also found that the price of hospital care rose
less in less concentrated markets, a result that at first glance seems inconsistent with
Lynk's (1995) study. However, Lynk distinguished the effect of output concentration by
hospital ownership. Gruber considered the effects of concentration on price irrespective
of ownership.
5.5. Diffusion of technology
A few empirical studies have investigated diffusion of technology in hospitals. How-

ever, there are few well-controlled studies, and most are now old. A large variety of
innovations have been examined, including process and more often product innovations
[Romeo et al. (1983), Russell (1979), Sloan et al. (1986a), Sloan et al. (1997)]. With rare
exceptions, there were no statistically significant differences in adoption between pri-
vate not-for-profit and for-profit hospitals. Schlesinger et al. (1997) studied technology
adoption patterns among psychiatric hospitals. He found that increased competition had
a greater impact on the for-profit psychiatric specialty hospitals' propensity to adopt in-
novative services (measured by an index of four services-treatment of eating disorders,
treatment of substance abuse, geriatric services, and adolescent treatment) than it did
for private not-for-profit psychiatric specialty hospitals.
5.6. Quality of care
In some hospitals models, quality (Y) is included as an argument for the reason that
rather than distribute profit to stockholders, not-for-profit hospitals may use profit to
1162 EA. Sloan
cross-subsidize a higher than profit-maximizing level of quality. Although economists

include quality in their conceptual work, noneconomists have conducted most of the
empirical work. This review considers some of the more important studies on the ef-
fect of ownership on hospital quality. Several different measures are used to measure
quality: structure or inputs to care; process or measures of how care is delivered; and
outcomes. Most private and public regulation of quality to date has focused on struc-
ture and process. The lower staffing levels in for-profit hospitals, especially lower nurse
staffing, discussed above, are structural indicators of lower quality in such hospitals.
However, the relationship between hospital staffing and other measures of quality in-
cluding outcomes has not yet been established [Institute of Medicine (1996)]. On other
structural measures, such as the percent of hospitals with national accreditation and per-
cent of hospitals with cardiac and intensive care, ownership does not make a difference
[Herzlinger and Krasker (1987)].
The most rigorous and extensive study of large-scale empirical study of quality pub-
lished to date that permits comparisons of quality by hospital ownership is by Keeler
and co-authors (1992). They used two process measures of quality based on reviews of
14,000 medical records for five diseases in five states. One of this "explicit process"
gauged the extent to which the charts showed that specific diagnostic and therapeutic
procedures were performed competently. Rather than focus on particular elements of
care as explicit process did, a second process measure, "implicit process", measured the
care process overall. For example, one of their implicit process questions to physician
reviewers was: "Based on what you now know about this case, would you send your
mother to this hospital?" In addition, they gauged quality on an outcome measure - the
difference between actual mortality and the rate that would be expected based on the
patient's characteristics.
They found no difference in quality between private not-for-profit and for-profit hos-
pitals on two indicators, excess mortality and explicit process; public hospitals fared
worse on both criteria. However, on a third measure, implicit process, there was a statis-
tically significant difference between quality of care of private not-for-profit hospitals
and the other two ownership types, indicating higher quality levels for the for-profits.
The authors appear to have been more persuaded by the results on the first two indi-
cators, stating that "nonprofit and for-profit hospitals provide similar quality overall"
(p. 1712).
In their national study of 981 hospitals in 1983-1984, Shortell and Hughes (1988)
found no difference in quality measured in terms of mortality by ownership. However,
using fewer covariates, Hartz et al. (1989) did find that mortality was higher in for-profit
than in private not-for-profit hospitals.
Sloan and co-authors (1998a, 1998b) and Sloan and Taylor (1999) examined out-
comes of care of elderly persons hospitalized for one of four conditions: hip fracture;
stroke; coronary heart disease; and congestive heart failure. They analyzed the first ad-
mission for these conditions since patients with a first unanticipated major health shock
are less likely to shop among hospitals. Their outcome measures were survival, func-
tional status, cognitive status, and living arrangements (probability of living in a nursing
Ch. 21: Not-For-Profit Ownership andHospital Behavior 1163
home). Although, on some measures, patients admitted to major teaching hospitals did
better, a result consistent with Keeler et al. (1992), there were no statistically signifi-
cant differences in outcomes between non-teaching private not-for-profit and for-profit
hospitals. On some measures, elderly patients admitted to non-teaching government
hospitals had worse outcomes, holding a large number of other factors constant.
Mark's (1996) study of psychiatric hospitals assessed process measures of quality -
violations and complaints reported to the Medicare and Medicaid programs. Comparing
means, she found some differences in quality that were unfavorable to for-profits. She
also estimated separate quality equations for profit and private not-for-profit hospitals.
The signs on some of the coefficients differed between the profit and not-for-profit mod-
els that she interpreted as evidence that the two ownership types have different objective
functions. For for-profits, market concentration was negatively related to the number of
violations. As competition increased, violations increased. But for not-for-profits, the
relationship was positive. In more competitive markets, quality improved. From these
results, Mark inferred that not-for-profit hospitals compete by offering a higher level of
quality. By contrast, for-profit hospitals respond to competition by lowering quality.
Schlesinger et al.'s (1997) study of psychiatric hospitals found that private not-for-
profit hospitals provided greater access than for-profits when competition is limited,
measured by uncompensated care. With increased competition, ownership-related dif-
ferences in uncompensated care were reduced, a consequence worth noting as competi-
tion among hospitals increases.
5.7. Hospitalcapitalfunds and investment
Ownership has an important bearing on sources of capital funds. Public and private not-
for-profit hospitals are precluded from obtaining funds from the sale of equity. All eq-
uity comes from either philanthropic contributions or internally in the form of retained
earnings. All hospitals have access to debt capital, but public and private not-for-profit
hospitals have better access to revenue from sale of tax exempt bonds and to direct
subsidies from private donors and from governments.
Two studies have examined capital structure of hospitals with emphasis on hospital
ownership. In Wedig et al.'s (1988) analysis, for-profit hospitals solve a one period prob-
lem of value maximization. Not-for-profit hospitals maximize profits to serve a nonpe-
cuniary objective, such as providing a higher level of service than the profit maximizing
level. The former pays taxes on debt and equity income and benefits from non-debt tax
shields. The tax advantage of debt is a primary motivation for issuing debt. The private
not-for-profit hospital pays no tax. Thus, other reasons, such as the extension of cash
flow for nonpecuniary objectives must be a reason for issuing debt. Bankruptcy risk is
a negative consideration in issuing debt for both ownership types. Both types receive
payments from cost- and charge-based payers. A cost-based payer reimburses the hos-
pital for "its" share of interest and depreciation expense. A charge payer bases payment
to the hospital in part of the price billed to a covered patient.
The authors' theoretical analysis showed that when cost-based revenue as a share
of total hospital revenue increases, private not-for-profit hospitals will issue more debt.
1164 FA. Sloan
Because of more complex offsetting effects, including the effects of cost-based payment
on the value of debt as a tax deduction, the direction of effect from an increase in the
cost-based revenue share on capital structure could not be determined theoretically for
for-profit hospitals.
Empirically, the study showed that the effect of the share of revenue from cost-based
sources on the hospital's debt to asset ratio was positive and higher for for-profit than
for not-for-profit hospitals. Other than the effect operating through the cost-based payer
share, there were no differences in hospital capital structure according to ownership.
Perhaps, as my discussions with persons in the hospital industry suggested was true
at the time, for-profit hospitals were more adept at maximizing reimbursement. The
payment system rewarded maximizing reimbursement rather than efficiency.
Although private not-for-profit hospitals do not benefit from using debt to shield prof-
its from the corporate income tax, they do benefit from the personal income tax reduc-
tions reflected in the yields they pay on municipal bonds. In the second article on capital
structure, Wedig et al. (1996) found that the not-for-profit hospital selects taxable and
tax exempt debt levels jointly with fixed investment to maximize end-of-period returns
which in turn are used to fund its nonpecuniary objectives. The hospital can earn an
indirect arbitrage by issuing low-yield tax exempt debt rather than spending its accumu-
lated cash reserves. Offsetting this incentive to issue debt are several factors causing the
marginal cost of borrowing to increase with higher leverage, such as agency costs and
bankruptcy risk. The balance of these factors leads to an internal optimum consisting of
an optimal mix of equity and debt. Also, use of debt may be constrained by a "project
financing" rule that requires that fixed investment exceed the value of tax-exempt debt
flows. Consequently, the hospital may be induced to undertake low-yield investments to
increase access, to tax-exempt debt. Conversely, the number of high-yield investments
may exceed debt capacity. In such cases the constraint against issuing equity may cause
the hospital to issue tax-exempt debt beyond the optimal leverage point. Empirically, the
authors found that private not-for-profit hospitals behave as if they have target levels of
tax-exempt debt. Debt targeting is constrained by availability of capital projects. Excess
debt capacity increases investment.
Payment policies of insurers have not only influenced hospital capital structure, but
also have affected hospital ownership composition. Wedig et al. (1989) assessed effects
of variations in cost recovery policies of insurers on hospital location. They found that
for-profit hospital market shares were higher in states in which payment policies were
more generous and conversely for private not-for-profit hospitals. That is, the latter sat-
isfied patient demand by locating in those areas that for profit hospitals were not willing
to serve,
In principle, having access to private donations could be an advantage for private
not-for-profit hospitals. But in recent years, philanthropy has become an unimportant
source of revenue for private not-for-profit hospitals. In 1983, such hospitals derived
only 0.4 percent of revenue from this source. For state and local government hospitals,
the corresponding percentage was 1.0 [Institute of Medicine (1986, p. 100)]. In earlier
years, this was a much more important revenue source of revenue [Sloan et al. (1990)].
One reason for the decline in private giving to hospitals is the growth of health in-
surance coverage. However, as Sloan and colleagues (1990) demonstrated, the direction
of effect of increased insurance coverage on donations to hospitals cannot be deduced.
On the one hand, increased coverage crowds out donations by providing an alternative
revenue source. However, insurance also raises patient willingness to pay for hospital
output, and high output is plausibly valued by donors. Thus, increased coverage al-
lows a dollar's worth of donations to buy more hospital output. In effect, the insured
patient provides matching funds to the donor's gift. Lump-sum subsidies by govern-
ments should crowd out donations, but matching grants may increase them. Using both
time series and cross section analysis, Sloan and co-authors found strong evidence that
increased health insurance coverage crowded out private donations to hospitals.
Given that the only source of private external equity to private not-for-profit organi-
zations such as hospitals is philanthropy, it is difficult to know what the cost of such
capital is. Several authors have argued that such cost equals the return on equally risky
tax-free-securities [Conrad (1984), Silvers and Kauer (1986), Herzlinger and Krasker
(1987)]. But determining "equally risky" is not a straightforward task. One approach
used by Sloan and colleagues (1998a, 1998b) was to take the beta (from the capital
asset pricing model) for publicly-traded for-profit hospital companies, compute the un-
levered beta, and then compute a levered beta for private not-for-profit hospitals, using
their debt-to-equity ratio, and finally adjust for the fact that not-for-profits do not pay
taxes. They found that the cost of equity capital was 1.1 to 2.5 percentage points lower
for not-for-profit than for for-profit hospitals. The weighted cost of capital (cost of debt
and equity capital weighted in proportion to their shares of total capital) was from 0.5
to 1.0 percentage points lower. This is not a very large competitive advantage for not-
for-profits. Morrisey et al. (1996), using a sample of California hospitals, calculated
that interest rate subsidies per not-for-profit hospital in that state amounts to about one-
fourth of the total tax subsidy such hospitals received. Corporate income and property
tax subsidies accounted for the rest.
There is very limited published empirical research on hospital investment [see Wedig
et al. (1989), Calem and Rizzo (1995)], and this research did not explicitly compare in-
vestment behavior by hospital ownership type. An unpublished study by Hoerger (1995)
investigated hospital investment decisions during the 1980s in California, Florida, and
Tennessee. The most noteworthy finding on ownership was that cash flow, the sum of ac-
cumulated hospital accounting profits and accumulated depreciation, had a much greater
positive effect on investment by not-for-profits than for for-profit hospitals. This result
is plausible because the former cannot issue stock.
5.8. Bottom line
Given the vast amount of evidence on hospital performance, one can find some contra-
dictory results on about any of the dimensions. Overall, one is struck by the similarity
between private not-for-profit and for-profit performance, except in areas, such as cap-
ital structure where there must be differences for institutional reasons. In reviewing the
1166 EA. Sloan
literature, I have emphasized comparison between the types of private hospitals rather
than between public and private hospitals. Public hospitals clearly have a different ori-
entation, for example, in treating much higher proportions of patients without insurance.
Some of the results on quality differences between public and private hospitals - those
showing the public facilities to be worse - are troubling to the extent that one expects
single-tier care. Quality, even of the sort that only experts observe, does not appear to
be a sound reason for public provision of hospital care in the United States. This result,
as many of the others, such as on uncompensated care, is likely to be country-specific.
6. Ownership conversions
Changes in ownership form have been fairly common in the United States. For exam-
ple, between 1990 and 1993, out of 6,015 hospitals, 183 changed ownership [Needle-
man et al. (1997)]. Of the 183, only 37 were conversions from private not-for-profit to
for-profit status. 42 were conversions in the reverse direction. Still more, 52 hospitals,
were conversions from public to private not-for-profit status, but 32 hospitals changed
from being private not-for-profits to being government-run. Various reasons have been
advanced for converting: to avert closure and continue the hospital's mission; to ob-
tain access to capital; to improve efficiency; to preserve or expand market share; and
to reduce regulatory constraints [Duke University (1998, p. 5)]. The rationale to reduce
regulatory constraints is often a reason for converting from public status. A desire for
efficiency gains and and/or access to equity capital may be a reason for seeking a change
to for-profit status. The reasons for simultaneous changes of ownership in all directions
is not at all well understood and is not explained by recent theoretical work on hospitals
Isee, in particular, Lakdawalla and Philipson (1998)]. Perhaps in some cases, conversion
was coincident with other changes, such as a merger.
Sloan and co-authors (2000) analyzed effects of hospital conversions on various out-
comes. They found that private not-for-profit hospitals that converted to for-profit status
experienced an increase in their profit rate, but so did hospitals that converted in the
opposite direction. Hospitals that converted from not-for-profit to for-profit status in-
creased the uncompensated care they provided, but this also occurred for some other
types of ownership conversions. The pattern of adoption of specific services was mixed,
even for services that have important public health benefits. The empirical findings were
based on a sample of hospitals in three states. Further empirical analysis is needed to de-
termine whether these results generalize and to better document the stylized facts about
changes in hospital behavior following conversions.
An important issue in conversions is what the transactions price should be. There is
widespread concern that sellers of hospitals obtain too little for their assets. An evalua-
tion of rates of return on hospital purchases by Sloan and co-authors (2000) found that
private rates of return to acquirers were mostly at competitive levels, especially if the
acquirer was a for-profit firm. However, whatever the private value is, the social value
may be different to the extent that some services are provided at a price below marginal
cost. Goddeeris and Weisbrod (1998) noted that to approximate social value for the for-
profit sector in a conversion from public or private not-for-profit, one should add private
bids, the present value of future tax liabilities, and adjust for the social value of oblig-
atory community services. As they indicated, estimating the marginal social value of
unprofitable social services is far from a trivial task.
7. Role of competition
Competition in the hospital sector is complex for several reasons. First, at least for some
types of patients, price is set by a government agency. Second, patient out-of-pocket
payments for hospital care are minimal. Third, in recent years, public and private in-
surance plans have increasingly selected hospitals on behalf of their insured [see, e.g.,
Robinson and Phibbs (1989), Vistnes (1994)]. To the extent that price is set administra-
tively by an outside governmental authority and/or the cost to the patient of selecting
a more expensive hospital is minimal, competition among hospitals must take place
on a nonprice basis. Bilateral negotiations between insurers and hospitals are likely to
involve both price and nonprice issues.
Pope (1989) analyzed nonprice competition among hospitals when the government
sets a fixed price per case and patients pay nothing for hospital care out-of-pocket. In
his model, hospitals value profits and slack. Profit is a proxy for objectives funded with
excess of revenues over costs. Hospitals compete for patients by offering higher quality,
which given the fixed price, reduces profit. There is a critical difference depending on
whether hospital profits are positive or zero. When profits are positive, increased com-
petition among hospitals (more competitors in the market area) raises expenditures on
quality, thereby raising costs, but does not affect efficiency (i.e., slack). When profits
are driven to zero, increased competition continues to raise quality expenditures, but
reduces slack while not affecting costs. To raise quality, hospitals are forced to become
more efficient. To encourage additional quality, the government can offer a mixed re-
imbursement system in which a combination of a fixed price and a payment based on
hospital cost is paid [also see Ellis and McGuire (1986)]. Although paying partly on
a cost basis may improve quality, it also subsidizes (increases) slack. Substituting any
other nonpecuniary objective for slack would yield the same conclusions. This implies
that increased competition should force private not-for-profit hospitals to be increas-
ingly similar to their for-profit counterparts.
In fact, the empirical evidence reviewed above suggests that this is so. Norton and
Staiger (1994) found no difference between private not-for-profit and for-profit hospitals
in provision of uncompensated care, once they controlled for hospital location. A study
by Banks and co-authors (1997) obtained empirical support for the view that for-profit
hospitals provide some uncompensated care as a cost of doing business. Most directly
to the point, is Gruber's (1994) finding of higher rates of decrease in uncompensated
care in the face of increased competition in California. The stability of the shares by
ownership - private not-for-profit hospitals were 60 percent of nonfederal short-term
1168 EA. Sloan
general and other specialty hospitals in 1965 and in 1965 attests to the flexibility of
the private not-for-profit sector. Of course, an alternative interpretation is that for-profit
hospitals, which were 15 percent in 1994 and 14 percent in 1994, were no better in
satisfying consumer wants [Gray and Schlesinger (1997, p. 16)].5
8. Conclusions and future research agenda
The hospital industry provides an interesting laboratory for studying the impact of own-
ership form on performance. Overall, the evidence suggests that for-profit and private
not-for-profit hospitals are far more alike than different. If private not-for-profit hospi-
tals are to distinguish themselves in terms of some nonpecuniary objective, they will
have to define specifically what that focus is. In an environment of increased compe-
tition, to finance this objective, they will have to secure sources of funding other than
patient dollars. On the other hand, given the transformation of health care in the United
States toward price and quality competition that has occurred during the 1980s and
1990s, one might have anticipated much more relative growth of the for-profit sector
than has actually occurred. Thus the evidence also calls into question the "knee-jerk"
reaction that for-profit automatically means greater efficiency.
One of the questions we posed at the outset was whether or not other ownership forms
than the for-profit form are more efficient ones for the hospital sector. The answer de-
pends on much more than technical efficiency and allocative efficiency in choice or
inputs. Viewed in such narrow terms, the for-profit form performs about as well as
private not-for-profits. The answer also depends on whether or not hospitals with par-
ticular ownership forms produce the socially optimal combination of outputs. On this
score, my review also suggests not much difference and, if anything, under competition,
differences may be expected to narrow. Private not-for-profit hospitals will have less lat-
itude than previously to produce outputs they deem to be socially worthy. In the United
States, private donations are too small a part of the entire hospital revenue stream to
make individual donors efficient monitors.
My review has intentionally not focused on hospitals with major missions of medical
education and research. Such hospitals produce public goods that are not likely to be
produced by a for-profit institution, unless, of course, dedicated subsidies exist for these
activities. Also, I did not explicitly consider public hospitals, especially those serving
disadvantaged populations in large cities. If such hospitals were to go, it would be nec-
essary to invent new organizations to serve these roles, especially in the United States,
where large numbers of persons lack health insurance.
5 Most empirical literature on nonprice competition among hospitals has not examined differences in be-
havior by ownership. Rather it has dealt with the "medical arms race", in which hospitals paid by insurers on
the basis of retrospective costs or charges compete for patients by offering higher quality [Robinson and Luft
(1987, 1988), Noether (1988), Dranove et al. (1992)].
This chapter's overall conclusion about the effect of ownership is supported by empir-
ical comparisons from other sectors that have questioned the supposition that the profit
motive leads to greater efficiency. In fact, within an industry, firms tend to be more
alike than they are different or at least the evidence is mixed [Borcherding et al. (1982),
Boardman and Vining (1989), Peters (1993), and Vining and Boardman (1992)]. For
example, based on their literature review, Vickers and Yarrow (1988) concluded that
competition in the firm's product market may be a more important determinant of per-
formance than is ownership.
Although much is known about effects of ownership on hospital behavior, there is
much to be learned. These are important priorities.
First, one response to this chapter's results is that it failed to consider some valuable
outputs that, if considered, would reverse the findings. Such responses are common
among noneconomists in health policy discussions, but among some very knowledge-
able economists as well. One issue that comes up is local control as a benefit of not-for-
profit ownership. If the hospital were owned by a stock company, local control would
be lost. Perhaps local control leads to a better perception of community preferences.
But more broadly, the question about local control is who the potential beneficiaries
are. Possible candidates are the managers and employees; less likely beneficiaries are
patients. More generally, the challenge is identify these heretofore unmeasured outputs
and evaluate the effects of ownership on their provision. Just believing that the outputs
are out there somewhere is not terribly compelling. Perhaps it would be useful to start
with an informal survey of practitioners in the health field. I leave the particulars to you,
the reader.
Second, empirical evidence on the performance of hospitals in countries other than in
the United States is badly needed. There is considerable heterogeneity among countries
in how care is organized and financed. In some, for example, patient choice of hospi-
tal is very limited. Complete lack of competition that exists in some countries where
patients do not have a choice of hospital should plausibly affect performance. The re-
sults reviewed above for the United States probably do not generalize. Nevertheless, the
studies do provide a useful point of departure for scholars of other health systems.
Third, research should not be confined to differences in performance. Although there
has been some interesting new research on the topic, there still is no really definitive
answer to the question as to why, in countries throughout the world, the for-profit own-
ership hospital is in the minority. To understand ownership composition of hospital
sectors, it may be necessary to think about other sectors as well.
Fourth, if there is one dimension of performance that is likely to be persuasive, in
comparing hospitals, it is likely to be quality. As often noted, quality of health care is
difficult for consumers to monitor. But it is even difficult for the experts to measure,
although much progress has been made in this field in recent years. To the extent that
ownership indeed affects quality of care, this is critical to know. The better studies of
quality use several alternative measures and are able to track outcomes for some time
after admission. To isolate effects of hospital characteristics on quality, it is necessary
to hold other factors, such as casemix, constant.
1170 EA. Sloan
Fifth, empirical research to date has varied in sophistication of the econometrics em-
ployed. There is probably no pressing need for another case-control study of hospital
ownership. Yet even some of the better studies may be plagued by endogeneity and
omitted heterogeneity. Patient choice of hospital or hospital choice of organizational
form are not exogenous to the behavior that has been studied. Data from randomized
controlled trials will never be available for this topic. Therefore, it will be necessary for
economists to deal with endogeneity in some other way. Precise ideas are beyond the
scope of this chapter. Readers are encouraged to read the chapter in this Handbook by
Jones (2000) on econometric applications in health economics.
Finally, there are important links between ownership and the constraints under which
hospitals operate. As stressed above, increased competition may be expected to seri-
ously limit the latitude hospitals have to engage in nonprofit maximizing behaviors. In
this sense, hospital ownership as a topic in the United States may be a declining indus-
try. On the other hand, hospitals are integrating vertically in the United States, largely
because of market changes. In many cases, the partners may be of a different organi-
zational form than the hospital. This opens up the possibility of all kinds of blended
combinations, for example, a for-profit health maintenance organization linked to a pri-
vate not-for-profit hospital which may produce a variety of outcomes. Studies of effects
of health system ownership on various dimensions of performance should be conducted.
References
Alchian, A., and H. Demsetz (1972), "Production, information costs, and economic organization", American
American Hospital Association (1995), Hospital Statistics, 1995-1996 (American Hospital Association,
Chicago).
53(5):941-973.
Banks, D.A., M. Paterson and J. Wendel (1997), "Uncompensated hospital care: Charitable mission or prof-
itable business decisions?", Health Economics 6:133-143.
Becker, E.R., and F.A. Sloan (1985), "Hospital ownership and preference", Economic Inquiry 23(1):21-36.
Boardman, A.R., and A.E. Vining (1989), "Ownership and performance in competitive environments: A com-
parison of the performance of private, mixed, and state-owned enterprises", Journal of Law and Economics
32:1-33.
Borcherding, T.E., W.W. Pommerehne and E Schneider (1982), "Considering the efficiency of private and
public production: The evidence from five countries", Zeitschrift for Nationalokonoemie/Journal of Eco-
nomics: Public Production (Supplement 2): 127-156.
Breyer, F., D. Paffrath, W.Preuss and R. Schmidt (1988), Die Krankenhaus-Kostenfunktion: Der Einfluss von
Diagnosespecktrum und Bettenauslastung auf die Kosten im Krankenhaus (AOK-Verlag, Bonn).
Brooks, J.M., A. Dor and H.S. Wong (1997), "Hospital-insurer bargaining: An empirical investigation of
appendectomy pricing", Journal of Health Economics 16(4):417-435.
Calem, P.S., and J.A. Rizzo (1995), "Financing constraints and investment: New evidence from hospital in-
dustry data", Journal of Money, Credit, and Banking 27(4):1002-1014.
Clark, R.C. (1980), "Does the nonprofit form fit the hospital industry?", Harvard Law Review 93:1416-1489.
Conrad, D.A. (1984), "Returns on equity to not-for-profit hospitals: Theory and implementation", Health
De Meza, D., and B. Lockwood (1998), "Does asset ownership always motivate managers? Outside options
and the property rights theory of the firm", Quarterly Journal of Economics 113(2):361-387.
Dranove, D. (1988), "Pricing by non-profit institutions: The case of hospital cost-shifting", Journal of Health
Economics 7(1):47-58.
Dranove, D., M. Shanley and C. Simon (1992), "Is hospital competition wasteful?", Rand Journal of Eco-
nomics 23(2):247-262.
Duke University Center for Health Policy, Law and Management (1998), A Guide for Communities Consid-
ering Hospital Conversion in the Carolinas (Center for Health Policy, Durham, NC).
Easley, D., and M. O'Hara (1983), "The economic role of the nonprofit firm", Bell Journal of Economics
14(2):531-538.
Ellis, R.P., and T.G. McGuire (1986), "Provider behavior under prospective reimbursement: Cost sharing and
supply", Journal of Health Economics 5:129-151.
Feldstein, M.S. (1971), "Hospital cost inflation: A study of nonprofit price dynamics", American Economic
Review 61(5):853-872.
Frank, R.G., and D.S. Salkever (1991), "The supply of charity services by nonprofit hospitals: Motives and
market structure", The Rand Journal of Economics 22(3):43-45.
Frank, R.G., D. Salkever and J. Mitchell (1990), "Market forces and the public good: Competition among hos-
pitals and provision of indigent care", in: R. Scheffier and L. Rossiter, eds., Advances in Health Economics
and Health Services Research (JAI Press, Greenwich, CT).
Ginsburg, P.B., and EA. Sloan (1984), "Hospital cost-shifting", New England Journal of Medicine
310(14):893-898.
Glaser, W.A. (1987), Paying the Hospital. The Organization, Dynamics, and Effects of Differing Financial
Arrangements (Jossey-Bass Publishing, San Francisco, CA).
Goddeeris, J.H., and B.A. Weisbrod (1998), "Conversion from nonprofit to for-profit legal status: Why does
it happen and should anyone care?", Journal of Policy Analysis and Management 17(2):215-233.
Grannemann, T.W., R.S. Brown and M.V. Pauly (1986), "Estimating hospital costs: A multiple-output analy-
sis", Journal of Health Economics 5(2):107-128.
Gray, B.H. (1997), "Conversion of HMOs and hospitals: What's at stake?", Health Affairs 16(2):29-47.
Gray, B.H. (1991), The Profit Motive and Patient Care: The Changing Accountability of Doctors and Hospi-
tals, 2nd edn. (Harvard University Press, Cambridge, MA).
Gray, B.H., and M. Schlesinger (1997), "The profit transformation of the hospital and HMO fields", unpub-
lished.
Grossman, S., and O. Hart (1986), "The costs and benefits of ownership: A theory of vertical and lateral
integration", Journal of Political Economy 94(4):691-719.
Gruber, J. (1994), "The effects of competitive pressure on charity - hospital response to price shopping in
California", Journal of Health Economics 13(2):183-212.
Hadlock, C.J. (1998), "Ownership, liquidity, and investmentf, Rand Journal of Economics 29(3):487-598.
Hansmann, H.B. (1996), The Ownership of Enterprise (The Belknap Press of Harvard University Press, Cam-
bridge and London).
Hansmann, H.B. (1980), "The role of nonprofit enterprise", The Yale Law Journal 89(5):835-901.
Hansmann, H.B. (1998), "Ownership of the firm", in: P. Newman, ed., The New Palgrave Dictionary of
Economics and the Law, Vol. 2 (McMillan Reference Limited/Stockton Press, London and New York)
735-742.
Hart, O. (1995), Firms, Contracts, and Financial Structure (Oxford University Press, Oxford).
Hart, O., and J. Moore (1990), "Property rights and the nature of the firm", Journal of Political Economy
98(6):1119-1158.
Hart, O., A. Shleifer and R.W. Vishny (1997), "The proper scope of government: Theory and an application
to prisons", Quarterly Journal of Economics 112(4): 1127-1162.
Hartz, A.J., H. Krakauer, E.M. Kuhn et al. (1989), "Hospital characteristics and mortality rates", New England
Journal of Medicine 321:1720-1725.
1172 EA. Sloan
Herzlinger, R.E., and W.S. Krasker (1987), "Who profits from nonprofits?", Harvard Business Review
(January/February):93-106.
Hoerger, T.J. (1991), "'Profit' variability in for-profit and not-for-profit hospitals", Journal of Health Eco-
nomics 10(3):259-289.
Hoerger, T.J. (1995), "Hospital investment in the 1980's", unpublished paper, March.
Hoffmeyer, U., K. Ullrich and T.R. McCarthy (1994), Financing Health Care (Kluwer Academic Publishers).
Institute of Medicine (1986), For-Profit Enterprise in Health Care (National Academy Press, Washington).
Institute of Medicine (1996), Nursing Staff in Hospitals and Nursing-Homes: Is it Adequate? (National
Academy of Sciences Press, Washington).
Jensen, M.C., and W.H. Meckling (1976), "Theory of the firm: Managerial behavior, agency costs and own-
ership structure", Journal of Financial Economics 3:305-360.
Jones, A.M. (2000), "Health econometrics", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health
Economics (Elsevier, Amsterdam) Chapter 6.
Joskow, P. (1980), "The effects of competition and regulation on hospital bed supply and the reservation
quality of the hospital", Bell Journal of Economics 11(2):421-447.
Keeler, E.B., G.L. Melnick and J. Zwanzinger (1999), "Changing effects of competition on non-profit and
for-profit hospital pricing behavior", Journal of Health Economics 18(11), February.
Keeler, E.B., L.V. Rubinstein, K.L. Kahn, D. Draper, E.R. Harrison, M.J. McGinty, W.H. Rogers and R.H.
Brook (1992), "Hospital characteristics and quality of care", Journal of the American Medical Association
268(13):1709-1714.
Koop, G., J. Osiewalski and M.F.J. Steel (1997), "Bayesian efficiency analysis through individual effects:
Hospital cost frontiers", Journal of Econometrics 76:77-105.
Kuttner, R. (1996a), "Columbia/HCA and the resurgence of the for-profit hospital business" (first of two
parts), New England Journal of Medicine 335(6):362-367.
Lakdawalla, D., and T. Philipson (1998), "A neoclassical theory of nonprofit production", July.
Lewin, L.S., R.A. Derzon and R. Margulies (1981), "Investor-owned and nonprofits differ in economic per-
formance", Hospitals 55(13):52-58.
Lynk, W.J. (1995), "Nonprofit hospital mergers and the exercise of market power", Journal of Law and Eco-
nomics 38(2):437-461.
Mann, J., G. Melnick, A. Bamezai and J. Zwanziger (1995), "Uncompensated care: Hospitals' response to
fiscal pressures", Health Affairs 14(1):263-270.
Mark, T.L. (1996), "Psychiatric hospital ownership and performance", Journal of Human Resources 1(3):631-
649.
Morrisey, M.A. (1994), Cost-Shifting in Health Care: Separating Evidence from Rhetoric (The American
Enterprise Institute Press, Washington).
Morrisey, M.A., GJ. Wedig and M. Hassan (1996), "Do nonprofit hospitals pay their way?", Health Affairs
15(4):132-144.
Needleman, J., D.J. Chollet and J. Lamphere (1997), "Hospital conversion trends", Health Affairs 16(2): 187-
195.
Newhouse, J.P. (1970), "Toward a theory of nonprofit institutions: An economic model of a hospital", Amer-
ican Economic Review 60(1):64-74.
Newhouse, J.P. (1994), "Frontier estimation: How useful a tool for health economics?", Journal of Health
Economics 13(3):317-322.
Norton, E.C., and D.O. Staiger (1994), "How hospital ownership affects access to care for the uninsured",
Rand Journal of Economics 25(1):171-185.
Pattison, R.V., and H.M. Katz (1983), "nvestor-owned and not-for-profit hospitals", New England Journal of
Medicine 309(6):347-353.
Pauly, M.V. (1980), Doctors and Their Workshops (University of Chicago Press, Chicago, IL).
Pauly, M.V. (1971), Medical Care at Public Exsistense (Praeger Publishing, New York).
Pauly, M.V. (1987), "Nonprofit hospitals in medical markets", American Economic Review 77:257-282.
Pauly, M.V., and M. Redisch (1973), "The not-for-profit hospital as a physicians' cooperative", American
Peters, L.L. (1993), "For-profit and non-profit firms: Limits of the simple theory of attenuated property rights",
Review of Industrial Organization 8(5):623-634.
Pope, G.C. (1989), "Hospital nonprice competition and medicare reimbursement policy", Journal of Health
Economics 8(2):147-172.
Relman, A.S. (1980), '"The new medical-industrial complex", New England Journal of Medicine 303:963-
969.
Robinson, J.C., and H.S. Luft (1987), "Competition and the cost of hospital care, 1972 to 1982", Journal of
the American Medical Association 257(23):3241-3245.
Robinson, J.C., and H.S. Luft (1988), "Competition regulation and hospital costs, 1982 to 1986", Journal of
the American Medical Association 260(18):2676-2681.
Robinson, J.C., and C.S. Phibbs (1989), "An evaluation of Medicaid selective contracting in California",
Romeo, A.A., J. Wagner and R. Lee (1983), "Prospective reimbursement and the diffusion of new technolo-
gies in hospitals", Journal of Health Economics 3(1):1-24.
Rosenberg, C.E. (1987), The Care of Strangers: The Rise of America's Hospital System (Basic Books, New
York).
Russell, L. (1979), Technology in Hospitals: Medical Advances and Their Diffusion (Brookings Institution,
Washington).
Schlesinger, M., R. Dorwart, C. Hoover and S. Epstein (1997), "Competition ownership and access to hospital
services", Medical Care 35(9):974-992.
Shalit, S.S. (1977), "A doctor-hospital cartel theory", Journal of Business 50(1):1-20.
Shortell, S., and E.FX. Hughes (1988), "The effects of regulation, competition, and ownership on mortality
rates among hospital inpatients", New England Journal of Medicine 318(17):1100-1107.
Silvers, J.B., and R.T. Kauer (1986), "Returns on equity for not-for-profit hospitals: Some comments", Health
Sloan, FA., and B. Steinwald (1980a), Hospital Labor Markets: Analysis of Wages and Work Force Compo-
sition (D.C. Heath-Lexington Books, Lexington, MA).
Sloan, EA., and B. Steinwald (1980b), Insurance, Regulation, and Hospital Costs (D.C. Heath-Lexington
Books, Lexington, MA).
Sloan, EA., and D.H. Taylor (1999), "Does ownership affect the cost of Medicare?", in: A.J. Rettenmaier and
T.R. Saving, eds., Medicare Reform: Issues and Answers (University of Chicago Press, Chicago).
Sloan, EA., and R.A. Vraciu (1983), "Investor-owned and not-for-profit hospitals: Addressing some issues",
Health Affairs 2(1):25-37.
Sloan, EA., J. Valvona and J.M. Perrin (1986a), "Diffusion of surgical technologies: An exploratory study",
Sloan, EA., J. Valvona and R. Mullner (1986b), "Identifying the issues: A statistical profile", in: EA. Sloan,
J.E Blumstein and J.H. Perrin, eds., Uncompensated Hospital Care: Rights and Responsibilities (The Johns
Hopkins University Press, Baltimore, MD) 16-53.
Sloan, EA., J. Valvona, M. Hassan and M.A. Morrisey (1988), "Cost of capital to the hospital section", Journal
Sloan, EA., T.J. Hoerger, M. Morrisey and M. Hassan (1990), "The demise of hospital philanthropy", Eco-
nomic Inquiry 28(1):725-734.
Sloan, FA., D.H. Taylor, G. Picone and S.Y. Chou (1998a), "Hospital ownership and cost and quality of care:
Is there a dime's worth of difference?", Working Paper #W6706 (National Bureau of Economic Research).
Sloan, EA., G. Picone, D.H. Taylor and S.Y. Chou (1998b), "Does where you are admitted make a difference?
An analysis of Medicare data", in: A.M. Garber, ed., Frontiers in Health Policy Research (MIT Press,
Cambridge, MA), forthcoming.
Sloan, EA., K. Whetten-Goldstein and A. Wilson (1997), "Hospital pharmacy decisions cost containment,
and the use of cost-effectiveness analysis", Social Sciences and Medicine 45(4):523-533.
1174 EA. Sloan
Sloan, EA., D. Taylor and C. Conover (2000), "Hospital conversions: Is the purchase price too low?", in:
D.M. Cutler, ed., The Changing Hospital Industry: Comparing Not-for-Profit and For-Profit Institutions
(University of Chicago Press, Chicago, IL) 17-44.
Starr, P. (1982), The Social Transformation of American Medicine (Basic Books, New York).
Thorpe, K.E., and C.E. Phelps (1991), "The social role of not-for-profit organizations: Hospital provision of
charity care", Economic Inquiry 29(3):472-484.
US Medicare Payment Advisory Commission (1998), Health Care Spending and the Medicare Program: A
Data Book (The Commission).
US Prospective Payment Assessment Commission (1996), "Medicare and the American Health Care System",
Report to Congress (The Commission).
Vickers, J., and Yarrow, G. (1988), Privatization: An Economic Analysis (MIT Press, Cambridge, MA).
Vining, A.R., and A.E. Boardman (1992), "Ownership versus competition: Efficiency in public enterprise",
Public Choice 73:205-239.
Vistnes, G. (1994), "An empirical investigation of procurement contact structures", Rand Journal of Eco-
nomics 25(2):215-241.
Vitaliano, D.F., and M. Toren (1996), "Hospital cost and efficiency in a regime of stringent regulation",
Eastern Economic Journal 22(2):161-175.
Wedig, G.J., EA. Sloan, M. Hassan and M.A. Morrisey (1988), "Capital structure, ownership, and capital
payment policy: The case of hospitals", Journal of Finance 43(1):21-40.
Wedig, G.J., M. Hassan and F.A. Sloan (1989), "Hospital investment decisions and the cost of capital", Journal
of Business 62(4):517-537.
Wedig, G.J., M. Hassan and M.A. Morrisey (1996), "Tax-exempt debt and the capital structure of nonprofit
organizations", Journal of Finance 5(4):1247-1284.
Weisbrod, B.A. (1977), The Voluntary Nonprofit Sector (D.C. Heath, Lexington, MA).
Weisbrod, B.A. (1988), The Nonprofit Economy (Harvard University Press, Cambridge, MA).
Wilson, G.W., and J.M. Jadlow (1982), "Competition, profit incentives, and technical efficiency in the provi-
sion of nuclear medicine services", Bell Journal of Economics 13(2):472-482.
Young, G.J., K. Desai and C. Lucas (1997), "Does the sale of nonprofit hospitals threaten health care for the
poor?", Health Affairs 16(1):137-141.
Zuckerman, S., J. Hadley and L. Iezzoni (1994), "Measuring hospital efficiency with frontier cost functions",
Zweifel, P., and F. Breyer (1997), Health Economics (Oxford University Press, New York).
Chapter 22
ECONOMICS OF GENERAL PRACTICE*
ANTHONY SCOTT
Health Economics Research Unit, University ofAberdeen
Contents
Abstract 1176
Keywords 1176
2. The GP-patient relationship 1178
3. Utilisation of GP services 1180
3.1. Choice of practice 1181
3.2. Choice to seek medical care 1182
3.3. Choice of general practitioner 1183
3.4. Choice of treatment 1183
4. Models of GP behaviour 1184
5. Explaining variations in referrals 1185
6. GP payment systems 1187
7. GPs as firms 1190
7.1. Partnerships and group practice 1190
7.2. Vertical integration, the boundaries of the firm and the balance of care 1191
8. Future research 1193
References 1194
*I am very grateful to Hugh Gravelle for his role in developing this chapter. Wynand van de Ven, Joe New-
house and Tony Culyer also provided useful comments. The Health Economics Research Unit is funded by
the Chief Scientist Office of the Scottish Executive Health Department. Any errors or omissions are the re-
sponsibility of the author.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.R Newhouse
1176 A. Scott
Abstract
General (or family) practice and its role within primary care is increasingly regarded as
the key to achieving efficiency and equity in many health care systems. This is partic-
ularly relevant where general practitioners (GPs) act as gatekeepers to specialist care.
This chapter outlines the main economic issues in general practice. Within the context
of gatekeeping, the first half of the chapter examines literature on agency, patient choice
and preferences for GP services, and the utilisation of GP services. Given that much
demand is determined by supply, this is followed by an examination of the determinants
of referral behaviour, the effects of payment systems, and GPs as firms (partnerships
and vertical integration). Overall, there has been little research by economists in these
areas. This needs to be rectified giving the growing importance of primary care in many
health care systems.
Keywords
agency relationship, utilisation, referral, financial incentives, vertical integration
JEL classification:D21, I11, 100, J44, L21

Ch. 22: Economics of General Practice 1177
1. Introduction
The definition, financing and regulation of general (or family) practice differs widely
across countries [Boerma et al. (1998), Fry and Horder (1994), Gervas et al. (1994),
Kristiansen and Mooney (1993a)]. Common to many definitions, however, are issues
related to continuity of care, the doctor-patient relationship, ease of access, and the
provision of services covering a broad range of health needs. In terms of financing and
regulation, there is an increasing international trend toward strengthening primary care
[Van de Ven (1996)]. This is widely seen as the key to enhancing equity and efficiency
in health care systems. The main characteristic which gives general practice a unique
and pivotal position in many health care systems is the General Practitioner's (GP's)
role as "gatekeeper".
The gatekeeping role of GPs operates where the insurance system generates low or
zero money prices for care at the point of consumption, and general practice is the first
and only point of contact for individuals for all types of non-emergency medical condi-
tion. The only way patients can access specialist (secondary) care is to obtain a referral
from their GP. This has significant implications for cost control given that specialist
care accounts for over half of health care expenditure in most developed countries. It
has been suggested that gatekeeping leads to lower health care costs and a reduction in
"unnecessary" medical interventions [Martin et al. (1989), Roos (1979), Franks et al.
(1992), Starfield (1994)]. With GP gatekeeping, specialists are said to be used more
efficiently because patients who have problems more appropriately treated in primary
care are screened out by GPs. Further, GPs usually have better information than pa-
tients about the quality of care available from secondary care providers, since they can
aggregate the experience of all their referred patients. GPs may therefore act as more
effective agents for their patients. The gatekeeping role also highlights the importance
of the interface between primary and specialist care (in particular factors influencing
GPs' decisions to refer and vertically integrate) in controlling costs and quality. Con-
sequently, the regulation of GPs and the incentive structures they face which influence
their behaviour have significant implications for costs, health outcomes and equity in
health care systems.
The chapter outlines the main economic questions in general practice and summarises
the relevant theories and empirical evidence. Since the focus is on GPs as gatekeepers,
the studies reported are mainly from those countries where such a system dominates,
although many of the issues discussed are relevant to countries with a reduced or no
gatekeeping role. The reviews of empirical work are not systematic but aim to illustrate
the important economic issues and research on general practice.
The chapter begins with a discussion about agency, patient choice and the demand
for and utilisation of GP services. Recognising that much demand and utilisation is in
fact determined by supply, the second half of the chapter focuses on factors influencing
GP decision making, particularly decisions to refer. Theories and evidence of the ef-
fects of GP payment systems on GP behaviour are then summarised. The organisation
of GPs into "firms" is then examined, particularly the role of partnerships and incen-
1178 A. Scott
tives within them. The issue of vertical integration is also examined since this is again
relevant to the gatekeeping role of GPs. Finally, the common themes that have been
researched to date and those that need to be examined in the future are summarised and
discussed.
2. The GP-patient relationship
The doctor-patient relationship is most frequently seen as one of agency. This arises
because patients (principals) are less informed than doctors (agents) about the relation-
ship between health care and health status. Potential inefficiency can arise since the
doctor has his or her own objectives to maximise that may conflict with those of the
patient. Much of the economic literature about the doctor-patient relationship has there-
fore focused on the ways in which the potential inefficiencies can be mitigated, and the
consequences for use of health services. In the absence of an explicit contract between
doctor and patient, systems of professional ethics, licensure, regulation and contracts
have been examined as means of reducing agency costs. The main agency problem has
most often been seen as one of supplier-induced demand: the doctor induces the patient
to consume more health care than patients would if they had the same information as
the doctor. However, agency costs may also take the form of under-provision of care or
doctor effort where the financial regime is such that additional care reduces the doctor's
income or requires additional effort.
Much of the literature about agency has therefore focussed on the consequences of
asymmetry of information in the doctor-patient relationship, i.e. the effects of regulation
and incentives and doctors' behaviour.
However, there are other aspects to the doctor-patient relationship that have impor-
tant implications for efficiency. The distinctive feature of general practice agency is that
the doctor-patient relationship is usually long term and more likely to be characterised
by repeated transactions. Patients consult their GPs much more frequently than they use
secondary care services. For example in the UK, the average number of GP consulta-
tions is five per year and some age and sex groups consult more frequently [Rowlands
et al. (1997)].
Long term relationships between patients and health professionals are referred to in
the health service literature as "continuity of care". This has been espoused by GPs as
one of the fundamental and most important characteristics of general practice [Starfield
(1994)], where GPs are defined as providing, "... personal, primary and continuing
medical care to individuals and families", and where long term relationships, "... are
the defining element of generalist practice" [Burke et al. (1993)]. The long term rela-
tionship has also been characterised by care provided by the same GP, for a wide range
of medical problems, and in the same location [Rogers and Curtis (1980)].
Long-term agency relationships are more likely when GPs are responsible for a de-
fined population of patients. This is especially the case when patients are enrolled with
single-handed or partnerships of GPs and in areas where there are few GPs per capita.
Ch. 22: Economics of GeneralPractice 1179
The existence of such long term relationships has implications for patient choice, the
nature of competition, and the demand for GPs' services.
Increased frequency of transactions between parties has been argued to reduce the
costs of information transmission and may also reduce monitoring costs [Williamson
(1993)]. Several principal-agent models have shown that repeated transactions between
principals and agents can lead to more efficient contracts [Rubenstein and Yaari (1983)]
when rewards are conditioned on the history of observable outcomes, not just on cur-
rent outcomes. In general practice repeated transactions are also potentially beneficial
because the GP becomes more aware of the context of the patients' health problems, and
has more information about the patients' medical history, social circumstances, values
and preferences. However, the theory underlying the effects of repeated transactions has
not yet been applied to general practice.
There has, however, been some empirical work. One study did find some evidence of
a negative association between the length of physician-patient "ties" and the cost and
intensity of medical care, after controlling for demography, socio-economic character-
istics and self-assessed health status [Weiss and Blustein (1996)]. Another study found
GPs having the most knowledge about their patients' medical history were more likely
to have shorter consultations, less likely to order laboratory tests, more likely to adopt a
"wait and see" approach to management of illness, more likely to prescribe, more likely
to issue a sickness certificate, and more likely to refer [Hjortdahl and Borchgrevink
(1991)]. Previous knowledge of patients' medical history therefore influenced the use
of resources in positive and negative directions.
It has been shown that patients both prefer and are more compliant with treatment
recommendations, if they see the same GP at each visit [Dietrich and Marton (1982),
Hjortdahl and Laerum (1992)]. Nonetheless, they may not be willing to wait longer than
two days if there is another GP available in a group practice [Freeman and Richards
(1993)]. No studies could be found which examined effects on patients' health status.
The emphasis on the communication of information rather than the self-interested
bargaining that is assumed in principal-agent theory, means that information transmis-
sion is likely to play a central role in meeting the objectives of the principal at minimum
cost. The long-term agency relationship facilitates two-way information transmission
between GP and patient. Doctors provide information to the patient about health prob-
lems and the options available for treatment, and the patient informs the doctor about
symptoms, social context, and their values and preferences attached to different treat-
ment options. There is much evidence suggesting that doctors who have been trained in
communication skills leave patients more satisfied, more compliant with treatment rec-
ommendations, and in better health [Ong et al. (1995), Kaplan et al. (1989), Wartman
et al. (1983), Stewart (1995)]. The role of information transmission in market transac-
tions has generally been ignored in the standard principal-agent literature, although such
issues are becoming more relevant in the analysis of contractual relationships between
firms [Casson and Wadeson (1997), Arrighetti et al. (1997), Burchell and Wilkinson
(1997)] and in the health sector [Goddard and Manion (1998)].
1180 A. Scott
However, one cannot use these studies to argue that better communication leads
to increased efficiency because they took no account of costs. More communication
and therefore presumably better care could be associated with longer consultations and
therefore higher costs. The issue then becomes the value of this additional time com-
pared with its marginal cost, neither of which is revealed in health care markets. Some
commentators have suggested that length of consultation should be an indicator of qual-
ity of care in general practice, although the costs of such a policy have not been exam-
ined [Howie et al. (1991)].
The evidence on long-term agency relationships and the nature of doctor-patient conm-
munication within them suggests that ways of improving the efficiency of general prac-
tice require further analysis, Research into the source of asymmetry of information
rather than its symptoms, in particular the "black box" of agency relationships, is re-
quired.
3. Utilisation of GP services
In health systems where GPs are gatekeepers, the demand for non-emergency health
care reflects a sequence of decisions involving the GP and the patient. In a typical list
system in which the patient registers with one and only one practice (as in the UK),
patients first choose whether to join the list of some practice. In most systems, the
practice has the choice of whether to accept the patient, which raises the possibility
of cream-skimming [see van de Ven and Ellis (2000)]. When ill, the patient decides
whether to seek care or advice from the practice or from alternative providers, such
as pharmacies and accident and emergency departments at local hospitals. (In systems
without patient registration, the choice of practice can occur after the decision has been
made to seek medical care and in conjunction with the choice of GP.) If they seek
care from the practice, they may face a choice of which doctor in the practice they
consult. The consultation may lead to either a referral to a specialist for diagnosis or for
treatment, to diagnosis and treatment in the practice, or simply reassurance.
This sequence has two important characteristics. First, asymmetry of information
and the doctor's own objectives means that the demand for health care is influenced by
the preferences of both doctor and patient. Second, decisions at one stage are affected
by expectations about decisions at later stages. For example, choice of practice may
depend on whether patients can choose a GP when consulting, and the likelihood that a
consultation for particular conditions leads to referral or treatment within the practice.
Decisions to seek care are influenced by the availability of services in an area. Thus,
referral decisions are likely to be influenced by secondary care supply conditions, which
will affect the time, money, and distance costs incurred by patients and GPs. Hence,
decisions at all stages in the sequence may affected by secondary care supply and GPs'
preferences.
As a consequence of these relationships, care is required both in making inferences
about patients' preferences from decisions at any stage, and in modelling the deter-
minants of decisions at each stage. An issue in much of the literature examining GP

utilisation is therefore dealing with this simultaneity problem.
3.1. Choice of practice
In countries with a system of patient registration, the first decision individuals usually
make is to choose a general practice. Several patient surveys have examined the deter-
minants of practice choice. Patients have indicated that they are influenced by waiting
times, access (by telephone, ease of making an appointment, distance, and ease of get-
ting to the practice), condition of premises, availability of out of hours cover, number
of services provided on site, and relationships with staff [Wensing et al. (1996), Baker
and Streatfield (1995), Grogan et al. (1995), van der Kar et al. (1992), Salisbury (1989),
Calnan et al. (1994a), Khayat and Salter (1994), Calnan et al. (1994b), Hjortdahl and
Laerum (1992), Smith and Armstrong (1989), Williams and Calnan (1991), Steven and
Douglas (1988)]. Few of these studies have examined the relative importance of these
factors but those that do generally show that aspects of the doctor-patient relationship
(e.g. the doctor listening) are more important determinants of satisfaction than prac-
tice characteristics. These studies have either used rankings of factors or have regressed
ratings of satisfaction with specific attributes on total satisfaction scores [Smith and
Armstrong (1989), Williams and Cainan (1991), Calnan et al. (1994b)].
However, such studies suffer from the drawback that they are not choice based, so
that inferences about the relative importance of practice attributes are made in a con-
text where respondents are not comparing alternative bundles of practice attributes. A
more appealing approach, at least to economists, is to use data on choice of practice.
Economic methods such as discrete choice conjoint analysis are beginning to be used
in this area. For example, Ryan et al. (1998) used this method to examine the relative
importance of a patient-held data card in comparison with waiting time until doctor was
seen, waiting time in reception, and whether patients saw their doctor of choice. Scott
et al. (1998) examined preferences for GP out of hours care. They found that "whether
the doctor seemed to listen" was the most important attribute, and that the majority of
respondents had a dominant preference for this attribute, i.e. they did not appear to be
willing to trade it off for a shorter waiting time, a home visit, or to see a doctor they
knew. Mooney et al. (1999) examined the relative importance of travel time to surgery,
opening hours, cost, waiting time, whether they saw doctor of choice, and whether GP
had additional training.
Other studies have examined data on revealed, rather than stated preferences. Dixon
et al. (1997) examined the determinants of the rate at which patients left practices in
three English health authorities. This study focused on patients who changed practice
without changing their home address and found that patients were more likely to leave a
practice if it was small, was a greater distance away, had shorter opening hours, provided
fewer clinics, or was a multifund (a group of fundholding practices with a common con-
tract with secondary care providers). The study also found that 38 percent of individuals
were registered with their nearest practice, suggesting factors other than distance were
important in choosing a practice.
!182 A. Scott
3.2. Choice to seek medical care
General practice is an experience good [Nelson (1970)]: patients may find it difficult
to judge the quality of different practices when registering but learn more about qual-
ity after a number of consultations. The decision to seek medical care and visit the GP
may be one alternative in the choice set when the patient requires non-emergency health
care, since they can choose between a visit to the GP or to other providers such as ac-
cident and emergency departments or pharmacies [Campbell (1994)]. Thus, the relative
ease of access to these substitute providers (time costs and distance) will also affect the
demand for GP consultations. The decision to seek health care most closely reflects the
demand for health [see Grossman (2000)]. Whether or not an individual decides to visit
depends upon the individual's beliefs about the severity of their health problem, and
their expectations as to whether anything can be done about it. These factors are high-
lighted by sociologists in the "Health Belief Model" [van der Kar (1992)]. Individuals
may also decide to visit their GP for information and reassurance rather than to improve
health per se, especially when the patient has a chronic disease (which make up much
of GP care).
The determinants of the utilisation of GP services has been examined using secondary
data from health or household surveys about socio-economic factors, age, sex, informa-
tion on illness, measures of access and availability to services, and the characteristics
of health care providers (although often there are few data on these in population sur-
veys). Generally, studies find that access factors (distance and availability of services),
socio-economic characteristics, measures of morbidity, and age and sex influence utili-
sation. Studies usually find that utilisation is negatively correlated with distance; higher
rates of consultation are found for those in poorer health, with chronic conditions or
in low income groups; older people use general practices more, with some evidence
of a u-shaped relationship between age and utilisation; and that females use services
more than men [Parkin (1979), Carr-Hill et al. (1996), Windmeijer and Santos Silva
(1997), Pohlmeier and Ulrich (1995)]. Other studies examining the determinants of GP
utilisation are summarised in other chapters in this volume relating to tests of the Gross-
man model [see Grossman (2000)], the effects of price on utilisation [see Zweifel and
Manning (2000)], equity and the utilisation of health services [see Wagstaff and van
Doorslaer (2000)], and supplier induced demand through the effect of GP density and
fee changes on utilisation [see McGuire (2000)1.
For utilisation to reveal patients' preferences the influence of supply on utilisation
needs to be accounted for in empirical work. Econometric studies have recently begun
to tackle this issue directly by attempting to distinguish between utilisation initiated by
the patient and that initiated by the GP in order to make inferences about the patients'
or GP utility functions. Several studies have used the likelihood of a follow up visit
as the dependent variable, to reflect GP decision making more accurately [Tussing and
Wojtowicz (1986), Scott and Shiell (1997a, 1997b)]. More sophisticated econometric
techniques are also being introduced to model the demand for health care as a two-part
decision process: in which the patient decides to visit the GP and the GP decides on
Clh. 22: Economics of General Practice 1183
the amount of services to be consumed [Pohlmeier and Ulrich (1995), Windmeijer and
Santos Silva (1997)]. These studies have used count data and hurdle models to examine
these sequential decisions, with two-stage estimation methods used to account for en-
dogeneity [see Jones (2000)]. However, these models define the first stage (the decision
to visit) as determined by the patient and do not account for the possibility that a pro-
portion of these visits could be follow up visits arranged by the doctor. Nevertheless,
the authors of these studies conclude that empirical approaches that ignore the two-part
nature of the decision can lead to inconsistent econometric estimates and consequent
misinterpretation of results. Other suggested developments include the recognition of
the importance of defining utilisation in terms of episodes of illness, rather than only
numbers of visits [Beland (1982), Santos Silva and Windmeijer (1997)].
3.3. Choice of generalpractitioner
Once they have decided to visit a GP, many patients are faced with a further choice
of which GP to see within the practice. Although some patients prefer to see the same
GP at each visit, the growth of group practice and partnerships in some countries has
arguably eroded the concept of continuity of care [Freeman and Richards (1993)]. The
decision about which GP to visit is also likely to be determined by patients' knowl-
edge of each GPs' characteristics such as their age, sex, and inter-personal skills. The
studies mentioned in Section 3.1 using surveys of satisfaction suggest those factors of
the doctor-patient relationship which are likely to be important. An Australian study
surveyed patients who had visited more than one GP [Veale et al. (1995)]. Respon-
dents were more likely to see more than one GP if they had more visits, were dissat-
isfied at their last visit, were younger, female and highly qualified. Respondents who
experienced good communication at their last visit were less likely to have seen more
than one GP. However, there are few studies that have directly examined choice of GP.
Scott and Vick (1999) and Vick and Scott (1998) used this decision context to exam-
ine the relative importance of the various attributes of the doctor patient relationship.
They modelled GP choice as a function of waiting time and attributes of the doctor-
patient relationship ("being able to talk to the doctor", "information about your health
problem and its treatment", "doctor's explanation of information", "who chooses your
treatment", and "number of days you waited for an appointment"). A discrete choice
experiment was used to test the relative importance of these factors. The results showed
that "being able to talk" was the most important attribute and that "who chooses your
treatment" was the least important. These factors may also influence the choice of gen-
eral practice to register with if patients already have information about the quality of
care provided by specific GPs.
3.4. Choice of treatment
It is assumed that most decisions about treatment are made by the doctor on behalf of
the patient, because the patient is less well informed than the doctor about the technical
1184 A. Scott
relationship between health care and health status. However, informational asymmetry
works two ways, since the patient has more information than the doctor about his or
her preferences, values and circumstances that could influence treatment options. Pa-
tients' preferences for treatment options have been elicited using decision analysis and
stated preference methods, such as Quality Adjusted Life Years (QALYs), and contin-
gent valuation. These methods and results are discussed in detail in Dolan (2000). The
various influences on treatment decisions are discussed in the remainder of this chapter
examining supply side issues.
However, there is some evidence that patients do state their preferences and expec-
tations to GPs about whether they want to be referred or prescribed medication [Arm-
strong et al. (1991)1. In a European study of referrals, Fleming (1992) reported that
pressure from patients about whether they should be referred, as perceived by the GP,
"influenced" between 30 percent and 60 percent of referrals. However, this study re-
lied on self-reports from GPs, rather than actual observation of consultations. Rochaix
(1989), in a theoretical model of physician behaviour, suggested that patients' search for
treatment is a monitoring technique that can lead to an improved agency relationship.
She suggested that only a small number of patients need to be informed for physicians
to behave as better agents.
4. Models of GP behaviour
GPs make many different types of decisions that influence the amount, type and loca-
tion of care received by patients. These include decisions to refer to a specialist or other
health professional, prescribe medication, arrange follow-up, and order tests. Many dif-
ferent theoretical models of GP behaviour have been specified in the literature (some
of which refer to physicians, not specifically GPs). These have often served the pur-
pose of the specific study question, and so are varied in the contents of utility functions,
constraints and predictions.
Common to many models is a basic income-leisure framework. The vast majority of
theoretical models have not modelled GPs as firms (i.e. with profit as a single objec-
tive) but as self-employed individuals supplying their own labour and with their own
objectives such as leisure and the consumption of other goods [Evans (1974), Dionne
and Contandriopoulos (1985)]. Thus, most theoretical models have included income
(or net income) and leisure as the main arguments. Several models also have workload
as an argument, although this is rarely defined in any detail [Evans (1974), Wilensky
and Rossitter (1983), Boardman et al. (1983), Grytten et al. (1995), Dranove (1988),
Lerner and Claxton (1994)]. The inclusion of workload is likely to have arisen directly
from principal-agent theory where "effort" (or the actions of the agent) is assumed to
be negatively related to utility.
Other models have included an "ethical" argument (sometimes referred to as an ethi-
cal constraint on inducement behaviour) which has been used to represent the doctors'
regard for not only professional codes of conduct but also for a more altruistic concern
for patients' welfare [Feldstein (1970), Zwiefel (1981), Dionne and Contandriopoulos
(1985), Woodward and Warren-Boulton (1984)]. Patients' utility or welfare has also
been included [Evans (1974), Farley (1986), Lerner and Claxton (1994)], while others
have included patients' economic well being [Richardson (1981)] and the interests of
society (i.e. a concern for the welfare of groups or populations of individuals) as ar-
guments in the GP's utility function [Blomqvist (1991), Kristiansen (1994)]. These ap-
proaches therefore try to explicitly model the trade-off between the GP's and patient's
objectives, although most do not succeed in doing this empirically.
Many models have also been constructed in the context of supplier-induced demand.
These studies have included "inducement" in the utility function which is assumed to
represent the disutility from inducing unnecessary demand [Evans (1974), Wilensky and
Rossiter (1983), Tussing and Wojtowytz (1986), Grytten et al. (1995)]. By including
inducement as an argument, it is implicitly assumed that doctors know that they are
inducing demand for health care that is "unnecessary".
Several models have also suggested doctors' reputations and status [Dionne and
Contandriopoulos (1985), Kristiansen (1994)], practice characteristics [Boardman et
al. (1983)], intellectual satisfaction [Feldstein (1970), Kristiansen (1994)] and auton-
omy [Kristiansen (1994), Lerner and Claxton (1994)] as arguments in the utility func-
tion.
The majority of models have examined treatment decisions as the main decision
variable (i.e. referrals, prescribing, etc.). As was noted earlier, there are many other
decisions GPs make that have not been modelled, but are nevertheless relevant when
attempting to find out what motivates GPs and when examining other issues apart
from SID, such as labour market behaviour.
The next few sections examine the influence of supply-side factors on GP decisions.
We first examine research about GP referral behaviour, which is perhaps the most im-
portant decision in relation to overall costs in health care systems where GPs are gate-
keepers. The following two sections consider the role of payment systems and the way
GPs organise themselves on the various decisions they make.
5. Explaining variations in referrals
Where GPs are gatekeepers to health systems, they are indirectly responsible for the use
and cost of hospital and other services through their power to refer patients to specialists
and other providers. In the UK, around 5 percent of all GP consultations (or 12 percent
of all patients) result in a referral [Coulter (1992)]. The referral behaviour of GPs has
been the focus of much analysis and debate.
The existence of large variations in rates of referral has been the main impetus for
research into GP decision making. Rates of referral can vary between practices by a
factor of three or four [Wilkin (1992)]. The extent to which such variations remain
unexplained, after accounting for clinical and diagnostic factors has been interpreted
as evidence of inefficiency. Where variations cannot be wholly explained by age, sex,
1186 A. Scott
geography, case-mix, socio-economic factors or patient demand, then the focus shifts
onto the clinical decisions being made by GPs as the key determinants of variation.
Furthermore, where patients with the same illness of the same severity, and with similar
characteristics, are treated differently by GPs, then this shifts the focus further onto non-
clinical factors, such as GPs' own objectives and preferences, knowledge, experience
and uncertainty.
Evidence about these factors has been gathered in studies examining factors influ-
encing aggregate referral rates of general practices, and studies examining the two-part
decision of whether to refer, and if so, which referral destination. Most is from the UK.
Generally, these studies have used patient, provider and health system characteristics as
independent variables. Evidence suggests that patient related factors such as age, sex,
social class and case-mix do not explain the variations between GPs in their rates of re-
ferral [Wilkin (1992)]. Various proxies for the knowledge and experience of GPs (age,
years of experience, postgraduate qualifications) and characteristics of practices have
also been used. The evidence about the influence of GP and practice characteristics on
referral rates is inconclusive. This is not to say that provider influences are unimportant,
but that the variables used in these studies are likely to be poor measures of non-clinical
influences on decision making. Evidence for the influence of health system variables,
such as proximity to hospitals and waiting times is also mixed and scarce. Such factors
are more likely to account for regional and area variations in referral rates, rather than
variations between GPs in the same area where health system variables are the same for
all GPs [Wilkin (1992)].
The paucity of findings from the use of aggregate data has led to research that has
examined the individual GP's decision to refer [Dowie (1983), Newton et al. (1991),
Jones (1992), King et al. (1994), Bailey et al. (1994)]. These have drawn upon psycho-
logical and sociological decision making theories, and have used qualitative methods
to examine the process of decision making. However, economists have also examined
the decision to refer. Healey and Ryan (1992) examined factors in the GPs' utility func-
tion, including those relating to the GP as agent for the patient (patient health status,
provision of information, reassurance, patients time and access costs), and factors re-
lated to GPs' preferences (leisure, prestige and status, income, and attitudes to risk).
Each of the factors was incorporated into a questionnaire as a series of statements with
which a small sample of GPs (68 responded) were asked to agree or disagree with.
Most agreed that health status was important, but there was no consensus about whether
patient accessibility was relevant, and 70% would refer a patient for reasons of reas-
surance. 66% did not consider NHS resource constraints to be a factor, and there was
no clear consensus about attitudes to risk (for example, the risk of being sued) in the
referral decision. Although this was a preliminary study, it did highlight many factors
that are likely to influence variation in referrals that had not been considered explicitly
in aggregate studies, and provides a link with other psychological and sociological the-
ories [Jones (1992)] which emphasize the importance of knowledge, social context, and
environmental factors on the referral decision,
When GP fundholding was introduced in the UK, practices were given a budget for
outpatient and inpatient referrals and were charged for each referral made. 1 This in-
troduced price as a factor potentially influencing the decision to refer. In a review of
the evidence, Gosden and Torgerson (1997) found that fundholding did not change the
level of referral rates. However, these studies were cross-sectional and did not control
adequately for factors influencing referrals other than fundholding, since being a fund-
holder reflects many other factors that could influence referrals. Croxson et al. (1999),
however, were able to control for GP fixed effects using four years of data. They ex-
amined the effect of GP, practice, area and specialty characteristics on admission rates
in one health authority. They found that fundholding was associated with increased ad-
missions in the preparatory year before they became a fundholder (thus increasing their
historically-determined fundholding budget), and with reduced admissions once they
became a fundholder (thus leading to savings on their fundholding budget).
Whynes (1996) examined the importance of price to GPs when selecting referral
destinations. Using data from a survey of fundholders, GPs were asked to rate hospi-
tal performance along 13 dimensions, for their first and second choice hospital for a
general surgery referral. A probit model estimated the effect of the 13 dimensions on
whether the hospital was the GP's first or second choice of referral destination. The
most important factor in the reduced model was "personal confidence in the consul-
tant", followed by "speed of notification of death", and "patient convenience". Distance
from hospital was also a significant predictor of hospital choice. Price of treatment was
not a statistically significant predictor of hospital choice. Similar results were found by
Miller (1997).
6. GP payment systems
We turn next to the impact of more direct financial incentives on the referral and other
behaviour of GPs. The three methods of GP payment: salary, capitation, fee-for-service,
are predicted to have different effects on behaviour, and provide different opportunities
to influence behaviour [Maynard et al. (1996), Rosen (1989), Donaldson and Gerard
(1989)].
Salary payment (where the GP is paid a fixed amount per hour, session, or year),
provides little opportunity to influence specific behaviours, such as prescribing and re-
ferral. Salaried contracts specify the length of time worked and criteria for promotion to
higher grades. Performance might not be explicitly defined, and in practice may relate
to pleasing superiors, which in turn may not necessarily be related to improvements in
efficiency [Maynard et al. (1996)]. GPs have an incentive to minimise effort at work,
by referring to other agencies and prescribing medication to reduce consultation length.
However, salaried payment is administratively simpler for payers and GPs compared to
1 Patients continued to face only out of pocket costs and time prices.
1188 A. Scott
other forms of payment which involve billing and claims for fees and payments. It also
has implications for the autonomy of GPs as under salary they will be employees, rather
than a partner or owner of a small business.
Under capitation payment a GP receives a payment for every patient for which they
provide care, where patients are registered or enrolled with a specific GP or practice. The
payment is ex ante: it does not vary with the amount of care provided although it may be
risk adjusted, so higher payments are received for patients that are likely to visit more
often and generate higher costs, such as the elderly [see van de Ven and Ellis (2000) on
risk adjustment]. Capitation requires either a system of patient registration or that the
GP is responsible for a defined population. Capitation payment provides incentives to
attract and compete for patients, as income is related directly to the number of patients
under the care of GPs. However, it also provides incentives to attract low cost patients
(cream-skim) and to minimise effort in consultations by referring and prescribing to
reduce workload [Iversen and Luras (1997)].
Fee for service payment links income directly to the volume of services provided and
has been studied most widely in the context of supplier induced demand [see McGuire
(2000)]. This can lead to problems in controlling expenditure (which has been attempted
through fee freezes, clawbacks, and alterations to fee schedules), and an incentive to in-
crease volume that is not necessarily compatible with improving patient outcomes. The
extent to which fee for service leads to high volumes of service is partly dependent on
the level of fees, although attempts to reduce fee levels have not influenced expenditure
as there is evidence that physicians have increased volume to compensate [see McGuire
(2000)].
There are also mixed systems of payment. In the UK, GPs are paid via annual al-
lowances, capitation, fee for service, and target payments, where payment is related
to achieving targets for population coverage for cervical screening and childhood im-
munisation [Hughes (1993), Silcock and Ratcliffe (1996)]. GPs in Denmark, Norway,
Austria, Ireland and Spain are also paid using a mixed system of remuneration [Gervas
et al. (1994), Kristiansen and Mooney (1993a)]. Recent reforms in the UK have given
the option to GPs to be paid by salary, in addition to the current mixed payment system.
Many empirical studies are based on natural experiments and are therefore oppor-
tunistic, and few studies have been able to control adequately for the many other differ-
ences between GPs that may influence their behaviour. The most notable omissions are
patient characteristics, case-mix and GP characteristics [Scott and Hall (1995)]. These
omissions are particularly relevant from a theoretical perspective.
The empirical evidence on the effects of different payment systems in general practice
has been reviewed in several papers [Donaldson and Gerard (1989), Scott and Hall
(1995), Gosden et al. (1999)]. Gosden et al. (1999) concluded that salaried payment
leads to lower levels of tests, referrals and patient throughput compared to FFS and
capitation payment. The authors argue that this is consistent with the hypothesis that
salaried doctors cannot increase income but can minimise effort. They found only one
study in which doctors were randomised physicians to different payment mechanisms.
Hickson et al. (1987) randomised physicians to either FFS or salaried payment and
reported that FFS payment led to greater use of services than salaried payment.
One of the more rigorous studies of GP payment was conducted in Denmark [Kras-
nik et al. (1990)]. A before and after design was supplemented by a control group of
GPs who were not subjected to the change in remuneration and who were followed up
through the same period. Data were collected at three points in time (one before and
two after). The authors evaluated the introduction of a mixed system of fee for service
and capitation, introduced for all GPs in Copenhagen city. This was compared with
the standard capitation system used to pay GPs in Copenhagen county. This study pro-
vided strong evidence that changing from capitation payment to FFS payment presented
income-generating opportunities of which GPs took advantage through increasing ser-
vice intensity. Compared to pure capitation payment, diagnostic and curative services
increased dramatically over the period, and prescribing and referral levels fell. The au-
thors concluded that GPs appeared to be doing more work themselves, rather than re-
ferring on to specialists or prescribing medication (both of which would be expected in
a capitation system where GPs may minimise effort).
In the United States, Steams et al. (1992), using a before and after design, found a
reduction in the rate of hospitalisations due to a change from FFS to capitation payment.
The authors showed that other potential confounding factors did not change over the
study period. In Canada, however, Hutchinson et al. (1994) used a controlled before
and after study to demonstrate no difference in hospital utilisation between capitated
and FFS practices.
Three studies by Kristiansen and Hjortdahl (1992), Kristiansen and Mooney (1993b),
and Kristiansen and Holtedahl (1993) used a cross sectional survey of GPs that provided
information on consecutive consultations and thus were able to collect data on patients.
They showed that, while controlling for GP and patient characteristics, GPs paid by FFS
were more likely to order certain tests, provide shorter consultations and visit patients at
home. The effects of remuneration were small compared to other factors such as patient
age and sex (proxies for patients' health status).
Consultation-based data were also used by Scott and Shiell (1997b) to examine a
change in the fee schedule where fees were paid for four different lengths of consulta-
tion, to fees being related more closely to the content and complexity of consultations.
The fee schedule was changed because time-based payment provided incentives to in-
crease prescribing, follow up, and test ordering. There was, however, no evidence that
the GPs likelihood of making these decisions had changed due to the introduction of
content-based payment.
The studies by Kristiansen and Scott and Shiell were notable in that they modelled
the actual clinical decision in the context of data gathered directly from consultations,
linked to GP and practice characteristics, rather than self report data on past visits col-
lected via population surveys of health care utilisation.
The bulk of research has been on the effects of payment systems on behaviour of
doctors in post and has ignored the implications for the medical labour market. The
focus has been on the marginal incentives for different types of activity (referrals, pre-
1190 A. Scott
scribing, etc). As well as influencing clinical decisions, payment systems are major
determinants of professional job satisfaction and morale, and can therefore influence
supply and labour market behaviour. For example, recent payment reform in UK gen-
eral practice, which has introduced more flexible payment systems and a wider variety
of NHS contract has been introduced to improve professional morale, job satisfaction,
recruitment and retention. The argument was that the supply of doctors to general prac-
tice depended both on the level of payment and on the method. There has been very little
research on the influence of different payment systems on labour market participation
in general practice.
Such research should also include recognition of the role of other social science dis-
ciplines to help enrich economic models. Several studies have used sociological and
psychological concepts and theories in the study of physician remuneration. Giacomini
et al. (1996), emphasised the social context of financial incentives and their use as a
method of communication and signalling. Qualitative methods were used to analyse
a range of funding changes in the Canadian health care system. The incorporation of
psychological theories into the economic analysis of financial incentives and motiva-
tion has also been led by Frey (1997), who uses the concepts of intrinsic and extrinsic
motivation in the analysis of individual's behaviour, The "carrot and stick" approach to
payment systems as espoused in principal-agent theory should therefore be strengthened
by the consideration of other contextual factors and theories that influence the effects of
payment systems.
7. GPs as firms
7.1. Partnershipsand group practice
Much of the analysis of payment systems and factors influencing the behaviour of
GPs is couched in terms of influences on the behaviour of the individual GP using a
basic income-leisure framework. However, where GPs practise in partnerships, mod-
els of their behaviour need to take account of the incentives within partnerships. Many
decisions are made by the group rather than the individual. Furthermore, the effects of
external incentives and regulation on individual GP behaviour may be modified by the
economic and professional relationships between partners.
Several economic models of group practice have been specified. These are based on
the trade-off between efficiency and risk-spreading, as specified in agency models of
worker and team behaviour [Alchian and Demsetz (1972), Holmstrom (1982), Holm-
strom and Milgrom (1994)]. There is an incentive for partners to shirk and free ride on
the effort of others if revenues are shared amongst partners, or where joint output is the
only indicator of inputs [Wilson and Bartlett (1994)]. Although such a system spreads
risks it reduces efficiency. If compensation of an individual physician is directly linked
to output (and therefore effort) there will be incentives to be efficient but risk is not
shared. Crucial to these models has therefore been the extent that risk aversion is a de-
terminant of internal compensation method and of the behaviour (effort) of physicians.
For example, Gaynor and Gertler (1995) examined the relationship between the de-
gree of risk sharing, compensation and effort in primary care physicians in the US in
medical group practice. They specify a model where demand is uncertain, and where
physicians choose "effort" to maximise utility in response to the incentives in the firm's
compensation structure. The utility maximising level of effort is where the marginal
revenue product of effort is equal to its marginal disutility. They go on to derive com-
parative static results of the effect of changes in internal compensation on the number
of patients seen (defined as "effort"). They also examine the effect of risk aversion on
choice of compensation structure. The empirical results found that a stronger link be-
tween compensation and productivity leads to more office visits per week (effort), and
that the greater the risk aversion of physicians the less strongly the compensation struc-
ture is related to productivity.
Bradford (1995) suggested that malpractice risk helped to explain why partnerships
may dominate solo practice. He argued that when risk premia are paid by the group and
shared equally amongst partners, then increased malpractice risk will increase shirking.
This will lead to increased administrative costs within a partnership to reduce shirking,
thus placing partnerships at a competitive disadvantage. Therefore partnerships become
less probable as malpractice risk increases. The predictions were rejected for primary
care but confirmed for surgical specialities where the financial consequences of being
sued are much greater. Encinosa et al. (1997) examined why compensation may not
always be linked to productivity in medical groups, using the sociological concept of
"group norms" incorporated into an economic framework of risk sharing and multi-
tasking. Group norms are defined as the social interactions resulting from comparisons
of effort and pay within groups. They demonstrate that group income and effort norms
make small groups more likely to adopt equal sharing rules than large groups, and that
risk aversion and multi-tasking make equal sharing more likely in large groups. Using
the same dataset as Gaynor and Gertler (1995), they find evidence that group norms do
influence choice of compensation method, in addition to the usual factors analysed in
principal-agent models (risk aversion and multitasking).
7.2. Vertical integration, the boundariesof thefirm and the balance of care
The decision to refer also includes the decision as to whether to treat the patient in the
practice. This can involve organisational decisions as to whether to provide services
in primary care rather than secondary care. This has led to debate about the extent
to which GPs should provide a wider range of services which can be competently be
performed by a GP, or a GP with special interests, while maintaining the gatekeeping
role of the GP. 2
2 Clearly, this is less of an issue in countries where there is a reduced or no gatekeeping role, and where
medical groups (e.g. HMOs) comprise physicians from a range of specialties, therefore providing a much
wider range of services and higher degree of integration [see Glied (2000)].
1192 A. Scott
For example, there has been a considerable policy emphasis and debate in the UK
about a "primary care led" NHS, which involves treating patients closer to their own
homes (i.e. in general practice or small community hospitals) and GPs having more
"control" over the whole episode of patient care. This has been encouraged by financial
incentives in the 1990 GP Contract, where GPs were paid to undertake minor surgery
(amongst other things), and by the GP fundholding scheme, which made GPs purchasers
of care for their patients, thereby giving GPs more control over the location and quality
of care. The shifting of services from secondary to primary care then raises questions
about the efficiency of such shifts, and the reasons why general practices decide to
integrate vertically.
There have been two main reviews of the literature in the UK context, each examining
the efficiency of shifts in the balance of care [Godber et al. (1997), Scott (1996)]. Both
of these reviews examine the changes in costs and changes in benefits to patients of the
shift of service provision. The evidence was scarce and mixed in terms of methodologi-
cal quality and results. For example, Godber et al. reviewed 23 studies, of which 11 were
randomised controlled trials. Ten studies examined changes in costs only. The remain-
der of studies examined multiple outcomes (clinical outcomes, waiting times, access
and patient satisfaction), thus making it difficult to examine the overall welfare effects
of shifts in the balance of care. Other methodological problems were also highlighted
(limited specification of costs, limited sensitivity analysis, uncontrolled differences in
case mix and severity, and generalisability). As a result, no clear or consistent conclu-
sions about the appropriate balance of care could be made.
Other literature has examined the possible economic determinants of vertical inte-
gration. These include technological economies, transactional economies, and market
imperfections [Perry (1989)]. Technological economies arise when integrated process
is less costly. Transactional economies refer to the transactions costs associated with the
process of exchange, where the exchange is based on a negotiated contract [Williamson
(1989)1. Applied to general practice, vertical integration may therefore reduce the costs
of negotiating and monitoring the contract with the hospital.
Market imperfections are also hypothesised to determine the extent of vertical inte-
gration. Supply assurance implies the existence of some market imperfection which, in
the absence of vertical integration, suggests that the firm could not obtain the quantity
of inputs it wished to purchase. For example, one reason why firms may not be able
to purchase the quantity of inputs they want to is inflexible prices due to regulation or
monopoly. Information asymmetry is another source of market imperfection that may
lead to vertical integration, where firms may vertically integrate to monitor more closely
the production process.
GPs may wish to enhance their autonomy and control over the "production" process
or to ensure better access, reduced waiting times and reduced travel costs for patients.
Supply assurance as a reason for vertical integration may be valued because of the ben-
efits accruing to patients, such as access and reduced waiting times. Similar arguments
may also apply to vertical integration for the purposes of better monitoring of outcomes
and quality. However, untangling these issues from the GP's desire for increased auton-
omy and increased future income has still to be addressed (e.g. investing in providing
more services so that the capital value of the practice is increased in preparation for
retirement).
It may also be important to distinguish between the integration of finance and pro-
vision. The preceding discussion has been concerned mainly with the integration of
provision. However, the integration of finance may be an important precursor to the
integration of provision. For example, GPs who chose to become fundholders were es-
sentially making a decision to vertically integrate financially, i.e. hold a new budget with
which to but hospital services and drugs. The decision to vertically integrate in terms of
provision is a separate decision and may reveal different aspects of the utility function.
However, there has been little empirical evidence of the determinants of the deci-
sion to vertically integrate by GPs in a gatekeeping system, although even in non-gate-
keeping systems, the evidence is scarce and inconclusive. Mitchell and Sass (1995)
examined physician ownership of ancillary facilities. It suggested three hypotheses for
vertical integration; altruism, indirect demand inducement, and quality assurance. Al-
truism emphasises the concern of physicians for ensuring access to services for patients,
but the authors discount this given the pattern of profits and location of physician-owned
facilities. Indirect demand inducement highlights the role of physician profits as the
main factor influencing ownership. Quality assurance arises from the asymmetry of in-
formation rationale for vertical integration, in that physicians can better monitor the
activities of ancillary services if they own them. The authors found no support for any
of the hypotheses.
Other limited evidence from the UK examined the decisions of GPs to become fund-
holders, which implies greater control over services [Ennew et al. (1998), Robinson
and Hayter (1995)]. Based on interviews with GPs, some mentioned the opportunity
to increase income, while others objected to possibility of personal financial gain. Sev-
eral GPs welcomed the opportunity to spend more time managing the activities of the
practices. Autonomy was also mentioned in terms of power, the freedom to refer and ex-
erting control over hospital doctors. Patients' interests were also mentioned as reasons
to become fundholding. These ranged from general statements about quality, through
to waiting times, patient choice, and access. Interestingly, no mention was made of pa-
tients' health status.
8. Future research
This chapter has attempted to summarise the main economic questions, theories and
empirical evidence in general practice. Focussing on GPs' role as gatekeepers, a range
of research by economists and non-economists has been reviewed.
To date, there has been little research by economists into general practice, even
though primary care is being recognised as the mainstay of many health care systems in
developed countries. It is difficult to prioritise a research agenda on the basis of specific
areas of work, since most areas in this chapter still require major research effort. Hy-
potheses that remain to be conclusively tested, however, are whether gatekeeping and
1194 A. Scott
shifting the balance of care towards primary care lead to a more efficient health care
system.
It is perhaps more fruitful to examine some broad research directions that economists
should attempt to move towards. Two central issues are the objective functions of pa-
tients and GPs. Much of GP care can be broadly labelled "social" care, rather than medi-
cal care. Many consultations do not lead to improvements in health, but are nevertheless
of value to patients. For example, the management of chronic disease does not always
involve "curing" but "caring". The longitudinal nature of the GP-patient relationship
is highly valued by GPs and many patients, yet does not feature in empirical economic
analyses. In their models and empirical work economists should be attempting to clarify
(and quantify) what it is that patients and society value. Eliciting preferences for factors
other than health outcomes is therefore an important area of research. It is difficult to
design regulation and payment systems unless it is known what objectives they should
be trying to meet.
Similarly, it is important to enrich the economic model of the GP who is assumed to
maximise income and leisure. This has happened in several theoretical models that in-
clude "patients interests" in the GPs' utility function, but has not been followed through
to empirical work, primarily because such broader factors are more difficult to measure.
Although much evidence suggests that GPs do respond to financial incentives, there are
many more factors that determine behaviour that do not fit easily into the economic
model. This does not mean they should be ignored by economists. For example, recent
changes to GP remuneration and out-of-hours arrangements in the UK were introduced
to improve the non-pecuniary aspects of GPs working lives. Similarly, GPs working as
firms in partnerships have important implications for GP behaviour that have not yet
been addressed.
References
Alchian, A.A., and H. Demsetz (1972), "Production, information costs and economic organisation", American
Economic Review 62:777-795.
Armstrong, D., J. Fry and P. Armstrong (1991), "Doctors' perceptions of pressure from patients for referral",
British Medical Journal 302:1186-1188.
Arrighetti, A., R. Bachmann and S. Deakin (1997), "Contract law, social norms and inter-firm co-operation",
Cambridge Journal of Economics 21:171-198.
Bailey, J., N. King and P. Newton (1994), "Analysing general practitioners' referral decisions II: Do high and
low referrers differ in factors influencing their referral decisions?", Family Practice 11:9-14.
Baker, R., and J. Streatfield (1995). "What type of general practice do patients prefer. Exploration of practice
characteristics influencing patient satisfaction", British Journal of General Practice 45:654-659.
Beland, E (1982), "The utilisation of health services. Sequences of visits to general practitioners", Social
Science and Medicine 16:2065-2072.
Blomqvist, A. (1991), "The doctor as double agent: Information asymmetry, health insurance, and medical
care", Journal of Health Economics 10:411-432.
Boardman, A.E., B. Dowd, J.M. Eisenberg and V. Williams (1983), "A model of physicians' practice attributes
determination", Journal of Health Economics 2:259-268.
Boerma, W.G.W., P.P Groenewegen and J. van der Zee (1998), "General practice in rural and urban areas:
The range of curative services", Social Science and Medicine 47:445-453.
Bradford, W.D. (1995), "Solo versus group practice in the medical profession. The influence of malpractice
risk", Health Economics 4:95-112.
Burchell, B., and E Wilkinson (1997), "Trust, business relationships and the contractual environment", Cam-
bridge Journal of Economics 21:217-238.
Burke, W., et al. (1993), "Training generalist physicians: Structural elements of the curriculum", Journal of
General Internal Medicine 9:S23-S30.
Calnan, M., P.P. Groenewegen and J. Hutten (1992), "Professional reimbursement and management of time
in general practice. An international comparison", Social Science and Medicine 35:209-216.
Calnan, M., V. Katsoyiannopolous, V.K. Ouchavev, R. Prokhorskas, H. Ramic and S. Williams (1994a),
"Major determinants of consumer satisfaction with primary care in different countries", Family Practice
11:468-478.
Calnan, M., J. Coyle and S. Williams (1994b), "Changing perceptions of general practitioner care", European
Campbell, J.L. (1994), "General practitioner appointment systems, patient satisfaction, and use of accident
and emergency services. A study in one geographical area", Family Practice 11:438-435.
Carr-Hill, R.A., N. Rice and M. Roland (1996), "Socioeconomic determinants of rates of consultation in
general practice based on fourth national morbidity survey of general practices", British Medical Journal
312:1008-1013.
Casson, M., and N. Wadeson (1997), "Communication costs and the boundaries of the firm", Paper presented
to a Conference to celebrate the work of Brian Loasby, University of Stirling.
Coulter, A. (1992), "The interface between primary and secondary care", in: M. Roland and A. Coulter,
Hospital Referrals, Oxford General Practice Series 22 (Oxford University Press, Oxford).
Coulter, A., and J. Bradlow (1993), "Effect of NHS reforms on general practitioners' referral patterns", British
Medical Journal 306:433-437.
Croxson, B., C. Propper and A. Perkins (1999), "Do doctors respond to financial incetives? UK family doctors
and the GP fundholder scheme", Paper presented to the 2nd International Health Economics Association
Conference, Rotterdam.
Dietrich, A.J., and K.I. Marton (1982), "Does continuous care from a physician make a difference?", Journal
of Family Practice 15:929-937.
Dionne, G., and A. Contandriopoulos (1985), "Doctors and their workshops", Journal of Health Economics
4:21-33.
Dixon, P., H. Gravelle, R. Carr-Hill and J. Posnett (1997), "Patient movements and patient choice", Report
for National Health Service Executive (York Health Economics Consortium, York).
Dolan, P. (2000), "The measurement of health-related quality of life for use in resource allocation decisions",
in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 32.
Donaldson, C., and K. Gerard (1989), "Paying general practitioners: Shedding light on the review of health
services", Journal of the Royal College of General Practitioners 39:114-117.
Dowie, R. (1983), General Practitioners and Consultants: A Study of Outpatient Referrals (Kings Fund, Lon-
don).
Dranove, D. (1988), "Demand inducement and the physician-patient relationship", Economic Inquiry
26:281-298.
Encinosa, W.E., M. Gaynor and J.B. Rebitzer (1997), "The sociology of groups and the economics of incen-
tives: Theory and evidence on compensation systems", Working Paper 5953 (National Bureau of Economic
Research).
Ennew, C., T. Feighan and D. Whynes (1998), "Entrepreneurial activity in the public sector: Evidence form
UK primary care", in: P. Taylor-Gooby, ed., Choice and Public Policy. The Limits to Welfare Markets
(Macmillan, London).
Evans, R.G. (1974), "Supplier-induced demand: some empirical evidence and implications", in: M. Perlman,
ed., The Economics of Health and Medical Care (International Economics Association, Macmillan).
1196 A. Scott
Farley, P.J. (1986), "Theories of the price and quantity of physician services. A synthesis and critique", Journal
of Health Economics 5:315-333.
Feldstein, M.S. (1970), "The rising price of physicians' services", Review of Economics and Statistics
(May):121-133.
Fleming, D. (1992), "The interface between general practice and secondary care in Europe and North Amer-
ica", in: M. Roland and A. Coulter, Hospital Referrals, Oxford General Practice Series 22 (Oxford Univer-
sity Press, Oxford).
Franks, P., C.M. Clancy and P.A. Nutting (1992), "Gatekeeping revisited - protecting patients from overtreat-
ment", New England Journal of Medicine 327:424-429.
Freeman, G.K. (1984), "Continuity of care in general practice: A review and critique", Family Practice 1:245-
252.
Freeman, G.K., and S.C. Richards (1993), "Is personal continuity of care compatible with free choice of
doctor? Patients' views on seeing the same doctor", British Journal of General Practice 43:493-497.
Frey, B.S. (1997), "On the relationship between intrinsic and extrinsic work motivation", International Journal
of Industrial Organisation 15:427-439.
Fry, J., and J. Horder (1994), Primary Health Care in an International Context (Nuffield Provincial Hospitals
Trust, London).
Gaynor, M. (1989), "Competition within the firm: theory plus some evidence from medical group practice",
Rand Journal of Economics 20:59-76.
Gaynor, M., and P. Gertler (1995), "Moral hazard and risk spreading in partnerships", Rand Journal of Eco-
nomics 26:591-613.
Gaynor, M., and M.V. Pauly (1990), "Compensation and productive efficiency in partnerships: Evidence from
medical group practice", Journal of Political Economy 98:544-573.
Gervas, J., M.P. Fernandez and B.H. Starfield (1994), "Primary care, financing and gatekeeping in Western
Europe", Family Practice 11:307-317.
Getzen, T. (1984), "A 'brand name' firm theory of medical group practice", Journal of Industrial Economics
33:199-215.
Giacomini, M., J. Lomas, J. Hurley, V. Bhatia and L. Goldsmith (1996), "The devil in the details: Some
conclusions about how funding changes translate into financial incentives in the Canadian health system",
Working Paper Series 96-14 (McMaster University Centre for Health Economics and Policy Analysis)
September.
Glied, S. (2000), "Managed care", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics
(Elsevier, Amsterdam) Chapter 13.
Godber, E., R. Robinson and A. Steiner (1997), "Economic evaluation and the shifting balance towards pri-
maty care: Definitions, evidence and methodological issues", Health Economics 6:275-294.
Goddard, M., and R. Manion (1998), "From competition to co-operation: New economic relationships in the
national health service", Health Economics 7:105-120.
Gosden, T., and D. Torgerson (1997), "The effect of fundholding on prescribing and referral costs: A review
of the evidence", Health Policy 40:103-114.
Gosden, T., L. Pedersen and D. Torgerson (1999), "How should we pay doctors? A systematic review of salary
payments and their effect on doctor behaviour", Quarterly Journal of Medicine 92:47-55.
Grogan, S., M. Conner, D. Willits and P. Norman (1995), "Development of a questionnaire to measure pa-
tients' satisfaction with general practitioners' services", British Journal of General Practice 45:525-529.
Grossman, M, (2000), "The human capital model", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of
Grytten, J., E Carlsen and R. Sorensen (1995), "Supplier inducement in a public health care system", Journal
Healey, A., and M. Ryan (1992), "Factors influencing general practitioners' decisions to refer: A preliminary
step towards explaining variations in GP referrals", Discussion Paper 06/92 (Health Economics Research
Unit, University of Aberdeen).
Hickson, G.B., W.A. Altemeier and J.M. Perrin (1987), "Physician reimbursement by salary of fee-for-
service: Effect on physician practice behaviour in a randomized prospective study", Pediatrics 80:344-350.
Hjortdahl, P., and C.F Borchgrevink (1991), "Continuity of care: Influence of general practitioners' knowl-
edge about their patients on use of resources in consultations", 303:1181-1184.
Hjortdahl, P., and E. Laerum (1992), "Continuity of care in general practice: Effect on patient satisfaction",
Holmstrom, B. (1982), "Moral hazard in teams", Bell Journal of Economics 13:324-340.
Holmstrom, B., and P. Milgrom (1994), "The firm as an incentive system", American Economic Review
84:972-991.
Howie, J.G.R., A.M.D. Porter, D.J. Heaney and J.L. Hopton (1991), "Long to short consultation ratio: A proxy
measure for quality of care for general practice", British Journal of General Practice 41:48-54.
Hughes, D. (1993), "General practitioners and the new contract: Promoting better health through financial
incentives", Health Policy 25:39-50.
Hutchinson, B., S. Birch, J. Hurley, J. Lomas and F. Stratford-Devai (1994), "Effect of a financial incentive
to reduce hospital utilisation in capitated primary care practice", Paper 94-2 (Centre for Health Economics
and Policy Analysis, McMaster University).
Iversen, I., and H. Luras (1997), "The effect of payment system on GPs' referral decisions", Paper presented
to the Nordic Health Economists' Study Group meeting, Helsinki.
Jones, R. (1992), "Decision making and hospital referrals", in: M. Roland and A. Coulter, eds., Hospital
Referrals, Oxford General Practice Series 22 (Oxford University Press, Oxford).
Kaplan, S.H., S. Greenfield and J.E. Ware (1989), "Impact of the doctor-patient relationship on the outcomes
of chronic disease", in: M. Stewart and D. Roter, eds., Communicating With Medical Patients (Sage Publi-
cations, Newbury Park, CA).
Khayat, K., and B. Salter (1994), "Patient satisfaction surveys as a market research tool for general practices",
British Journal of General Practice 44:215-219.
King, N., J. Bailey and P. Newton (1994), "Analysing general practitioners' referral decisions I: Developing
an analytical framework", Family Practice 11:3-8.
Krasnik, A., P.P. Groenewegen, P.A. Pedersen, P Scholten, G. Mooney, A. Gottschau, H.A. Flierman and
M.T. Damsgaard (1990), "Changing remuneration systems: Effects on activity in general practice", British
Kristiansen, I.S. (1994), "What is in the doctor's utility function? A theoretical and empirical investigation
into what influences doctors' decision making", Ph.D. Thesis (University of Tromso).
Kristiansen, I.S., and P. Hjortdahl (1992), "The general practitioner and laboratory utilization: Why does it
vary?", Family Practice 9:22-27.
Kristiansen, I.S., and K. Holtedahl (1993), "The effect of the remuneration system on the general practitioner's
choice between surgery consultations and home visits", Journal of Epidemiology and Community Health
47:481-484.
Kristiansen, I.S., and G. Mooney (1993a), "Remuneration of GP services: time for more explicit objectives?
A review of the systems in five industrialised countries", Health Policy 24:203-212.
Kristiansen, I.S., and G. Mooney (1993b), 'The general practitioner's use of time: Is it influenced by the
remuneration system?", Social Science and Medicine 37:393-399.
Lerner, C., and K. Claxton (1994), "Modelling the behaviour of general practitioners. A theoretical foundation
for studies of fundholding", Discussion Paper 116 (Centre for Health Economics, University of York).
Lyons, B., and J. Mehta (1997), "Contracts, opportunism and trust: self interest and social orientation", Cam-
bridge Journal of Economics 21:239-258.
Martin, D., P. Diehr, K.E Price and W.C. Richardson (1989), "Effect of a gatekeeper plan on health services
use and charges: A randomized trial", American Journal of Public Health 79:1628-1632.
Maynard, A., M. Marinker and D. Pereira Gray (1996), "The doctor, the patient and their contract mII. Alter-
native contracts: are they viable?", British Medical Journal 292:1438-1440.
1198 A. Scott
Miller, P. (1997), "Are GP fundholders wasting money?", Health Service Journal (March 6):28-29.
Mitchell, J.M., and T.R. Sass (1995), "Physician ownership of ancillary facilities; Indirect inducement or
quality assurance?", Journal of Health Economics 14:263-290.
McGuire, T.G. (2000), "Physician agency", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health
Mooney, G., S. Jan, M. Ryan, K. Bruggerman and K. Alexander (1999), "What the community prefers, what
it values, what health care it wants. A survey of South Australians", Final Report to South Australia Health
Commission (Department of Public Health and Community Medicine, Sydney).
Nelson, P. (1970), "Information and consumer behaviour", Journal of Political Economy 78:311-329.
Newton, J., V. Hayes and A. Hutchinson (1991), "Factors influencing general practitioners' referral deci-
sions", Family Practice 8(4):308-313.
Ong, L.M., J.C. Haes, A.M. Hoos and F.B. Lammes (1995), "Doctor-patient communication: A review of the
literature", Social Science and Medicine 40:903-918.
Parkin, D. (1979), "Distance as an influence on demand in general practice", Epidemiology and Community
Health 33:96-99.
Perry, M.K. (1989), "Vertical integration: Determination and effects", in: R. Schmalensee and R. Willig, eds.,
Handbook of Industrial Organisation, Vol. I (North Holland, Amsterdam).
Petchy, R. (1995), "General practitioner fundholding: weighing the evidence", The Lancet 346:1139-1142.
Pohlmeier, W., and V. Ulrich (1995), "An econometric model of the two-part decision process in the demand
for health care", Journal of Human Resources 30:339-361.
Richardson, J. (1981), "The inducement hypothesis: that doctors generate demand for their own services",
in: J. Van der Gaag and M. Perlman, eds., Health, Economics and Health Economics (Amsterdam, North-
Holland).
Robinson, R., and P. Hayter (1995), "Why do GPs choose not to apply for fundholding?", Occasional Paper
(Institute for Health Policy Studies, University of Southampton).
Rochaix, L. (1989), "Information asymmetry and search in the market for physician services", Journal of
Rogers, J., and P. Crtis (1980), "The concept and measurement of continuity in primary care", American
Roos, N. (1979), "Who should do the surgery? Tonsillectomy and adenoidectomy in one Canadian province",
Inquiry 16:73-83.
Rosen, B. (1989), "Professional reimbursement and professional behaviour: Emerging issues and research
challenges", Social Science and Medicine 29:455-462.
Rowlands, O., N. Singleton, J. Maher and V. Higgins (1997), Living in Britain. Results from the 1995 General
Household Survey (Office for National Statistics, The Stationery Office, London).
Rubenstein, A., and M. Yaari (1983), "Repeated insurance contracts and moral hazard", Journal of Economic
Theory 30:74-97.
Ryan, M., E. Mclntosh and P. Shackley (1998), "Using conjoint analysis to assess consumer preferences in
primary care: An application to the patient health card", Health Expectations 1:117-129.
Salisbury, C.J. (1989), "How do people choose their doctor?", British Medical Journal 299:608-610.
Santos Silva, J., and F. Windmeijer (1997), "Stopped sum models of health care demand", Paper presented at
the 6th European Workshop on Econometrics and Health Economics, Lisbon.
Scott, A. (1996), "Primary or Secondary care? What can economics contribute to evaluation at the interface?",
Journal of Public Health Medicine 18(1):19-26.
Scott, A., and J. Hall (1995), "Evaluating the effects of GP remuneration: Problems and prospects", Health
Policy 31:183-195.
Scott, A., and A. Shiell (1997a), "Analysing the effect of competition on general practitioners' behaviour
using a multilevel modelling framework", Health Economics 6:577-588.
Scott, A., and A. Sliell (1997b), "Do fee descriptors influence treatment choices in general practice? A mul-
tilevel discrete choice model", Journal of Health Economics 16:323-342.
Scott, A., and S. Vick (1999), "Patients, doctors and contracts: An application of principal-agent theory to the
doctor-patient relationship", Scottish Journal of Political Economy 46:111-134.
Scott, A., S. Watson, S. Ross and N. Torrance (1998), "Eliciting patients' preferences for different models of
out of hours care", Discussion Paper No. 04/98 (Health Economics Research Unit, University of Aberdeen).
Silcock, J., and J. Ratcliffe (1996), "The 1990 GP Contract: meeting needs?", Health Policy 36:199-207.
Smith, C.H., and D. Armstrong (1989), "Comparison of criteria derived by government and patients for eval-
uating general practitioner services", British Medical Journal 299:494-496.
Starfield, B. (1994), "Is primary care essential?", The Lancet 344:1129-1133.
Steams, S.C., B.L. Wolfe and D.A. Kindig (1992), "Physician responses to fee-for-service and capitation
payment", Inquiry 29:416-425.
Steven, I.D., and R.M. Douglas (1988), "Dissatisfaction in general practice: What do patients really want?",
Medical Journal of Australia 148:281-282.
Stewart, M.A. (1995), "Effective physician communication and health outcomes: A review", Canadian Med-
ical Association Journal 152:1423-1433.
Tossing, A.D., and M.A. Wojtowycz (1986), "Physician-induced demand by Irish GPs", Social Science and
Medicine 23(9):851-860.
van de Kar, A., A. Knottnerus, R. Meertens, V. Dubois and G. Kok (1992), "Why do patients consult the
general practitioner? Determinants of their decision", British Journal of General Practice 42:313-316.
van de Ven, W.P.M.M. (1996), "Market oriented health reforms: Trends and future options", Social Science
and Medicine 43:655-666.
van de Ven, W.P.M.M., and R.P. Ellis (2000), "Risk adjustment in competitive health plan markets", in:
A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 14.
Veale, B.M., J. McCallum, D.C. Saltman, J. Lonergan, Y.J. Wadsworth and R.M. Douglas (1995), "Consumer
use of multiple general practitioners: An Australian epidemiological study", Family Practice 12:303-308.
Vick, S., and A. Scott (1998), "Agency in health care. Examining patients' preferences for attributes of the
doctor-patient relationship", Journal of Health Economics 17:511-644.
Wagstaff, A., and E.K.A. van Doorslaer (2000), "Equity in health care finance and delivery", in: A.J. Culyer
and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 34.
Wartman, S.A., L.L. Morlock, EE. Malitz et al. (1983), "Patient understanding and satisfaction as predictors
of compliance", Medical Care 21:886-891.
Webb, S., and M. Lloyd (1994), "Prescribing and referral in general practice: A study of patients' expectations
and doctors' actions", British Journal of General Practice 44:165-169.
Weiss, L.J., and J. Blustein (1996), "Faithful patients: the effect of long-term physician-patient relationships
on the costs and use of health care by older Americans", American Journal of Public Health 86:1742-1747.
Wensing, M., R. Grol, P. van Montfort and A. Smits (1996), "Indicators of the quality of general practice care
of patients with chronic illness: A step towards the real involvement of patients in the assessment of the
quality of care", Quality in Health Care 5:73-80.
Whynes, D.K. (1996), "General practitioners' choice of referral destination: A probit analysis", Managerial
and Decision Economics 17:587-594.
Whynes, D.K., D.L. Baines and K.H. Tolley (1997), "Prescribing costs in UK general practice: The impact of
hard budget constraints", Applied Economics 29:393-399.
Wilensky, G.R., and L.F. Rossiter (1983), "The relative importance of physician induced demand in the de-
mand for medical care", Milbank Memorial Fund Quarterly 61(2):252-277.
Wilkin, D. (1992), "Patterns of referral: Explaining variation", in: M. Roland and A. Coulter, eds., Hospital
Referrals, Oxford General Practice Series 22 (Oxford University Press, Oxford).
Williams, S.J., and M. Calnan (1991), "Key determinants of consumer satisfaction with general practice",
Family Practice 8:237-242.
Williamson, O.E. (1993), "Opportunism and its critics", Managerial and Decision Economics 14:97-107.
Williamson, O.E. (1989), "Transactions cost economics", in: R. Schmalensee and R. Willig, eds., Handbook
of Industrial Organisation, Vol. I (North Holland, Amsterdam).
1200 A. Scott
Wilson, D., and W. Bartlett (1994), "Partnerships and incentives in health care: The case of general practice
in the UK", Paper presented at the Health Economist's Study Group Meeting, London School of Hygeine
and Tropical Medicine, January.
Windmeijer, F., and J. Santos Silva (1997), "Endogeneity in count data models: An application to demand for
health care", Journal of Applied Econometrics 12:281-294.
Woodward, R.S., and E Warren-Boulton (1984), "Considering the effects of financial incentives and profes-
sional ethics on appropriate medical care", Journal of Health Economics 3:223-237.
Zweifel, P. (1981), "Supplier induced demand in a model of physician behaviour", in: J. van der Gaag and
M. Perlman, eds., Health, Economics and Health Economics (North-Holland, Amsterdam).
Zweifel, P., and W.G. Manning (2000), "Moral hazard and consumer incentives in health care", in: A.J. Culyer
Chapter 23
WAITING LISTS AND MEDICAL TREATMENT:

ANALYSIS AND POLICIES*
JOHN G. CULLIS, PHILIP R. JONES,

University of Bath
CAROL PROPPER
University of Bristol
Contents
Abstract 1202
Keywords 1202
1.1. An example of waiting lists: the UK 1205
2. Waiting: theoretical issues 1205
2.1. Demand and consumer surplus dissipation 1212
2.2. Demand side: Lindsay's approach 1214
2.3. Supply side: Iverson's approach 1215
2.4. Supply side: other considerations 1218
2.4.1. Medically "interesting" cases 1218
2.4.2. Maintaining your empire 1218
2.4.3. Fostering private practice 1220
2.4.4. Waiting as part of least cost supply 1221
3. Waiting: empirical matters 1221
3.1. Demand side: estimating the "costs" of physical waiting 1221
3.2. Demand side: estimating the "costs" of administered waiting 1222
3.2.1. Costs of waiting from market data 1222
3.2.2. Estimates of costs using contingent valuation methods 1225
3.3. Supply side: estimating the impact of supply variables 1228
3.4. Supply side: inter-sectoral effects 1230
4. Waiting: policy issues 1231
4.1. A taxonomy of policy options 1232
4.1.1. Demand rationing and reduced waiting 1233
*We are grateful for comments made on this Chapter by Tony Culyer, Pat Danzon, Ulf Gerdtham, Joe New-
house and Alan Williams. Thanks are due to Katherine Green for excellent research assistance. Remaining
errors are our own.
Handbook of Health Economics, Volume I, Edited by A.J. Culyer and J.P.Newhouse

1202 J.G. Cullis, P.R. Jones and C. Propper
4.1.2. Supply expansion and reduced waiting 1234

4.1.3. Subsidies to reduce waiting 1235
4.1.4. Encouraging private provision 1238
4.2. Efficient waiting 1239
4.2.1. Indices for prioritisation 1240
4.2.2. Quasi (internal) markets 1242
4.2.3. Reducing uncertainty 1242
5. Conclusions 1243
References 1245
Abstract
A number of health care systems use waiting time as a rationing device for access to
inpatient care. However, a considerable amount of research has focussed in particular
on the UK's National Health Service and its perceived problem of waiting "lists". In this
chapter a theoretical discussion addresses the issue of the optimum wait in the context
of Paretian welfare economics. However, reference is also made to public choice anal-
ysis and to queuing theory. Empirical literature that explores the various dimensions of
waiting costs is reviewed and evaluated. Different methods of estimation are illustrated
and these include contingent valuation, implied valuation and econometric modelling.
The policy section assesses various "solutions" to the waiting list "problem". Options
are classified in terms of their impact on excess demand and the issue of waiting list
management is addressed. In the absence of an over-arching welfare analysis both em-
pirical work and policy recommendations are inevitably piece-meal and open to debate.
Given the inherent weaknesses of applied welfare economics the challenge is to find a
framework which would attract a broader consensus.
Keywords
waiting lists, waiting times, waiting costs, NHS inpatient queues
JEL classification:H11, I111, 1118

Ch. 23: Waiting Lists and Medical Treatment 1203
1. Introduction
Demanders of health care frequently wait in person to see a physician or to receive

treatment. Such waits are common in all health care systems. However, in some health
care systems, demanders of hospital care are allocated to explicit waiting lists. Typically,
demanders of care will not allocate themselves to a list, but will be placed on the list
by a physician acting as the patient's agent. This chapter examines the economics of
these lists. These explicit waiting lists are predominantly found in health care systems
where there is tax-financed insurance and where there is a global budget on expenditure.
In tax-financed systems consumers do not pay the full price of their health care at the
point of demand, so unless capacity exceeds demand when price is approximately zero,
demand must be limited by means other than price. Explicit waiting lists are the most
commonly used means of limiting demand in these systems. This is not to imply that
tax finance is a sufficient condition for the existence of lists: publicly funded systems
that reimburse providers by fee for service do not have explicit lists. There is also a
potential waiting list problem in all insurance-funded systems where the supply side is
constrained (for example, by managed care measures) not to meet all demand. However,
measures adopted to limit the impact of moral hazard in private insurance do not make
use of persistent, involuntary waiting lists. So lists occur where there is a combination
of tax finance and global budgets. Countries in which lists are used include the UK, the
Nordic countries, Canada, Australia and New Zealand.
This chapter examines the theoretical, empirical and policy issues in the use of wait-
ing lists. What economic theories explain the use and persistence of lists and what are
the efficiency and equity consequences of allocation of medical care by lists? To what
extent does the evidence on lists support these theories? What policies will ameliorate,
if not resolve, efficiency and equity problems that arise from use of waiting lists? The
central question addressed in this chapter is what is lost and what gained by opting for
non-market resource allocation and not adopting explicit money prices. If prices are
"wrong" or absent then introductory texts predict problems of signalling, incentives and
rationing. The questions addressed here relate to the role of waiting lists with respect
to these functions. What economic signals should and do they emit? What economic
incentives should and do they provide? How should and do they ration medical care?
From a consumer perspective, the existence of waiting lists raises the issue of time as
a price of health care. The chapter examines when time on a waiting list is a price and
the nature of that price. The notion of time as a price raises the interrelated issues of
optimal waits, input stocks, queuing theory and their associated welfare effects. How-
ever, the fact that waiting lists are a feature of health care systems where the state is the
financial mediator means that the political economy of waiting lists is important. There
is a "public choice" dimension to allocation of care by lists. The political economy of
waiting lists means that they cannot be understood simply as an exercise in applied neo-
classical welfare economics or statistical queuing theory. Waiting time is not simply the
outcome of unrelated demand and supply: to understand waiting lists and to put for-
ward policy proposals requires understanding the links between demand and supply in
a tax-financed health care system.
So any comprehensive economic treatment must take into account not only the de-
manders of care, but the place played by waiting lists in the allocation of funds into
health care, the suppliers of care, and the interaction of demander and supplier of care
with the elected representative and the agents of government.
Other commentators would go further. The approach that underlies almost all the
analyses to date of waiting lists is that of neo-classical welfare economics. As such,
Paretian value judgements are either explicitly or implicitly invoked. There are argu-
ments for an alternative approach. In a health care system where the funding is from
the state, an explicit decision has generally been made to reject the market place. To
adopt conventional welfare economics may look like re-introducing a rejected per-
spective by the back door. One line of argument would suggest that the government
is elected partly to be responsible and accountable for a state health care system, so
government rather than consumer sovereignty should be the criteria by which such a
health care system is evaluated. This would be consistent with a rationale for state inter-
vention rooted in "merit wants" where individual preferences are over-ridden because
it is argued individuals act irrationality and/or on sub-optimal information. In addition
it could be argued that individuals via the voting mechanism have made their choice
"government choice". Whilst such arguments have some currency in health economics
a government sovereignty framework seems insufficiently articulated to form a basis of
sustained evaluation. In addition, the public choice perspective argues that government
behaviour must be seen as the outcome of self-interested decision making by individual
voters, consumers and producers.
Another line of argument that follows from the observation that funding is public is
that the efficiency of the system should be judged in terms of whether those in greatest
need are served soonest. Waiting lists should be evaluated in terms of how they meet
"need" for health care. So the yardstick against which waiting lists would be judged is
do they, or do they not, ensure that this criteria is met. This perspective highlights the
role of the producer and as we shall see below, producers play a key role in the alloca-
tion of patients to lists. A list is actually individual inpatient health care "investments"
that physicians have accepted, in their professional judgement, as offering a positive net
present value from hospital treatment. However, complete deference to medical expert
views only makes sense in a world where medical care is viewed as a purely technical
matter closed to interference from lay opinion. This view is not easy to accept uncriti-
cally in a system financed by the general taxpayer. In a collectively financed system the
evaluation of waiting lists and waiting list policy might be expected to have some col-
lective input. Further, adopting a perspective which gives large weight to professional
judgements of physicians could be seen as giving an already powerful producer group
even more power.
The remainder of this chapter employs the conventional welfare economics approach.
This is for two reasons. First, while there may be alternatives in principle, articulated
government or producer sovereignty models do not currently appear in economics. Sec-
ond, almost all of the existing literature adopts the conventional welfare approach: to
adopt another framework would rule out most of the existing analysis from this survey.
The organisation of this chapter is as follows. It begins with an example of how
waiting lists operate in a health care system in which waiting lists play a key role. We
present data on the length and nature of lists in the UK National Health Service. In
Section 2 we examine the factors that have to be taken into account in determining the
optimal waits in an NHS type health care system. Sections 3 and 4 focus on empirical
matters and policy issues, respectively. Within each section the material, where possible,
is organised in the sequence "demand", "supply" and "demand-supply interactions".
1.1. An example of waiting lists: the UK
In the UK NHS allocation to a waiting list operates as follows. The patient will consult
their primary care physician who, if he or she deems it necessary, will refer the patient
to see a hospital-based physician as an outpatient. For those cases which are not emer-
gencies but which require treatment, the patient may be put on a list for admission as
an inpatient. These hospital based physicians are employed by the NHS, but, whilst in
the employ of the NHS, may also work in the private sector, which specialises in the
treatment of non-urgent conditions (mainly elective surgery) for which there are long
waiting lists.
Chart 1 presents the total number of patients on inpatient waiting lists for England and
Wales from 1951 to 1998. These data exclude any wait to see a primary care physician
and any wait between seeing the primary care doctor and the hospital doctor to whom
the patient has been referred. The chart shows that total number of patients on a list
was around 0.5 million for the first half of this period. The numbers then rose, and
were around 0.7 million for most of the 1980s. In the early 1990s, numbers on lists fell
sharply, but began again to rise in the mid 1990s. Numbers on lists in relation to total
cases treated fell in the early part of the period, remained flat during the 1960s, then
rose again from the mid 1970s to early 80s, since when they have basically fallen.
From 1976 a separate count was made of numbers waiting less than and more than
a year. The chart shows that the number waiting under a year grew up to 1994 then
fell thereafter. The numbers waiting over a year (a significant minority of whom waited
over 2 years) remained relatively constant until 1992. The picture has changed since
then, with an elimination first of the number waiting over 2 years and then a major
reduction in those waiting over 1 year. These changes are not random and the reasons
for them will be explored further below. Chart 2 breaks down waiting lists by speciality.
The chart shows clearly that lists are more important in some specialities than others.
This is not chance and we examine reasons for this below.
2. Waiting: theoretical issues
Whilst Samuelson's (1964, p. 56) parrot cum learned political economist had to learn
the two words "Supply" and "Demand", his or her public sector counterpart must learn
1206 J.G. Cullis, PR. Jones and C. Propper
IIcfiPr;'
00
.5
I I
3
00
c
I
I2
.
0
to
'.z
0
iU
.;a
ut
Ea
, E
0
9
4
i
0
.9
.0
a
'
o
*0
S
*0
U
a
I
00
9 d30 H
§ R
R 9§ 9,,, § 9 §- ' 0
Chart 2
Distribution of patients waiting by speciality, England, September 1996
20
18
16
14
LIZ12
C 10
@ 8
0.
6
4
2
0
2>1 1 co - e2E ) a )o
- o E : c
a) F- C Ot
0 0 0 c
o0
"Costs" and "Benefits". At its simplest, given the public funding of health care, the
waiting list issue can be seen as one of the optimal timing of medical treatment. The
optimal inpatient treatment wait (which is unlikely to be zero for non-urgent cases)
is one that equates marginal (social) cost and marginal (social) benefit so that the net
benefit of inpatient treatment is maximised. 1 In Figure 1 total costs and benefits are
illustrated for three waiting scenarios. The essential question being addressed is the
optimal capacity size of the health care system (HCS). The shorter the wait, other things
equal, the larger the capacity of the HCS will need to be. The total cost curves (TC)
represent the present value of resources required to advance the treatment date of a
particular case type2 (or speciality "representative" patient) and they have a negative
slope. The negative slope reflects the fact that discounted cost declines with the deferral
of treatment, as less real resources need to be allocated to the HCS at present date. (The
precise shape of TC will of course reflect the nature of the production function and the
prices of factor inputs.) Many different capacity structures are possible and as well as
the associated short run marginal cost curves, capital costs of different sizes of HCS
need to be incorporated. It is a case of locating the constant annuity, say k, which over
I The argument could be presented in terms of expected values but the main analysis would be unaffected in
any significant way.
2 These costs can be thought of as being calculated on the basis that the remainder of the specialties in the
health service are optimally adjusted.
PV(TC)
PV(TB)
TC1
to Waiting time (t)

(a) NB 1
PV(TC)
PV(TB)
-TC 2
to tm Waiting time (t)

(b) NB2
PV(TC)
PV(TB)
to ,' tl - . Waiting time (t)

(C) NB3
Figure 1. Determining optimal waits.
the planning period constitutes a time stream of outlays whose present value is the base
year capital cost (K). This is found by the familiar discounting formula:
Kr
k= ' (1)
1- [1/(1 + r)n]
where r is the rate of interest and n is the number of years of the planning period
[see Millward (1971)]. TC then indicates the minimum cost of various capacity levels
and involves summing annual equivalent capital costs with annual running or operating
costs. When capacity is optimally adjusted, SRMC = LRMC and LRAC = SRAC.
As indicated above, the underlying assumption is that the longer the wait for inpatient
treatment the lower TC will be. The interpretation of the costs is that these are for cases
of a given type or cases "representative" of a given speciality. If incurred, such costs
would, in a "certain" world (i.e. one in which the number of cases per period is deter-
ministic), offer the wait indicated on the x-axis. The total benefit curve (TB) represents
the discounted benefits of treatment, which falls as the treatment day is delayed. Total
benefit can be thought of as a "QALY" score converted into money terms [but there is
some debate about the relevance of discounting in this context, e.g. see Cairns (1992)
and Parsonage and Neuberger (1992)]. The final optimal capacity requires a measure
of marginal benefit (typically price in market discussions). Who is to define benefits
(and costs) in a non-market setting is debatable (in a collective institution some type of
democratic mechanism may well seem appropriate) but here it is sufficient to note the
optimal volume of output to provide is given by MB = SRMC = LRMC (with the latter
equality indicating optimal capacity adjustment for that level of output has taken place).
Equivalently it is a matter of choosing for each case type a waiting time that maximises
net benefit.
Depending on the empirical location of TC and TB, clearly differential optimal waits
are dictated. Case type 1, depicted in Figure 1 panel (a), is the case where treatment
yields benefit now (to) or not at all. Case type 1 individuals (treatment now) need not be
urgent. It is just that, with treatment delay, the benefit of treatment falls and disappears
quickly (perhaps the body self-corrects the health status disequilibrium itself). Case
type 2, illustrated in panel (b), indicates a "medium" wait (tin), whereas in case type 3
the total benefit is not much affected by delay and a long wait (tl) appears optimal. In
such circumstances it would be optimal when initially assessed if people were given a
certain date when they are to become an inpatient. That is, there would be a record of
"supply of patient input" dates for the inpatient health production process. In the case
where TB < TC everywhere no offer of inpatient admission would be optimal and the
implied inpatient wait would be infinite. In this simple world it is a question of providing
the public health care inpatient capacity that maximises the net benefit from treatment.
Cases for which TB < TC everywhere set the limit to the inpatient episodes of treatment
to be provided in an optimal capacity HCS.
Assume there are M possible case types and hence optimal waiting times with their
associated discounted benefits. If there were N patients of each type the total benefit of
the HCS would be given by EN I CY_ I TBij . Similarly the total costs involved would
be EN=I yM TCij. The net benefit secured would be EN1I EMI(TBij - TCij).
In this abstract world there would be no waiting list problem as such but simply a
record of the stock of case types (patients' names) that will be delivered at the end of
a wait that maximises the net benefit that health service inpatient resources generate.
To draw an analogy with the unemployment literature, the pool of unemployed (wait-
ing list) is a stock where for each occupation (speciality) there is a stationary register
(constant list) when inflows into unemployment (additions to the list) equals outflows
to employment (discharges) in any period. In the present analysis the duration of wait-
ing time (unemployment) as measured by the difference between entry date (name put
on list) and exit date (discharge) would be optimal in that it serves to maximise the
net benefit of the HCS. As with the unemployment stock the waiting list comprises an
inflow (additions) and a duration (wait) component that in principle can be disaggre-
gated. Note that for each speciality the flow and duration will be given but different, so
that optimal list length varies by speciality but nevertheless would add to an aggregate
waiting list that would be optimal. This implies that without undertaking detailed (and
difficult) empirical work to determine what is the optimal list length for each speciality,
observers looking at the data on waiting lists cannot say for any period that the waiting
list is sub-, supra-, or actually optimal.
In some ways this stock/flow inventory approach has the same structure as a queuing
one. If the HCS underestimates the demand for inpatient care it will not provide valued
inpatient treatments. If it overestimates demand it will either have idle capacity or be
treating on an inpatient basis cases for which MC > MB. Can the waiting list be viewed
as an optimal queue in this sense?
A simple model of a queue [see, for example, Cox and Smith (1961), Wagner (1969)
and Cooper (1981)] can be adapted to offer some insight into this question, especially if
the certainty assumption underlying the discussion so far is relaxed. Elements in a very
simple queuing model of the HCS inpatient treatment process would be:
D = mean number of individuals demanding inpatient treatment
(reflecting, say, a Poisson distribution);
S = mean number of individuals being treated per period;
d = the total number of individuals being treated or waiting;
1 = the total number of individuals waiting on the list;
Pd = probability of d individuals being treated or waiting per period;
po = probability of zero individuals being treated or waiting per period;
t = time a new joiner must wait before being treated;
E (d) = expected number of individuals in the queue or being treated;
E(1) = expected number of individuals waiting;
E(t) = expected waiting time for each joiner to the list.
Key relationships are3:
D
E(d) = (S D)' (2)
D2
S(S - D)'
E(t) = ) (4)
If reality could be adequately captured in this way (subsequent sections cast doubt on
this conjecture) then values for E(d), E(l), and E(t) could easily be obtained. Such a
model would readily provide valuable types of information. For example, if the period
is a week, then
total time lost (TTL) in the queue per year = E(l) E(t) 52 = TTL
total cost of queuing (TCQ) = E(1) E(t) 52 - w = TCQ,
where w is the opportunity cost of time waiting.
Two inferences that arise from these, albeit simple, theoretical considerations are
summarised in a static way by Figure 2. In what is described above as an optimal ca-
pacity/waiting system, the occurrence of a type of ill health (j) will involve an optimal
wait for treatment of t, . The effect of this on the utility of a risk averse individual can be
captured in the downward shift and flattening of the utility function. Utility is assumed
to depend on both full income (Yf) and health status (hs) [so that U = U(Yf, hs)], with
U(Yf, w) > U(Yf, i), where w = well or healthy and i = ill. In addition, the flattening
suggests U/ Yfl > U/ Yfli so that the marginal utility of income is lowered by ill-
ness. If Yfo is the individual's income level and t is guaranteed then there is a utility
cost of waiting equal to Uw - Ui, which in terms of full income is the distance Yfo - Yfi.
If this argument is accepted, such a cost is part of a Pareto efficient resource allocation
(as it arises when t obtains). If, however, the waiting list is not run on a "guaranteed
booked case" system and the uncertainty raised by queuing theory comes into play then
E(t J ) = t. That is, the expected value of the optimal wait, E(tJ), equals the optimal
wait, tJ , only on average. For risk averse individuals this imposes an additional cost
equal to Ue - Ui in utility terms and Yfe - Yfi in income terms. Given that demand for
inpatient treatment is inherently uncertain, this cost is also part of the Pareto efficient
configuration. In this perspective the waiting list problem is one of non-optimal cost
levels associated with two issues and their combination: viz. sub- or supra- optimal in-
patient capacity that dictates t t, or an administrative system that does not minimise
the variance of E(t), or both.
3 Other include:
PO = 1-(D/S), (In)
Pd = (D/S)dpo = (D/S)pdl (2n)
Total
(expected)
utility
. . : .11, I
Uw (,Yf,w)
ui '
U=U(Yfi)
Ue
I I
l l
l lf l~~~~
O Yfe Yfi Yfo Full income (Yf)
Figure 2. Waiting and costs.
Whilst the questions directed at the waiting list issue are neatly encapsulated by the
optimal capacity/waiting and associated queuing theory implications, there are limita-
tions to this approach. For example, with respect to queuing theory, the waiting list
context differs in a number of important ways. The queue is not a physical one and
waiting may impose no direct opportunity cost of time; D is not a stochastic process
but is much influenced by the behaviour of hospital and primary care physicians' deci-
sions whose interaction create the waiting list stock (in contrast the recorded stock of
unemployed is largely the result of an impersonal labour market profitability test); S is
influenced by the number of direct admissions to hospitals; the queue is not organised
on a first come first served basis - an everyday "equity" rule - but rather certain patients
are selected from the list depending on their personal characteristics. For the moment,
however, it is worth considering the physical queue analogy as it helps identify the cost
of waiting, which has to be compared with the benefit.
2.1. Demand and consumer surplus dissipation
Both optimal wait and queuing theory imply waiting costs for some types of potential
inpatient. In the queuing discussion w was designated the opportunity cost of time spent
waiting. Such costs have been viewed as prices and their ability to reduce consumer
surplus analysed. Barzel (1974) describes a natural experiment in which a local bank
had a one-off "sale" of money - selling notes at less than their nominal value on a "first-
come first-served" basis. To limit their financial commitment, only the first 35 persons
get the maximum gain of £20, the next 50 each gain £10 and the next 75 each gain £4.
The method of rationing was a physical queue. If the first person in the queue waited
17 hours and the thirty-fifth 9 hours, economic theory could be used to predict the wage
Price
O qg qt
Quantity of
Medical Care/t
Figure 3. The demand and supply of inpatient care.
characteristics of these two individuals. Waiting in the queue involves costs to the extent
of the waiting time (t) multiplied by the opportunity cost of time, say a wage rate (w).
With £20 as the maximum possible gain, w times t should not exceed this figure except
by accident, misinformation or economic irrationality. If an individual could earn £3 per
hour, it would not pay him or her to wait in a queue for 9 hours in order to get a "free"
£20, since they could have used the time to earn £27. For the first person waiting the
limiting value for w is approximately £1.20 per hour, and for the thirty-fifth, w = about
£2.20.
More generally, assume Dm in Figure 3 is an individual's demand curve for medical
care. Zero money prices are charged so that the individual would demand Oqt per period.
Suppose the government restricts quantity so that Oqg is the maximum amount of care
an individual can obtain per time period. In these circumstances the maximum the indi-
vidual would be prepared to pay for Oqg health care is 0 1 2 qg, which sets an upper bound
on the value of the length of wait the individual will endure to secure Oqg. If the waiting
costs [again the product of the opportunity cost of time (w) and length of waiting time
(t)] exceed 01 2 qg the individual will not join the queue. Consumer surplus is dissipated
completely by a cost of wait equivalent to 0 1 2 qg. In a market the consumer surplus
gained would be the equivalent to triangle p1 2 , assuming for convenience the market
clearing price is Op. Consumer surplus under the market system of allocation has to be
greater than consumer surplus achieved under a time price system. Under the time price
system the marginal receiver of care enjoys no consumer surplus. Under a price system
the marginal consumer receives no consumer surplus only on the marginal unit of care
they buy. (This assumes that individuals vary in the time prices that they are willing to
pay so that the market demand curve for a ration of 'Oqg' is standard downward slop-
ing. If individuals do not differ then consumer surplus is dissipated for all individuals.)
Unlike a market price, 012qg (= wt) is a deadweight cost and does not fulfill the incen-
tive function of compensating factor input owners. The rationing function is, however,
achieved.
Is it possible to view waiting patients as "paying" for their care via a deadweight
time price involving such intangible costs as pain, discomfort, uncertainty, etc., and
tangible costs such as lost earnings? The answer is "yes" where patients wait in person
(e.g. in a physician's surgery), as there is an opportunity cost involved in having to
wait physically at the surgery. However, the answer is a qualified "no" for administered
inpatient waiting where the alternative to waiting for care from the public health care
system is to use the private sector (assuming a private sector alternative exists). The
individual will take this decision if the expected consumer surplus gain from alternative
care outweighs the additional costs associated with private care. Should this not be a
relevant option the potential inpatient simply waits until they get to the top of the list,
carrying on, in the meantime, as best they can. Not being on the list simply makes
securing care from the HCS impossible: in other words, for a large number of waiting
patients it is not economically rational to avoid the costs of waiting. Hence the marginal
waiting individual is not the one whose value of waiting equals the valuation of the
benefits of inpatient care but rather the individual who sees any benefit to care on the
anticipated future date when it is received. (This abstracts from the time costs of getting
on the waiting list and the pain and discomfort of treatment in itself which clearly are
avoidable by not being on the list.)
Culyer and Cullis (1976) explored the waiting time as a price argument in more de-
tail. They found its main implications to be: a negative association between throughput
capacity and waiting times; a positive association between capacity and waiting lists
and a negative association between waiting lists and waiting times. These implications
found little empirical support. The type of assumption needed to make this "naive" time
price hypothesis hold up is that doctors, as patients' agents, act "as if" waiting times
were a price and make decisions on behalf of patients on this basis. As long as the loca-
tion of the demand curve for inpatient treatment is unaffected by waiting, the potential
consumer surplus from inpatient treatment is not dissipated by the wait for treatment
(which, however, imposes costs on the individual that arise because the delay of the
benefits of inpatient treatment reduces the present value of the benefits) because the po-
tential patient cannot undertake costly activities (unlike in the bank example above) that
will qualify him or her for inpatient treatment.
2.2. Demand side: Lindsay's approach
Lindsay (1980) [also see Lindsay and Feigenbaum (1984) and Spicer (1982)] builds a
model of waiting lists that is equilibrated via waiting time on the list by attacking the
assumption made in the theory above that the demand curve for inpatient care remains
unaffected throughout the wait. In Lindsay's model waiting time matters because the
Ch. 23: Waiting Lists andMedical Treatment 1215
value of inpatient care decays the longer care is postponed after the diagnosis day. More
specifically, in Lindsay's model waiting time equilibrates the queue by rising or falling
until the numbers joining the queue equal the number of patients treated per time period.
Marginal queue joiners satisfy the condition:
C = Ve - t , (5)
where C is the lump sum costs of joining the queue, V is the valuation of inpatient
treatment now that decays at a rate, d, until treatment occurs after a wait of t. Two main
comparative static results are derived from this model.
First, if the typical case of two different diseases with different decay rates, say di <
d2, and C and V are assumed to be constants (C and V, respectively) then for
= Ve - d ltl
and C= Ve -d 2t2 (6)
to hold simultaneously tl > t2 and waiting times for inpatient admission for each dis-
ease should be inversely related to the demand decay rate. Second, the responsiveness
of waiting time to capacity changes should be negatively related to the decay rate. The
logic behind this result is illustrated in Figure 4. In panel (a) the effect of increasing
the supply capacity (rate of treatment) from SCI to SC 2 is to decrease waiting time
from t to t2 and expand the number of joiners along J(t). Potential patient V 2e - d2t
is not a joiner at waiting time t because such a long wait reduces the value of treat-
ment to V2 < C, however, at waiting time t2, V 2 e- d2t = C and the individual becomes
the marginal patient (as opposed to the di case initially). Equilibrium waiting time is
depressed because with relatively low rates of decay (dl and d2) few potential deman-
ders are discouraged (because of lack of possible substitutes) by the long waiting time
and a relatively large reduction in it can be achieved by a capacity increase. Now con-
sider panel (b) that illustrates the position of high decay rate diseases. Here treatment
is needed quickly or not at all and if inpatient treatment is not available other sources
of treatment (if any) must be sought. Given this, many potential patients are not in the
queue and V3 e -d 3t can be viewed as a valuation curve typical of many patients. In
these circumstances an increase in the treatment rate from SCI to SC2 induces very lit-
tle change in waiting time (none in Figure 4) as there are large numbers attracted from
the various substitutes to inpatient treatment by the addition of supply capacity. As long
as the range of decay rates for given diseases is not too great then low decay rates will
always be associated with longer waiting time and vice versa.
2.3. Supply side: Iverson 's approach
Iverson (1993) outlines a supply side model in which "long waits" may result. In his
model, observed waiting lists and waiting times for medical treatment are placed in
the context of a political bargaining process over resources. The heart of his model is
captured by Figure 5. He describes production possibilities between the number of ex-
pected treatments supplied E(s) and the expected wait E(t) before treatment. A key
Waiting time (t)
:h
es
SC2 SC 1
Ie
II I .
with
ates
Joiners and SC 2 SC1 0 C V Cost and value

throughput
Figure 4. Lindsay's waiting list model.
assumption is that the number of expected treatments is affected adversely by increased

waiting lists and times. As the waiting list lengthens, staff resources that could be used in
providing treatment are absorbed into waiting list management tasks (keeping records,
prioritising those listed, responding to enquiries, perhaps repeating diagnostic tests ren-
dered invalid by the passage of time). For a given capacity constraint, SCo, increasing
waits initially have a positive impact on expected treatments via improved capacity util-
isation but after a critical wait E (to) the negative impacts described above dominate and
the number of expected treatments falls (hence the shape of OSCo in Figure 5). To the
left of E(to) waits are described as "short" to the right of E(to) as "long". A waiting
time neutral decision taker would have a horizontal indifference curve map between
E(s)
;EPg
O E(t o) E(t)
Figure 5. Iverson's waiting list model.
E(s) and E(t) (not shown) and choose point 1 on SCo maximising expected treatment
numbers for that and other levels of SC. As SC increases, e.g. SC 1, SC2 , the "short-
long" wait boundary is delineated as the "numbers maximising" equilibrium, point 2
for SC1 , point 3 for SC2 migrates to the right. Increases in supply capacity decrease
the level of capacity utilisation for each level of E(t). Given this there is an increase in
the number of expected treatments caused by a marginal increase in the expected wait
as supply capacity increases. The slope of each production possibility arch reflects this
being steeper at a given E(t) thereby generating the positive slope of the maximum
treatments curve (M-T).
Why might a long wait represent equilibrium? Intuition suggests that the expected
wait is a "bad" and the expected number of treatments is a "good" so that indiffer-
ence maps shaped like II, 12, 13 would be relevant and equilibrium (points 4, 5 and
6) found on a capacity expansion path like the curve labelled, CEPhg. Here the hospital
doctors/administrators and the sponsoring agent (government at some level) both simul-
taneously act altruistically, or perhaps more accurately non-instrumentally indicated by
the subscripts 'h', and 'g'. Given this "long" waits will not arise.
Iverson then introduces the idea of bargaining for resources to highlight the circum-
stances in which long waits would result. If the budget decision is sequential [the gov-
ernment sets SC and the hospital actors set E(s) and E(t)] then the result turns on the
shape of the government's capacity expansion path (CEPg) and the bargaining solution
concept invoked. If the CEPg is positively sloped to the right of M-T then a Stackelberg
equilibrium with the hospital as leader can be found at point 7. CEPg is the "follower"
reaction function and I h is the highest utility level the hospital actors can achieve subject
to that "reaction". Point 7 is associated with long waits as it is to the right of M-T and
must be Pareto inefficient (like all Stackelberg equilibria) in that point 1 compared to
point 7 (both on SCo) involves a larger E(s) and a lower E(t). For this solution to arise
the government must base its supply capacity decision on the expected wait. In particu-
lar, the valuation of the (negative) effect of an increase in E (t) on a patient's health must
be larger than the (negative) effect on the cost of treating a patient as E(t) increases.
Hence the slope of the government indifference curve I g at point 7 (the marginal benefit
of treating a patient as waiting time increases is larger (smaller in absolute value) than
the marginal cost of treating a patient when the waiting time increases).
2.4. Supply side: other considerations
Iverson (1993) deliberately assumes altruistic agents but recognises there may well be
deviations from this assumption. The public choice perspective in economics raises the
question of the economics of processes and the argument that all economic actors are
maximising their utility subject to constraints. In this context it means attention has to
be directed towards the narrower incentives of producers (hospital physicians, managers
and the like), those in the government (politicians) and central government bureaucracy
(civil servants). What does such a perspective imply for waiting lists and times?
2.4.1. Medically "interesting" cases
Feldstein (1970) suggested that physicians may get utility from maintaining a queue
for their services because it enables them to increase their utility by selecting more
interesting or urgent cases. Pauly (1990) also puts "interesting" cases in the physician's
utility function. The context is a market one. Physicians are assumed to have a utility
function of the form:
U = U(Y, a), (7)
where Y is money income and a the proportion of interesting cases in the workload.
Given this, physicians "trade-off" income against more interest by charging less than
market clearing prices and having a "list" to choose from. (A parallel might be an em-
ployer paying above market clearing wages to induce a queue of potential employees.)
The desire for more interesting cases seems particularly plausible in a teaching hospi-
tal context where trainee doctors need to have some familiarity with a wide variety of
cases. In such circumstances, where there are joint output benefits of treating certain
cases, a shorter optimal waiting time may be consistent with net benefit maximisation.
In a less attractive light Frankel (1989) also takes up the theme of interesting cases. He
sees the problem of waiting in terms of the lack of interest by the medical profession in
the treatment of the "day to day" complaints that await treatment; research interests be-
ing more active in other fields. This leads to lists building up in the routine uninteresting
fields (and as evidence cites patterns such as those of Chart 2).
2.4.2. Maintainingyour empire
Being on the waiting list is usually the result of a primary care referral to a hospital
outpatient clinic. Within a limit set by the number of referrals and the bounds set by
Ch. 23: Waiting Lists and Medical Treatmnent 1219
hospital managers, the hospital doctor can (at least in an "agency" model 4) have any
length of waiting list he or she wants. In practice the situation is more complex than
this and certain factors can be isolated that will help explain the decision to place a
potential patient on a waiting list. It is well known that waiting lists predominantly arise
in surgical specialities, largely because of the point at the heart of Lindsay's analysis
that for some illnesses inpatient treatment is needed now, in the near future or not at
all. Hence, for medical specialities if inpatient treatment is indicated but a bed is not
available then an alternative treatment regime must be adopted. So waiting lists tend to
build up for delayable "cold" surgical cases rather than for other treatments. Therefore
lists are argued to be beneficial to the extent they represent a stock of available work
ensuring that the scarce and skilled resources of surgeons and other theatre staff can be
fully utilised (accounting for the positively sloped sections of the SC curves in Figure 5).
Additionally as noted above a waiting list allows for a balance of cases of differing
nature and complexity to be chosen facilitating the teaching function of many hospitals.
Frost (1980) pointed out that on average in the 1970s each senior NHS hospital doctor
had 160 individuals on his waiting list representing approximately two months' work.
It was argued above that there was an optimal waiting time from the point of view
of optimum resource allocation. However, unless hospital doctors fully internalise the
costs of delay for potential patients, they may choose waiting lists that are too long.
Lindsay (1980) questions the notion of a largely doctor determined waiting list. He
argues that a direct test of an "agency" effect in the NHS can be made. Using Figure 6,
which has a similar interpretation to Figure 4, suppose to is the initial equilibrium wait
and that additional hospital doctors are appointed. Via the "agency" relationship, these
new doctors increase the demand for inpatient care shifting the V-function to the right
and the joining function to the left J(t) to J (t). The effect is to raise the equilibrium
wait of the marginal joiner to tl as long as throughput capacity does not change. By
contrast, if increasing the number of hospital doctors is viewed simply as increasing the
supply capacity SCo to SCI, other things equal, the equilibrium wait should fall to t2
associated with point 3 on J(t). Lindsay's empirical evidence lends support to the latter
proposition.
But this test is not robust. As Frost and Frances (1979, p. 195) point out, consultants
are not only the gatekeepers determining the length of the waiting list but also inputs in
the hospital sector's production function. Once this dual perspective is recognised the
unaffected throughput assumption made by Lindsay is too strong. Indeed, if hospital
doctors are both "agents" and inputs, changes in their number can simultaneously shift
the V-curve to the left, the J(t) to the right and increase supply capacity, thereby mak-
ing a fall in waiting time conceivable under both scenarios (e.g. t3 in Figure 6 consistent
with point 4). Hence it seems empirically testing for the "agency" aspect of waiting list
construction in the way suggested by Lindsay is not decisive.
4 Agency effects raise the question of supplier induced demand. Here no assumption is made about the
welfare consequences of the agency effect.
1220 J.G. Culis, PR. Jones and C. Propper
Waiting time (t)
Vle-dt
2'
- _- - - -- - -ti -------
-to --------
4 -------------- t3
J1 (t)
3 t2
J(t)
SC1 SCO 0 C V1 V2
Joiners and throughput Cost and value
Figure 6.Lindsay's "test" of the agency effect.
2.4.3. Fosteringprivate practice
It is commonly alleged that long waiting lists where physicians are also allowed private
patients encourages the growth of private practice and results in increases in physicians'
income. McAvinchey and Yannopoulos (1993) use a cost shares model to investigate
the impact of hospital doctors on the shares of NHS acute care and insurance financed
private acute care. Their econometric results are consistent with a direct effect of an
increase in doctors employed in the NHS on costs (and so activity) in both sectors,
as opposed to an increase only in public sector health care. However, there is also an
indirect effect of waiting time on the public-private mix of care in their model. They
derive a price index (P) for public sector care that is based on the cost of being on a
waiting list:
P = E(I + i) t , (8)
where E = a measure of average labour earnings, i = the short-term interest rate; t =

the time wait on an NHS waiting list. Given that hospital doctors can control the wait-
ing lists, and hence t, they can also affect the public-private mix indirectly via this
admittedly "subjective index" (p. 179) of the price of public sector care.
2.4.4. Waiting as part of least cost supply
There are possible avenues via which gains from waiting can arise. Medical care con-
texts are generally uncertain ones and waiting may afford an opportunity to obtain more
information about a patient and improve diagnosis and treatment. On seeing a patient in
a hospital clinic the hospital doctors can refer the patient back to his or her GP, suggest
an alternative hospital speciality, treat them more or less immediately or, assign them
to a waiting list. In considering the last two options the hospital doctor may in uncer-
tain circumstances be weighing up two costs, the cost of choosing the wrong treatment
now as against the cost of waiting for new information which raises the probability of
a correct choice of treatment at a later date. If the latter delay cost is small, waiting list
assignment offers a potential net benefit over treatment now. However, given that wait-
ing lists are predominately composed of cases involving a small number of conditions
not associated with difficult surgery (see, for example, Chart 2) this argument cannot be
used to explain much of the waiting list.
In fact, it appears that supply increases designed to reduce waiting lists and/or wait-
ing times seem to induce behavioural changes that results in fairly stable numbers on
lists. In the UK case for example, waiting lists seem to remain at around half a mil-
lion (Chart 1). Given that the stock of illness amongst patients does not change rapidly
over time, this fact suggests that referral practices by primary care physicians and the
willingness of hospital doctors to assign outpatient clinic attendees to the waiting list is
not independent of supply. This has been referred to as Say's Law of Hospital Beds or
"feedback" [Worthington (1987)]. The motivations that underlie such behaviour have
been explored above, and do not seem consistent with making patients wait in case new
information is revealed. In addition, this feedback mechanism highlights the problem
of reducing recorded waiting lists: funds spent to decrease the list may simply result in
higher referrals. It may or may not be efficient to treat the increased number of cases.
3. Waiting: empirical matters
This section discusses empirical tests of the issues raised in Section 2. The empirical
research can be grouped into tests of the impact of lists on demand, some of which
attempt to estimate the cost of waiting lists for demanders of care or for society; tests
of the association between public sector supply and waiting lists, and tests of the asso-
ciation between the rest of the medical care system and waiting lists. These empirical
tests are often based on limited data and as a consequence, often provide only partial or
rather indirect tests of the hypotheses discussed above.
3.1. Demand side: estimating the "costs" ofphysical waiting
First, we briefly look at analyses of waiting in person. The costs of waiting in person are
not the focus of this review, but recognition of the cost of such waiting is widespread,
and it has generated a large literature. One of the earliest analyses of the economic costs
of waiting in person was Acton (1975) who proposed a model of the demand for medical
care that incorporated time as an input. The consumer maximises utility, derived from
the consumption of medical care and all other goods, subject to total income and total
time constraints. Consumption of medical care in the Acton model has both a time
and money cost. The comparative static results indicate that the elasticity of demand
for medical care with respect to time depends on both the time and the money price;
further, the absolute value of this elasticity is a positive function of the relative size of
the time price. The implication is that time price will be more important in determining
demand in health care systems in which care is allocated primarily by means of time,
rather than money. This study and others that followed [e.g. Acton (1975), Phelps and
Newhouse (1974), Coffey (1983), Gertler et al. (1987)] treated time as an input into the
production of health care. However, although some of this research identified different
types of time as having different cost, these studies did not estimate the cost of the
allocation of care by waiting list.
3.2. Demand side: estimating the "costs" of administeredwaiting
Waiting on a list has a cost because a good received later is worth less than one received
now [the point made by Lindsay and Feigenbaum (LF)]. In addition, the individual on
a list for treatment is generally in less good health than they would be after treatment
[Propper (1995)]. The cost of waiting will generally be a function of both these compo-
nents.
The medical literature accepts that waiting lists are costly to users of medical care
services. For example, in the UK, General Practitioners, as agents for patients, state that
waiting times are one of the factors they take into consideration when making refer-
rals [e.g. Kennedy and O'Connell (1993)]. There have been relatively few attempts to
estimate the magnitude of these costs and the pattern of variation in cost across individ-
uals. 5 Because consumers cannot buy shorter waiting time for public sector treatment,
there are no direct measures of willingness to pay for shorter waits. When market data
are missing there are basically only two alternative data sources. The first is to try to in-
fer prices from observed behaviour in markets which interact with the public health care
sector, i.e. to use revealed preference data. The second is to use direct inquiry via ques-
tionnaires (contingent valuation) or some form of utility experiment, i.e. to use stated
preference data.
3.2.1. Costs of waitingfrom market data

In the UK potential NHS patients can buy private care. Can the behaviour of individuals
who choose private care be used to infer the costs of waiting? The answer is a qualified
yes if total costs are sought, but no if the variation in costs across individuals is required.
5 Gribben (1992) measured the impact of waiting times on utilisation of GP services in New Zealand and
found that long waiting times reduced utilisation. Regidor et al. (1996) who found that time spent waiting for
doctor consultations in Spain varies by socio-economic status.
Cullis and Jones (1986) used the LF framework to estimate the costs of waiting using
data on length of lists and the price of alternative actions the demander of care may
take. Like LF, they assumed the only cost to waiting arises because medical care is
worth less if received later than sooner. Let P be the price of medical care in the private
sector, V be the value of health care if received now, and V(t) be the value if received
at time t(V(t) < V for all t > 0). The "submarginal waiter" is the individual for whom
V - V(t) < P. She will not buy private care but will wait. The cost of waiting is less
than P. The "marginal waiter" is the individual for whom V - V(t) = P, and the cost
of waiting = P. Finally, the individual for whom V - V(t) > P will not wait, but
will purchase immediate medical care on the private market and experience no cost of
waiting.
Given this taxonomy, Cullis and Jones concluded that the upper bound of the cost of
waiting was equal to the average price of private medical care. The lower bound was
zero: for some individuals the decay rate is 0. Using data on the prices of private medical
care, and making the assumption of a uniform distribution of the costs of waiting, they
multiplied the annual number of patients waiting for hospital admissions in the NHS by
P/2 to derive an estimated cost between £1,205 to £2,155 million (the range depending
on low and high estimates of P). This was equal to a cost per month of £110-220:
equivalent to 9 to 16 percent of the NHS budget (0.5 and 0.8 percent of GDP) in the
year of their analysis.
If there is a positive cost to waiting per se, then the lower bound may be an under-
estimate. If the list is artificially inflated by people who no longer need treatment, or if
individuals are subsidised to wait for treatment (for example, if time off work is covered
by disability insurance individuals may overvalue V(t), and so wait in line when they
would pay), the number on lists is too high, and the costs are an overestimate. If the
value of time in bad health waiting is similar to the value of time in good health and
treatment imposes costs, then bringing treatment forward will not necessarily increase
welfare for all patients. In this case, the estimates will also be too high.
To estimate the productivity cost (lost output) of waiting lists in two Canadian
provinces, Globerman (1991) used the numbers on waiting lists in two provinces mul-
tiplied by the average wait to derive the total time waited in one year. He then deflated
this by a factor reflecting the fact that only a proportion of those on waiting lists are in-
capacitated by the wait. This was then multiplied by the average wage rate to derive an
estimate of productivity loss of between 0.1 and 0.2 percent of the total wage and salary
bill in the two provinces. The sum was comparable to the income lost from strikes and
lockouts during the same period.
Feldman (1994) adopted a similar framework to Cullis and Jones to estimate the costs
of waiting that would arise if the US health care system was replaced by a "US NHS"
in which care was funded by taxation and excess demand rationed by waiting list. He
used estimates of the elasticity of demand from the Rand Health Insurance Experiment
to estimate the efficient level of health care expenditure (after adjustment for the impact
on prices of both provider monopoly and employer subsidisation of health insurance). 6

He made the assumption that this estimate was the level of care that would be funded
by the "US NHS'. At this level of funding, there would be individuals waiting for care
since the price is zero at point of demand.
To derive the costs of waiting for all medical expenditures 7 Feldman estimated the
percentage of total medical expenditure accounted for by those who would wait. By
definition, the marginal waiter is the individual for whom the cost of waiting is the
market price of treatment. As more expenses fall into the marginal category, so will the
cost of waiting rise. Feldman made the following assumptions. First, the percentage of
US NHS expenditure that was marginal could be 25, 50 or 75 percent. Second, the cost
of waiting for the marginal waiters had a uniform distribution with mean p/2. Third,
the cost of waiting for sub-marginal users had a uniform distribution with mean p/4 (to
allow for the fact that the cost of waiting for a sub-marginal waiter, by definition, is less
than the market price).
Under these assumptions he calculated the cost of waiting per family ranged from
around $720-1000 per year (1984 dollars), where the largest figure assumes that 75
percent of the waiters are marginal and the smallest figure that 25 percent are marginal.
Offsetting this cost of rationing by lists is the consumer benefit from the reduction in
risk from full insurance under the "US NHS". Using estimates of the risk premium
(derived from the price elasticity estimates for health care under insurance), Feldman
estimated the social cost per family of rationing by waiting were an NHS system to be
introduced in the US to be in the order of $540-830 per annum (or 1-1.5 percent of 1984
GPD). 8 While this appears low, this figure may underestimate the cost of waiting. First,
it ignores the cost of uncertainty that is associated with use of waiting lists. Second, it
ignores the costs of other forms of rationing that arise whenever price is below marginal
costs. 9
These papers have all estimated costs at a system level. Is it possible to use data on
individuals who are covered by public insurance who then subsequently buy private
care to infer their costs of waiting and so make estimates of variation across individuals
in costs? The answer is generally not. First, the price which individuals pay for private
care is not just the price of avoiding a wait. The consumer who chooses private care is
also buying choice of location, choice of time for treatment, better hotel facilities, and
6 Adjustment for employer subsidisation of health insurance may be inappropriate since in the long run these
costs are largely borne by employees in the form of lower wages.
7 The Feldman estimate is of all care, not just the hospital care costed by Cullis and Jones and Globerman.
8 Feldman argues that this is between 45-69 percent of the social cost from the over-utilisation of medical
care by the 6/7th of the US population who have close to full insurance.
9 In systems where there are no waiting lists, but price is below marginal costs for the demander of care, other
forms of price or non-price rationing do occur. Other non-price forms of rationing entail direct administrative
costs, which may be lower or higher than the costs of waiting lists. Danzon (1992) made a rough estimate of
the costs of non-price rationing by waiting in Canada and the administrative procedures used to limit demand
in US private insurance, and concluded that in fact the US system was less costly.
possibly better information. Second, even if the price were adjusted for this higher qual-
ity, typically the analyst does not observe how long those who chose private treatment
would have had to wait. Third, any individual covered by private insurance will only
pay the copayment. If copayments are zero, the demander may choose private sector
treatment for any length of wait. l°0
One method of deriving valuations of non-directly marketed public or private goods
is to estimate hedonic prices. This assumes that observed prices reflect variation in the
characteristics that are not marketed. Estimates of the value of goods such as noise,
air pollution and climate have been made by assuming that spatial variation in housing
prices reflect spatial variation in these goods across different communities. In theory,
as waiting times tend to vary considerably across communities, observations on house
prices could be used to infer the value of shorter lists. However, use of such a method
would require that correction could be made for differences in house prices due to fac-
tors other than the length of waiting lists, for factors which might drive both lists and
house prices but did not link them causally, or for endogeneity of waiting times. The
second possibility would arise if lower income individuals both lived in cheaper areas
and had higher demand for health care services. There would be a negative association
between prices and lists that was not due to the valuation of waiting time, but the fact
that housing is a normal good and poorer individuals are sicker. The third case would
arise if an influx of poorer people into a cheap housing area caused lists in that area to
rise. In practice, this method has not been used to value costs of time spent on waiting
lists.
3.2.2. Estimates of costs using contingent valuation methods
The contingent valuation method uses sample surveys to elicit the willingness of respon-
dent's to pay for projects or programs. The method has been widely used in resource and
environmental economics, where it is used to elicit preferences for public goods, such
as national parks, clean air, or to value the "existence value" of goods which individu-
als may not use but may value (such as the preservation of species). Outside resource
economics, it has been used to value a large number of non-environmental policies or
programs, including health related ones such as reduced risk of death from heart attack
[Acton (1973)], reduced risk of respiratory disease [Krupnick and Cropper (1992)],
highway safety [Jones-Lee, Hammerton and Phillips (1985)]. Contingent valuation can
also be used as a means of eliciting preferences for private goods. It has, for example,
been used extensively to value time savings in travel [e.g. Bates (1988)] and in health
care has been used to value time spent on waiting lists and the value of other goods
which are non-marketed but are mainly private [Ryan (1996)].
While there is no standard approach to the design of a contingent valuation survey,
the elements of an application are the following [Pourtney (1994)]. First, the survey
10 Clearly the price of insurance is not zero. But once insured the demander of private care faces zero marginal
cost if co-payments are zero.
must contain a description of the good that the respondent is being asked to value.
Second, the survey must contain a mechanism for eliciting the values or choice from the
respondent. These mechanisms may take a number of forms, which may be either open
ended (e.g. "how much would you be willing to pay for x") or closed ended forms where
individuals make discrete choices between fixed alternatives. Using the close-ended
form of the survey instrument, individuals are asked to make discrete choices between
goods that contain different amounts of each attribute. Assuming a random utility model
for individual preferences, standard techniques for binary choice can be used to estimate
a willingness to pay distribution function [Hanemann (1994)]. Third, the survey usually
elicits socio-economic information on respondents, so that willingness to pay functions
can be estimated which includes these characteristics as explanatory variables.
Contingent valuation is currently the subject of controversy, particularly where used
to elicit preferences for public goods where there are no close marketed substitutes and
contingent values cannot be compared to observed market transactions. l In the case of
waiting times, the good is private, but as for public goods it is not generally possible to
observe trade-offs individuals would make between time and money.
Propper (1990, 1995) adopted an economic framework used to study the value of the
utility of time in transit economics to estimate the value of waiting time in the UK NHS.
The framework assumes utility is derived from a vector of commodities, plus a vector
of time spent in various activities [e.g. de Serpa (1971), Truong and Hensher (1985)].
Where the consumer chooses between discrete uses of her time, her problem is:
max U(G, q, t),

s.t. pG + Ejdjcj Y [], (9
q +Ejdjtj T [],
tj > t forall j [y],
where G is the quantity of a generalised consumption good, q is the time spent in

a generalised activity, tj and cj are the time and costs, respectively, associated with
alternative choices, dj, j = 1, ... , J are dummies such that dj = 1 if choice j is made
and 0 otherwise, t is the technologically determined amount of time for activity j, and
/I, and are the shadow prices of the income, total time and time spent in activity j
constraints, respectively.
Taking a first order approximation to the direct utility function (9) and substituting
in the first order conditions, the total time and the money budget constraints, the condi-
tional indirect utility function for choice j can be derived as a function of income, costs
and time of each alternative. Given an indirect utility function, a probabilistic choice
model can be used to allow estimation of the parameters of interest. Using a random
utility model, terms common to both alternatives drop out of the specification of the
l See, for example, the Journal of Economic Perspectives (Fall 1994), Hausman (1993), National Oceanic
and Atmospheric Administration (1993).
deterministic component of (indirect) utility, and the deterministic element of indirect

utility can be specified as
Vj = -- c - jtj. (11)
The coefficients . and 9j can be interpreted as scale transformations of the marginal

utility of cost and time respectively, and the ratio 99j/)X interpreted as the value of saving
time in health care choice j. The indirect utility function (10) contains only linear terms
in income, and so (11) contains no terms in income. Therefore income does not enter
directly into the estimation. In estimation of this model, Propper allowed the constraints
A and 99 to be functions of income and other socio-economic attributes. (For example,
the marginal utility of income, X.may be expected to fall as income increases.)
Data to estimate (11) were collected using a contingent valuation exercise, in which
individuals were asked to trade-off waiting time against money for a non-urgent medical
treatment (one with a low decay rate in the LF terminology). The hypothetical situation
was designed to be as close as possible to the existing market. 12 The estimates indicated
that the value of waiting time was of the order of £35 per month (1987 prices).13 In
addition, there was systematic and significant variation by income: those with higher
incomes had higher values of time.
The total costs of waiting for treatments with zero decay rates was estimated to be
around 0.1 percent of GDP. These estimates were below those of Cullis and Jones,
who estimated the GDP cost of waiting to range between 0.5 and 0.85 percent of GDP.
However, the difference was in the expected direction. The contingent valuation exercise
explicitly considered only zero decay conditions while the Cullis and Jones analysis
valued waiting lists for all conditions. In addition, given that the purchase of private
care depends on ability to pay and so income, and that income has a log normal rather
than a normal distribution, the Cullis and Jones assumption of a uniform distribution
of waiting values may be an overestimate. The estimates from the contingent valuation
exercise were also lower than those of the value of transit time [M.V.A. Consultancy
et al. (1988)]. Again, the difference was in the expected direction: time spent in transit
is closer to an exclusive use of time, whereas those on waiting lists can perform some
of their normal activities.
Johannesson et al. (1998) used contingent valuation exercise to derive willingness
to pay for reductions in waits in Sweden. With a waiting time guarantee of 3 months
given by the public health care system in Sweden taken as the benchmark, individuals
in a survey in 1995 were asked how much they would pay for each of two different
insurance plans that would reduce waiting to one week and to six weeks, respectively.
In estimation, the authors allowed the probability that individuals would pay zero to
12 The contingent valuation design conformed to the practices listed as desirable in the National Oceanic and
Atmospheric Report (1993).
13 Rebasing to 1991 prices gave an estimate of approximately £50 per month.
avoid a wait to be strictly positive. The results suggested that on average individuals
were willing to pay about SEK 2,000 per annum for an insurance plan which reduced
waiting from three months in the state health care system to six weeks. In comparison
to other estimates presented here, the results suggest individuals were willing to pay in
the order of £95 to £100 per month (in 1991 prices) for a reduction of a month's waiting
time.
3.3. Supply side: estimating the impact of supply variables
More research effort has been directed to assessing the relationship between waiting lists
and the supply of public sector care (primarily again in the UK NHS). In general, most
studies that examine the association between list length and some measure of the level
of supply (such as number of consultants, number of beds, hospital expenditure) find no
clear pattern. Much of this literature simply examines bivariate relationships between
some measure of supply (often at a geographically aggregate level) and numbers on
waiting lists. Such an approach does not take into account the possible endogeneity of
either demand or supply, and perhaps not surprisingly therefore, few clear results have
emerged.
Investigating the impact of changes in supply on changes in numbers on lists, Frost
and Francis (1979) and Frost (1980) tested Frost's (1980) assertion that consultants
(hospital doctors) adjust their waiting list/admissions thresholds in order to maintain
a constant waiting list. The hypothesis implies that the elasticity of numbers on the
waiting list with respect to supply of consultant numbers is unity. Using time series data
on total numbers on the list and the number of consultants Frost (1980) found that a 1%
increase in the supply of consultants led to approximately 1% increase in the size of
lists. Using cross section data [Frost and Francis (1979)] found similar results. Findings
such as these that are used either to support the hypothesis that "nothing can be done
about lists", or to support "supplier induced demand": an increase in resources will
simply lead to greater demand [Pope (1992), Roland and Morris (1988)].
The Frost (1980) evidence has been subject to critical comment. McPherson (1981),
whilst conceding that some unknown element in waiting lists will be "consultant in-
duced", pointed out that on average Districts in any Region with greater population
will, in a centralised health care system, have greater numbers of both consultants and
waiting patients. In short, waiting lists adjusted for population are likely to be very sim-
ilar between districts. McPherson (1981, p. 194), having made a crude adjustment for
population size (and ignoring two data points), commented ". . . to argue that in aggre-
gate the numbers of consultants determines waiting list size would be foolhardy in the
extreme".
Buttery and Snaith (1979) used cross sectional data for 1977 and found no clear as-
sociation between lists and surgical provision. Yates (1987) found no linear relationship
between shortage of beds and long waiting times using data from the 1970s and early
1980s. Goldacre et al. (1987) used time series data for 1974 to 1983 and found no re-
lationship between admissions and length of list. Newton et al. (1995) used time series
data for 1987-1994 to examine the relationship between changes in the number of ad-
missions to hospital and changes in list size. They found that changes in the number of
admissions were inversely correlated with changes in list size, so an increase in admis-
sions did not reduce list length because additions to the list tended to increase at the
same rate (presumably because as lists fell more patients were referred onto the lists).
But waiting times (as distinct from numbers on a list) appeared to fall as a result of
increased admissions.
So much of this research is inconclusive. Many have examined numbers on lists rather
than average time (or the distribution of time) spent on lists. But it is the latter factor
which affects the behaviour of demander or their agents (the family doctors who make
referral decisions). Failure to model the demand-side response to a change in waiting
times reduces the relevance of the research. Martin and Smith (1999) attempted to esti-
mate a simultaneous model of demand and supply in which waiting times are the price
of elective surgery. They argued on the demand side, an increase in waiting deters the
marginal joiner of the queue, and on the supply side, an increase in waiting times leads
either to additional resources and so supply of care, or on pressure to improve efficiency
and so to more care. From this they argued that correct estimation requires estimation
of a model in which demand, supply and waiting times are endogenous.
They do not directly estimate the possible models of Section 2, but argued that de-
mand and supply of elective surgery will be affected by the following variables:
demand = f(waiting time (-), medical need (+), GP supply (?), provision of
private sector supply (-));
and
supply = g(waiting time (+), provision of NHS beds (+), length of stay in
hospital (-), share of elective surgery done as day cases (+),
proportion of admissions that are elective (?)).
where their priors as to the direction of the associations were as given in parentheses.
They estimated these demand and supply functions using small area level data for
two years (1991 and 1992). The unit of observation was 4985 "synthetic wards", (small
areas) with an average population of 10,000 people in each, for which they had over
2 million records on elective episodes of treatment in routine surgery or gynaecology.
(Elective episodes are those for which waiting lists are used to allocated care.) They
allowed for endogeneity of utilisation, waiting lists, the proportion of surgical cases
that are elective, the proportion of admissions that are day cases and the average length
of stay in elective surgery.
The results for the demand equation indicated that the direct impact of waiting time
on demand for elective surgery is small. The long run elasticity estimates suggest that a
1% decrease in waiting times will be associated with a 0.09% increase in demand. From
this the authors conclude that the induced demand effect of a reduction in waiting lists
is very small. In the supply equation, there is a larger association between waiting times
and the supply of elective surgery: a 1% increase in elective surgery is associated with
a 0.35% increase in waiting times. Solving for the equilibrium relationship between
waiting times and supply, Martin and Smith found the net effect (i.e. after taking into
account both supply and demand responses) of an increase in NHS resources on waiting
times to be negative but not large.
In recognising endogeneity of demand, supply and waiting times, the paper is a direct
attempt to estimate the impact of the behaviour of demanders and suppliers of care. It
therefore addresses the issues raised by the theoretical models of waiting lists. However,
the results of the study cannot be used to distinguish between the different hypotheses
advanced above. For example, the authors cannot test whether increases in lists lead to
more resources with no change in efficiency (an Iverson type effect) or whether longer
waiting lists lead to higher effort and greater efficiency.
3.4. Supply side: inter-sectoraleffects
In a study that allowed for interaction between the public and private sector,
McAvinchey and Yannopolous (1993) estimated a dynamic cost shares model with three
goods: public sector medical care (the NHS), private sector medical care and all other
consumption (see Section 2.4.3). The price of NHS care was the waiting time cost (see
Section 2.4.3) and the price of private care was the price of private insurance. From their
estimates from aggregate annual data for England, Scotland and Wales for 1955-1987
they derived both short and long run own-price and cross-price elasticities for NHS and
private care. The own-price short run elasticities were small, and were smaller for the
NHS than for private care (for NHS care they ranged from -0.29 to -0.68 and for pri-
vate care from -0.79 to -0.85). The computed long run elasticities were, as is usual,
much higher than the short run elasticities. The long run own-price elasticities in the
two sectors are similar and highly elastic, being around -4.5 for the two sectors. These
are perhaps rather high in comparison to price elasticities estimated for other health care
markets. The movement of patients into the private sector following a rise in the costs
of waiting (elasticity estimate = 0.6%) is estimated to be slightly less than that into the
NHS following a rise in insurance premia (elasticity estimate = 0.82%).
The results indicate both that waiting lists appear to act as a price, but also that the
demand for care will depend on changes in both the public and private sectors. If, for
example, a 1% rise in waiting lists is followed by a 1% increase in premia the net
demand for NHS care will change very little. So if increases in NHS lists are offset
by rises in private sector costs, then NHS demand may be little affected by waiting
list changes. Further, the authors note a positive correlation between the number of
consultants and the waiting time on a waiting list which is consistent with a "supplier
induced" effect on the share of private health care.
Besley et al. (1999) examined the impact of the quality of NHS care, as measured
by the length of waiting lists, on the demand for private care. They advanced a simple
model of the demand for private insurance in a market in which individuals have au-
tomatic entitlement to public sector care. They argued that the consumer's decision to
buy private health insurance is a function of the quality of care in the two sectors. Un-
der the assumption that the quality of NHS care is always lower than quality of private
care (since any private supplier who had lower quality than a public provider would
have zero demand since the public sector is free at point of use) they derive analytical
results for the impact of income and quality of NHS care on demand for insurance. The
analytical results show that selection into private insurance, and that a lower level of
relative quality in the public sector increases demand for private insurance. Empirically,
this model suggests investigating the determinants of demand for private health insur-
ance as a function of the relative quality of public sector provision and other individual
characteristics, particularly income.
The model was estimated using 5 years of micro data from a cross sectional survey,
the British Social Attitudes Survey (BSAS). These data cover the period 1983-1991
(i.e. prior to the NHS health service reforms). The quality of the NHS was measured in
a number of ways: the total numbers on waiting lists, the total number of staff, spend-
ing on headquarters, spending on support staff, spending on treatment and the numbers
waiting over a year. NHS quality is measured only at a regional level, so regional mea-
sures were matched to individual data using location data of individual respondents.
The effect of the regional quality indicators is identified from deviations from regional
and time means, which constitutes a stiff test of the NHS quality measures. Strictly
speaking, the quality measures ought to have been entered as the differences between
the private and public sector quality measures. As waiting lists are zero in the private
sector, the measures of waiting time are de facto in differences, but the other measures
of quality are not.
The results indicate that only one of the NHS quality measures appears to determine
private insurance purchase (conditional on individual level covariates, time and regional
dummies): the proportion of individuals waiting longer than a year. This gives qualita-
tive support to the hypothesis that private demand is affected by waiting lists.
4. Waiting: policy issues
Section 2 addressed the question "what waiting time is appropriate?". Waiting is neces-
sary when resources are insufficient to furnish immediate medical treatment; "optimum"
waiting time occurs when a socially optimal allocation of resources has been provided
in the NHS and when patients are ordered optimally in the list. However, even if an
efficient waiting time is not achieved, it remains possible to be cost effective in the man-
agement of those who wait. In this case the aim is essentially X-efficiency; maximising
physical output from inputs employed and choosing least cost input combinations.
Optimum waiting time is unlikely to be zero (for if the optimum were zero this would
imply that sufficient resources had been made available to ensure immediate treatment
for all case types). If the analysis of Section 2 is accepted the application of much of
the empirical work to policy is not direct. For example, the demand studies effectively
measure, in the view of potential patients, the fall in the total benefit curves from to to
the actual or anticipated treatment date in parts (b) and (c) of Figure 1. Some waiting
costs (loss of total benefits) are optimal for illustrated case types 2 and 3. However,
without knowing these amounts for each case type, empirical estimates of actual costs
sit in a welfare vacuum. In the absence of further assumptions and/or empirical work it
is impossible to look at waiting costs of, say, £x and draw strong normative conclusions.
In the absence of an over-arching primary efficiency calculus, policy has been directed
at a secondary level: at reducing the waiting numbers and their associated waits on the
assumption that this is welfare improving.
In practice, governments focus on ill-defined, almost ad hoc, target waiting lists
and/or "guaranteed" waiting times. In the UK for example, waiting lists have assumed
the same importance as other indicators of government performance, such as public
sector borrowing [Cullis and Jones (1983, 1985)]. In recent years promises to "guaran-
tee" waiting times have emerged in general election campaigns [Yates (1991)]. But the
targets implicit in a reduction of waiting lists and waiting times are not rationalised in
terms of welfare maximisation and, as explained in Section 2, the relationship between
waiting lists and welfare is far from obvious.' 4 In just the same way as governments
pursue public sector borrowing targets, waiting list targets are established without a
strong welfare foundation. Yet from an economic perspective, the key consideration is
how a reduction in waiting lists or times enhances welfare. 1 5 In this section a taxonomy
of alternative policy options is presented and in each case efficiency and equity aspects
are considered. Efficiency, defined in Paretian terms, is achieved when it is impossible
to reallocate resources to make one individual better off without making another worse
off.
4.1. A taxonomy of policy options
As policy has been premised on the assumption that waiting list reduction is desirable,
the following policy options have been considered.
14 Analysis of waiting lists [e.g. see Lee et al. (1987), Yates (1991)] show that they comprise: patients who
have already died; patients who would refuse treatment if offered it immediately and patients who benefit
clinically from waiting. For example, Hemingway and Jacobson (1995, p. 819) note that, while one third
of patients continue to face operations, "the natural course of recurrent throat infection, the main indication
for tonsillectomy, may be one of improvement". They note that, " ... a prospective study to determine the
morbidity caused by a delay in tonsil surgery found that a fifth of patients grew out of their condition and
were spared surgery". Though views on tonsil surgery differ [Yates (1995)].
15 In the UK, 52% of those on inpatient waiting lists wait for General Surgery, Trauma and Orthopaedics
and Ophthalmology (see Chart 2) and suffer from a variety of complaints, e.g. hernias, varicose veins, joint
complaints. As patients also vary according to personal and social characteristics, it is unlikely that common
targets for waiting lists and/or waiting time will be easily reconciled with maximisation of welfare. Pope
(1992, p. 577) comments: "The length of time people wait to be admitted is undoubtedly important, but
factors like severity, urgency and social and physical circumstances of the person waiting may be equally
important in assessing the situation".
4.1.1. Demand rationingand reduced waiting
Waiting has been perceived as a reflection of excess demand at zero user cost. One re-
sponse is to introduce mechanisms to ration demand.16 Seldon's (1967) answer would
be to introduce a price, so that supply is rationed to those with the highest valua-
tion. 17 Private (market-based) insurance often relies on utilisation reviews, protocols
and provider-financial systems (e.g. capitation fees for physicians) to restrain demand
pressures. Buchanan (1965) interprets the problem of excess demand with reference to
an inconsistency inherent in non-market decision making. When individuals vote for a
tax contribution that pays for the care of others they attach less weight to others than
they do to themselves. The supply of health care resources is less than would emerge
when individuals demand private medical insurance in a market scenario (when each in-
dividual is conscious of the benefits to be personally experienced). Hence waiting lists
are endemic in the absence of institutional reform. Buchanan's policy solution is that
health services be rationed and distributed in a manner that would ape the outcome of a
competitive market.
The policy of aping competitive markets is difficult to implement. To set a price, or
to ration efficiently, would require that the point at which marginal benefit of medical
treatment equals marginal cost be identified. It would be necessary to assess individ-
uals' preferences accurately to know marginal benefit. It is far from obvious that the
allocation of a ration and the removing of waiting would enhance welfare. Some pa-
tients may prefer waiting to the prospect of no further treatment. 18 Also, on efficiency
criteria there is a case for considering "option demand" [Weisbrod (1964)]. Even if indi-
viduals make no use at all of hospital services, they still feel better off to know that such
services are at hand if required. An allowance for option demand would be necessary
when determining the appropriate ration.
Turning to equity considerations, constructing a competitive market solution in the
public sector is likely to be less attractive than instituting "fair" procedures. Frankel
and West (1993, p. 129) argue: "... at the limit we probably all accept that there are
conditions which are not recognised as diseases and as a consequence which do not
deserve to be treated by the NHS". It may not be fair to offer all treatments via the public
sector but, if equity arguments carry force, who should determine priorities? Frankel and
West (1993) note that one of the implications of waiting lists is that it permits policy
makers to blur such rationing issues.
16 As Globerman (1991) notes, non-price rationing in other markets (e.g. rent control) has been criticised
when price is set "too low".
17 The Telegraph, Wednesday, 8 October 1997, reported that the British Medical Association favoured a small
fee to consult general practitioners in order to ration demand.
18 Cullis (1993) notes individuals in the UK may be better off waiting than the non-insured in the USA who
do not get a chance to join a list.
4.1.2. Supply expansion and reduced waiting
If waiting arises from excess demand, an alternative to rationing demand is to increase

supply. Whether under-provision is the problem is a moot point. For example, while
Yates (1987) argues that there is under-provision in the UK by comparison with other
European countries, he also notes that there are countries, which spend less than the UK
but which do not have a waiting list. Within the NHS he reports that waiting lists appear
in both efficient and inefficient health districts.
It is always possible to respond to a waiting list problem by commending better use
of existing resources [e.g. Yeates (1980), Ingram (1980), Cottrell (1980), Yates (1987),
Mills and Heaton (1991)]. To be more specific, Frankel and West (1993) call for better
use of operating theatres and clinical staff. Also, in the UK context, they call for greater
incentives for hospital consultants to spend more time treating NHS patients (rather than
private patients). However, there can be dangers if better use of resources is interpreted
as more intensive use of resources. Globerman (1991) warns that measures to speed
up treatment to reduce waiting can have adverse consequences on quality of treatment.
Moreover, if increasing throughput of patients leads to an increase in re-admissions the
impact on waiting lists is not obvious. Even if the quality of care is not affected there
remains the question of whether more productive use of resources will reduce waiting
time; reduction of waiting time may simply stimulate demand from new patients. De-
mand for medical care is affected by the availability of new treatments and (as argued
in Sections 2.4.2 and 3.3) by additional suppliers of medical care who have an incentive
to foster waiting lists. Empirical studies do not show a clear association between in-
creases in admissions and length of lists. For example, in the UK Goldacre et al. (1987)
did not find a significant negative correlation between the number of admissions in one
period and the length of waiting list in the next period, while Martin and Smith (1999)
(Section 3.3) reported that an increase in supply would reduce waiting time.
If additional funds are made available, it matters how they are used. Increased funding
per se is less likely to reduce waiting than using funds to create direct incentives to
reduce waiting. For example:
(a) In the Australian State of Victoria three sets of changes were introduced in 1991
[Duckett and Street (1996)]. These were: (i) directives for surgeons to better cate-
gorise patient need for treatment; (ii) hospital funding linked to activity rather than
unmet demand (lists); (iii) funding linked specifically to treatment of patients on
the list. The intention of the last measure was to ensure that, as hospitals increased
their throughput in response to payment based on activity, they did not allow lists
for routine operations to grow. The impact of the policy was a fall in both waiting
times and numbers on lists.l 9 However, while the results appear encouraging, the
follow-up period (analysed by Duckett and Street) was only a year and the longer
run policy impact has yet to be evaluated.
19 These changes were not simply due to re-categorising of patients.

(b) In Sweden sponsors of health care services set a 3-month maximum waiting time
guarantee for the year 1992. Waiting lists dropped substantially between 1991 and
1992. Providers undertook actions to bring down lists when paid to do so, but once
funding stopped they no longer took these actions [Hanning 1996].
(c) In the UK a "Waiting Times Initiative" ran from 1987/88 to 1993/94 inclusive. Over
this period a total of £252 million was allocated to reduce long waits. The Inter-
Authority Comparisons and Consultancy (1990) reported that mainstream funding
had limited success and recommended that money be paid to health authorities who
succeeded in reducing waiting time [see Iverson 1993]. Between 1991 and 1994
the percentage waiting over 12 months was established as a performance measure
and there was a reduction of long waits. However, after March 1996 there was an
upturn in patients waiting over 12 months and Edwards (1997) questions the long-
term success of the initiative.
While there is some support for the proposition that funding must be linked explic-
itly to reducing waiting there are problems. Once hospitals are (even partly) penalised
for having long lists, list composition is likely to be managed more effectively [Han-
ning (1996), Druckett and Smith (1996)]. This may have some benefits, for example,
removing patients who no longer need treatment. However, shortening of lists does not
necessarily signal an equal increase in productivity if there is an incentive to admit fewer
patients to waiting lists. The equity implications of increased funding also require con-
sideration. Critics have argued that individuals from the middle class are better able to
manipulate systems such as the NHS [Le Grand 1982]. The impact of additional fund-
ing may be to reduce waiting for those most able to secure treatment and the policy may
not be progressive in terms of income distributional impact. However, more recent UK
evidence [e.g. O'Donnell and Propper (1991)] suggests that the middle class bias is no
longer evident.
4.1.3. Subsidies to reduce waiting
Rather than increase funding for public sector treatment, waiting for public sector treat-
ment may be reduced by using additional public expenditure to finance a subsidy to
those who will leave the waiting list to purchase treatment in the private sector [Cullis
and Jones (1983, 1985)]. In Figure 7, Do represents the demand for treatment of a med-
ical condition. At OP1 the private sector deals with 0ql. The public sector, at zero price,
reduces the remaining demand of ql q3 by ql q2, leaving a waiting list of q2q3. Whether
this constitutes an "optimum" wait is a moot point. However, assume that a reduction to
q4-q3 is desired and that a budget q212q4 has been allocated.
A first response to this situation (described in Section 4.1.2) would be to increase
public expenditure to increase output in the public sector by q2q4 cases. The cost of
this expansion is q212q4 (assuming that private and public sector provision are equally
cost effective at MC = AC). The same result might be achieved with a lower budget if
demand for treatment is price elastic. If a subsidy were offered to those willing to pay
something for private treatment, the cost of the subsidy will be P134P 2 when the private
sector expands by ql -q5. Assuming supply is infinitely price elastic in the private sector,
this switch to the private sector releases resources in the public sector, so that the waiting
list is reduced to q4-q3 20 The cost of the subsidy is less than the increase in expenditure
on the NHS of q212q4. (It should be noted that as ql53q5 equals q212q4, the resource
cost of expanding capacity are identical so that the previous comparison is only in terms
of "exchequer" budget costs.2 1)
While the subsidy scheme ensures that those with the highest individual marginal
valuations are treated, no such guarantee applies to the alternative strategy. However,
the case for choosing a subsidy requires, in part, that demand for treatment is price
elastic. To ascertain price elasticity, Cullis and Jones show that elasticity (e.g. at point 5
in Figure 7) is approximated for a linear demand function by the ratio:
cases treated in the NHS + waiting list

1= (12)
cases treated privately
When this ratio exceeds one, the option of providing a selective subsidy to private con-
sumers of this type of medical treatment proves attractive but there are problems.
First, there is the effect of "supplier-induced demand" (SID). With increased avail-
ability of resources, doctors (with asymmetric information) can respond by simply stim-
ulating demand. As noted in Section 3.3, Frost and Francis (1979) and Frost (1980)
found that the size of the waiting list responded positively to an increase in the number
of consultants (a 1% increase in the number of consultants resulted in a 1% increase
in the size of the waiting list). It follows that a growth in the number of surgeons may
simply be reflected in a growth in the waiting list. The implications for the choice of
private sector subsidy, rather than additional public sector funding, depend on the way
in which SID operates. One possibility (Case A) is that the availability of additional
resources may simply shift the demand curve in Figure 7 to the right, i.e. to D 1. Alter-
natively (Case B), the availability of additional resources specifically increases demand
in the public sector, so that there would now be a kink in the demand curve (which be-
comes Do59q6).2 2 In Case A, if the government increases public expenditure there is an
increase in demand (Do to D 1) but there is an increase in treatment in the private sector
and the waiting list reduction is achieved (i.e. the waiting list is qg-q6 which is equal
to the target q4-q3). In Case B, demand increases only in the public sector and even if
there is an increase in public expenditure the waiting list will remain q4-q6. Now con-
sider the effect of using a subsidy (Case C). If a subsidy is used, when Do increases to
D 1 the waiting list will remain on target q4-q3 q8-q6 but the financial costs increase
20 As is evident in Figure 7 there are assumptions, e.g. concerning costs of treatment in the private sector. It
is assumed, for example, that the costs of treatment do not rise in the private sector as a result of increased
demand and that quality of treatment is not lower in the private sector (and conversely in the public sector).
21 Other costs (e.g. the dead-weight loss of taxation) are assumed constant in raising the same exchequer
funds.
22 This second impact appears consistent with some empirical evidence [e.g. Frost (1980)].
Price
D1
P1
1 2
1 2 MC=AC
P2
I II
I I
I I I
,I , ,I II I
I I
I, I I I
O_I~_I~~ ,I .I~I
q, q5 q9 q7 q2 q4 q3 q8 q6
Private
Psectore Public sector Waiting list Cases treated/t
I I
sector Waiting list
' d Private sector ubic

P sector Public sector Waiting list
' '-' C;CaseA
. I
I I I
iseictore: Public sector , Waiting list

' ' - b' Case
Case B
Figure 7. Effects of subsidies and expansion af the NIHS on waiting lists.
(P1 78P2 rather than P1 34P7). It follows that, if there is a need to allow for SID, the
subsidy only operates with the risk that a budget constraint will be breached. Of course,
this risk can be monitored. As the demand curve shifts to the right, demand becomes
inelastic and by reference to Equation (12) there is a signal that the subsidy is no
Price
longer as attractive.
The assumption that there is an infinitely elastic supply of resources for the private
sector is important. A second potential problem arises if the expansion of the private
sector is only possible because resources are taken from the public sector. If resources
were diverted from the public sector the subsidy would have two effects. The first is
a positive impact on waiting by removing demand. The second is a negative effect by
switching resources away from the public sector to the expanding private sector. If both
effects operate, the net effect of private provision on waiting in the public sector is
indeterminate. Iverson (1997) examines a situation in which the more price elastic the
demand for public treatment the more that waiting time is likely to increase (if the
private sector option is pursued).
A third potential problem of using a price subsidy concerns the quality of output in
the public sector. Expansion of the private sector is likely to remove those patients best
able to exert pressure for improvements in the delivery of public sector medical care.
Besley and Gouveia (1994) note that private sector expansion may reduce the size of the
political coalition supporting the current level of public provision. The "exit" of some
patients will affect the potency of "voice" in the public sector [Hirschman (1970)], if,
as Besley et al. (1999) confirm, it is the wealthier and better educated who express
dissatisfaction with public sector provision.
Equity considerations are also important. The scheme will appear regressive if higher
income "waiters" (more able to purchase private sector treatment) qualify for a sub-
sidy. 23 However, there is no presumption in the above analysis that they will get a re-
fund for their contribution to the NHS and those remaining in the public sector may
have shorter waits. Income-related subsidies could be used to alter the distribution of
benefits. However, the transactions costs of administering such schemes will also need
to be considered.
4.1.4. Encouragingprivateprovision
Offering subsidies to patients to purchase private sector treatment encourages demand

in the private sector. An alternative policy response is to encourage private sector supply.
While not looking explicitly at medical care, Shmanske (1996) considers a policy option
to reduce waiting by encouraging private sector supply. In Figure 8 the demand curve
is DD'. If the public sector supplies the good at a price of Pg (below average cost ACg)
there is a loss of Oqg(ACg - Pg) and to mitigate this loss, public sector output is reduced
to Oq'. Average cost is now AC' and the loss is Oq'(AC' - Pg). However, with a shortfall
of q'-qg individuals must queue. The value of the good to the consumer at the margin
is Pv, and a waiting cost of Pv - Pg is incurred to equilibrate the market.
If a private sector firm were to offer units at a price of Pp intermediate to Pv and
AC' (and therefore above Pg), the firm cannot supply the whole market. No one will
23 Iverson (1986) shows that private practice can reduce waiting time for patients but with an increase in
inequality (as the demand for private health care depends on income).
Price
AC
ACg
Pv
AC'
Pg
0 q' qg QX/t
Figure 8. Waiting in a non-contestable market.
patronise the private firm for those waiting know that, if other demanders switched to
the private supplier, costs of waiting in the public sector will be reduced. With this free
rider problem the market is non-contestable (i.e. the private firm is unable to supply the
whole of the market).
If the market were contestable an entrant could obtain the whole market by setting
a price just below Pv. However, if the full price of Pv is composed of a "rigid" price
Pg and a "flexible" price of Pv - Pg, the residual demand curve facing the private firm
entrant (allowing for the existence of the public sector supply and the waiting problem)
is Pv 12D'. The segment Pv1 is parallel to the original demand curve because, for any
price P between Pv and Pg, the public sector supplier sells the first q' units. When
a queue of waiters causes marginal waiting cost of Pv - Pg, the segment Pv 1 of the
residual demand is below average cost.
If there were no subsidy for losses (i.e. AC' - Pg per unit) incurred by the public
sector supplier, then this supplier would compete like any private business (as far as
covering costs is concerned). The public sector supplier would have to charge a price at
least equal to the height at the minimum point on the average cost curve. In this case,
entry of a second firm would not be precluded. Shmanske (1996) argues that, if queuing
is created by the subsidy, the policy option is to "privatise" public sector provision (i.e.
remove the public sector subsidy) and thereby encourage competition from the private
sector. However, if government policy also addresses under- consumption of the good
(as in the case of medical care) there is a case to retain a subsidy in some form.
4.2. Efficient waiting
In this sub-section we address policy options that aim to manage existing lists. The
objective is to reduce costs for existing patients and the focus of attention is on
measures to prioritise patients. Of course, different rules of prioritisation may stim-
ulate or deter demand from new patients [e.g. Bowles (1982), Worthington (1987,
1991)]. For example, Goddard and Tavakoli (1994) use a queuing model to assess the
impact of: (i) treating complaints in the order that they join the queue; (ii) affording
priority to achieve equality between the total level of "suffering" experienced by all
complainants; (iii) offering rapid rationing to as many seriously ill complainants as the
system can cope with. The method of prioritising affects demand (e.g. in the third case
considered, many with minor complaints would not bother to join the queue as there
would be no hope of treatment).
While the focus of attention in this section is on cost-effective management of exist-
ing lists the impact on overall demand cannot be ignored. Nor can equity considerations
be dismissed. Prioritisation of patients based on explicit and consistent criteria is com-
mendable for waiting list management but, for equity, this also applies to decisions to
refer patients to waiting lists. 24
4.2.1. Indicesforprioritisation
A number of different criteria can be employed to prioritise waiting list patients. A

"first come first served" approach or "a common guaranteed waiting time" approach
may seem "fair" ways of proceeding [Higgins and Ruddle (1991)],25 but both these
approaches imply that individuals who have minor ailments but come earlier will be
treated while others who come later with more substantive ailments (or social need)
would continue to wait.
Naylor et al. (1990) use clinical guidelines to determine the urgency of patients with
angiographically proven coronary disease and in need of revascularisation procedures. 2 6
One comment on British experience is that, implicitly, doctors have adopted a forgone
present value of patient earnings criterion [Aaron and Schwartz (1984)].27 Of course, a
broader set of criteria than either clinical guidelines or present value of patient earnings
is likely to prove more acceptable. Culyer and Cullis (1976) suggested that selection of
patients who are waiting should depend on: (i) time already spent on the waiting list;
(ii) urgency based on the expected deterioration of the patient's condition; (iii) urgency
based on the patient's health status; (iv) urgency based on the social productivity of the
patient and the number of economic dependants; (v) urgency based upon other social
factors. An attempt to apply these broader criteria (by allocating points for each ele-
ment on a range between 0 and 4) was adopted by a consultant anaesthetist at Salisbury
District Hospital in the UK [Edwards (1994)].
Edwards (1997) argues that waiting list and waiting time information should be pub-
lished not only by clinical speciality but also by medical condition so that, if national
24 Hicks (1972), Forsyth and Logan (1968) reported evidence of wide variation in referral rates.
25 "In many ways, a system of rationing by queuing is a fairer more open way of restricting access to scarce
resources than some of the alternatives used in other countries" [Higgins and Ruddle (1991, p. 18)].
26 Factors considered were severity and stability of symptoms of angina, coronary anatomy from angio-
graphic diagnostic studies and the results of non-invasive tests for the risk of ischaemia.
27 The implication is that the narrow objective of maximising national output applies.
Clh. 23: Waiting Lists and Medical Treatment 1241
maximum waiting times are to be stated, there should be a gradient of "clinically appro-
priate times". More difficult is to ascertain how social considerations are to be integrated
and, on this, Edwards calls for greater information about the views of relevant interested
parties (patients, members of the general public, general practitioners, purchasers and
politicians).
Decisions will be sensitive to the precise formula adopted and the question arises as to
who should determine and implement criteria. In New Zealand in the late 1990s one of
the tasks of the National Health Committee is to generate criteria for the prioritisation
of elective waiting lists. Both clinicians and members of the public are involved in
determining criteria [Dixon and New (1997)].28 When considering who will implement
selection criteria, the study by Brattberg (1988) of an experiment at a Department of
Anaesthesia at Sandvikken Hospital in Sweden is relevant. Rankings of patients made
by the secretary and nurse (based on a questionnaire completed by the patients) were
compared with assessments by the doctor (based on consultation with the patient). The
general result was that the secretary and nurse were inclined to overestimate urgency of
treatment.
One selection criterion employs Quality Adjusted Life Years (QALYs). Williams
(1988) suggests that, for each medical speciality, patients should be ranked periodi-
cally by a predicted QALY score. Selection should then be carried out to maximise total
predicted QALYs secured. Assume that there are two specialities, A and B, and that cur-
rently the same level of resources are allocated to both specialities. If in A the QALY
per pound spent for the last patient treated exceeds that of the last patient treated in B,
the implication is that resources should be moved away from speciality B to speciality
A until QALY per pound is equalised. 2 9 Attempts have been made to operationalise
this procedure. Gudex et al. (1990) calculated the QALY gain from treatment versus no
treatment and the QALY gain from treatment now versus treatment one year later for 22
common medical conditions on Guy's Hospital's general surgical waiting list [see also
Edwards and Barlow (1994), James et al. (1996)].
Some prioritisation processes concentrate simply on costs (e.g. the Duthrie Report
for the UK recommended that points be allocated to waiting patients according to the
resources required, so that estimates could be used to assess the number of operating
theatres required to meet pressure from different the waiting lists3 0). By comparison,
Williams' measure has the advantage that it also incorporates an estimate of the produc-
tivity of treatment time. Williams' suggestion also means that the doctor no longer has
the same incentive to foster long waiting lists (as length of waiting list is no longer the
28 The criteria included both clinical and social considerations but not length of time waited.
29 As Williams (1988, p. 240) notes, a speciality with a small list of people waiting a short time for very
beneficial treatment may still have priority over a speciality with a long list of patients who have waited a
long time. Yet, as Edwards (1994) comments, patients have the same opportunity of being treated (given their
expected total health gain per pound of treatment) regardless of which clinical speciality they required.
30 For the Duthrie Report see Department of Health and Social Security (1981). Note that Donaldson and
Stoyle (1987) argue that using the waiting list to assess theatre time is more informative.
sole criterion for resource allocation). Moreover, patients who can expect little benefit
from treatment soon realise that the only course is to move to the private sector (i.e.
patients form correct expectations of what might be possible if they join waiting lists).
The approach is not based on Paretian efficiency; the objective is the maximisation of
health not of welfare per se. However, it is questionable that the measure is sufficiently
robust for waiting list prioritisation [Coast et al. (1996)]. With respect to equity, Broome
(1987) notes that individuals have "claims" to the publicly funded health care system,
and that claims should be met in proportion to their strength but never be completely
over-ridden. It would follow that people should always be entitled to treatment as long
as they are prepared to wait an appropriate length of time. 3 1
4.2.2. Quasi (internal)markets
In 1990 NHS reforms created an "internal" or quasi-market. While there were broad
policy objectives, Frankel and West (1993, p. 130) assert that "... the reforms can ...
be best understood as the definitive waiting list initiative". The emphasis again is on
cost efficiency [Jones and Cullis (1996)]. Quasi-markets make a distinction between
purchasers and providers. A health authority or a GP (as "purchaser") acts as the agent
of the patient to secure the best quality of hospital treatment (from the "provider"). The
greater knowledge of agents about the availability of treatment at different hospitals is
expected to assist the principal (i.e. taxpayer/patient). For example, agents may allo-
cate funds to purchase treatment for their patients from hospitals and, when choosing
hospitals, waiting time will be an important consideration. Health authorities and GPs
are not tied to local hospitals; they are able to negotiate contracts with hospitals lo-
cated further afield but with lower waiting times. In Working for Patients [Department
of Health (1989)] and associated documents it is asserted that increased efficiency from
competition would decrease waiting lists [Mullen (1993)].
While hospitals compete for patients and respond to incentives to reduce waiting
times, the transactions costs associated with operating the internal markets cannot be
ignored [Bartlett (1991), Jones and Cullis (1996)]. Also, critics suggest that, if the incen-
tives are to reduce waiting lists, this may be achieved in part by refusing patients access
to lists. Mullen (1993) argues that, if hospitals are judged by waiting times or length of
waiting list then, provided hospitals have dealt with the contract numbers agreed with
the purchasers, they have an incentive to decline admissions to their waiting lists.
4.2.3. Reducing uncertainty
Propper (1990, 1995) valued a reduction of one month on a UK waiting list for non-
urgent treatment at £50 (in 1991 prices). The average value of the disutility of uncer-
31 Stronger normative criteria may give weight to the numbers treated. Culyer and Cullis (1976, p. 262) note
that "... a greater reduction in total need is possible if, say, two persons with relatively short expected lengths
of inpatient stay and low index scores are admitted instead of one person with a long expected stay and a high
index score".
tainty of admission date was, additionally, around £30 (in 1991 prices). One objective
of stipulating priorities is to reduce uncertainty for patients. For example, in the UK,
the "Patient's Charter" launched in April 1991 stated an entitlement: there was to be
guaranteed admission for treatment no later than two years from the date when the con-
sultant places the patient on a waiting list [Mullen (1993)].32 However, such guarantees
can prove inefficient. Using theoretical queuing models, Goddard and Tavakoli (1994)
comment that the Patient's Charter performs poorly in terms of efficiency because those
who are treated are not those who will receive the greatest benefit, as some with minor
ailments will have to be treated before others in order to guarantee the maximum wait-
ing time for all. Mullen's criticism also applies. If health authorities or GP fundholders
try to secure treatment with another provider to meet their guaranteed target, the other
provider may not have an incentive to add to their list in the absence of additional re-
sources.
It has already been noted that in Sweden after 1991 waiting time (for patients for any
of 12 different procedures) was to be limited to 3 months from the physician's deci-
sion to treat/operate. 3 3 However, only in the first year of the scheme was the guarantee
associated with extra resources. Waiting time was reduced and this was a result of in-
creased production, of improved administration of the waiting list and of a change in
attitudes toward waiting lists [Hanning (1996)]. Increased funding and a change in in-
centives achieved a more intensive use of medical resources as well as a reduction of
uncertainty.
Booking appointments can mitigate patient uncertainty. Some classes of patients in
the USA may wait longer than patients in the UK [Light (1990)], but patients in the
US who have definite appointments do not see themselves as "waiting" [Frankel and
West (1993)]. Of course, the administration of a booking system requires a relatively
predictable length of stay and, when booking systems have been used, emergency ad-
missions will fall if facilities for dealing with emergencies have already been assigned
[Devlin (1980) and Southam and Talbot (1980)].
5. Conclusions
Following an essentially Paretian perspective, theoretical considerations relating to

waiting lists and waiting times involve a number of deceptively simple arguments. In a
normative context it was suggested that optimal waiting times should be computed for
32 In 1995 the Patient's Charter promised that the waiting time guarantee of 18 months that covered hip, knee
replacements and cataract operations was to be extended to all admissions to hospital [Department of Health
(1995)]. Moreover, it guaranteed that 9 out of 10 patients can expect to receive an outpatient consultation
within 13 weeks of referral by the general practitioner.
33 It was understood that, if the hospital with primary responsibility for the patient cannot offer treatment
within such a waiting time, the patient would have the right to be treated in another hospital, or by a private
clinic, at the expense of the home hospital.
each case type and a booked inpatient date be offered. Given the stochastic nature of in-
patient demands, a further requirement was that the variance of the actual inpatient date
about the booked date be minimised. This normative perspective takes the goal of sup-
pliers and demanders as maximising welfare. But in practice, the actions of suppliers,
demander and governments are not necessarily to maximise social welfare, and theory
on waiting lists reflects this. It has developed in a more positive context in which central
contributions have focused on the role of waiting times in the decay of the benefit of
inpatient care, bargaining over budgets to be allocated to medical care and arguments
consistent with consultants seeking their self-interest.
Many of these theories have not been rigorously tested: in part the result of poor data.
However, in a more positive light, the relatively few estimates of the costs on consumers
imposed by waiting lists that have been made indicate magnitudes that may not be all
that great at the individual level. Econometric investigations of the relationship between
waiting lists and resource allocation to the public health care system have provided few
definitive results. Early work on the UK NHS suggested an increase in resources had
no impact on lists; later work has suggested an increase in supply may decrease lists.
Empirical work also indicates connections between the public sector in which there are
waiting lists and the private sector that operates alongside large publicly funded systems
on both the demand and the supply side. But given the absence of work related to the
normative theoretical considerations, it is difficult to assess the welfare significance of
these empirical findings.
A review of policy issues for dealing with waiting lists reveals that none is obvious
in terms of practical simplicity. Each option poses difficulties. However, the way in
which the policy issue is approached is of critical importance. Policy discussion should
be conducted using an explicit set of criteria against which options can be assessed.
The issue of the optimum waiting time for different medical procedures also needs to
be addressed. When addressing the issue of the optimum wait, questions of efficiency
and equity stand in sharp relief. The optimum wait is unlikely to be the same for every
medical condition and for every patient. Much policy analysis has ignored this. But these
attempts to "treat" the "symptoms" of waiting lists by relying on ad hoc methods of
waiting list reduction do not result in changes which are necessarily welfare improving.
More broadly, it was noted at the outset of this chapter that there is not necessarily
a consensus over the evaluative framework to be used to assess the significance of, and
costs and benefits of waiting lists. Whilst economics offers individual valuations as the
natural benchmark, this, as discussed above, does not always command wide assent in
the health care arena. However, without a resolution of the question of the yardstick by
which the costs and benefits of the delay of inpatient treatment is to be measured and
the construction of that yardstick, the further question of what is the optimal waiting
list or time for a state health care system to aim for is left in limbo. Suggestions for
waiting list policy that make sense by reference to one set of criteria do not necessarily
make sense by reference to another and almost certainly are not commensurate. Hence
the onus is on advocates for policy changes to make clear their underlying assumptions
Ch. 23: Waiting Lists andMedical Treatment 1245
and subsequent analysis. To date the implicit framework adopted by most researchers
in economics has been a Paretian one.
In summary, despite the considerable range of work that has been carried out on
waiting lists and related issues, the interpretation of data on waiting lists is difficult.
Lord Kelvin (1889) is often quoted as saying "When you cannot express it in numbers
your knowledge is of a meagre and unsatisfactory kind". The general conclusion here is
that even when the event can be expressed in numbers (for example, the NHS waiting
list) knowledge is of a meagre and unsatisfactory kind. The challenge is to root policy
discussion in productive, theoretical and empirical soil.
References
Aaron, H.J., and W.B. Schwartz (1984), The Painful Prescription: Rationing Hospital Care (Brookings Insti-
tution, Washington, DC).
Acton, J. (1973), "Evaluating public progress to save lives: The case of heart attacks", RAND Research Report
R-73-02 (RAND, Santa Monica).
Acton, J. (1975), "Nonmonetary factors in the demand for medical services: Some empirical evidence", Jour-
nal of Political Economy 83:595-614.
Alderman, H. (1987), "Allocation of goods through non-price rationing mechanisms, evidence on distribution
by willingness to wait", Journal of Development Economics 25:105-124.
Babes, M., and G.V. Sarma (1991), "Out-patient queues at the Ibn-Rochd Health Centre", Journal of the
Operational Research Society 42:845-855.
Bartlett, W. (1991), "Quasi markets and contracts: A markets and hierarchies perspective on NHS reform",
Public Money and Management 11:53-61.
Barzel, Y. (1974), "A theory of rationing by waiting", Journal of Law and Economics 18:73-95.
Bates, J. (1988), "Econometric issues in SP analysis", Journal of Transport Economics and Policy 22:59-70.
Besley, T., and M. Gouveia (1994), "Alternative systems of health care provision", Economic Policy 19:199-
258.
Besley, T., J. Hall and I. Preston (1999), "The demand for private health insurance: Do waiting lists matter?"
Journal of Public Economies 72:155-181.
Bowles, R.A. (1982), "Delay as a rationing device", International Journal of Social Economics 19:90-104.
Brattberg, G. (1988), "Priority setting with regard to placement on waiting lists to a pain clinic: The feasibility
of a delegated ranking procedure", Scandinavian Journal of Social Medicine 16:173-179.
Brookshire, D.S., M.A. Thayer, W.D. Schulze and R.C. D'Arge (1982), "Valuing public goods: A comparison
of survey and hedonic approaches", American Economic Review 72:165-177.
Broome, J. (1987), "Good, fairness and QALYs", in: M. Bell and S. Medus, eds., The Proceedings of the
Royal Institute of Philosophy Conference on Philosophy and Medical Care (Cambridge University Press,
Cambridge).
Buchanan, J.M. (1965), "The inconsistencies of the NHS", Occasional Paper No. 7 (Institute of Economic
Affairs, London).
Buttery, R.B., and A.H. Snaith (1979), "Waiting for surgery", British Medical Journal, Supplement (2):403-
404.
Buxton, M. (1992), "Are we satisfied with QALYs?", in: A. Hopkins, ed., Measures of the Quality of Life and
the Uses to which Measures may be Put (Royal College of Physicians, London).
Cairns, J. (1992), "Discounting and health benefits: Another perspective", Health Economics 1:76-80.
Coast, J., J. Donovan, J. Frankel and S. Frankel (1996), Priority Setting: The Health Care Debate (John Wiley
and Sons, Chichester).
Coffey, R. (1983), "The effect of time price on the demand for medical care", Journal of Human Resources
18:406-424.
Cottrell, K. (1980), "Count the cost of day cases", Health and Social Service Journal 90:1319-1322.
Cox, D.R., and W.L. Smith (1961), Queues (Chapman &Hall, London).
Cooper, R.B. (1981), Basic Queueing Theory (North-Holland, New York).
Cullis, J.G. (1993), "Waiting lists and health policy", in: S. Frankel and R. West, eds., Rationing
and Ratio-
nality in the National Health Service (Macmillan, London) 15-41.
Cullis, J.G., and PR. Jones (1983), "Inpatient waiting: A discussion and policy proposal", British Medical
Journal 287:1483-1486.
Cullis, J.G., and P.R. Jones (1985), "NHS waiting lists: A discussion of competing explanations
and a policy
proposal", Journal of Health Economics 4:119-135.
Culllis, J.G., and P.R. Jones (1986), "Rationing by waiting lists: An implication", American Economic
Review
74:250-256.
Cullis, J.G., and P.R. Jones (1997), Public Finance and Public Choice, 2nd edn. (Oxford University Press,
Oxford).
Culyer, A.J., and J.G. Cullis (1976), "Some economics of hospital waiting lists", Journal of Social
Policy
5:239-264.
Dansky, K.H., and J. Miles (1997), "Patient satisfaction with ambulatory healthcare services: Waiting
time
and filling time", Hospital and Health Services Administration 42:165-177.
Danzon, P.M. (1992), "Hidden overhead costs: Is Canada's system really less costly?", Health
Affairs
(Spring):21-43.
Davidge, M., M. Harley, L. Vickerstaff and J. Yates (1987), "The anatomy of large in-patient waiting
lists",
Lancet 1:794-796.
Department of Health (1989), "Working for patients", White Paper on the NHS CM555 (HMSO, London).
Department of Health (1995), The Patient's Charter and You (HMSO, London).
Department of Health and Social Security (1981), "Orthopaedic services: Waiting time for outpatient
appoint-
ments and in-patient treatment", Report of a Working Party to the Secretary of State for Social
Services
(HMSO, London).
de Serpa, A.C. (1971), "A theory of the economics of time", Economic Journal 81:828-846.
Devlin, H.B. (1980), "Programmed elective surgery in waiting for hospital treatment", in: Harrogate
Seminar
Reports (DHSS, London) 18-25.
Diamond, P.A., and J.A. Hausman (1994), "Contingent valuation: Is some number better than no number?",
Journal of Economic Perspectives 8:45-64.
Dixon, J., and B. New (1997), "Setting priorities New Zealand-style", British Medical Journal 314:86-87.
Donaldson, L.J., and T.F. Stoyle (1987), "Resource implications of orthopaedic waiting lists: Implementing
the Duthrie proposals", Public Health 101:17-22.
Duckett, S.J. (1995), "Hospital payment arrangements to encourage efficiency - the case of Victoria,
Aus-
tralia", Health Policy 34:113-134.
Duckett, S., and A. Street (1996), "Are waiting lists inevitable?", Health Policy 36:1-15.
Dunn, E., C. Black, J. Alonso, J.C. Norregaard and G.F. Anderson (1997), "Patients' acceptance
of waiting
for cataract surgery: What makes a wait too long?", Social Science and Medicine 44:1603-1610.
Edwards, R.T. (1994), "An economic perspective of the Salisbury Waiting Points Scheme", in: M.
Malek, ed.,
Setting Priorities in Health Care (John Wiley and Sons, Chichester) 56-69.
Edwards, R.T. (1997), NHS Waiting Lists; Towards the Elusive Solution (Office of Health Economics,
Lon-
don).
Edwards, R.T., and J. Barlow (1994), "Rationing health care by waiting lists: An extra welfarist perspective",
Discussion Paper 114 (Centre for Health Economics, University of York, York).
Feldman, B. (1994), "The cost of rationing medical care by insurance coverage and waiting", Health
Eco-
nomics 3.
Feldstein, M.S. (1967), Economic Analysis for Health Service Efficiency (North-Holland, Amsterdam).
Feldstein, M.S. (1970), "The rising price of physicians' services", Review of Economics and
Statistics
52:121-131.
Forsyth, G., and R. Logan (1968), Gateway or Dividing Line - A Study of Hospital Outpatients in the 1960s
(Oxford University Press, London).
Frankel, S.J. (1989), "The natural history of waiting lists - some wider explanations for an unnecessary
problem", Health Trends 21:56-58.
Frankel, S.J., and R.R. West (1993), "What is to be done?", in: S.J. Frankel and R.R. West, eds., Rationing
and Rationality in the National Health Service: The Persistence of Waiting Lists (Macmillan, London)
115-132.
Frankel, S.J., A. Farrow and R.R. West (1989), "Non admissions or non invitation? Care control study of
failed admission", British Medical Journal 299:598-600.
Frost, C.E.B. (1980), "How permanent are the NHS waiting lists", Social Science and Medicine C 14:1-11.
Frost, C.E.B., and B.J. Francis (1979), "Clinical decision making: A study of general surgery within Trent
RHA", Social Science and Medicine A 13:193-198.
Gerchak, Y., D. Gupta and M. Henig (1996), "Reservation planning for elective surgery under uncertain
demand for emergency surgery", Management Science 42:321-334.
Gertler, P., L. Locay and W. Sanderson (1987), "Are user fees regressive? The welfare implications of health
care financing proposals in Peru", Journal of Econometrics 36:67-88.
Globerman, S. (1991), "A policy analysis of hospital waiting lists", Journal of Policy Analysis and Manage-
ment 10:247-262.
Goddard, J., and M. Tavakoli (1994), "Rationing and Waiting List Management - Some Efficiency and Equity
Considerations", in: M. Malek, ed., Setting Priorities in Health Care (John Wiley and Sons, Chichester) 71-
92.
Goldacre, M.J., A. Lee and B. Don (1987), "Waiting list statistics I: Relation between admissions from waiting
list and length of waiting list", British Medical Journal 295:1105-1108.
Gribben, B. (1992), "Do access factors affect the utilisation of general practitioner services in South Auck-
land?", New Zealand Medical Journal 105:453-455.
Grossman, M. (1972), "On the concept of health capital and the demand for health", Journal of Political
Economy 80:223-255.
Gudex, C. (1994), "The QALY: How can it be used?", in: S. Baldwin et al., eds., Quality of Life: Perspectives
and Policies (Routledge, London).
Gudex, C., A. Williams, M. Jourdans, B. Mason, A. Maynard, R. O'Flynn and M. Rendall (1990), "Prioritising
waiting lists", Health Trends 22:103-108.
Hanemann, W.M. (1994), "Valuing the environment through contingent valuation", Journal of Economic Per-
spectives 8:19-43.
Hanning, M. (1996), "Maximum waiting-time guarantee - an attempt to reduce waiting lists in Sweden",
Health Policy 36:17-35.
Harrison, S., and G. Wistow (1992), 'The purchaser/provider split in English health care: Towards explicit
rationing?", Policy and Politics 20:123-130.
Hausman, J.A., ed. (1993), Contingent Valuation: A Critical Assessment (North-Holland, New York).
Hemingway, H., and B. Jacobson (1995), "Queues for cure? Let's add appropriateness to the equation", British
Hicks, D. (1972), "Waiting lists - a review", unpublished report prepared for the OR Branch of DHSS.
Higgings, J., and S. Ruddle (1991), "Waiting for a better alternative", Health Services Journal 101:18-19.
Higgins, J., and R. Wiles (1992), "Study of patients who chose private health-care for treatment", British
Journal of General Practice 42:326-329.
Hirschman, A.O. (1970), Exit, Voice and Loyalty (Harvard University Press, Cambridge, MA).
Ho, C.J., H.S. Lan and J. Li (1995), "Introducing variable-interval appointment scheduling rules in in-service
systems", International Journal of Operations and Production Management 15:59.
Ingram, R.M. (1980), "Day surgery in ophthalmology", in: Waiting for Hospital Treatment (DHSS, London)
Chapter 6.
Inter-Authority Comparisons and Consultancy (1990), "Cutting waiting time", in: Half Year Report (Health
Services Management Centre, Birmingham).
Iverson, T. (1986), "An interaction model of public and private health services: Surgical waiting lists", in:
A.J. Culyer and B. Jonsson, eds., Public and Private Health Services (Basil Blackwell, Oxford) 131-151.
Iverson, T. (1993), "A theory of hospital waiting lists", Journal of Health Economics 12:55-71.
Iverson, T. (1997), "The effect of a private sector on the waiting time in a national health service", Journal of
James, M., S.St. Leger and K.V. Rowsell (1996), "Prioritising elective care; a cost utility analysis of or-
thopaedics in the North West of England", Journal of Epidemiology and Community Health 50:182-189.
Johannesson, M., P. Johansson and T. Stderqvist (1998), "Time spent on waiting lists for medical care: An
insurance approach", Journal of Health Economics 17:727-744.
Jones, P.R., and J.G. Cullis (1996), "Decision making in quasi markets: A pedagogic analysis", Journal of
Jones-Lee, M.W. (1976), The Value of Life: An Economic Analysis (Martin Robertson, Oxford).
Jones-Lee, M.W., M. Hammerton and P.R. Phillips (1985), "The value of safety: Results of a national sample
survey", Economic Journal 95:49-72.
Kennedy, F., and B. O'Connell (1993), "General practitioner referral patterns", Journal of Public Health
Medicine 15:83-87.
Krupnick, A., and M. Cropper (1992), "The effect of information on health risk valuation", Journal of Risk
and Uncertainty 2:29-48.
Le Grand, J. (1982), The Strategy of Inequality (Allen and Unwin, London).
Lee, A., B. Don and M.J. Goldacre (1987), "Waiting list statistics: An estimate of inflation of waiting list
length", British Medical Journal 295:1197-1198.
Light, D. (1990), "Medical house arrest", The Health Service Journal 100:1648-1649.
Lindsay, C.M. (1980), National Health Issues: The British Experience (Roche Laboratories, Nutley).
Lindsay, C.M., and B. Feigenbaum (1984), "Rationing by waiting lists", American Economic Review 74:405-
417.
Martin, S., and P. Smith (1999), "Rationing by waiting lists: An empirical investigation", Journal of Public
Martin, S., and P. Smith (1996), "Explaining variations in inpatient length of stay in the National Health
Service", Journal of Health Economics 15:279-304.
McAvinchey, I.D., and A. Yannopoulos (1993), "Elasticity estimates from a dynamic model of interrelated
demands for private and public acute health care", Journal of Health Economics 12:171-186.
McPherson, K. (1981), "Clinical decision making - A response", Social Science and Medicine N3c, 15:193-
196.
Mills, R.P., and J.M. Heaton (1991), "Waiting list initiatives: Crisis management or targeting or resources?",
Journal of the Royal Society of Medicine 84:405-407.
Millward, R. (1971), Public Expenditure Economics (McGraw-Hill, London).
Moutlon, B.R. (1987), "Diagnostics for group effects in regression analysis", Journal of Business and Eco-
nomic Statistics 5:275-282.
Mullen, P. (1992), "Waiting lists and the NHS: Review reality and myths", HSMC Research Report 29.
Mullen, P. (1993), "The future of waiting lists", Journal of Management in Medicine 7:60-70.
M.V.A. Consultancy (1988), The Value of Time Savings (Policy Journals, Newbury, Berks).
National Oceanic and Atmospheric Administration (1993). Report of the NOAA Panel on Contingent Valua-
tion, Federal Register 58:4602-4614.
Naylor, C.D., R.S. Baifrie, B.S. Goldman and A. Basinski (1990), "Assessment of priority for coronary revas-
cularisation procedures", Lancet 335:1070-1073.
Newton, J., J. Henderson and M.J. Goldacre (1995), "Waiting list dynamics and the importance of earmarked
funding", British Medical Journal 311:783-785.
O'Donnell, O., and C. Propper (1991), "Equity and the distribution of NHS resources", Journal of Health
Economics 10:1-20.
Parker, B., and V. Srinivasan (1976), "A consumer preference approach to the planning of rural primary
health-care facilities", Operations Research 24:991-1025.
Parsonage, M., and H. Neuburger (1992), "Discounting and health benefits", Health Economics 1:71-75.
Pauly, M.V. (1990), Doctors and their Workshops (University of Chicago Press, Chicago and London).
Phelps, C., and J. Newhouse (1974), "Coinsurance, the price of time, and the demand for medical services",
Review of Economics and Statistics 52:334-342.
Pope, C. (1992), "Cutting queues or cutting comers: Waiting lists and the 1990 NHS reforms", British Medical
Journal 305:577-579.
Propper, C. (1990), "Contingent valuation of time spent on NHS waiting lists", Economic Journal 100:193-
199.
Propper, C. (1995), "The disutility of time spent on the United Kingdom's National Health Service waiting
lists", Journal of Human Resources 30:677-700.
Propper, C. (1998), "Who pays for and who gets health care: Equity in the finance and delivery of health care
in the UK" (Nuffield Hospital Trust, London).
Propper, C., and A. Eastwood (1991), "Health insurance purchase: An example of enterprise culture?", in: J.
Hutten et al., eds., From the Culture of Enterprise to the Culture of Dependency (Routledge, London).
Pourtney, P.R. (1994), "The contingent valuation debate: Why economists should care", Journal of Economic
Perspectives 8(4):3-17.
Regidor, E., S. Demateo, J.L. Gutierrezfisac, K.E Delahoz and C. Rodriguez (1996), "Socioeconomic differ-
ences in the use and accessibility of health-care services in Spain", Medicina Clinica 107:285-288.
Robinson, R. (1995), "Are the NHS reforms working?", Economic Review 13:9-19.
Roland, M., and R. Morris (1988), "Are referrals by general practitioners influenced by the availability of
consultants?", British Medical Journal 285:1368-1369.
Ryan, M. (1996), "Using willingness to pay to assess the benefits of assisted reproductive techniques", Health
Samuelson, P.A. (1964), Economics (McGraw-Hill, New York).
Seldon, A. (1967), "National or personal health services", Lancet 1.
Shmanske, S. (1996), "Contestabilty, queues and governmental entry deterrence", Public Choice 86:1-15.
Southam, J.A., and R.W. Talbot (1980), "Planned surgical admissions in a district hospital", British Medical
Journal 280:808-809.
Spicer, M.W. (1982), "The economics of bureaucracy and the British National Health Service", Milbank
Memorial Fund Quarterly 6:657-672.
Truong, P., and D.A. Hensher (1985), "Measurement of travel time values and opportunity cost from a
discrete-choice model", Economic Journal 95:438-451.
Wagner, H.M. (1969), Principles of Operations Research (Prentice-Hall, Englewood Cliffs, NJ).
Wesibrod, B.A. (1964), "Collective-consumption services of individual-consumption goods", Quarterly Jour-
nal of Economics 78:471-477.
Williams, A. (1988), "Applications in management", in: G.E Teeling Smith, ed., Measuring Health: A Prac-
tical Approach (Wiley, London).
Wind, Y., and L. Spitz (1976), "Analytical approaches to marketing decision in health care organisations",
Operations Research 24:973-990.
Worthington, D. (1987), "Queueing models for hospital waiting lists", Journal of the Operational Research
Society 38:413-422.
Worthington, D. (1991), "Hospital waiting list management models", Journal of Operational Research Society
42:833-843.
Wright, J. (1993), "Fundholding practices get preference", British Medical Journal 306:206.
Yates, J. (1987), Why Are We Waiting? (Oxford University Press, Oxford).
Yates, J. (1991), "Lies, damned lies and waiting lists", British Medical Journal 303:802.
Yates, J. (1995), Private Eye, Heart and Hip (Churchill Livingstone, London).
Yeates, W.K. (1980), "Waiting lists and urological cases", in: Waiting for Hospital Treatment (DHSS, London)
26-28.
Chapter 24
ECONOMICS OF DENTAL SERVICES*

HARRI SINTONEN and ISMO LINNOSMAA
Departmentof Health Policy and Management, University of Kuopio, 70211 Kuopio, Finland
Contents
Abstract 1252
Keywords 1252
2. Markets for dental services 1254
2.1. Is dental care different? 1254
2.2. Dental care inputs and outputs 1255
3. Demand for and utilisation of dental care 1257
3.1. Theoretical considerations 1257
3.1.1. Modelling the individual's demand for dental care 1257
3.1.2. Supplier inducement (SID) 1259
3.2. Empirical work on the demand for dental health and services 1260
3.2.1. Measurement of variables 1261
3.2.2. Empirical models 1265
3.2.3. Results 1268
4. Productivity, technical efficiency and economies of scale in dentistry 1274
4.1. Introduction 1274
4.2. Theoretical concepts 1275
4.3. Methods 1276
4.4. Measurement and results 1280
5. Economic evaluation in dentistry 1284
5.2. Measurement of costs 1285
5.3. Measurement of outcomes in CEA 1286
5.4. Measurement of outcomes in CBA 1287
5.5. Results 1288
References 1292
*Authors would like to thank Joseph Newhouse, Anthony Culyer, Willard Manning, Charles Phelps, and the
participants of Health Economics Handbook Conference at the University of Chicago, July 1998, for very
helpful comments. The remaining errors are our responsibility. Financial support from the Yrj6 Jahnsson
Foundation is gratefully acknowledged.
1252 H. Sintonen and L. Linnosmaa
Abstract
The purpose of this chapter is to review dental economics in three major areas: (i) de-
mand for and utilisation of dental care, (ii) productivity, technical efficiency and
economies of scale in dental care production, and finally (iii) economic evaluation of
dental care procedures and programmes. As a background to the review, we consider
what makes dental care different from other health care, describe briefly dental care
systems in a number of countries and present data on inputs and outcomes in terms
of dental health. Within each major area, we review developments in conceptual and
theoretical thinking, consider developments in the measurement of key variables and
methods, and present some major results. We also draw lessons to be learnt concerning
the state of the art in terms of theory, methodology and results, and outline directions
for future research.
Keywords
dental care, prevention, cleanings, fillings, examinations, extractions, demand,

utilisation, productivity, economies of scale, technical efficiency, inefficiency, input
factor productivity, total productivity, economic evaluation, efficiency
JEL classification:II, D1, D2, C2

Ch. 24: Economics of Dental Services 1253
1. Introduction
This chapter reviews three major areas of dental economics: (i) demand for and util-
isation of dental care, (ii) productivity, technical efficiency and economies of scale in
dentistry, and (iii) economic evaluation in dentistry. Within each section, we review the
theoretical and conceptual developments in the area, developments in measurement is-
sues and methodological solutions and present major results. We also draw lessons and
outline directions for future research.
We look first at dental care markets in Section 2. We consider whether it is different
from other health care and, if so, in what respects. We also describe briefly dental care
systems in a number of countries by presenting data on inputs and outcomes in terms of
dental health.
In Section 3 on demand and utilisation, we emphasise the shift from regarding dental
care as any other market good to considering it as an input in oral health production
as a consequence of Grossman's (1972) work. Since the Grossman model and its sub-
sequent development are discussed thoroughly elsewhere, we only briefly present the
basic theoretical role of price, income and time as determinants of dental care demand
and utilisation. In the subsection on measurement issues we focus on exploring how
these and some other key variables have been empirically measured. In the subsection
on empirical modelling we review developments in empirical modelling and the esti-
mation of demand and utilisation from single equation OLS models through tobit and
binary choice models to more sophisticated probability-based two-part models includ-
ing count models and times series models. This presentation is also relatively short since
these techniques are thoroughly discussed in other chapters of this Handbook. The sec-
tion concludes with a brief account of empirical results and lessons for the future.
Section 4 covers issues dealing with production side of dentistry such as productivity,
technical efficiency and economies of scale. These concepts are discussed from a the-
oretical point of view and empirical methods of analysing these entities are reviewed.
We consider production function models, cost function models and Data Envelopment
Analysis. The section concludes with a review of how the measurement problems asso-
ciated with the use of these models have been solved, the kind of empirical results that
have been obtained and the lessons that can be learned.
In Section 5 we explore how dental services have been evaluated by economists. We
very briefly introduce the main methods and principles of economic evaluation, since
these issues are dealt with in detail in other chapters, in order to focus on measurement
issues specific to dental care and to present major results obtained for dentistry. Again
some directions for future work are delineated.
1254 H. Sintonen and I. Linnosmnaa
2. Markets for dental services
2.1. Is dental care different?
In his seminal paper, Arrow (1963) analysed the differences between health care and
other goods and services. The characteristics of dental illness and dental care and dif-
ferences between dental care and general health care have received less attention in the
literature [Feldstein (1973), Yule and Parkin (1985)]. One explanation for this might be
that dental illnesses and organisation of dental care are thought to differ conceptually
very little from health care in general. This is not entirely correct, since there are some
features in dental care that make it different from general health care.
One feature of health care that distinguishes it from other goods and services is that
individual's demand for health care is unpredictable and intensifies when a person is ill.
Another feature is that individuals may not necessarily know much about the quality of
services [Arrow (1963)].
These features are not so strongly present in dental care. First, the number of dental
diseases is relatively few and their occurrence is more predictable than is the case with
many others. Second, individuals usually experience the same dental procedure several
times during their lifetimes and are therefore able to learn from experience about the
quality of service. Third, dental diseases are relatively easy to diagnose and almost all
relevant information for treatment decisions can be obtained from X-rays and photos.
Fourth, there is probably a wider variety of alternative treatments available to treat a
given disease than in most other cases. Fifth, there are extensive prevention possibilities
and, in dental care, prevention may actually save resources (see Section 5), which is
often not the case in other forms of medical care. Sixth, except for dental accidents and
toothache, dental care is seldom "emergency" care and untreated dental illnesses rarely
have dramatic consequences on an individual's health. Because of this the individual
can more freely plan and time treatment decisions.1 This also gives the individual more
freedom in the choice of service provider which, in theory, increases the price elasticity
of the individual's demand for dental care.
A well-known feature of market economies is that in the presence of externali-
ties markets fail to allocate goods and services efficiently. An example of externali-
ties in health sector is communicable disease [see Arrow (1963)]. Dental disease is
non-communicable which implies that risks of dental disease across individuals can be
treated as independent. In this regard, one would expect market mechanism to perform
better in dental care and insurance than in other health services.
However, the special features of dental disease and care have other implications for
dental insurance, which often operates quite differently from other types of medical in-
surance. The non-emergency nature of most dental care, the ease of access to relevant
information for treatment decision via X-rays and photos, and different treatment alter-
natives with varying costs make it possible for dentists working in an insurance office to
Authors thank Willard Manning and Charles Phelps for making these points.
validate or reject the recommendations of the treating dentist and thus to control costs.
Many insurance plans have stiff requirements for prior authorisation, commonly based
on X-rays and photos before treatment. The copayments selected for dental insurance
are often much larger than for other types of medical care, which may be explained by
higher demand elasticity and lower intrinsic risk (variance of financial outcomes). How-
ever, many dental insurance plans have a much lower copayment (even zero) for routine
prophylaxis than for other dental services, which would be irrational unless prevention
actually saves money. The relative freedom to plan and time treatments may result in
surges in dental utilisation. For an experimental example see Manning et al. (1985),
whose results are briefly introduced in Section 3.2.3.
However, prior authorisation may not work as desired. A good current example of
this is in Sweden, where public dental insurance was established in 1974 [Widstr6m and
Barenthin (1997)]. In the scheme some fraction or all of dental expenses are reimbursed
to the patient. Two things have been observed in Sweden. First, improving oral health
has been accompanied by dental patients consuming more expensive dental services in
spite of prior authorisation and, second, it has been argued that demand inducement by
underemployed dentists has increased. This behaviour tends to increase dental expendi-
tures and has forced the Swedish government to reduce insurance indemnities on several
occasions. Currently Sweden is reducing its public dental insurance coverage, leaving
room for the emergence of private dental insurance. This is currently happening in other
parts of Europe as well. The Netherlands and England are also moving towards a dental
insurance system with more emphasis on private insurance [Widstr6m et al. (1996)]. In
the US most of dental insurance is private and comes in the form of employment benefit
plans. About half of the US population has some form of dental care reimbursement
plan [see American Dental Association (1998)].
The Swedish case indicates the possibility of moral hazard and supplier inducement
in the presence of dental insurance so that demand and supply may not be independent
of one another, but it is difficult to say whether moral hazard is more (or less) of a
problem in dental insurance than in other health insurance. It has been shown that bet-
ter insurance coverage induces people to use more dental services and, moreover, that
insurance changes the structure of demand towards more costly services [see Manning
et al. (1986), Mueller and Monheit (1988)]. This probably also occurs in other forms
of health insurance and a priori it is difficult to assess whether the effect of insurance
on the demand for dental care is larger (or smaller) than the effect of insurance on the
demand for other forms of health care.
A final point is that a high proportion of dental care expenditures comes directly from
patients. This applies also to countries in which health care and insurance are mostly
public [see Parkin (1992)].
2.2. Dental care inputs and outputs
Table 1 presents data on dental care inputs and outcome in different countries. It con-
tains information on dental care expenditures, number of dentists, and dental health.
1256 Sintonen and I. Linnosmaa
16.
Table 1
Dental care inputs and outcome
Country PCE DEHE D OH
Australia 80e 4 .9e 0 .4e 1.0f

Austria 1 35b 10.7 b 0 .5g 3.0 d
e
6.8 e h
Canada 135 0.5 1.7 a
Denmark 119g 6, 6g 0,5 f 1.8 g
Finland 94g 6,8 g 0.9 g 1.2 e
France 111 5.6g 0.7f 2.1d
f
Germany 1 69 g 7 .4g 0.7g 2.3
h e
Iceland 159 7.9h 1.0 1.6 d
Ireland 44 c 4.4C 0.4g 1.5g
Japan 108f 6.9 f 0.6e 4.3 b
Luxemburg 52g 2.4g 0. 5d 2.3 f
Netherlands 72g 4.0 g 0. 5f 0.9 C
New Zealand 51d 4. 6 d 0.4 g 1.4f
b
Norway 54 3.6 b 0 .8 g 1 .8 g
Spain 68 b 7 .6 b 0.4g 2.3d
Sweden 147c 9.8c 1.0f 1.4f
Switzerland 197 f 8.2 f 0.5 f na
f d
UK 63c 5 .5C 0.4 1.4
US 179 g 4.6g 0.6 g na
Source: OECD (1998).

Abbreviations: PCE= Total expenditure on dental services per capita, US dollars, in purchasing
power parity, DEHE = Total expenditure on dental services as a percentage of total expenditure
on health services, D = Number of operating dentists per 1000 population, OH = Oral health
measured by average DMFT at the age of 12.2
Superscripts: a = 1990, b = 1991, c = 1992, d = 1993, e = 1994, f = 1995, g = 1996, h = 1997.
Monetary variables are presented in US dollars and dental health is measured using
DMFT-index which counts the number of decayed, missing and filled teeth. These data
from the OECD health database cover the years 1990-1997.3
Per capita expenditure is highest in Switzerland, United States and Germany and
lowest in Ireland and New Zealand among the countries in Table 1. The ranking based
on the proportion of dental care expenditure of total health care expenditure is slightly
different from that based on per capita expenditure. The ratio of dental care expenditures
to total health care expenditures is high in Austria and Sweden and low in Luxemburg
and Ireland.
2 DMFT (DMFS)-index measures the number of decayed, missing and filled permanent teeth (tooth sur-
faces) for an individual. Corresponding measure for children is dmft (dmfs)-index, which measures the num-
ber of decayed, missing and filled teeth (surfaces) in the primary dentition.
3 Because the data are from different years, they are comparable only if the measures do not change a lot
over the time period 1990-1997. This should be kept in mind when comparing figures for different countries.
Some of the variation in dental care expenditure may be explained by the way the
dental care markets are organised in different countries. Public dental care emphasises
equal access to dental care, typically through subsidised dental fees. Private dental care
markets, on the other hand, may suffer from market failures, like monopolistic service
production and incomplete information, tending to raise dental care fees above the com-
petitive level.
European systems may be split into two main classes: mixed systems, in which pub-
lic dentists are an alternative to private and systems based purely on private dentistry.
Mixed systems prevail mainly in the Nordic countries, whereas in other European coun-
tries services are provided by private dentists [see Widstr6m et al. (1996)]. In both cases,
however, public interventions in private markets are common. In the US the organisation
of service production and insurance are mostly private.
The third column of Table 1 shows that Sweden and Iceland have the highest number
of dentists per population (1 dentist per 1000 population). At the other end of the scale,
Australia, Ireland, New Zealand, Spain, and United States have only 0.4 dentists per
1000 population.
The last column presents data on dental health measured by the DMFT-index. The
Netherlands, Australia, and Finland have low index values indicating good dental health.
On the other hand, Japan and Austria score highest suggesting less good dental health.
Some might expect that high inputs in dental care would produce good oral health. If
this is true, one should observe a negative correlation between the measures of expendi-
ture and oral health. Ignoring two countries with no available data on dental health over
the period 1990-1997, the correlation between per capita expenditure and dental health
was 0.23 and that between the proportion of dental expenditure of total health expen-
diture and dental health 0.30. Neither differed significantly from zero. These estimates
do not support the hypothesis that high expenditure on dental care would be (linearly)
associated with good oral health. 4
3. Demand for and utilisation of dental care
3.1. Theoretical considerations
3.1.1. Modelling the individual'sdemandfor dental care
Early literature (see e.g. Andersen and Benham (1970) and Maurizi (1975)] on the de-
mand for dental services made no distinction between a typical commodity purchased
in the market and dental services [Yule and Parkin (1985)]. Dental services were treated
like any other market good.
4 Due to paucity of data and the fact that the data are not necessarily comparable one should interpret this
conclusion with caution. Moreover, due to lack of data, this simple analysis cannot take into consideration
variation in productivity and other factors that influence dental health, which may explain the obtained results.
1258 H. Sintonen and 1. Linnosmaa
Grossman's (1972) insight was that people basically demand health, and the demand
for health services is derived from the demand for health. Thus health services are not
demanded as direct sources of utility (in fact most health services produce direct disutil-
ity), but because of their potential for improving health. Grossman suggested that indi-
viduals produce their own health by using their own time and market goods, and that the
stock of health directly enters the utility function with all other goods the individual con-
sumes. Health has similar properties as capital stock in the traditional economic theory,
it tends to depreciate over time and the stock can be increased by making investments
in it. Grossman (2000) describes the approach in detail.
Analysis of the demand for dental health has been mostly based on static theoretical
models incorporating similar features to the Grossman model. The basic ideas were
presented by Holtmann and Olsen (1976) and Hay et al. (1982). 5 Their papers analyse
the behaviour of a utility maximising individual whose welfare is affected by dental
health and the other commodities the individual consumes. The individual can produce
dental health and other commodities by using own time and other market goods and
services. For example, individual's own time spent on self-care, dental care services
and toothbrushing enters into the individual's production function for dental health.
Individual's use of market goods and services is constrained by income, affected by
market prices (including the wage rate) and wealth.
Holtmann and Olsen (1976) showed that the demand for dental care is decreasing in
the price of care and in the time needed to produce a unit of dental health. 6 Pedersen and
Petersen (1980) reached similar conclusions. Moreover, they showed that, in general,
the effect of the wage rate on the demand for dental care is ambiguous and depends
on the relative magnitudes of income and substitution effects. Hay et al. (1982) also
predicted 7 that the time used for self-care is increasing in the price of dental services
and decreasing in the wage rate.
One shortcoming of the static models is that they omit the time-varying nature of
the dental health stock. It is quite natural to think that oral health depreciates over time
and that in each time period the individual can invest by purchasing dental care and
allocating time to self-care. This observation would justify the use of dynamic models
in the analysis of dental health determination. From the perspective of modelling, there
is a feature in oral health that makes it different from general health. In Grossman's
model death occurs when individual's stock of health falls below some minimum level.
Such a connection between death and minimum dental health does not exist. 8
Another weakness of static models is that they assume that health production pro-
cesses involving time to self-care, market goods, dental care, and other commodities
exhibit constant returns to scale. Although constant returns to scale are often assumed
to retain tractability, this assumption can be criticised in the production of dental health.
5 See also Pedersen and Petersen (1980).

6 Which is treated as a fixed parameter of the production function in the analysis.
7 Hay et al. (1982) do not present the derivation of theoretical predictions.
8 The authors thank Martin Forster for making this point.
Intuition strongly suggests that the production of dental health exhibits decreasing re-
turns to scale. For example, additional units of own time allocated to brushing add dental
health at a decreasing rate, because frequent brushing damages the enamel. 9 Similarly,
one expects that the use of dental care services has similar features because there are
finite number of teeth and tooth surfaces.l°0
3.1.2. Supplier inducement (SID)
The above discussion concentrates on analysing the individual's demand for dental ser-
vices under perfect information and certainty. In most cases, however, the patient is not
fully aware of the dental services needed, nor of the optimal amount of service.
If dentists act as imperfect agents, they may intentionally influence the patients' use
of services in two directions. If at any price level they shift the demand curve to the right
(left), there is supplier inducement (rationing). According to Ryan and Mooney (1992)
the standard definition of supplier induced demand is anything that the doctor orders that
the fully informed and knowledgeable patient would not order. Other definitions also
exist, see e.g. Labelle et al. (1994) and Ryan and Mooney (1992). Supplier influence
is different from the traditional economic analysis in which the market demand and
supply of a good are based on independent decisions made by consumers and firms
in that particular market. In this, dental care may be no different from health care in
general.
Due to its potential importance, considerable efforts have been exerted to measure
the presence of SID in health care generally [see Labelle et al. (1994)]. Direct measure-
ment is quite difficult, since in practice there are few patients with the same information
as their physicians.12 Several indirect hypotheses and empirical tests have been carried
out but, due to the lack of theoretical model and the presence of econometric and mea-
surement problems, results concerning the existence of SID remain still controversial
and inconclusive [Labelle et al. (1994)].
The existence of supplier inducement in dentistry has also been subject to empirical
investigation. The principal approach to testing for the existence of SID has been to look
for a positive correlation between dentist density and utilisation of dental care [see e.g.
Manning and Phelps (1979), Grytten et al. (1990), Mueller and Monheit (1988)].
9 The authors thank Charles Phelps for this point.

10 The intuition behind these examples is not sufficient to prove that there are decreasing returns to scale in
the production of dental health. Using the Cobb-Douglas production function with two inputs, it is relatively
easy to construct examples in which the production function exhibits decreasing returns to scale when only
one input is allowed to change and the other is kept fixed and, simultaneously, the production function exhibits
increasing returns to scale. Therefore, decreasing returns to scale with respect to one input when other inputs
are kept fixed is not sufficient for decreasing returns to scale in the whole production process.
l ! SID may have impact on health care expenditures, health status and the allocation of income between
patients and physicians. For a broader discussion, see Labelle et al. (1994).
12 However, at least two studies have measured the behavior among physicians and their families [Bunker
and Brown (1974) and Hay and Leahy (1982)]. Both studies found little or no support for SID.
1260 1f. Sintonen andI. Linnosmaa
The target income hypothesis holds that physicians respond to an increasing supply
of doctors by creating more demand to maintain income levels [Evans (1974)]. The
hypothesis predicts a positive correlation between supply of doctors and utilisation of
services. The hypothesis has been criticised, because its predictions are consistent with
the neoclassical theory of competitive markets, which also predicts a positive correla-
tion between the supply of physicians and utilisation of services. Increased supply of
physicians shifts the supply curve to the right, resulting in a decrease in service fees
along the demand curve. Increased supply of physicians also reduces the time price
of services, shifting the demand curve to the right. Therefore the neoclassical theory
predicts that the increased supply is associated with increased utilisation of services.
The above discussion focuses on unregulated markets. Birch (1988) explored the ex-
istence of SID in a fee-regulated dental market, where the dentists work on a fee-for-
service basis. He argued that an increase in the number of dentists leads to a greater
total supply of services, since more dentists want to satisfy their income desires. It also
leads to a reduction in time price and thus in the total shadow price of dental visits,
and consequently to a rise in demand for dental care. This movement along the demand
schedule is also predicted by neoclassical theory. If the increase in demand equals that
in supply, a new equilibrium is established at the original price. If the increase in de-
mand falls short of that in supply, dentists have an incentive to induce demand up to
the income-leisure optimum, since fees are fixed. In the neoclassical theory fees would
decrease until the optimum is reached, and no supplier inducement takes place. Birch's
model predicts an increase in dental visits per capita and an increase in the average con-
tent per visit provided that the net benefits of additional utilisation induced at the given
number of visits exceeds the net benefits of the additional visits induced. On this ground
Birch (1988) concluded that a positive correlation between the number of dentists per
capita and treatment content per visit provides sufficient (but not necessary) evidence
for the existence of SID in a fee-regulated market environment.
Sintonen and Maljanen (1995a, 1995b) made a distinction between individual and
general inducement. The former are actions affecting patients. The latter is a continu-
ous, strong and systematic effort by dental regulators, dental associations and individual
providers to make people adopt a regular pattern of visiting a dentist (once or twice a
year). How these forms of inducement were operationalised will be discussed later.
3.2. Empirical work on the demandfor dental health and services
There are rich data sets ranging from routine data to survey data and even to experi-
mental survey data. For this reason data and measurement issues are given a section of
their own. After that we sketch the development that has taken place in the empirical
modelling of the demand for and utilisation of dental care. Finally, some results are
reviewed.
3.2.1. Measurement of variables
Measures of demand and utilisation. Conceptually the demand for dental care can be
differentiated from utilisation, the latter being the use of dental care as determined by
the interaction of demand and supply. Demand for dental care is the amount of den-
tal care that patients would use independently of any dentists' influence. In practice,
what is observed is utilisation and it is usually difficult to disentangle demand from it.
A problem is that in the literature the concepts of demand and utilisation are not used
clearly and consistently. Sometimes demand and utilisation are used interchangeably,
sometimes with different meanings.
Two main variables have been used to measure the demand for and utilisation of
dental care: the number of dental visits and total expenditure on dental care. The number
of visits was used in the early literature [Holtmann and Olsen (1976), Manning and
Phelps (1979), Pedersen and Petersen (1980), Hay et al. (1982)], but the variable has
seen a new revival in the latest literature [Sintonen and Maljanen (1995b), Rosenqvist
et al. (1995) and Arinen et al. (1996)]. The number of visits is not a flawless measure of
utilisation of dental care because it does not take into account the quality and quantity
of services actually purchased [see e.g. Yule and Parkin (1985), Sintonen and Maljanen
(1995a) and Hu (1981)] and does not differentiate between visits initiated by patients
and dentists [Yule and Parkin (1985)].
Total expenditure on dental services as a measure of demand for and utilisation of
dental care has been used by Andersen and Benham (1970), Upton and Silverman
(1972), Manning et al. (1986), Conrad et al. (1987), Mueller and Monheit (1988), Gryt-
ten et al. (1990) and Sintonen and Maljanen (1995a). Total expenditure captures the
amount and quality 13 of services used better than the number of visits, but is not a good
measure of demand and utilisation if individuals are charged different prices for the
same service [Yule and Parkin (1985)]. Ideally, however, one would like to measure
the use of specific services, like cleanings or fillings, provided by dentists. A practical
problem encountered here is that service-specific data rarely exist. 14
In a time series analysis based on aggregate data Parkin and Yule (1988) used several
measures for demand and utilisation. As a demand measure, they used the number of
patient-initiated contacts. The authors admit, however, that in dentistry providers may
have more scope than in other health care to influence initial contacts. As a utilisation
measure, Parkin and Yule (1988) used the sum of dentists' annual gross fees computed
using a fixed fee schedule. Besides these two variables, Parkin and Yule measured the
utilisation of specific treatments.
Measures of money price. Ways of measuring money price empirically vary consid-
erably. Manning and Phelps (1979) studied seven specific dental treatments: fillings,
13 If higher quality is more costly.

14 An exception here is Manning and Phelps (1979).
1262 H. Sintonen and I. Linnosmaa
cleanings, extractions, examinations, dentures, crowns and orthodontia. The authors had
average retail prices for cleanings, fillings and extractions, but no explicit price data for
other services. They approximated the missing prices by constructing a price index as
a weighted average of the available prices. As observed by the authors, the use of such
a price variable may introduce a measurement error into the analysis and lead to biased
estimates of the unmeasured price variables.
Studies explaining the variation in total expenditure have used a price index for the
total amount of services purchased. Mueller and Monheit (1988) used the mean price
of a routine maintenance visit in the individual's sampling unit. Hu (1981) constructed
a weighted price index from service-specific prices. Sintonen and Maljanen (1995a)
constructed an out-of-pocket price for an individual by taking into account the price
difference between the private and public sector, the reimbursement rate from insurance,
and tax deductibility of the remaining expenses through marginal tax rates.
Studies having the number of visits as the dependent variable have used different
measures of the price of dental services. Holtmann and Olsen (1976) and Pedersen and
Petersen (1980) created a price variable by dividing the total expenses by the number of
visits. As pointed out by Manning and Phelps (1979), Newhouse et al. (1980) and Yule
and Parkin (1985), this construction may not reflect the price variation, only changes in
the quality of services. Another problem with this price variable is that if the number of
visits is measured with error the parameter estimate for price variable obtained from the
regression analysis is biased towards zero.15
Arinen et al. (1996) and Rosenqvist et al. (1995) created a price variable by ask-
ing individuals for their estimate of the public sector price for a treatment costing 100
Finnish marks (FIM). The price variable was given a value of the individual's estimate,
if the individual used (or would have used) public sector services. If the individual used
(or would have used) private sector services, then the value of the price variable was
set equal to 100 FIM. According to Arinen et al. (1996), this price construction reflects
the perceived differences in price level and quality between the public and private sec-
tor. Indeed, in circumstances where the consumers do not know the actual price, the
individual's demand is affected by perceived money price.
Measures of time price. The time price associated with the use of dental care has
been measured in several ways. Sintonen and Maljanen (1995a), Arinen et al. (1996),
and Rosenqvist et al. (1995) measured the time needed for a dental visit directly with
travel, waiting and treatment time included. Hay et al. (1982) used distance and Grytten
et al. (1990) travel time as a measure of time price. Thus, only a part of total time
was measured. In none of the above studies was time valued explicitly. This is done
in Mueller and Monheit (1988), Holtman and Olsen (1976), and Conrad et al. (1987).
Mueller and Monheit (1988) multiplied travel time by hourly wages. Holtmann and
15 In multiple regression the other coefficients are also biased, although in unknown directions [see Greene
(1993)1.
Olsen (1976) and Conrad et al. (1987) had both waiting time and travel time in their
analysis and the latter was multiplied by the estimated value of the person's own time.
Theoretically it is important for the models used to explain demand or utilisation to
include the money price and the full time cost covering travel time, waiting time at the
office and the actual treatment time. If price and full time cost are missing from the
covariates, it is not possible to distinguish the effect of more dentists lowering price and
full time cost from the effects of SID. This also implies that in the absence of those
variables it may be particularly difficult to measure SID by using dentist density. 16
It is questionable whether total time should be valued explicitly by the wage rate
or whether it should be left for implicit valuation. Cauley (1987) points out and also
shows empirically that a number of factors may break the equality between wages and
marginal value of time. These include not working for market wages (e.g. housewives
or children), paid sick leave, and direct utility or disutility of time spent consuming
medical care. In addition, an illness or injury may reduce the opportunity cost of time.
Supplier inducement. As mentioned earlier the main approach to testing for the exis-
tence of SID has been to look for a positive correlation between utilisation of dental care
and dentist density measured in slightly different ways [see e.g. Manning and Phelps
(1979), Mueller and Monheit (1988), Grytten et al. (1990)].
Sintonen and Maljanen (1995a, 1995b) attempted to measure supplier inducement
explicitly in terms of individual and general inducement. They measured individual
inducement by a dummy with a value of 1, if the patient was recalled by the dentist and 0
otherwise. To approximate the effectively unobservable variable of general inducement,
they [and Arinen et al. (1996)] constructed a variable (1 - )RDA, where RDA is Regular
Dentist Attendance, taking the value 1 if the respondent visits a dentist regularly (at least
once in two years) and 0 otherwise. it is a prediction from a logit model in which regular
dentist attendance is regressed on variables affecting the regularity of visits. Therefore,
for an individual who regularly visits a dentist, general inducement has a value 1 - a
and for an individual with no pattern of regular dental visits, general inducement is
zero. The authors believe that the most important omitted variable in the model that
explains RDA is general inducement, and therefore the residual 1 - approximates
general inducement well enough.
Insurance and experimental data. Manning et al. (1985), Manning et al. (1987), and
Mueller and Monheit (1988) studied the effect of insurance coverage on the demand for
medical and dental care. From theory one expects the insurance terms have an effect on
the amount and the mix of services used. In the above studies, the insurance terms were
operationalised by dummy variables. Table 2 contains the variables used by Mueller and
Monheit (1988).
Mueller and Monheit (1988) used nonexperimental survey data, which may suffer
from the fact that observed insurance may be the result of the optimising behaviour of
16 The authors thank Willard Manning for making this point.

Table 2
Insurance variables in Mueller and Monheit (1988)
Variable Description
LOW = 1 if no deductible and coinsurance rate <0.10

HIGH = 1 if deductible and coinsurance rate >0.10
DED = I if a plan requires a payment of a deductible
Table 3
Insurance variables in Manning et al. (1985)
Variable Description
POO = 1 if free plan

P25 = 1 if dental coinsurance rate = 25%
P50 = I if dental coinsurance rate = 50%
P95 = 1 if dental coinsurance rate = 95%
IDP = 1 if dental deductible plan
a consumer. Therefore the insurance variable is endogenous and, as is known from the
estimation theory of simultaneous equation models [see e.g. Greene (1993)], this leads
to biased and inconsistent parameter estimates.
One of the main objectives of the Health Insurance Experiment (HIE) by The Rand
Corporation was to correct this possible flaw and collect purely exogenous insurance
data. This was done by setting up a social experiment in which families were enrolled in
six sites between the years 1974-1982 in the USA. Families were assigned to an insur-
ance plan with varying degrees of cost sharing. Each plan defined a coinsurance rate 17
and an upper limit on out-of pocket expenses. 18 A detailed description and the results of
the study are given in Manning et al. (1987), Manning et al. (1988) and Newhouse et al.
(1993). Results concerning dental care are presented in Manning et al. (1985), Manning
et al. (1986), Newhouse et al. (1993), and Zweifel and Manning (2000).
Manning et al. (1985) had 5 dummy variables describing the terms of dental insur-
ance. These variables are listed in Table 3.
When comparing the effects of price and insurance variables, one has to bear in mind
variations in dental prices and insurance arrangements. US scholars usually try first to
measure this variation with a price variable (usually with relatively poor success as seen
above) and then they typically introduce separate dummies to account for various insur-
ance schemes and coinsurance rates. In Finland, as in many other European countries,
the original price for a service is fixed (no variation in the original price paid) and this
17 One of 0%, 25%, 50%, 95%.

18 5%, 10% or 15% of family income, up to a maximum $1000.
fixed price is reimbursed by insurance to varying degrees. Thus the price variable of
Sintonen and Maljanen (1995a) reflects the final out-of-pocket price for the patient (the
varying coinsurance rates have been incorporated in the price variable), whereas Gryt-
ten et al. (1990) had no price variable, since the prices of dental services are fixed and
dental insurance covers only a small proportion of population in Norway.
Health and socioeconomic variables. Individual's health status has been frequently
measured by dummy variables. Manning and Phelps (1979) had 5 dummy variables
describing the dental health status of the individual. "Good health", "fair health", and
"poor health" were dummy variables obtaining a value of 1 if a person's health was
good, fair or poor, respectively. Toothache and bleeding gums in Manning and Phelps
(1979) were similar dummy variables. Mueller and Monheit (1988) and Grytten et al.
(1990) used one health status variable, Sintonen and Maljanen (1995a) had two health
variables. Dummy variables have also been frequently used for describing the indi-
vidual's level of education or socioeconomic situation [Manning and Phelps (1979),
Grytten et al. (1990) and Sintonen and Maljanen (1995a)].
3.2.2. Empiricalmodels
Modelling techniques in the literature on the demand for dental care have evolved from
traditional single equation regression models [Holtmann and Olsen (1976)] to more
sophisticated probability models [Manning and Phelps (1979), Manning et al. (1981),
Duan et al. (1983), and Arinen et al. (1996)]. This section outlines this development.
Most of the existing literature concentrates on cross-section techniques, but we also
briefly review time-series and simultaneous equation studies in a separate subsection
[see also Jones (2000)1.
Cross-section studies. We start with the following empirical specification of theoreti-

cal demand function
I~~ m
fo(ci) = o + jf(zji)+ E Pjfj(wji) + i, (1)
j=1 j=l+1
where c measures the demand for dental services, z = (zl, Z2, , l) is a vector of
economic variables, w = (wl+l . ..., w) is a vector of noneconomic variables, and is
a statistical disturbance term.
Using ordinary least squares (OLS) methods, Holtmann and Olsen (1976) estimated
four specifications of the above empirical model: linear, quadratic, log-linear, and semi-
log forms. A high proportion of households in the sample had no dental visits. This
causes a problem because the logarithm of zero is not defined. The authors evaded the
problem by replacing zeros by small numerical values like 0.01, 0.02, 0.1, and con-
ducted a sensitivity analysis showing that the results were insensitive to the values cho-
sen. A more serious econometric problem is that when zero observations are present in
the dependent variable, the linearity assumption is violated rendering the estimation of
the linear model invalid. These problems are not unique to dental or health economics.
More sophisticated econometric models, like discrete choice and tobit models, have
been developed to tackle them.
The basic structure of the discrete choice model is as follows. Consider a sample
of n individuals for which either c = 1 or c = 0. Here c = I and c = 0 mean that an
individual uses and does not use dental care, respectively. These observations may arise
from an underlying model
Yi = 0lXi + i, (2)
where y is a latent variable, x is a vector of both economic and noneconomic vari-

ables affecting the latent variable y and 0 is a parameter vector. A convenient economic
interpretation of the latent variable is the difference between the marginal benefit and
marginal cost of using dental care services for an individual. Therefore we observe that
the individual uses dental care services, or c = 1, if the marginal benefit of using dental
care services exceeds marginal cost (y > 0), and that the individual does not use dental
care services, or c = 0, if the marginal benefit of using dental care services is less than
marginal cost (y < 0). Manning and Phelps (1979) applied the discrete choice model in
their analysis on the demand for dental care.
When the dependent variable is continuous but censored, the OLS estimation of the
linear model (3) is again inappropriate. The tobit model was developed to handle this
problem. To display the basic set-up of the tobit model, suppose that the sample contains
only nonnegative observations on the demand for dental care c. Consider a model
Yi = 'xi + i, (3)
where y is a latent variable, x is a vector of independent economic and noneconomic

variables, B is a parameter vector, and e is a disturbance term. What is observed is that
c = 0, if y < 0 and c = y, if y > 0. Manning and Phelps (1979) and Hu (1981) used the
tobit model in the estimation of demand for dental care measured by dental visits.
It can be argued that the demand for medical and dental services is a two-stage pro-
cess, in which the individual first decides whether to seek care (decision 1 = D 1) and,
conditional on the decision to have care, the individual decides on the amount (decision
2 = D2 ). It is possible that the same variables affect decisions 1 and 2 differently (for
example, price may have a different effect on seeking care than on the amount of care)
or that a different set of variables affects the two decisions (for example, remuneration
system of dentists may not have an effect on the individual's decision to seek care, but
may influence decision 2).
A shortcoming of all single equation models is that none captures the two-stage nature
of the demand for dental care. If we denote the relationship of these decisions to a
vector of explanatory variables by D 1 = b'xl and D2 = d'x 2, the single equation models
implicitly assume that xl = x2 and b' = d'. This hardly holds in practice, since clients
and dentists have a different degree of influence on these decisions as was argued above.
Moreover, the tobit model is not robust against violations of the assumption that is
normally distributed.
Manning et al. (1981) and Duan et al. (1983) developed a two-part model that in-
corporates the sequential nature of the individual's demand for medical or dental care.
Because of this feature and the fact that the assumptions xl = x2 and b' = d' can be re-
laxed, the model has been applied frequently in the subsequent health economics stud-
ies, including dental economic studies. The empirical model consists of two equations.
The first equation is a discrete choice equation (decision 1)
Ii = O'xil + ui, (4)
in which I = 0 (individual does not consume any dental care) or I = 1 (individual con-
sumes a positive amount of dental care), xl is a vector of independent variables affecting
the decision to use dental care, 0 is a parameter vector, and u is a disturbance term. The
second equation of the model is linear for positive observations only (decision 2)
Wi = I'xi2 + Vi, (5)
where wi = log(ci Ii > 0), P is a parameter vector, x2 is a vector of variables affecting

the utilisation of dental care, and vi is a disturbance term. Mueller and Monheit (1988),
Grytten et al. (1990) and Sintonen and Maljanen (1995a, 1995b) applied a two-part
model in dentistry. A more detailed discussion on the model can be found in Jones
(2000) in this Handbook.
Recent research has focused on models in which the dependent variable is the num-
ber of dental visits. Sintonen and Maljanen (1995b), Rosenqvist et al. (1995) and Ari-
nen et al. (1996) are examples of the most recent research. These papers used modified
count data models to construct a probability model for the discrete dependent variable.
Modified count data models differ from count data models in their treatment of zero ob-
servations. In the case of dental visits zero observations come from a distribution that is
different from positive observations, and therefore the use of discrete distributions treat-
ing all observations similarly may not be suitable. Manning et al. (1986) were the first
to apply count data models to dental visits. Again, Jones (2000) provides a presentation
of count models and their applications.
Time series and simultaneous equation models. Compared to the number of cross-
section studies, there are very few studies on the demand for dental care based on time
series data. The reason for this is that data on individuals' dental health status or so-
cioeconomic variables present in cross-section studies are not readily available in time
series. One exception is Parkin and Yule (1988). They regressed the measures of demand
for dental care on price and income variables and variables describing the availability
of dentists plus dummy variables capturing the changes occurred over time in the NHS
dental care charging system in Scotland. The authors used the linear model (1). Some of
1268 iH.Sintonen and I. Linnosmaa
the results suffered from autocorrelation which was solved by imposing AR(1) structure
on the disturbance:
et = pet-! + vt. (6)
Whenever the results from the linear model did not suffer from autocorrelation, the
OLS estimates of the linear model were used.
A common problem with cross-section studies is simultaneous-equations bias. For
example, demand for dental care is affected by dental health status, but at the same time
dental health status is influenced by the use of dental services. This makes both dental
health status and dental services jointly determined. To illustrate the situation, consider
a following model:
c= az + Bh + El, (7)
h = Tc + E2, (8)
where c is the demand for dental services, h is dental health status, z is a vector of
exogenous variables, and Eifor i = 1, 2 are disturbance terms of the model. This simple
model captures the basic idea that dental health and demand for dental services are
jointly determined by the exogenous variables of the system. The reduced form of the
model is given by
c= 1-fz
i + l 2 + 81' (9)
1 - PT 1PT 1 - P1
ra + e
1
h= -z -- l + 1-- r2. (10)
1- 1 -f 1-
Assuming that E(ei) = 0 for i = 1, 2, E(EIE 2) = 0, and that E(e 2 ) = Oa2, then it is
easily shown that
T 2
cov(h, e) = --
and therefore OLS estimation of Equation (7) provides an inconsistent estimate for B.
Two examples of studies that estimate the full system like (7) and (8) are Pedersen and
Petersen (1980) and Hay et al. (1982).
3.2.3. Results
Pricesand income and demandfordental care. Holtmann and Olsen (1976) estimated
several linearised models using households as units of observation. The estimated price
Ch. 24: Economics of DentalServices 1269
elasticities fell into the range -0.032 to -0.19, depending on the model. Income elas-
ticities ranged from 0.12 to 0.41. When interpreting these results one has to recall that
the linear estimates may be biased and inconsistent by the presence of zero observations
in the dependent variable.
Manning and Phelps (1979) estimated the demand for 7 specific services and com-
puted price and income elasticities for each service separately for white adult females,
males and children. The demand for cleanings 19 was found to be more price elastic
for adult females and children than for adult males. The estimated price parameter was
significantly different from zero for adult females and children but not for adult males.
Income elasticities for the demand for cleanings had the expected signs and the parame-
ters associated with income variables were statistically significant for all age and gender
groups. In regard to fillings and examinations, Manning and Phelps obtained similar re-
sults to those for cleanings. 20 The other services had price and income elasticities with
unexpected signs and magnitudes and several parameters turned out to be statistically
insignificant. This may be due to the fact that Manning and Phelps had explicit price
variables for cleanings, fillings and extractions only whereas the prices for the rest of
the services were computed as weighted averages of three available prices.
Manning and Phelps (1979) also regressed the number of visits on several economic
and noneconomic variables and computed elasticities for price and income variables.
For adult males, estimated income and price elasticities were 0.61 and -0.65, respec-
tively. Corresponding elasticities for adult females were 0.55 and -0.78 suggesting that
females' demand for dental care is more price elastic and less income elastic than that
of males. Income and price elasticities for children were 0.87 and -1.40. They com-
puted income and price elasticities in different income groups and found that the price
elasticity became larger as income rose. The authors also found evidence for the the-
oretical result that price elasticity approaches zero when the price of the dental care
approaches zero. When considering these results it is to be noted that tobit model may
not be appropriate when estimating a model based on count data like visits.
Keeler and Rolph (1988) estimated "pure price elasticities" for medical spending
from the HIE data. When the coinsurance rate was in the range of 0-25%, the price
elasticity for dental spending was -0.12 and when the range was 25-95%, the elasticity
was -0.39.
Mueller and Monheit (1988) found that the price of dental care had a negative and
significant impact on the likelihood of using dental care (elasticity at mean -0.18), but
that the price was positively related to the number of visits and expenditure (elasticity
at mean 0.67). The authors' explanation for this unexpected result was that the money
prices become less important once the decision on the use of services has been made.
Another explanation may be the nature of their price variable, which does not reflect the
net price facing the consumer.
19 Manning and Phelps (1979) used the logit model and therefore a more precise term would be the elasticity
of the probability that the individual uses the cleaning service with respect to price.
20 The price elasticity of fillings was statistically significant for adult males as well.
1270 H. Sintonen and1. Linnosmaa
Sintonen and Maljanen (1995a) found that out-of-pocket price had a small, negative
and statistically significant effect on the probability of use and also on gross expenditure.
Price elasticity was found to be -0.069. They found a positive, although statistically
insignificant, effect of income on the likelihood of use and on the expenditure. When
visits were used as the utilisation measure and explained by two-part logit-negative bi-
nomial regression (negbin) model, the elasticities were even smaller and insignificant
[Sintonen and Maljanen (1995b)]. Conrad et al. (1987) obtained in two-part model elas-
ticities comparable to Sintonen and Maljanen (1995a) for expenditure on "basic" dental
services (by primary insurance subscribers).
In their time series analysis Parkin and Yule (1988) found, as might be expected, that
the different measures of dental care demand were negatively related to the correspond-
ing price variables. An exception was the number of non-denture treatments, which was
positively related to price, although the estimated positive price elasticity was small.
Depending on the model, the estimated price elasticities ranged from -0.024 to -0.75.
They also found that price variables had minor negative effects on the initial contacts,
but a greater influence on the volume of treatment. Surprisingly, the authors found a neg-
ative income elasticity suggesting that dental care is an inferior good. As they pointed
out, it may be erroneous to interpret the elasticities obtained from time series data in a
conventional way, because tastes and technologies are likely to change over time. Some
unmeasurable part of the time trend in demand variables may be due to the increasing
trend in income, which possibly causes the income effects to be underestimated.
Arinen et al. (1996) analysed the effect of the 1986 subsidisation reform for the young
adults on the demand for dental care in Finland. The reform essentially decreased the
consumer price of dental care in public and private sectors and, in addition, increased
dentist availability in the public sector. The authors first tested the hypothesis that zero
and positive observations came from the same distribution. The results did not sup-
port the hypothesis. Then the authors estimated models in which the binary decision
to seek care was modelled using a logit model and the decision on the number of vis-
its was modelled using either 0-truncated Poisson or 0-truncated negbin model. The
logit-negbin two-part model clearly fitted the data better. The results indicated that the
subsidisation reform changed mainly the number of visits and had no significant effect
on binary decision to seek care. The reform also increased the probability of choosing a
public sector dentist by three percentage points from 82% to 85% [Arinen and Sintonen
(1994)].
To summarise the results the studies mentioned above, all income and price elas-
ticities exhibited the expected signs and therefore the results are consistent with the
predictions of theory. An exception is the result by Mueller and Monheit (1988), sug-
gesting that price may also have a positive effect on the demand for care once the choice
to seek care has been made. The result may be attributable to the nature of their price
variables.
Time cost and demand for dental care. Holtmann and Olsen (1976) found that in-
creased waiting time per visit reduced the demand for dental care. The time cost elastic-
Ch. 24: Economics ofDental Services 1271
ities of demand were higher than the price elasticity regardless of the model used. They
also discovered that travel time is not as significant as waiting time as an explanatory
variable. Hay et al. (1982) found the distance to a dentist to be positively related to
the number of visits. The associated parameter estimate was not statistically significant,
however. Mueller and Monheit (1988) estimated a two-part model and found that the
time cost had a significant negative effect on the dental visits and a marginally signif-
icant negative effect on dental expenditure. Similarly, Sintonen and Maljanen (1995a)
found that the time required by a dental visit had a significant negative effect on the
probability of visiting a dentist, but a non-significant negative effect on dental expen-
diture for those who had visited or (in a 0-truncated negbin regression) on the number
of dental visits for those who had visited [Sintonen and Maljanen (1995b)]. Grytten
et al. (1990) found no significant effect of travel time on the likelihood of using dental
services.
A problem with time cost variables, in particular travelling time, is that they may
correlate with individuals' income or the dentists per population variable. A correlation
between travelling time and income may arise if travelling time is related to suburban
living and therefore to individual's income [Holtmann and Olsen (1976)]. Both income
and supply variables are present in the utilisation models as explanatory variables. Cor-
relation between explanatory variables of a regression model increases the variance of
parameter estimates, increasing the probability that the parameter estimates are insignif-
icant.
Dental insurance and demand for dental care. As a part of the HIE Manning et al.
(1985) examined the effect of insurance terms on the utilisation of dental care ser-
vices. On average, individuals with 0% coinsurance (free care) had 34% more visits
and 46% higher dental expenses than individuals assigned to a 95% coinsurance plan
with an out-of-pocket cap. They also found similar effects of insurance on the use of
specific services, such as diagnostic/preventive, restorative, prosthodontic and endodon-
tic/periodontic services. Moreover, increases in income led to higher utilisation. On the
other hand, the results also suggested that insurance had a larger effect on the use of
dental services in the low income groups than in the high income groups.
These results describe responses to complex insurance plans with different coinsur-
ance rates, deductibles, internal limits, stop-losses or limits to covered benefits. Another
interesting concept is the pure price response, where the consumer reacts to a constant
marginal cost of dental care, i.e. to a constant coinsurance rate or copayment rate. Keeler
and Rolph (1988) estimated such a quantity [see also Newhouse et al. (1993)]. For ex-
ample, with the 95% coinsurance rate the estimated spending on dental services turned
out to be 50% of that in the free plan. At the 50% coinsurance rate the spending was
68% of that in the free plan. They emphasise that the plan results underestimate pure
price responses.
Results of the HIE also indicated that people responded differently to insurance plans
at the beginning and at the end of the experiment than during its middle years. Manning
et al. (1986) showed that in the free, 25-, 50-percent and individual deductible plans the
intensity of use of dental care was significantly higher in the first year than in the second
year of the experiment. Similarly, although not so strong, a surge in the use of dental care
was observed at the end of the experiment. These results are best understood in terms
of dental health. Since individuals can freely plan and time their dental treatments, they
accumulate dental health by using dental services whenever the economic terms are
most advantageous.
Mueller and Monheit (1988) found that dental insurance had a positive and significant
effect on the demand for dental care. Moreover, they showed that insurance increased
the likelihood of obtaining dental care from 0.47 (no insurance) to 0.55 and 0.57 de-
pending on the terms of contract. Insurance also increased the per capita expenditure
from $96 (no insurance) to $157 (most generous insurance terms), suggesting a possi-
ble increase of 64% in per capita expenditure. In a separate analysis, they also showed
that insurance increased the individual's use of more expensive services, like crowns
and bridges, and therefore altered the mix of services purchased. One has to recall,
however, that the parameter estimates may be inconsistent if the insurance variable is
endogenous.
Supplier inducement and demand for dental care. Manning and Phelps (1979) ob-
tained results that are consistent with the conventional SID hypothesis, i.e. that there
will be a positive correlation between dentist density and utilisation. An alternative ex-
planation for the finding offered by the authors was that a higher number of dentists in
a geographical area reduces the access time for patients, thereby increasing the demand
for services. Due to lack of data on time costs, they were not able to test the two alter-
native hypotheses but, through an indirect analysis, they came to a conclusion that at
least part of the effect of increased dentist per capita is attributable to reductions in time
costs.
Conrad et al. (1987) assumed that the dentist/population ratio measures the incentive
for creating demand, but the coefficients did not turn out to be significant. Mueller and
Monheit (1988) reported highly significant elasticities of 0.23 for probability of use and
0.39 for conditional expenditure with respect to the ratio. They attributed the former
effect to diminished time cost and the latter to a concentration of specialists and other
high quality and high price providers in areas with a high dentist/population ratio.
Birch (1988) studied the existence of SID in a fixed price context in the United King-
dom by testing his hypothesis that a positive correlation between suppliers per capita
and content per visit constitutes a sufficient condition for the existence of supplier in-
ducement. He used data from the 1981 U.K. population census and National Health
Service statistics with primary health care districts as observation units. He measured
the content per visit by the average cost per treatment course in the districts and ex-
plained variation in it by variables describing population dental health, demographic
mix, income level, access cost, supply characteristics, and finally the population/dentist
ratio. The results showed that the coefficient of the population/dentist ratio was nega-
tive and highly significant - the elasticity of the ratio with respect to the average cost
per treatment course was -0.25. This was interpreted to provide strong support for the
inducement hypothesis.
Grytten et al. (1990) attributed the significant elasticities of 0.62 for demand (prob-
ability of use) and 0.28 for conditional expenditure with respect to the ratio (actually
they used population/dentist ratio so their original elasticities were negative) to supplier-
inducement. However, there are some problems with these results [Sintonen and Malja-
nen (1995a)]. First, although Birch (1988) argued that in a fixed-price setting a positive
correlation between dentist/population ratio and content per visit or per treatment course
is sufficient evidence of supplier-inducement, the argument does not necessarily apply
to the correlation between the ratio and content per year as estimated by Grytten et al.
Second, if dentists ration services in accordance with need, even a positive correlation
between the ratio and content per visit may not necessarily imply inducement but simply
the fact that, with more resources, more need can be met. Third, the authors erroneously
equate the elasticity of the odds of having used dental services with respect to the ra-
tio (0.62) with the elasticity of demand (probability of use), which is only about 0.16
(evaluated by Sintonen and Maljanen (1995a) from the data reported). How much of
this is attributable to inducement and how much to missing variables like full time cost,
remains unclear.
Sintonen and Maljanen (1995a) found that individual and general inducement ap-
peared to have a considerable effect on utilisation, but no systematic connection with
supply conditions (dentist/population ratio). This was interpreted to indicate that indi-
vidual inducement had been adopted by some dentists regardless of the market situation
they work in and general inducement took place at a more general level. In these results,
the dentist/population ratio variable reflected the net supply of dentists available to the
adult population, not the gross supply as in other studies. Also, and perhaps more im-
portantly, dental services are provided by private and salaried public dentists in Finland
and the latter have only a small incentive to induce.
Lessons from the review. Little attention has been paid to the effect of model specifi-
cation and estimation techniques in dental utilisation studies, or to examining whether
the distributional assumptions of the models are met. Exceptions are the HIE Study [see
e.g. Duan et al. (1983), Manning et al. (1986)] and Sintonen and Maljanen (1995a) in a
cross-sectional context and Parkin and Yule (1988) in a time series context. For exam-
ple, Sintonen and Maljanen (1995a) tested twelve generally used models, of which only
two met the distributional requirements. Not surprisingly the parameter estimates pro-
duced by the different models had a wide range. For example the highest total estimate
for price and income elasticity (evaluated at means) was about 4-fold and for visit time
elasticity 7-fold those produced by the appropriate two-part models.
The definitions and operationalisations of key variables, e.g. price, income and time
cost also vary considerably, resulting in a wide variation in elasticity estimates. As a re-
sult, great caution is required when comparing elasticities from different studies. Much
more attention should be paid to the specification and careful measurement of relevant
variables in demand equations. Omission of some key variables like price and full time
cost from the equations makes it very difficult to disentangle their effects from the pos-
sible effects of SID.
The existence and extent of SID in dentistry still remains uncertain. As we have noted,
plausible alternative explanations can be given for the positive correlation between den-
tist density and utilisation, which is often taken as indicative of SID. One is less likely
to observe SID in an environment where prices are flexible than in a fixed-price setting.
The results of two-part models suggest that the probability of visiting the dentist and
the amount of care for those who have visited are affected to quite a different extent
by the variables considered in the models. Moreover, the probability of visiting can be
explained quite well, but that is not the case with the amount of care. Thus, at least
partly xl 0 x2 and, for the same variables, b' 0 d'. As far as explaining the amount of
care is concerned, there are important omitted variables from the models. Perhaps a bet-
ter understanding and measurement of provider behaviour might suggest more relevant
variables. On the other hand the use of count data models seems to be highly relevant
when demand is measured as dental visits.
A better measure of oral health and proper modelling and measurement of own time
devoted to self-care (home production of oral health) may improve performance. After
all, our theoretical starting point claims that the prime driving force for demanding
dental care is health, so both self-care and utilisation of dental care should produce
measurable health improvements. Private and public insurance or government subsidies
to dental care are usually justified on the ground that dental care improves oral health.
The HIE showed that more generous dental insurance coverage improved oral health
for those younger than 35 years and especially for subgroups of the population with
the poorest oral health [Bailit et al. (1987)]. That no clear improvement was observed
in other groups may be due to the insensitivity of dental health measures used. This
highlights the importance of developing the measurement of oral health further.
The relatively poor performance of the models in explaining the variation in the
amount of care after a decision to visit may also have to do with the fact that in spite of
the dynamic nature of dental care utilisation and oral health development, the models
used have been static. One obvious reason for not using dynamic models is a lack of
appropriate panel data over a longer period of time.
4. Productivity, technical efficiency and economies of scale in dentistry
4.1. Introduction
Analysing productivity or technical efficiency might contribute to making better use

of scarce resources. From the consumer's point of view lower productivity may cause
longer queues, whereas improvement in productivity may decrease delays in access.
Productivity measurements are also regarded as significant for management, helping to
ensure that activities are well run and developed, and that managers and personnel face
suitable incentives.
In this subsection we first consider concepts, then review different methodological

approaches to productivity or technical efficiency, to be followed by a brief review of
empirical studies in the field in dentistry. Finally we consider the lessons from the review
and to what extent the existing research has fulfilled or might fulfill the expectations held
of this type of research.
4.2. Theoretical concepts
Productionand cost functions. To keep the presentation simple, we assume that only
one output y, is produced using m inputs x = (xl, x2, ... , Xm). An example of an output
produced in dentistry is "fillings" and inputs used in the production of fillings are, for
example, "dentist's own time", "time of ancillary personnel", "capital equipment" and
"materials needed in the production of fillings". Given the levels of inputs x and current
state of technology described by a parameter 0, the maximum production is given by a
production function f (x; 0).
Several concepts in the production studies can also be defined using a minimum cost
function c(w, y), where w is an input price vector. Because the production occurs at the
frontiers, the cost function contains the same information about the production technol-
ogy as the production function. 21
Productivity and technicalefficiency. The simplest productivity concept is input factor

productivity, IFP, which is defined as the ratio of output to the level of individual input
used, or IFP= y/xi, for some input xi. IFP measures the average output produced by
the input factor xi jointly with other inputs. Total factor productivity, on the other hand,
measures the average production of all input factors. Suppose that I(x) measures the
total usage of all inputs simultaneously. Then the total factor productivity, TFP, is a
ratio TFP = y/l (x).
Technical inefficiency exists when there is a deviation of output from the highest pos-
sible output. Actual production cannot be higher than the maximum production defined
by the production function. Production is said to be efficient if the actual production is
equal to the maximum production attainable. Otherwise production is inefficient. Usu-
ally inefficient production implies that actual costs exceed the minimum costs of pro-
duction defined by the cost function. This happens because it is possible to reduce the
use of at least one of the inputs and still produce the same level of output. With posi-
tive input prices this means that realised costs are also being reduced and therefore the
realised costs must have exceeded the minimum costs of production. All this assumes
of course that output is adequately measured and that the apparent inefficiency does not
reflect the (efficient) use of resources to produce a different, or higher quality, output. 2 2
21 See e.g. Varian (1992).

22 The authors thank Tony Culyer for this point.
Economies of scale. Economies of scale (ES) describe the sensitivity of output with
respect to input scalings. Production technology is said to exhibit increasing returns
to scale, if f(tx) > tf(x), decreasing returns to scale if f(tx) < tf(x), and constant
returns to scale if f(tx) = tf(x). Here t > 1 is a scaling factor. In empirical studies,
one is often interested in the local behaviour of production and inputs and therefore it is
useful to define a concept of local economies of scale [see e.g. Varian (1992)]:
ES(x)= f(tx ) t (11)

at f(x, 0)
ES(x) describes the sensitivity of output with respect to small input scalings around
input combination x. ES(x) is evaluated at t = 1 to measure local economies of scale.
The production function is characterised by increasing, decreasing, or constant returns
to scale at x as ES(x) is greater, less, or equal to 1.
Due to the duality of production and cost functions, local economies of scale can also
be defined using the cost function as follows
SE(w, y) c(w, y (12)

ay c(w,y)
SE(w, y) measures the elasticity of cost with respect to small output scalings at (w, y).
If SE(w, y) < (>) 1, then there are economies (diseconomies) of scale at (w, y). Oth-
erwise, technology exhibits constant returns to scale.
4.3. Methods
Estimation of production and cost functions and nonparametric Data Envelopment

Analysis (DEA) have been used in the measurement of productivity, economies of scale
and technical efficiency in dentistry. Studies that follow the production function ap-
proach [see, e.g. Nash and Wilson (1978), Scheffler and Kushman (1977), Sintonen
(1986)] use an empirical specification of the production function. A widely used specifi-
cation is the transcendental production function first used in health economics literature
by Reinhardt (1972)
ni
Yi = [x i
exp(bjxji)]expg(x)e 'i, (13)
j=l
where y is output, xj is input j, c is a parameter, a and b are parameter vectors, g

is some function of the input vector x and is disturbance term. The transcendental
function is flexible because it does not place a priori restrictions on the behaviour of
marginal products of inputs. 23 Using different specifications of g(x) one can estimate
23 For a definition of flexible functional forms see Lau (i986).

several forms of the function. In dental economics Scheffler and Kushman (1977) and
Sintonen et al. (1983) chose g to be a polynomial.
Studies following the cost function approach use similar methodology as the produc-
tion function studies by choosing a specification for the cost function. A flexible cost
function specification for the cost function, which was used in dental economics by
Grytten and Dalen (1997), is a translog cost function
m
n ci = o + al In yi + E fj In wji
j=l
+2 I C (lnyi )2 + PIklIn wli Ilnwki

1=1 k=l
+ yy lnwjiln yi + i, (14)
j=1
where c measures costs, wj is input of price j and /lis disturbance term. The advantage
of the translog cost function compared to the Cobb-Douglas functional form or to the
CES function is that it allows economies of scale to change with output.
Technical efficiency is a maintained assumption in the ordinary model. To test that
assumption, empirical production and cost function models have to be modified slightly.
This can be done by resorting to the econometric frontier models first introduced by
Aigner et al. (1977). The basic idea is to decompose the error term of a production or
cost model into two parts
Ei= i vi, (15)
in which r is treated like the ordinary disturbance term and v is the inefficiency term
dealing with inefficient observations. In the production function model v is constrained
to be nonpositive and in the cost model v is constrained to be nonnegative. Applications
of the production or cost frontier models in health economics can be found in the Jour-
nal of Health Economics, 1994, pages 255-300, including a critique of the method by
Skinner (1994) and Jones (2000), but so far the method has not been used in dentistry.
Data Envelopment Analysis (DEA) is a nonparametric method used to evaluate tech-
nical efficiency, productivity and economies of scale of decision making units. The theo-
retical virtue of DEA is that it does not require any a priori assumption on the functional
relationship between inputs and outputs. 24 The method can be applied to decision mak-
24 When estimating the production or cost function, one has to choose a specification for the production or
cost function. The flexibility of the functional form is often given a lot of emphasis to avoid placing strong a
priori restrictions on the form of production technology. Flexibility is not the only criterion, however. Besides
flexibility, Lau (1986) lists theoretical consistency, domain of applicability, computational facility, and factual
conformity of a functional form. When all these criteria are given equal weight, the choice of functional form
is not a straightforward matter.
ing units producing several outputs that can be measured in different "natural" units
without having to establish in advance a set of weights for aggregating them (or in-
puts). It is also possible to incorporate variables describing the quality aspects of inputs
and/or outputs (variables describing quality aspects can also be introduced in produc-
tion or cost functions). Applications of DEA in dentistry are Roos (1996) and Nordblad
et al. (1996).
The DEA method was introduced by Charnes et al. (1978). According to Seiford
and Thrall (1990), there are already several different DEA models emphasising either
input or output efficiency and making different assumptions on returns to scale. Sup-
pose that we have n dentists each producing I outputs and using m inputs, and let
xi = (xli, ... , xmi) and yi = (li, ... , yli) be the observed input and output vectors
for dentist i. The efficiency score for dentists j, 0, is obtained from a problem
min0, (16)
2,)ii Yrj, (17)

i=l
Oxsj - Q O a,i, (18)

i=l
ki > 0, (19)
where r = 1,2, 1, s = 1, 2, ... m, and i = 1, 2 ... , n. In relative terms, the effi-

cient dentists obtain a score of 1 and inefficient ones less than 1.25
The intuition of the input oriented model presented above can be illustrated graphi-
cally in the case of two inputs, xl and x2, and one output, y (Figure 1).
The production isoquant for the level of output y is presented by I (y). Now suppose
that dentist i produces output y using an input bundle of x. Obviously dentist i is not
efficient, because the same level of production can also be achieved by some input
bundle on the isoquant I (y). One possibility is to scale the input use down towards the
origin until the isoquant and x* are reached. A natural measure for inefficiency in this
case is the ratio of distances between x * and the origin and x and the origin. To measure
inefficiency, it is convenient to define a distance function
D(x, y) = max > 0X EV(y) , (20)
where V(y) is the input requirement set defining all input bundles that can produce out-
put y. The distance function D has a value of 1 if the input bundle lies on the isoquant;
25 The model presented above assumes constant returns to scale. The model can be extended to handle vari-
able returns to scale. More on this and other types of models, see Seiford and Thrall (1990).
X2
V(y)
xD(x,y)
I(y)
X1
Figure 1.
for any input bundle that belongs in the interior of the input requirement set D > 1. In
Figure 1, x* = 1/D(x, y)x. The measured technical efficiency (or the optimal value of
0) obtained as a solution of the problem (16) (subject to constraints) is an estimate for
1/D(x, y) 26 [see e.g. Fire and Grosskopf (1996)].
The distance function and the DEA method together with the Malmquist productiv-
ity index 2 7 turn out to be useful concepts when studying changes in productivity over
time. In the Malmquist index, the observed improvement in productivity between two
dates is attributed to improved technical efficiency or to improvements in the production
technology, or both. Given a distance function for technologies and production plans in
different time periods, the Malmquist productivity index uses the values of the distance
function, which in turn can be estimated using the DEA method. Roos (1996) utilises
these concepts in his study on productivity changes in public dentistry in Sweden.
It is well known that the DEA method is sensitive to imperfect output measurement,
variable selection, outliers and influential observations, which obviously are drawbacks
of the method. Moreover, the method tends to see the productivity of each unit in as
"favourable" light as possible. This means that if there are several outputs and a unit is
technically efficient in producing one of them but inefficient in producing the others, it
still obtains an efficiency score of 1. Thus, the score does not necessarily indicate a high
26 The optimal 0 is equal to the inverse of the value of distance function if efficient observations operate on
the production function. Otherwise this need not be the case.
27 For definition and applications of Malmquist index, see e.g. Fare and Grosskopf (1996).
overall efficiency of the unit. This tendency can be countered by placing restrictions to
the weights [Dyson and Thanassoulis (1988)].
The reliability of efficiency measurement both in the DEA and frontier methods
hinges on the availability of a full set of input and output measures. Moreover, cost fron-
tier estimation requires data on input prices, which are not always readily available. 28
Producing high quality is not costless and, if the output measures are not adjusted to
reflect quality changes, the extra cost may be erroneously interpreted as inefficiency.
Newhouse (1994) gives examples. In frontier estimation, in particular, observations that
"look" inefficient may also be due to the skewness of the error term r [see Skinner
(1994)].
4.4. Measurement and results
No matter which of the above approaches is chosen to study the productivity or tech-
nical efficiency of dentistry, one has to deal with the question of input and output mea-
surement. The ideal measure of output would be the change in oral health, or each
dentist's contribution to oral health. However, there is no generally accepted yardstick
for oral health. Moreover, one would immediately run into problems when measuring
an individual dentist's contribution to oral health. Therefore, output in dentistry has
been measured by several intermediate measures, for example the number of patients or
visits [see e.g. Scheffler and Kushman (1977), Sintonen (1986), Doherty and Hussain
(1975), Bentley et al. (1984)]; total market value of procedures [see e.g. Nash and Wil-
son (1978), Sintonen (1986)]; number of relative productivity units [Mitry et al. (1976)];
and the number of relative value units [Bentley et al. (1984)].
Productivity. Scheffler and Kushman (1977) estimated a transcendental production

function using data for over 29 000 US dentists from 1967 to 1970. As an output vari-
able the authors used the number of patient visits to the practice in a week. Labour
input was measured by the dentist's time working at the chair and capital input was
measured by the number of chairs. The production function was defined so that if either
(or both) labour or capital inputs were zero then the dentist's output was zero. They
added auxiliary personnel's working time, the dentist's working time away from chair,
the dentist's free time at the office, the dentist's age, and the number of dentists per
population variables.
The results showed that, apart from technicians, the marginal products of labour input
were positive. The negative marginal product of technicians may be due to the fact that
services provided by the technicians were not captured by the output variable.2 9 In
regard to the use of auxiliary personnel, they showed that the behaviour of US dentists
was approximately consistent with the profit maximising behaviour. They also estimated
the shadow price of dentists' time to be about $12 per hour in 1968.
28 The authors thank Miika Linna for this point.

29 Use of visits as a dependent variable may also bias the estimates.
Gray (1982) used a sample of 266 General Dental Practitioners from Scotland to ex-
amine their productivity. Productivity was measured by gross fees per dentist hour and
variation in this measure was explained by a transcendental production function with
hours worked by the dentist and auxiliary staff; variables describing the characteristics
of dentist, practice and working methods; and service mix as explanatory variables. The
best model was able to explain only 16% of the variation. The estimated model demon-
strated a declining marginal product per dentist hour worked. Case-mix was generally an
insignificant explanatory factor. The most surprising result was the insignificance of sur-
gical assistants and hygienists, although other ancillary staff were significant. Dentists
not performing four-handed dentistry had lower productivity levels. Capital equipment,
measured by the number of chairs used, had a significant effect on productivity. The
author attributed the low explanatory power at least partly to the output measure, which
may not have adequately reflected the probable substantial variation in the quality of
work. The somewhat surprising results may also have something to do with the size and
nature of the private work these dentists undertook.
Sintonen (1986) used three output measures: the number of visits, the number of
procedures, and the total value of procedures performed. Each procedure was valued at
its average fee. He measured the productivity of dentists by four measures: the number
of visits per dentist hour (VpD), the number of procedures per dentist hour (PpD), the
value of output per dentist hour (OpD), and the value of output per total duration of
visits (OpTD).
He found, using measures PpD and OpTD, a statistically significant higher produc-
tivity of public dentists. On average, the productivity of public dentists was also higher
than that of private dentists using measures VpD and OpD, but the difference was not
significant.
This picture changed when he used a transcendental model to explain differences in
dentists' productivity in Finland. The main independent variables were dentist hours,
working hours of assistants, sex, education of the dentist, variables characterising the
dental status of patients, and variables describing the mix of services offered by den-
tists. He found that dentists' hours and working hours of auxiliary personnel both had
a significant and positive effect on productivity of all dentists. Time off during working
hours for various reasons lowered productivity. Increases in the proportion of older pa-
tients and patients in poor oral health lowered productivity, presumably because these
patients are not easy to treat. A dummy variable distinguishing private (S = 0) and
public (S = 1) dentists yielded a statistically significant negative parameter estimate,
suggesting that, when differences between sectors in factors related to productivity (in-
put mix, practice and patient characteristics, service mix and market situation) were
controlled, the productivity of public dentists was on average 14% lower than that of
private dentists.
Jonsson et al. (1983) provided comparable results for Sweden. The authors com-
pared the productivity of private and public dentists using three measures: number of
completely treated patients per dentist hour, value of output for the completely treated
patients per dentist hour (two measures of dentists' productivity), and total costs per
completely treated patient (a measure of total productivity). The results suggested that,
apart from the first productivity measure, private dentists were more productive than
public. These results are similar to those obtained by Sintonen (1986).
Westerberg (1987) studied changes in productivity in Swedish dental care over 1975-
1984. He used the number of patients treated per dentist hour and total cost per treated
patient as productivity measures, the latter indicating total productivity. This study also
indicated that private dentists performed slightly worse by the former measure, but much
better (20-30% better) by the latter. Moreover, the total productivity of public dentists
declined over 1975-1981 but improved after that, whereas the total productivity of pri-
vate dentists remained stable over the whole study period.
Utriainen et al. (1994) explained the differences in the total productivity of public
dentists working in Finnish health centres using a transcendental production function.
Total productivity was defined as the total output of the dentist (measured by the num-
ber of procedures multiplied by the fixed fee schedule of private dentists) divided by
the total production cost. The variation in total productivity was explained by labour
inputs of the dentist and chair-side auxiliary personnel, capital inputs were measured by
office space and age of dental unit, while other variables described the characteristics of
the dentists and their working patterns. They found that the marginal product of chair
side auxiliary personnel evaluated at the mean was negative, suggesting that a marginal
increase in this input raised cost more than the value of output. The number of dentists
working in the same clinic also had a negative effect on total productivity. The results
suggested that the optimal number of chair side auxiliary personnel per dentist was 1.3
if the objective is to maximise total productivity.
In an earlier study Utriainen et al. (1993) found that total productivity (measured
by number of visits divided by total running costs) in Finnish public health centres
decreased 6.8% per year from 1982 to 1991. In a pilot study based on the Malmquist
Productivity Index approach in Sweden it was found that productivity declined in 1993-
1994 on average by 3.9% and in 1994-1995 by 6.7%. The -3.9% productivity change
in 1993-1994 could be split into -4.2% due to change in technology and +0.3% due to
change in technical efficiency, whereas in 1994-1995 the corresponding figures were
-4.7% and -2.0%, respectively. No quality-related variables were included in the
study. The study also clearly showed that the results are sensitive to the way outputs
are measured [Roos (1996)1.
It is difficult to say whether these results indicate real productivity declines or whether
they give an incorrect picture of the development due to measurement errors or inabil-
ity to address adequately or at all changes in factors like casemix, quality, economic
incentives and practice patterns over time.
Economies of scale. Scheffler (1979) reported results of the Research Institute Trian-
gle (RTI) study by Nash and Wilson (1978) in which a production function for dentists
was estimated. Three variables were used as output measures: number of patient vis-
its, gross billings and value added. Capital inputs were measured by capital costs after
depreciation, and labour inputs of dentists and auxiliary personnel were measured by
Ch. 24: Economics of DentalServices 1283
total annual hours worked. Other explanatory variables used were the dentists' age, size
of dental firms, use of four-handed dentistry, regional per capita income, and the den-
tist/population ratio. The RTI study estimated Cobb-Douglas and transcendental spec-
ifications of a production function. Results from the estimated Cobb-Douglas function
suggested that the average size dentistry unit experienced economies of scale.
Grytten and Dalen (1997) estimated economies of scale in Norwegian private den-
tistry by using a translog specification of the cost function. They measured the output
by the number of consultations and included four input variables: dentists' time, dental
surgery assistants' time, capital and supplies. Besides input prices and output variables,
the estimated cost function contained several control variables hypothesised to affect
costs. It was found, when evaluated at the sample mean, that the measure for economies
of scale obtained from the estimated translog function was 1.85. This is the inverse of
the measure in Equation (12). The results also suggested that in the average practice
the elasticity of substitution between dentist's and assistant's time is small, which is
consistent with Scheffler (1979).
Pietila et al. (1998) investigated the costs and productivity of orthodontic care pro-
vided by the municipal health centres in Finland with data for 1992. Output was mea-
sured by the estimated number of completely treated patients and total productivity by
the production cost of orthodontic care per completely treated patient (ATC). The ATC
varied within wide limits ranging from FIM 1299 to FIM 24751 (on average FIM 7358).
The variation was to some extent explained by the mix of labour input and whether
the treatment was started early or late (early treatment was associated with somewhat
lower ATC) but remained largely unexplained. The results also suggested that most or-
thodontic units operated under increasing returns to scale. Lowest average cost unit size
in terms of treated patients was 830 completely treated patients per year compared with
the observed average of 133. The lowest cost ATC clinic size would have been higher if
relevant costs items like capital costs and social security contributions on top of salaries
paid had been included.
Technical efficiency. Nordblad et al. (1996) used DEA to measure technical efficiency
in public health centres in Finland in 1992. Inputs were measured by the running costs
of dental care consisting mainly of salaries, material costs and rents. Output was mea-
sured by the number of patients examined and number of visits in three age categories
of patients. The results indicated that there was large variation in technical efficiency
(from 0.44 to 1) in dental care provided by the Finnish health centres and that this vari-
ation was not explained by the age structure of the patients.
Lessons from the review. The review reveals that, irrespective of the methodology
of analysis used, there appears to be great cross-sectional variation among individual
providers or dental care units in productivity and technical efficiency. The extent to
which the studies standardised for omitted variables, such as possible differences in
casemix or quality, varied. Less attention has been paid to the development of produc-
tivity or technical efficiency over time. Productivity studies rarely give clear indication
of what should be done by poor-productivity units to improve their performance.
1284 H. Sintonen and . Linnosmaa
The empirical evidence seems also to suggest consistently that dental care units and
clinics operate under increasing returns to scale. Why this is so remains an unanswered
question. One plausible explanation is that there is incomplete information in the mar-
ket about efficient production techniques, a variant of the extended problem of regional
variation in patterns of use of medical services [Phelps (2000)]. Seeking explanations
and ways of improving productivity is an important future challenge since these inef-
ficiencies have major resource implications. It is to be noted, however, that the least
production cost clinic size does not probably represent the optimal clinic size in terms
of social costs. To be able to estimate that, one would need to know, in addition to
the production costs, the travel and time costs of the patients and their accompanying
family members. With these costs included the socially optimal clinic size would prob-
ably vary geographically depending on population density, supply of dentists and other
factors affecting travel and time costs.
There has been a considerable development in methodology over the past ten years.
Irrespective of the method, however, the measurement of inputs and outputs needs to be
developed further. In particular, it is important that quality is addressed more seriously
than has been the case so far. Otherwise artificial productivity gains can be exhibited,
e.g. by diluting the services provided at the expense of quality and effectiveness.
When using dental services, consumers assume that they are beneficial to oral health.
Thus, from the consumer's point of view, the primary concern is allocative efficiency
(i.e. cost-effectiveness), not the productivity or technical efficiency of their production.
Productivity studies do not tell (at least have not told so far) what is the relationship
between productivity and allocative efficiency. They are not able to answer the question
of whether allocative efficiency in a poor productivity unit is better or worse than that
in a high productivity unit. If it can be assumed that dental services are generally bene-
ficial to oral health then productivity and allocative efficiency tend to go hand in hand.
However, rather than relying on an uncertain assumption we should focus our attention
and efforts on allocative efficiency, to which we turn next.
5. Economic evaluation in dentistry
5.1. Introduction
The main methods and principles of economic evaluation are dealt with in detail in
other chapters of this Handbook, in particular Garber (2000). A thorough treatise can
also be found in Drummond et al. (1987, 1997) and Gold et al. (1996). We focus on
measurement issues specific to dental care and also review some main results obtained.
Finally some directions for future work are outlined.
5.2. Measurement of costs
Weinstein and Stason (1977) propose the following general decomposition of the costs
of a health care programme
AC = ACRx + ACs - ACMorb ACRXALE. (21)
Here ACRp contains all direct costs due to the health care programme, ACs measure
the health care costs incurred by the adverse side-effects of a health care programme,
ACMorb contains all savings in health care costs due to the programme, and ACRALE
include all treatment costs borne due to the fact that the programme increases the aver-
age life-years of population.
The main components of direct costs, ACR, are variable costs, capital costs and
overhead costs [see Drummond et al. (1987)]. A significant part of variable costs in a
dental programme is labour costs. Other relevant variable costs are costs of materials,
such as chemicals in fluoridation programmes. Fluoridation equipment in fluoridation
programmes and teaching facilities in preventive care programmes provide examples of
physical capital. Capital costs can be further divided into opportunity cost and depreci-
ation. By definition, opportunity costs measure the benefits lost somewhere else when
the capital resources are used for the purposes of a particular programme and deprecia-
tion is the annual loss in the value of assets [for more thorough discussion of costs, see
Garber (2000)].
Yule et al. (1986) observed that many dental studies do not pay adequate attention
to the opportunity costs of resources. For example, Horowitz and Heifitz (1979) and
Manau et al. (1987) argue that services provided by school and water plant personnel in
fluoridation, toothbrushing and mouthrinsing programmes are costless.
This is false, since there is an opportunity cost involved in personnel time. Niessen
and Douglas (1984) showed how drastically the results may change when opportunity
costs of relevant resources are ignored. They used both CBA and CEA to analyse four
dental programmes in a population consisting of 7000 school-aged children. The pro-
grammes were community water fluoridation, school water fluoridation, weekly school-
based fluoride mouthrinse, and school-based sealant programme. The authors computed
annual and total costs for the school mouthrinsing programme with and without the op-
portunity cost of teachers' time. One hour of a teacher's time was valued at $10. The
total and annual costs of the programme with opportunity costs were approximately 4
times higher than corresponding costs without opportunity costs.
One important benefit of a dental care programme is the cost savings arising due to
the programme, ACMorb. In CBA studies benefits are often measured as cost savings
[White et al. (1989)] and the factor ACMorb is ignored as a cost component. In CEA
studies, however, it is rare that the cost savings component is included. An exception is
Donaldson et al. (1986) who estimated the cost savings component by multiplying the
reductions in DMFS attributed to the programme by the price of restoring one surface.
In several studies not all cost components considered are clearly reported. For ex-
ample, Horowitz and Heifetz (1979) do not report labour costs, and Nelson and Swint
(1976), Niessen and Douglas (1984) do not report overhead costs, and Davies (1973)
and Doessel (1985) do not report any of labour, capital or overhead costs. The fact that
costs are not reported does not mean that the costs are wholly ignored, since usually
some estimate for the direct total costs is given. On the basis of the information pro-
vided in the studies it is often difficult to assess whether all relevant costs have been
included.
5.3. Measurement of outcomes in CEA
A widely used physical outcome measure of dental programmes has been the DMFT
(DMFS)-index. The index has been used e.g. by Horowitz and Heifetz (1979), Manau
et al. (1987), and Birch (1990). The measure has been heavily criticised. One of its
drawbacks is that it ignores any changes in the quality of a tooth [Birch (1986)]. For
example, if the individual has one carious and one missing tooth, DMFT has a value
of 2. Suppose now that the individual goes to the dentist who fixes a carious tooth. Now
the individual has a filled and missing tooth and DMFT still has a value of 2, despite the
quality of the stock of teeth being evidently higher.
Birch (1986) improved the simple DMFT-index by developing a measure for the
quality-adjusted tooth years (QATYs). Let Vi and Ni (t) denote the value and the number
of type i teeth in period t for an individual. Possible tooth types are decayed, missing,
filled and sound. Let S = {d, m, f, s} be the set of possible types. Normalise the values
of sound and missing teeth to 1 and 0, respectively, or set Vs = 1 and Vm = 0. Let Vd
and Vf denote the individual's values for decayed and filled teeth. The total value of the
individual's dental stock at a particular point of time t, V(t), is obtained as the sum of
these values taken over individual's teeth. To put the same thing formally
V(t) = Vi Ni (t). (22)

iGS
Lifetime QATYs for the individual are then obtained by taking the sum over an indi-
vidual's life-time.
Foch (1981) criticises physical measures on the ground that they are not sensitive to
changes in dental health occurring after the evaluation period. This may be the case if,
for example, a programme delays rather than prevents dental decay. As pointed out by
White et al. (1989), postponement of deterioration in dental health may be an important
objective per se because this produces more high quality dental years for individuals.
Foch's criticism can be overcome by the use of measures of expected QATY
(EQATY). Such a measure can be formulated as follows. Let Vi (t) denote the utility
value of tooth i at year t and qt the probability that the individual loses tooth i in year t
given that he has had the tooth for x years. The expected quality-adjusted tooth years
for a tooth of age x can be computed using formula
d t-1
EQATY=>( -q
2
) (1-q 5 ) Vi (t). (23)
(23)
t=x S=x
Here d is the last whole living year for the tooth. If an individual never looses the
tooth, d is the dying year of an individual. Computation of EQATY is based on the
assumption that the tooth lives half of its last year. If the health status of the tooth
changes during year t, the utility value for the tooth is obtained as a sum
Vi (t) = Vrhrt, (24)

rcS
where hrt is the tooth's time spent in dental health state r in year t. The expected quality-
adjusted tooth years for the individual's dental stock is obtained by taking a sum over
the expected quality-adjusted tooth years of individual teeth. Antczak-Bouckoms and
Weinstein (1987) provide an example of a study using EQATYs as an effectiveness
measure.
5.4. Measurement of outcomes in CBA
CBA studies in dentistry typically measure outcome by the savings associated with the
health programmes being evaluated. Davies (1973) compared CB ratios of different flu-
oridation programs, including community water fluoridation, school water fluoridation,
fluoridate tablets, and fluoride solutions. The monetary benefits of the programmes were
measured as cost savings associated with the prevention of caries. Niessen and Douglas
(1984) measured the physical outcome of a programme as the difference between the
number of carious teeth per person in a test and control group. A monetary outcome
was obtained by attaching a cost of $20 to treating a carious surface. 3 0 Nelson and
Swint (1976) and Doessel (1985) also used cost savings for measuring the monetary
consequences.
However, the estimated cost savings as a measure of benefits is not capable of placing
any value on the quality difference between sound and restored teeth and thus under-
estimates true benefits of the programme. Yule et al. (1986) point out that cost savings
approach tends to underestimate true benefits to the society because it typically ignores
time and travel costs and reductions in physical uneasiness, like pain and discomfort.
A systematic downward bias in benefit estimates may nontheless be useful, if the result-
ing benefits still exceed costs.
30 Thus assuming that this represents the minimum willingness to pay.

Willingness-to-pay (WTP) is another way of quantifying benefits in monetary terms.

According to this approach, an individual's valuation of a good is given by the area un-
der a demand curve. The demand curve also reveals how much the individual is willing
to pay for an extra unit of that good. When applied to dental care, the individual's WTP
for a sound tooth might be the WTP for the dental care needed to restore the tooth. Ac-
cepting this approach implicitly assumes that the patient can always evaluate the value
of treatment needed to restore a tooth to full health and can assess the probability that
the treatment will be effective. Yule et al. (1986) argued that dentists are better judges
than patients on how much care is needed and the value of that treatment. If that is
the case, one faces a problem that dentists may place a different value on the WTP for a
treatment than the patient.3 1 Common to all WTP studies is, of course, the fact that will-
ingness is a function of ability to pay. Its acceptability as a method of valuing benefits
thus depends at least in part upon distributional value judgements.
5.5. Results
Reviews of economic evaluation studies of dental care programmes are: Yule et al.
(1986), Antczak-Bouckoms et al. (1989), White et al. (1989), and Mitchell and Mur-
ray (1989). There are four major types of programme in these studies: fluoridation
programmes [e.g. Vehmanen (1993), Birch (1990), O'Rourke et al. (1988), Doessel
(1985), Davies (1973), Manau et al. (1987), Niessen and Douglass (1984), Horowitz
and Heifitz (1979)], sealant programmes [e.g. Donaldson et al. (1986), Llodra et al.
(1993), Mitchell and Murray (1989), Lewis and Morgan (1994)], periodontal programs
[e.g. Antczak-Bouckoms and Weinstein (1987), Douglass and Fox (1996)], and health
education programmes [e.g. Doherty and Martie (1987), Manau et al. (1987), Niessen
and Douglass (1984), Horowitz and Heifitz (1979)].
Fluoridationprogrammes. Results of CEA and CBA studies suggest that water flu-
oridation programmes are socially worthwhile [Nelson and Swint (1976), Niessen and
Douglas (1984), and Doessel (1985)]. Doessel (1985) estimates a 28% rate of return for
a fluoridation programme in Australia. Nelson and Swint (1976) estimated benefits and
costs of fluoridation over 20 years period in Houston, Texas. They used 10% discount
rate and estimated the discounted net benefit to be $1,202,970. Niessen and Douglass
(1984) estimate $545,352 net benefit for a community water fluoridation programme.
They used a 5% real discount rate and the time horizon was 20 years.
When compared to other dental health programmes for the prevention of caries, flu-
oridation programmes tend to perform well. Niessen and Douglas (1984) compared the
CB and CE ratios of community water fluoridation, school water fluoridation, weekly
school-based fluoride mouthrinse, and school-based sealant programmes to each other.
31 The truth is probably that dentist can assess the effectiveness better than the patient, while the patient can
assess the value of this outcome better.
Ch. 24: Economics ofDental Services 1289
The community water fluoridation programme had the highest benefit-cost ratio among
the programmes studied. A similar conclusion was obtained from the CEA. The cost
of attaining one unit of caries reduction is lowest for the community water fluoridation
programme. The CE of fluoridation programmes depends on the prevalence of caries in
the population under study [Birch (1990)]. Achieving a one unit improvement in effec-
tiveness with a fluoridation investment was approximately four times more expensive in
low caries areas than high caries areas.
Manau et al. (1987) compared the CE of a community fluoridation programme with
school mouthrinsing and toothbrushing programmes. Horowitz and Heifetz (1979) com-
pared the CE of a fluoridation programme to school water fluoridation and fluoride tablet
programmes. Both studies reached a similar conclusion to that of Niessen and Douglass
(1984), namely that the community water fluoridation programme is superior to alter-
native programmes.
Vehmanen (1993) carried out an economic evaluation of three caries preventive strate-
gies among at-risk children aged 13-14 years in a randomised field study. Over a two
year treatment period the interventions in three randomised groups were normal preven-
tive care consisting of 1-2 applications of topical fluoride (fluoride solution or fluoride
varnish) a year (C-control) group), normal preventive care plus a recommendation to
use chlorhexidine-fluoride solution for mouth rinsing twice a day for 1-2 weeks every
fourth month (S-group), and fluoride varnish every fourth month (V-group). All groups
used fluoride tablets daily. The follow-up measurements took place at two and four
years. There was no statistical difference between the groups in net caries increments
and caries restoration costs over two year and four year periods. It was not reported
whether the prevention costs differed significantly between the groups, but the costs ap-
peared to be higher in the S- and V-groups than in the C-group. At best the intensified
prevention interventions produced an equivalent effectiveness at the same cost as the
normal preventive care; at worst an equivalent effectiveness at a higher cost.
Sealant programmes. Llodra et al. (1993) performed a meta-analysis on evaluation

studies of sealant programmes. The results indicated that the autopolymerized sealants
are more effective than light-polymerizing sealants. Since costs of sealants were not
included in the study, this does not tell anything about their relative CE. Moreover, they
found that the effectiveness of sealants is higher in populations exposed to fluoridated
water.
When compared with other dental care programmes, sealant programmes do not per-
form well. In their survey Mitchell and Murray (1989) concluded that early clinical trials
of the CE of fissure sealants found that sealants were not cost-effective when compared
with fillings. Niessen and Douglas (1984) reached a similar conclusion. When compared
with community water fluoridation, school water fluoridation, and weekly school-based
fluoride mouthrinse programmes, the sealant programme performed worst in terms of
CB and CE ratios. It is worth mentioning, however, that there is some evidence [see
Akehurst and Sanderson (1993)] suggesting that, when targeted to a subpopulation of
high-risk children, sealants can become a cost-effective way to protect pit and fissure
surfaces. This is due to high effectiveness of sealants at preventing decay in pit and
fissure surfaces.
In their review of methods used in the economic evaluation of fissure sealants Lewis
and Morgan (1994) found many methodological errors. Among them they named inap-
propriate choice of comparators influencing the conclusions (the controversy is whether
comparators should be other preventive measures or restorative procedures), equating
different outcomes without regard for their disparate utilities, lack of discounting and
lack of sensitivity analysis.
Donaldson et al. (1986) compared the CE of a preventive dental health programme
with the ordinary restorative treatment of carious teeth. The preventive programme con-
sisted of personal health education, oral fluoride supplements, applications of fluoride
gel, and pit and fissure sealants. The cost per unit improvement in DMFS 32 was higher
for the preventive programme than for the restorative treatment. Once the quality differ-
ences between the restored and healthy teeth was taken into account, however, the rank-
ing changed. Given a value of a healthy tooth as 1, they demonstrated that if the value
of a restored tooth is less than 0.85, the preventive programme is more cost-effective.
Periodontalprogrammes. Antczak-Bouckoms and Weinstein (1987) analysed the CE

of non-surgical and surgical procedures and the use of antimicrobiological agents in the
treatment of periodontal disease by using decision tree analysis. Results show that non-
surgical procedures have both lower cost than surgical procedures and higher outcomes
in terms of quality-adjusted tooth-years.
In another study Douglass and Fox (1996) combined CEA and decision analysis to
explore whether new diagnostic tests can be useful in the treatment of periodontal dis-
ease. The effectiveness of the test was measured by the increased probability of choos-
ing an appropriate treatment (thus also avoiding unnecessary costly treatment), which
presumably results in better patient outcome. A utility of 0.75 was assigned to appropri-
ate treatment (meaning that 75% of patients receiving an appropriate treatment would
improve) and a utility of 0.25 to an inappropriate treatment. The authors found that, for
patients under age 45, the test was not cost-effective. On the other hand, for the patients
older than 45 and/or with history of periodontal disease, the use of test was justified on
the basis of CEA.
Restorative treatments. Maryniuk et al. (1988) used a decision analysis model to esti-
mate the expected lifetime treatment costs of replacing a failed amalgam on a posterior
tooth either with another amalgam or with a crown for a 30 year old patient. The subse-
quent sequence of outcomes (more amalgams or crowns with/without need for endodon-
tic treatment) was modelled by using information on relative restoration longevity from
32 Preventive programme and restorative programme were intended for 4-6 years old children and 7-10 years
old children, respectively. Dental health was measured by the dmfs-index and the DMFS-index in these age
groups, respectively.
the literature. The probabilities of these outcomes were derived from a panel of dentists.
A sensitivity analysis was used to test the robustness of the conclusions. Depending on
the assumptions made, the initial choice of replacement by another amalgam appeared
to result in 13-24% lifetime cost saving relative to a crown. An implicit assumption was
that the treatment strategies were of equal value to the patient.
Lessons from the review. The number of economic evaluations in the dental field is
relatively small. The number of high-quality evaluations is even smaller. Many studies
suffer from methodological problems and deficiencies like omission of relevant cost
items, paucity and poor quality of data, lack of discounting and sensitivity analysis, and
above all, from a lack of an appropriate and sensitive measure of oral health. General
reporting standards are low. Authors should comply with the guidelines for economic
evaluations [Drummond and Jefferson (1996)].
While there have been methodological improvements in economic evaluation in the
dental area, many challenges remain. Decision analysis models have been used very
little in dentistry. When they have been adopted, they have strengthened and system-
atised analyses, and have also highlighted the lack of appropriate data on the duration
and value (utility) of various outcomes and the probabilities of their occurrence. A more
widespread use of decision analysis tools would draw analysts' and "clients" attention
to the identification and proper measurement of all the relevant variables, and would
thus improve the quality of evaluation in practice.
As far as outcome measurement to date is concerned, the EQATY concept represents
a clear step forward but the QATY measure covers only cari4s-related problems and
outcomes, not other oral health problems and outcomes such as those in the periodontal
and orthodontic areas. Thus, a future challenge is to develop the QATY framework in
this direction - the goal should be a generic, utility-based, single index number oral-
health-related quality-of-life measure that can be combined with duration data to end
up with quality-adjusted oral health years gained as an appropriate outcome measure.
A drawback of an oral-health-specific quality-of-life measure is that the outcomes of
investments in oral health area are not comparable with those of investments in other
areas of health care. Ideally, oral health outcomes should be reflected in generic health-
related quality of life measures. Some experimentation is going on to explore this possi-
bility [e.g. Arinen and Sintonen (1995)]. Nonetheless, one needs to match the measures
to the purposes of the analysis in question.
Partly due to the methodological weaknesses of the studies carried out and partly due
to paucity of economic evaluations of the major areas of dental care, we have very little
solid knowledge of CE in dentistry. The area offers numerous important topics for future
studies. Economic evaluations have concentrated heavily on preventive measures, which
is understandable if the ultimate aim of dentistry is to retain the teeth intact. Prevention
in dentistry can save money. Water fluoridation appears to be the most cost-effective
preventive strategy, but is often not implemented, apparently for reasons that lie well
beyond the analytical scope of most CEA/CBA work.
References
Aigner, D., K. Lovell and P. Schmidt (1977), "Formulation and estimation of stochastic frontier production
models", Journal of Econometrics 6:21-37.
Akehurst, R., and D. Sanderson (1993), "Cost-effectiveness in dental health. A review of strategies available
for preventing caries", Working Paper No. 106 (University of York, Centre for Health Economics, Health
Economics Consortium).
American Dental Association (1998), "American Dental Association looks at dental reimbursement: Respon-
sible consumption of appropriate dental services", Journal of American College of Dentists 65:7-16.
Andersen, R., and L. Benham (1970), "Factors affecting the relationship between family income and medical
care consumption", in: H. Klarman, ed., Empirical Studies in Health Economics (Johns Hopkins University
Press, Baltimore).
Antczak-Bouckoms, A.A., and M.C. Weinstein (1987), "Cost-effectiveness analysis of periodontal disease
control", Journal of Dental Research 66:1630-1635.
Antczak-Bouckoms, A.A., J.F.C. Tulloh, B.A. White and E.I. Capilouto (1989), "Methodological considera-
tions in the analysis of cost-effectiveness in dentistry", Journal of Public Dental Health 49:215-222.
Arinen, S.S., and H. Sintonen (1994), "The choice of dental care sector by young adults before and after
subsidisation reform in Finland", Social Science and Medicine 39:291-197.
Arinen, S.S., and H. Sintonen (1995), "Does the 15D register variation in oral health?", in: Abstract Book,
ISTACH Eleventh Annual Meeting, Stockholm, Sweden, June 4-7, 1995, 155.
Arinen, S., H. Sintonen and G. Rosenqvist (1996), "Dental Utilisation by young adults before and after sub-
sidisation reform in Finland", Discussion Paper No. 149 (Centre for Health Economics, University of York).
LIII:941-973.
Bailit, H., J. Newhouse, R. Brook, N. Duan, G. Goldberg, J. Hanley, C. Kamberg, V. Spolsky, A. Black and
K. Lohr (1987), "Does more generous dental insurance coverage improve dental health", RAND Note
N-2591-HHS (The RAND Corporation, Santa Monica).
Bentley, J.M., P. Green and I.I. Ship (1984), "Achieving health outcomes through professional dental care:
Comparing the costs of dental treatment for children in three practice modes", Health Services Research
19:181-96.
Birch, S. (1986), "Measuring dental health: Improvements on the DMF-index", Community Dental Health
3:303-311.
Birch, S. (1988), "The identification of supplier-inducement in a fixed price system of health care provision",
Birch, S. (1990), "The relative cost-effectiveness of water fluoridation across communities: Analysis of vari-
ations according to underlying caries levels", Community Dental Health 7:3-10.
Bunker, J., and B. Brown (1974),"The physician-patient as an informed consumer of surgical services", New
England Journal of Medicine 290:1051-1055.
Cauley, S.D. (1987),'"The time price of medical care", Review of Economics and Statistics 69:59-66.
Charnes, A., W.W. Cooper and E. Rhodes (1978), "Measuring the efficiency of decision making units", Eu-
ropean Journal of Operational Research 2:429-444.
Conrad, D.A., D. Grembowski and P. Milgrom (1987), "Dental care demand: Insurance effects and plan
design", Health Services Research 22:341-367.
Davies, G.N. (1973), "Fluoride in the prevention of dental caries, a tentative cost-benefit analysis", British
Dental Journal 135:131-134.
Doessel, D.P. (1985), "Cost-benefit analysis of water fluoridation in Townswille, Australia", Community Den-
tistry and Oral Epidemiology 13:19-22.
Doherty, N., and I. Hussain (1975), "Costs of providing dental services for children in public and private
practices", Health Services Research 10:244-253.
Doherty, N.J.G., and C.W. Martie (1987), "Analysis of the costs of school-based mouthrinsing programs",
Community Dentistry and Oral Epidemiology 15:67-69.
Donaldson, C., J.F Forbes, M. Smalls, EA. Boddy, K.W. Stephen and D. McCall (1986), "Preventive dentistry
in a health centre: Effectiveness and cost", Social Science and Medicine 23:861-868.
Douglass, C.W., and C.H. Fox (1996), "Cost effectiveness of periodontal services", Journal of Clinical Peri-
odontology 23:278-282.
Drummond, M.E, G.L. Stoddard and G.W. Torrance (1987), Methods for the Economic Evaluation of Health
Care Programmes (Oxford University Press, Oxford).
Drummond, M.E, and TO. Jefferson (1996), "On behalf of the BMJ economic evaluation working party.
Guidelines for authors and peer reviewers of economic submissions to the BMJ", British Medical Journal
313:275-283.
Drummond, M.E, B. O'Brien, G.L. Stoddard and G.W. Torrance (1997), Methods for the Economic Evalua-
tion of Health Care Programmes, 2nd edn. (Oxford University Press, Oxford).
Duan, N., W.G. Manning, C.N. Morris and J.P. Newhouse (1983), "A comparison of alternative models for
the demand for medical care", Journal of Economic and Business Statistics 1:115-126. See also: The Rand
Corporation Series, R-2754-HHS (The RAND Corporation, 1982).
Dyson, R., and E. Thanassoulis (1988), "Reducing weight flexibility in data envelopment analysis", Journal
of Operational Research 85:563-576.
Evans, R.G. (1974), "Supplier-induced demand: Some empirical evidence and implications", in: M. Perlman,
ed., The Economics of Health and Medical Care (Macmillan, London and Basingstoke).
Feldstein, P.J. (1973), Financing Dental Care: An Economic Analysis (D.C. Heath and Company, Lexington).
Foch, C. (1981), "The costs, effects, and benefits of preventive dental care: A literature review", RAND Note
N-1732-RWJF (The RAND Corporation, Santa Monica).
Fare, R., and S. Grosskopf (1996), Intertemporal Production Frontiers: With Dynamic DEA (Kluwer-Nijhoff,
Boston).
Garber, A.M. (2000), "Advances in cost-effectiveness analysis of health interventions", in: A.J. Culyer and
J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 4.
Gold, M.R., J.E. Siegel, L.B. Russell and M.C. Weinstein, eds. (1996), Cost-Effectiveness in Health and
Medicine (Oxford University Press, Oxford).
Gray, A.M. (1982), "The production function of dental care in the British National Health Service", Scottish
Journal of Political Economy 29:59-74.
Greene, W. (1993), Econometric Analysis, 2nd edn. (Macmillan, New York).
Grossman, M. (1972), "The demand for health: A theoretical and empirical investigation", NBER Occasional
Paper 119 (Columbia University Press, New York).
Grytten, J., and D.M. Dalen (1997), "Too many or too few? Efficiency among dentists working in private
practice in Norway", Journal of Health Economics 16:483-497.
Grytten, J., D. Holst and P. Laake (1990), "Supplier inducement. Its effect on dental services in Norway",
Hay, J., and M.J. Leahy (1982),"Physician induced demand: An empirical analysis of the consumer informa-
tion gap", Journal of Health Economics 1:231-244.
Hay, J.W., H. Bailit and D.A. Chiriboga (1982), "The demand for dental health", Social Science and Medicine
16:1285-1289.
Holtman, A.G., and E.O. Olsen (1976), "The demand for dental care: A study of consumption and household
production", The Journal of Human Resources XI:546-560.
Horowitz, H.S., and S.B. Heifitz (1979), "Methods for assessing the cost-effectiveness of caries preventive
agents and procedures", International Dental Journal 29:106-117.
Hu, T. (1981), "The Demand for dental care services, by income and insurance status", in: R.M. Scheffler,
ed., Advances in Health Economics and Health Services Research, Vol. 2, 143-195.
1294 H. Sintonen andI. Linnosmaa
Jnsson, B., T. Faresjb and I. Westerberg (1983), "Productivitet i privat och offentliga tandvard", Studies on
Health and Society, SHS 2 (University of LinkBping).
Keeler, E.B., and J.E. Rolph (1988), "The demand for episodes of treatment in the health insurance experi-
ment", Journal of Health Economics 7:337-367.
Labelle, R., G. Stoddart and T. Rice (1994), "A re-examination of the meaning and importance of supplier-
induced demand", Journal of Health Economics 13:347-368.
Lau, L.J. (1986), "Functional forms in econometric model building", in: Z. Griliches and M.D. Intriligator,
eds., Handbook of Econometrics, Vol. III (North-Holland, Amsterdam).
Lewis, J.M., and M.V. Morgan (1994), "A critical review of methods for the economic evaluation of fissure
sealants", Community Dental Health 11:79-82.
Llodra, J.C., M. Bravo, M. Delgado-Rodrigues, P. Baca and R. Galves (1993), "Factors influencing the effec-
tiveness of sealants - a meta-analysis", Community Dentistry and Oral Epidemiology 21:261-268.
Manau, C., E. Cuenca, J. Martinez-Carretero and L. Salleras (1987), "Economic evaluation of community
programs for the prevention of dental caries, Spain", Community Dentistry and Oral Epidemiology 15:297-
300.
Manning, W.G., and C.E. Phelps (1979), "The demand for dental care", The Bell Journal of Economics
10:503-525.
Manning, W.G., H.L. Bailit, B. Benjamin and J.P. Newhouse (1985), "The demand for dental care. Evidence
from a randomized trial in health insurance", Journal of American Dental Association 110:895-902.
Manning, W.G., H.L. Bailit, B. Benjamin and J.P. Newhouse (1986), "The demand for dental care. Evidence
from a randomized trial in health insurance", RAND Corporation Series R-3225-HHS (The RAND Corpo-
ration, Santa Monica).
Manning, W.G., C.N. Morris, J.P. Newhouse, L.O. Larry, D. Naihua, E.B. Keeler, A. Leibowitz, K.H. Marquis,
S.M. Marquis and C.E. Phelps (1981), " A two-part model of the demand for medical care: Preliminary
results from the health insurance study", in: Van der Gaag and J. Perlman, eds., Health, Economics, and
Health Economics (North-Holland, Amsterdam).
Manning, W.G., J.P. Newhouse, N. Duan, E. Keeler, B. Benjamin, A. Leibowitz, M.S. Marquis and
J. Zwanziger (1987), "Health insurance and the demand for medical care. Evidence from a randomized
experiment", American Economic Review 77:251-277.
Manning, W.G., J.P. Newhouse, N. Duan, E. Keeler, B. Benjamin, A. Leibowitz, M.S. Marquis and
J. Zwanziger (1988), "Health insurance and the demand for health care. Evidence from a randomized
experiment", RAND Corporation Series R-3476-HHS (The RAND Corporation, Santa Monica).
Maryniuk, G.A., S.O. Scweitzer and R.J. Braun (1988), "Replacement of amalgams with crowns: A cost-
effectiveness analysis", Community Dentistry and Oral Epidemiology 16:263-267.
Maurizi, A.R. (1975), Public Policy and the Dental Care Market (American Enterprise Institute, Washington,
DC).
Mitchell, L., and J. Mun-ay (1989), "Fissure sealants: A critique on their cost-effectiveness", Community
Dentistry and Oral Epidemiology 17:19-23.
Mitry, D.J., K. Johnson and N.W. Mitry (1976), "Specification of the production function for dentistry: Mea-
surement and the paraprofessional input", Inquiry 13:152-157.
Mueller, C.D., and A.C. Monheit (1988), "Insurance coverage and the demand for dental care. Results for
non-aged white adults", Journal of Health Economics 7:59-72.
Nash, K., and J. Wilson (1978), "Economies of scale and productivity in dental practices", Final Report
(Research Triangle Institute, Durham, NC).
Nelson, W., and M.J. Swint (1976), "Cost-benefit analysis of fluoridation in Houston, Texas", Journal of
Public Health Dentistry 36:88-95.
Newhouse, J. (1994), "Frontier estimation: How useful a tool for health economics?", Journal of Health
Newhouse, J.P., C.E. Phelps and M.S. Marquis (1980), "On having a cake and eating it too. Econometric
problems in estimating the demand for health services", Journal of Econometrics 13:365-390.
Newhouse, J.P., and the Insurance Experiment Group (1993), Free for All. Lessons from the RAND Health
Insurance Experiment (Harvard University Press, London).
Niessen, L.C., and C.W. Douglass (1984), "Theoretical considerations in applying benefit-cost and cost-
effectiveness analysis to preventive dental programs", Journal of Public Health Dentistry 44:156-168.
Nordblad, A., M. Linna, K. Luoma and T. Niskanen (1996), "Differences between cost efficiency scores in
oral health care in health centres in Finland in 1992", Journal of Social Medicine 33:307-314.
OECD (1998), OECD Health Data 1998. A Comparative Analysis of 29 Countries (Paris).
O'Rourke, C.A., M. Attrill and P.J. Holloway (1988), "Cost appraisal of fluoride tablet programme to Manch-
ester primary schoolchildren", Community Dentistry and Oral Epidemiology 16:341-344.
Parkin, D. (1992), '"The dental care system in the UK", in: S. Arinen, J. Grytten, D. Holst, D. Parkin, H. Sin-
tonen and N. Whitehouse, eds., Dental Care Systems in Europe, An Overview and the Cases of Finland,
Norway and UK, Occasional Paper No. 45 (Den Svenska Sjukvarden, June 1993).
Parkin, D., and B. Yule (1988), "Patient charges and the demand for dental care in Scotland, 1962-1981",
Applied Economics 20:229-242.
Pedersen, K.M., and P.E. Petersen (1980), "The demand for dental care among industrial workers: Construc-
tion and testing of a structural model", Odense Working Papers No. 6 (Odense University, Denmark).
Phelps, C.E. (2000), "Information diffusion and best practice adoption", in: A.J. Culyer and J.P. Newhouse,
Pietild, T., H. Sintonen, I. Pietild, E. Widstr6m, J. Varrela and P. Alanen (1998), "Cost and productivity
analysis of orthodontic care in Finland", Community Dentistry and Oral Epidemiology 26:283-288.
Reinhardt, U. (1972), "A production function for physician services", Review of Economics and Statistics
54:55-66.
Roos, P. (1996), "Effectivitet och produktivitet inom folktandvarden. En pilot-studie", IHE Working Paper
(Lund, Sweden).
Rosenqvist, G., S. Arinen and H. Sintonen (1995), "Modified count data models with an application to de-
mand for dental care", Working Paper 293 (Swedish School of Economics and Business Administration,
Helsinki).
Ryan, M., and G. Mooney (1992), "Supplier induced demand: Where are we and where should we go",
Discussion Paper 05/92 (Health Economics Research Unit, University of Aberdeen).
Scheffler, R.M. (1979), "Productivity and economies of scale in dentistry: Some recent empirical evidence",
in: Proceedings of a Conference on Modelling Techniques and Applications in Dentistry (Northwestern
University, Evanston, IL).
Scheffler, R.M., and J.E. Kushman (1977), "A production function for dental services: Estimation and eco-
nomic implications", Southern Economic Journal 44:25-35.
Seiford, L.M., and R.M. Thrall (1990), "Recent developments in DEA. The mathematical programming ap-
proach to frontier analysis", Journal of Econometrics 46:7-38.
Sintonen, H. (1986), "Comparing productivity of public and private dentistry", in: A.J. Culyer and B. Jonsson,
eds., Public and Private Health Services, Complementarities and Conflicts (Basil Blackwell, Oxford).
Sintonen, H., and T. Maljanen (1995a), "Explaining utililisation of dental care. Experiences from Finnish
dental care market", Health Economics 4:453-466.
Sintonen, H., and T. Maljanen (1995b), "Supplier inducement and price as factors in dental care use: Econo-
metric evidence from the Finnish dental market", in: A. Harris, ed., Economics and Health: 1994, Proceed-
ings of the Sixteenth Australian Conference of Health Economics, Australian Studies in Health Service
Administration No. 78 (School of Health Service Management, University of New South Wales).
Sintonen, H., T. Maljanen, M. Heinonen and A. Mynttinen (1983), "Economics of Finnish dental care"
(in Finnish), Official Statistics of Finland, Special Social Studies, SVT XXXII:90 (Helsinki, Finland).
Skinner, J. (1994), "What do stochastic frontier cost functions tell us about inefficiency?", Journal of Health
Upton, C., and W. Silverman (1972), "The demand for dental services", Journal of Human Resources 7:250-
261.
Utriainen, P., H. Sintonen and E. Widstrom (1993), "The productivity in dental care of Finnish health centres
in years 1982 and 1991", Administrative Studies 12:235-240.
Utriainen, P., H. Sintonen, J. Kentala and E. Widstrdm (1994), "Productivity of health centre dentists in Vaasa
province", Administrative Studies 13:38-46.
Yule, B.E., B.M. van Amerongen and M.C.M. van Schaik (1986), "The economics and evaluation of dental
care and treatment", Social Science and Medicine 22:1131-39.
Yule, B., and D. Parkin (1985), "The demand for dental care: An assessment", Social Science and Medicine
21:753-760.
Varian, H.R. (1992), Microeconomic Analysis, 3rd edn. (W.W. Norton, New York).
Vehmanen, R. (1993), "An economic evaluation of two caries preventive methods", Ph.D. Dissertation (Uni-
versity of Turku, Finland).
Weinstein, M.C., and W.B. Stason (1977), "Foundations of cost-effectiveness analysis for health and medical
practices", The New England Journal of Medicine 296:716-721.
Westerberg, . (1987), "Produktion, produktivitet och kostnader i svensk tandvard", Link6ping Studies in Arts
and Science 15.
White, B.A., A.A. Antczak-Bouckoms and M.C. Weinstein (1989), "Issues in the economic evaluation of
community water fluoridation", Journal of Dental Education 53:646-657.
Widstr6m, E., and I. Barenthin (1997), "Aikuisten hammashoidon tukeminen sosiaalivakuutuksesta paittyy
Ruotsissa", Suomen LAikiiirlehti 52:1467-1471.
Widstrdm, E., K. Eaton and J. van den Heuvel (1996), Dentistry in the EU and EEA, A Council of Euro-
pean Chief Dental Officers' Survey (National Research and Development Centre for Welfare and Health,
Helsinki, Finland).
Zweifel, P., and W.G. Manning (2000), "Moral hazard and consumer incentives in health care", in: A.J. Culyer
Chapter25
THE PHARMACEUTICAL INDUSTRY*
F.M. SCHERER
HarvardUniversity
Contents
Abstract 1298
2. Distinguishing characteristics 1300
3. Pharmaceutical industry structure 1303
4. Research, development, and the discovery of new drugs 1305
5. Government regulation of new drug introductions 1308
6. Patents and pharmaceutical innovation 1316
6.1. The unusual importance of drug patents 1317
6.2. Consequences of the Uruguay Round agreement 1318
7. Pricing 1319
7.1. Pricing branded drugs 1320
7.2. Tort liability risks and prices 1320
7.3. Generic drug competition 1321
7.4. Branded vs. generic drug price competition 1322
7.5. Stimulating generic substitution 1324
7.6. New institutions; New power relationships 1325
8. Profits and price controls 1328
9. Conclusion 1331
References 1332
*The author thanks Ernst Berndt, Randall Ellis, and Joseph Newhouse for helpful comments.
1298 EM. Scherer
Abstract
This Handbook chapter surveys the extensive body of research on the economics of
the pharmaceutical industry (with peripheral attention paid also to regulated medical
devices). Pharmaceuticals is one of the world's most research-intensive industries, gen-
erating a continuing steam of new products that save lives and raise the quality of life.
The discovery of new drugs has evolved over time from a decidedly empirical process
to one based to a considerable degree upon fundamental scientific knowledge. Rich
linkages have emerged between profit-seeking manufacturers and basic research per-
formers such as universities and national laboratories. The safety and efficacy of new
pharmaceutical products are stringently regulated in most industrialized nations, adding
to clinical testing costs. Because of high expenditures on research, development, and
clinical testing and because new products, once proven, might be imitated easily, patent
protection is unusually important. The extension of patent protection to third-world na-
tions under Uruguay Round Treaty mandates has precipitated vigorous policy debates.
Patents, first-mover advantages, and the lack of good substitutes for significant new
drugs often give rise to substantial monopoly power, against which many national gov-
ernments have counterpoised a diverse panoply of price control mechanisms. When
patents expire, however, generic substitutes often introduce vigorous price competition.
The extent to which generics capture market share from the branded original drugs de-
pends upon government regulatory policies, the reimbursement strategies of health care
insurers, and the organization of health care provider institutions.
JEL classification:Ill, L5, L65, 031, 034

Ch. 25: The PharmaceuticalIndustry 1299
1. Introduction
This chapter analyzes the supply of important inputs into the provision of health care -
notably, medicines, vaccines, and other drugs, which will be called "pharmaceuticals"
here. Because their supply is affected in analogous ways by regulatory institutions and
changes in technology, passing attention will also be devoted to medical devices such
as pacemakers, arterial grafts, prosthetics, and the like.
Pharmaceutical products account for sizeable but widely varying proportions of to-
tal expenditures on health care in the world's industrialized nations. In 1993 or 1994,
for instance, their share of total health care spending was 8.3 percent in the United
States, 15.4 percent in France, 18.5 percent in Germany, and approximately 29 per-
cent in Japan [PhRMA (1997, p. 48) and Neary (1995, p. 22)]. Some reasons for the
observed differences will emerge as our analysis proceeds. Variations in income levels
and national health care institutions lead to even larger differences in average annual ex-
penditures per capita on pharmaceutical products among the world's principal regions,
as the following United Nations staff estimates for 1990 (calculated in 1980 U.S. dollars
at prevailing exchange rates) indicate1:
North America $123.90
European Community 102.90
Other Western Europe 85.70
Japan 276.60
South and East Asia 5.00
China 4.80
Latin America 20.30
Sub-Saharan Africa 3.30
The boundaries delineating the pharmaceuticals industry, as it will be analyzed here,

are imprecise. An indication of the principal components is provided by U.S. Bureau of
the Census data (1997) on the sales (in millions of dollars) of domestic manufacturers
in diverse categories for 1995:
Finished pharmaceutical preparations
for human use 48,066
Vaccines and other biological products
for human use 4,986
Blood and blood derivatives for human use 1,624
Diagnostic products (used in vivo and in vitro) 8,271
Pharmaceutical and biological preparations
for veterinary and industrial use 2,601
Bulk medicinal and botanical products
(many used as inputs to other sectors) 7,111
The source is Ballance et al. (1992, pp. 30-31). Herbal and other naturally occurring traditional medicines
are excluded from the estimates. For 1995 estimates based upon purchasing power parity rather than prevailing
exchange rates, and showing France in the lead followed by the United States and Japan, see PhRMA (1998,
p. 75).
1300 FM. Scherer
The first category will be emphasized here, although peripheral attention will be de-
voted to analogous products. Manufactured products other than pharmaceuticals and
biologicals used primarily in health care had the following U.S. sales at wholesale (in
millions of dollars) during 1995:
Surgical, orthopedic, prosthetic, and therapeutic
appliances and supplies 10,566
Other surgical appliances and supplies 964
Electronic hearing aids 476
Surgical and medical instruments and apparatus 14,285
Dental equipment and supplies 2,117
Electromedical diagnostic and therapeutic equipment 6,852
Ophthalmic goods (including eyeglasses) 2,715
2. Distinguishing characteristics
The pharmaceuticals industry has a number of characteristics differentiating it from

most industries serving demands outside the health sector.
For one, pharmaceutical products can often be obtained legally only when the con-
sumer presents a prescription from a physician. Consumers in effect cede an important
decision-making role to physicians, who include in their decision calculus professional
responsibilities and malpractice liability risks and who, usually incurring no obligation
to pay for the consumer's purchase, may serve as imperfect agents for the consumer.
The prescription system is much more prevalent in industrialized nations than in de-
veloping nations [see Peltzman (1987)]. Its mandatory extension in the United States
to a wide range of new pharmaceutical products occurred through regulatory agency
action in 1938 without a clearly focused Congressional debate [Temin (1979a)]. Un-
der U.S. regulations, paralleled by those of many other nations, a distinction is made
between prescription drugs, which require a physician's intervention and which can be
purchased for out-patient use only from licensed pharmacists, and "over the counter" or
"proprietary" drugs, which can be obtained without prescription from a broader array
of retail outlets. In the United States, prescriptions are normally required for new drugs,
but chemical entities shown by substantial experience to have minimal toxicity risks and
whose use (often at less intense dosage levels) does not require the advice of a physician
can be accorded over-the-counter status. See Schachtel (1994) and Weintraub (1994).
Since 1972, many drugs such as acetaminophen, ibuprofen, numerous antihistamines,
the four leading anti-ulcer agents (Tagamet, Zantac, Pepcid, and Axid) and diverse top-
ical antibiotics and steroids have been shifted to OTC status in the United States. The
share of OTC drugs in sales of all U.S. pharmaceutical entities for 1988 was estimated
to be 23 percent [Ballance et al. (1992, p. 35)]. Comparable estimates for twelve other
nations ranged from six percent (for Italy) to 42 percent (for Saudi Arabia), with an
unweighted mean of 18 percent.
Second, consumers' purchases of prescription drugs, like their purchases of health
care services, are often reimbursed in whole or in part by insurance. As other contri-
butions in this volume analyze more fully, with insurance a wedge intrudes between
the demand curve derived from consumers' income and the full price paid and the
(higher) demand curve reflecting quantities consumed at prices net of insurance pay-
ments. This divergence of demand functions leads to reduced demand elasticity and
increased purchase of the insured item. See, e.g., Leibowitz et al. (1985) and Newhouse
(1993, pp. 165-171). Western European nations have tended to provide more generous
insurance coverage of pharmaceutical product purchases than the United States. How-
ever, convergence has been occurring. In the U.S., the share of outpatient drug costs
paid for out-of-pocket by consumers fell from 82.4 percent in 1970 to 33.9 percent in
1995, largely as a result of expanded private health insurance coverage. See PhRMA
(1998, p. 51) and U.S. Office of Technology Assessment (1993, pp. 238-263). Mean-
while consumers' payment shares have been rising in Europe as government health
authorities struggled to combat ever-rising costs by requiring substantially increased
co-payments. Gone, it is fair to say, are the days the author experienced in Germany
during the early 1970s when most consumers obtained prescriptions even for common
aspirin so that their health insurance would bear the cost.
Even if their outlays are not reimbursed by insurance, consumers, and especially
affluent consumers in industrialized nations, are willing to pay a considerable price to
combat a painful and debilitating infection, reduce the risk of a heart attack, alleviate the
pain of arthritis and muscle tears, or soothe the tensions of a neurotic world. What this
means, and especially when decision-making by physicians (which constricts substitu-
tion opportunities) and reimbursement by insurers are factored in, is that the demand for
many drug products is fairly inelastic up to rather high price levels before income effects
begin imparting appreciable elasticity. For instance, Berndt et al. (1995) estimated the
demand elasticity for H2 -antagonist anti-ulcer drugs as a group at -0.69 and for indi-
vidual molecular variants in the range of -0.74 to -1.03. See also (on cephalosporins)
Ellison et al. (1997).
These demand-side characteristics interact with the presence of monopoly power on
the supply side to support prices that commonly exceed drug production costs by a sub-
stantial margin. The newest ethical drugs are covered in many nations by patents, which
may limit the supply alternatives for a given chemical entity to a single firm. Even
when multiple alternatives exist, consumers' limited knowledge of efficacious substi-
tutes and insulation of decision-making physicians from the need to pay for the drugs
they prescribe can lead to the persistence of high prices for well-established pharmaceu-
tical products. Where MC is marginal cost, P the product price, and ep is the absolute
value of the price elasticity of demand, a monopolist will maximize its profits by set-
ting a price such that the price-cost margin (P - MC)/P = 1/ep. Thus, prices will be
raised until demand becomes price-elastic. This condition is inconsistent with evidence
of price elasticities below unity, but can be readily reconciled with low observed elas-
ticities if the producers of substitute drugs engage in price-eroding oligopolistic rivalry,
or if they exercise restraint to deter rival entry or avoid possibly adverse public rela-
tions repercussions from high prices. See Scherer and Ross (1990, pp. 227-229 and
Chapter 10).
1302 EM. Scherer
Using Census data, one can approximate the price-cost margins industries realize by
computing
sales - materials purchase costs - in-plant payroll costs

sales
See Scherer and Ross (1990, pp. 418-419). Such estimates are probably downward-
biased, since some payroll costs may be fixed and hence not part of true marginal cost.
Among 459 four-digit manufacturing industries covered by the U.S. Census in 1987,
pharmaceuticals (SIC 2834) had the sixth-highest price/cost margin at 61.4 percent; the
average for all manufacturing industries was 30.5 percent. 2 Public perception that drug
prices are set monopolistically has triggered numerous governmental interventions into
price-setting and other aspects of pharmaceutical marketing. See Comanor (1986) and
U.S. Office of Technology Assessment (1993). This is a subject to which we return later.
Patents are significant in part because the pharmaceutical industry is extraordinar-
ily dynamic technologically. Among the roughly 230 narrowly-defined U.S. industries
for which comparable data have been collected, pharmaceuticals has for at least two
decades devoted the largest fraction of its sales receipts to research and development
[U.S. Federal Trade Commission (1985, p. 31)]. The intensity with which it pursues
research has been rising over time. In 1975, companies with membership in the Phar-
maceutical Research and Manufacturers of America association had a weighted average
R&D/sales ratio of 11.6 percent in their ethical drug divisions; by 1995, the comparable
average was 19.4 percent [PhRlMIA (1997, p. 12)].3
From the industry's R&D efforts has emerged a stream of new pharmaceutical prod-
ucts, some of which have made significant contributions to the quality of health care.
Many afflictions that once were untreatable or even fatal can now be combatted with
routine success through drug therapy. From an analysis of matched U.S. prescription and
hospitalization records for the years 1980 and 1992, Frank Lichtenberg (1996) found
that a $1 increase on the purchase of pharmaceuticals was associated on average with
a $3.65 reduction in hospitalization expenditures. Innovative drugs had a particularly
marked impact in reducing hospitalization costs. A subsequent statistical investigation
by Lichtenberg (1998) revealed significant life-extending effects from the introduction
of important new drugs.
Product differentiation, which affects the drug makers' discretion in setting prices,
has been enhanced by vigorous marketing efforts. Under the so-called Dorfman-Steiner
2 Industries with higher PCMs than pharmaceuticals were cigarettes (67.6 percent), breakfast cereals, soft
drinks, cosmetics and toilet preparations, and chewing and smoking tobacco.
3 Company divisions selling over-the-counter drugs are much less R&D-intensive. In 1977, the latest year
for which comprehensive disaggregated data are available, among companies providing breakdowns for the
Federal Trade Commission's Line of Business reporting program (1985, p. 31), the average R&D/sales ratio
for prescription drug operations was 10.2 percent, but only 2.9 percent for over-the-counter drug operations. In
the surgical and medical instruments category, 1977 R&D outlays were 3.8 percent of sales. The median ratio
for 230 reporting lines of business spanning virtually all of the U.S. manufacturing sector was 1.0 percent.
theorem [Dorfman and Steiner (1954)], a profit-maximizing firm will seek a ratio
of advertising (or other promotional activity) to sales that satisfies the relationship
(A/S) = ea[(P - MC)/P], where A is the amount of advertising, S is sales volume,
(P - MC)/P is the price-cost margin explained earlier, and ea is the advertising elas-
ticity of demand. 4 Because price-cost margins are relatively high in pharmaceuticals
and because physicians and consumers are responsive to promotional activity, 5 appre-
ciable resources are devoted to it. Nevertheless, one finds here striking differences be-
tween prescription and over-the-counter drug sales and also among nations. Tradition-
ally, prescription drugs were sold in the United States primarily through "detailing"
- that is, through one-on-one encounters between drug makers' field sales representa-
tives and individual physicians. Advertising was confined largely to medical profession
journals. However, during the 1990s, and especially after regulatory changes in 1997,
companies attempted to cultivate end-consumer demand for their prescription products
through advertising in general readership publications. Such direct-to-consumer adver-
tising, which in the United States reached $1 billion during 1998, is prohibited by law in
most European nations.6 For over-the-counter drugs, the emphasis both in Europe and
the United States is strongly on advertising, and especially (in recent decades) on televi-
sion advertising. In 1977, OTC sellers in the United States devoted 20.2 percent of their
sales receipts to media advertising - the highest such ratio recorded in 225 reporting
industries [U.S. Federal Trade Commission (1985, p. 31)]. The corresponding media
advertising figure for prescription drug vendors was 4.0 percent (28th in rank among
225 industries). Combining outlays on media advertising and other forms of sales pro-
motion (including detailing, distribution of free samples, and in-store displays), total
selling costs in 1977 were 35.6 percent of sales for over-the-counter drug providers and
19.4 percent of sales in prescription drug operations. Total prescription drug advertising
and promotion outlays in the U.S. market during 1997 were estimated to be $12 billion,
or 18 percent of ethical drug sales [Zuger (1997)].
3. Pharmaceutical industry structure
The term "pharmaceutical industry" as used here will refer to the collection of com-
panies that manufacture ethical and over-the-counter drugs. Still one must recognize
that other actors are involved in significant ways - notably, the retail pharmacies from
which consumers purchase drugs taken outside health care institutions, the health care
4 That is, the percentage by which the quantity demanded shifts for a given percentage change in advertising
(or other promotional activity).
5 For estimates of elasticities for "detailing" and two kinds of advertising, see Berndt et al. (1995, p. 103).
Price elasticities may in turn be reduced by advertising. See Rizzo (1997).
6 See "Pill pushers", The Economist, August 9, 1997, pp. 58-59, and "Go on, it's good for you", The
Economist, August 8, 1998, pp. 51-52.
1304 EM. Scherer
institutions administering in-patient medications, and the wholesalers linking manufac-

turers with "downstream" outlets. In 1995, members of the Pharmaceutical Research
and Manufacturers of America made 78.9 percent of their ethical drug sales in the first
instance to wholesalers, 12.1 percent directly to retailers, 4.8 percent to hospitals (who
dispensed 14.3 percent of total prescription drugs by dollar value, securing the balance
of their supplies from wholesalers), and 4.2 percent to other institutions [PhRMA (1997,
pp. 30-31)]. Traditionally, retailers in the United States realized a 40 percent gross mar-
gin on the prescription drugs they sold to consumers, which meant that of the average
dollar spent by a consumer in 1966, 40 cents went to the retail pharmacist to cover op-
erating costs and profit, nine cents to wholesalers, and 51 cents to manufacturers. Task
Force on Prescription Drugs (1968, pp. 54-55). To defend their margins, retail pharma-
cists were among the strongest supporters of resale price maintenance - i.e., the fixing
by manufacturers of minimum prices below which their retailers could not sell. 7 See
Scherer (1997). Partly because the support in U.S. law for resale price maintenance was
eliminated in 1975, but even more because of events analyzed in a later section, these
relationships changed dramatically during the 1980s and 1990s, squeezing the relative
gross margins of retailers and wholesalers. Over the same period, chain outlets displaced
approximately 15,000 independent retail pharmacies.
The manufacturers of drugs, both ethical and OTC, are a diverse lot. The U.S. Census
Bureau (1992) reported 640 companies to be operating in the "pharmaceutical prepa-
rations" industry (SIC 2834) during 1987. The leading eight companies accounted for
36 percent of total industry sales; the leading 20 companies for 65 percent. Since the
early 1980s the U.S. industry has been augmented by hundreds of companies formed to
pursue opportunities opened up by the advance of biological science. By the mid-1990s,
only a dozen or so of these drug-oriented biotechnology companies had progressed to
the point of marketing profitable products.
Most of the U.S. pharmaceutical manufacturing leaders are multinational enterprises
(MNEs), operating not only in the United States but in many other nations. Incentives
to take advantage of unique and exclusive product franchises by making sales through-
out the world are so strong that pharmaceuticals is one of the most multinational of
industries. Markets outside a pharmaceutical maker's home territory are penetrated by
exporting, through the establishment of production facilities overseas, or through li-
censing of products to companies with already-existing sales channels and regulatory
relationships. According to United Nations estimates, pharmaceutical imports averaged
8.2 percent of domestic consumption during 1989 in developed nations and 19.8 per-
cent in less-developed nations [Ballance et al. (1992, p. 52)]. In 1980, approximately
27 percent of the world's pharmaceutical demand was satisfied through local production
by foreign-owned companies [Ballance et al. (1992, pp. 68-78)]. The extent of MNE
7 In some continental European nations, retail pharmacies are effectively cartelized, among other things
enjoying governmental protection against the formation of retail drug chains or the entry of nearby competing
pharmacies unless specified population density thresholds are exceeded.
operation is believed to have increased since then. In 1995, members of the Pharma-
ceutical Research and Manufacturers of America recorded ethical drug sales of $65 bil-
lion within the United States and $37 billion outside the United States [PhRMA (1998,
p. 100)].
The multinationality of leading pharmaceutical manufacturers increased during the
1980s and 1990s in part through mergers. Among the mergers that joined sizeable com-
panies with disparate national home bases were the following (with the acquiring firm
listed first):
Beecham (U.K.) - SmithKline Beckman (U.S.) 1989
Rhone-Poulenc (France) - Rorer (U.S.) 1990
Hoffmann LaRoche (Switzerland) - tGenentech (U.S.) 1990
Hoechst (Germany) - Copley (U.S.) 1993
Hoffmann LaRoche (Switzerland) - Syntex (U.S.) 1994
Bayer (Germany) - Sterling Drug (part) (U.S.) 1994
Hoechst-Roussel (Germany) - Marion Merrell Dow (U.S.) 1995
Pharmacia (Sweden) - Upjohn (U.S.) 1995
Hoffmann LaRoche (Switzerland) - Boehringer Mannheim (Germany) 1997
Other mergers linked companies of considerable scale with home bases in the same
nation. These included the fusion of two leading Swiss firms, Ciba-Geigy and Sandoz,
to form Novartis in 1996; the merger of two leading British companies, Glaxo and Bur-
roughs Wellcome, in 1996; and, in the United States, the 1989 merger of Bristol-Myers
with Squibb, and the 1994 acquisition of American Cyanamid by American Home Prod-
ucts. Altogether, pharmaceutical company acquisitions valued at well over $200 billion
occurred between 1989 and 1998. In wholesaling too, a merger wave shrank the number
of companies until the leading four firms accounted for 80 percent of U.S. wholesaling
activity. Attempts by the four leading wholesalers to merge down to two entities were
blocked in 1998 by antitrust action. 8 A significant motive for the merger wave among
drug manufacturers was the desire of companies to pool their research and development
project portfolios, yielding cross-fertilization of specialized know-how and hoped-for
risk hedging. On the former, see Henderson and Cockburn (1996).
4. Research, development, and the discovery of new drugs9

Research and the development of new and improved products play a central role in the
activities of modem pharmaceutical enterprises. This was not always so. There have
been several revolutions in the way new drug products are created. See, e.g., Schwartz-
man (1976, Chapter 2), Cooper (1970), and Temin (1979b).
As late as the 1930s, the use of scientific methods to develop new medications was
rare. Millennia of experience had identified naturally occurring substances with ther-
apeutic properties. Co-existing with genuinely efficacious medicines were numerous
8 "Judge rejects two separate drug mergers", New York Times, August 1,1998, p. B1.
9 Substantial parts of this section are adapted from Scherer (1996, pp. 343-346).
1306 FM.Scherer
quack medicines. Retail pharmacies' shelves were lined with bottles containing organic
and inorganic chemicals compounded on the spot to satisfy physicians' prescriptions or
patients' pleas for the druggist's own preferred recipe. Until 1938, prescriptions were
required in the United States only to obtain narcotic substances.
The synthesis of aspirin marked one of the first breakthroughs in the use of modem
chemistry for medicinal purposes. See Mann and Plummer (1991). A substance from the
bark of the white willow tree had long been used to relieve fever and pain. By the 1830s
German and French chemists had extracted its active ingredient and derived from it
salicylic acid. But salicylic acid caused ulcers and other gastric problems. Seeking new
markets for by-products of the synthetic organic dyes upon which its growth was based,
the Bayer company of Germany established in 1896 a laboratory devoted to creating
and testing dyestuff derivatives for medicinal effects. Early work led to the synthesis of
acetylsalicylic acid, which it named "Aspirin" (a trademark that even a century later can
be used in Germany only by Bayer) and sold with great success.
In the early years of the 20th Century a German professor, Paul Ehrlich, formulated a
conception of how small organic molecules interacted with proteins in the human body
as keys do with locks. Ehrlich found many new chemicals with useful therapeutic ef-
fects, including Salvarsan, the first drug effective against previously incurable syphilis.
Later research at the laboratory of Bayer's merged successor, I.G. Farben, led in 1935
to the discovery of a red dye derivative, sulfanilamide, that combatted lethal streptococ-
cal infections. Numerous sulfa drug variants were subsequently synthesized and tested,
leading to safer anti-infectives and the discovery of drugs with diuretic (blood pressure-
reducing) properties.
The antibacterial properties of a naturally occurring mold, penicillium notatum, were
first observed accidentally in 1929 by Alexander Fleming in England. Fleming failed to
follow through, but penicillin's therapeutic properties were identified by Howard Florey
and Emest Chain at Oxford University in time for that first antibiotic to play a life-
saving role in treating World War II casualties. Mass production methods using corn
steep liquor as a fennentation medium were devised at a U.S. Department of Agricul-
ture laboratory. Twenty U.S. companies participated in the top-priority wartime peni-
cillin production program. When the war ended, intense price competition among those
companies drove the wholesale price of penicillin in 300,000-unit doses from $3 in 1945
to $1 in 1948 and 10 cents in 1953. See U.S. Federal Trade Commission (1958).
The therapeutic success of penicillin suggested to Selman Waksman of Rutgers Uni-
versity that other naturally occurring spores might have antibiotic effects. By screening
and testing numerous soil samples during the early 1940s, he made two important dis-
coveries: a specific new antibiotic, streptomycin, and even more important, a systematic
method for finding new medicinal substances. Waksman insisted that the patent he ob-
tained on his purified form of streptomycin be licensed to all interested parties, and as
a result, the penicillin price competition experience was repeated, leaving the average
producer with negative profits.
Nevertheless, traditional pharmaceutical companies found in Waksman's methodol-
ogy a potent means of discovering new therapeutic entities on which they might obtain
exclusive patent rights. Several new broad-spectrum antibiotics - Aureomycin (chlor-

tetracycline), Chloromycetin (chloramphenicol), oxytetracycline, and tetracycline - fol-
lowed between 1948 and 1953. Each was patented, each was sold at prices high in rela-
tion to its production costs, and each proved to be very profitable for its suppliers. The
lure of high profits in turn induced many other firms to try their hand at the game. U.S.
pharmaceutical company R&D outlays (at constant 1990 price levels) rose from roughly
$263 million in 1951 to $1.7 billion in 1967, $3.1 billion in 1980, and $8 billion in 1990.
With few exceptions, the industry's approach to research and development during
this period entailed testing for therapeutic effects large numbers of newly synthesized
molecules (following the Ehrlich model) or naturally occurring substances (the Waks-
man model). In 1970 alone, U.S. pharmaceutical manufacturers are said to have con-
ducted more than 700,000 laboratory culture and animal screening tests, out of which
only a thousand chemical entities emerged with sufficiently interesting results to be car-
ried into higher animal and human tests [Schwartzman (1976, p. 60)]. To the extent that
this process was guided by systematic theory, it stemmed from prior discoveries that
certain molecules had recognized therapeutic effects, so chemists synthesized "me too"
variants of those molecules to see whether they might work better or at least circumvent
rival firms' patents.
As scientific knowledge has advanced, the industry has moved gradually from more
or less random screening to methods called "rational drug design". Extending the early
insights of Paul Ehrlich, it is known that each of the countless proteins in the human
body has specific functions, and that the functioning or malfunctioning of those pro-
teins is sensitive to the addition of chemicals at key receptor surfaces. The properties
of receptors can be gleaned through various analytic techniques, and molecules can be
designed to fit the receptors. Breakthroughs in genetic research and techniques for cre-
ating and multiplying new life forms have made it possible to manufacture proteins and
other substances that replace defective proteins in the body or modify their functioning.
Although much more highly focused than the "try every bottle on the shelf" methods
of the 1950s and 1960s, these modern techniques are more like shotgun blasts than rifle
shots. Considerable trial and error, including animal and then human tests, are required
to isolate among many possibilities new molecules or proteins that actually work with-
out serious adverse side effects. Failure is still more frequent than success. I
The science base upon which pharmaceutical manufacturers build in discovering new
drugs owes its growth in part to private research investments. Among the 24 U.S. indus-
try groups on which detailed survey statistics are published, pharmaceuticals devoted
the highest fraction - 16.6 percent - of its total 1992 R&D outlays to basic research
[U.S. National Science Foundation (1996, p. 44)]. For all other industries, the compa-
rable average was 5.3 percent. Even more important, however, is the role of govern-
ment, e.g., in the United States, the National Institutes of Health (NIH), which perform
10 On the extent to which scientific research can reduce drug development uncertainties, compare Gam-
bardella (1995) with Schwartzman (1997).
1308 F.M. Scherer
basic research and some applied work (including testing cancer drugs) intra-murally
and award grants supporting much of the relevant research at universities and medical
schools. In 1991, NIH research spending and grants totalled $7.7 billion, of which as
much as $4.8 billion was in fields germane to pharmaceuticals [Scherer (1995, p. 27)].
This work was supplemented by $260 million of National Science Foundation grants in
the biological sciences along with appreciable additional sums from other federal agen-
cies. Knowledge flows from basic research institutions to companies not only through
the medium of publications, but also through direct collaboration. Rich scientific paper
co-authorship interactions between drug company research staff on one hand and aca-
demic and governmental basic researchers on the other have been traced by Cockburn
and Henderson ( 19 9 8 ).l l
Edwin Mansfield (1995) questioned a panel of U.S. industrial R&D decision-makers
concerning the relevance of academic research to the development of specific new prod-
ucts. For the pharmaceutical industry, he learned that 27 percent of the new products
sampled could not have been developed absent underlying academic research, and an
additional 29 percent of the products were "significantly facilitated" by academic re-
search. Academic research was found by Mansfield to be more important to the emer-
gence of new products in pharmaceuticals than in the computer, instruments, electrical
equipment, and metals industries. An econometric study by Toole (1997) suggests that
17 years elapsed on average between basic biomedical research and the introduction of
commercial products building upon that research. See also (on oral contraceptives) IIT
Research Institute (1968, pp. 58-73).
Government agencies also provide important support for pharmaceutical research and
development through educational grants for prospective scientists. In 1989, the U.S.
National Institutes of Health had training grants of roughly $327 million outstanding,
supporting some 7,800 pre-doctoral candidates and 6,600 post-doctoral researchers in
the life sciences [U.S. Office of Technology Assessment (1993, p. 205)]. Many of the
researchers trained in this way subsequently accept positions with pharmaceutical and
biotechnology firms, carrying both the skills they have honed and the knowledge of
technological opportunities they have acquired into commercial research and develop-
ment activities.
5. Government regulation of new drug introductions
Most of the industrialized nations have created regulatory institutions to ensure that de-
sired safety and efficacy standards are met by new and existing drugs (and also by cer-
tain medical devices). The severity of regulatory control varies widely among national
11 Analyses of citations in published scientific articles to other work reveal that knowledge flows are interna-
tional, but with appreciable own-nation biases. See U.S. National Science Board (1998, p. A-327).
jurisdictions, with the United States leading chronologically in the implementation of

tough standards. We focus therefore initially on the U.S. experience. 12
Systematic regulation began in the United States with the Pure Food and Drug Act
of 1906, which prohibited adulteration, imposed labelling mandates, and required that
drugs sold in interstate commerce conform to U.S. Pharmacopeia standards. After some
hundred deaths occurred when a drug maker dissolved sulfanilamide in toxic diethylene
glycol, the 1938 Food, Drug, and Cosmetic Act was passed, giving the Food and Drug
Administration (the FDA, established in 1930) power to bar the interstate sale of new
drugs unless they were found to be safe. However, if the FDA did not act upon applica-
tions for new drug approvals (NDAs) within 180 days, the drug's maker was authorized
to begin commercial sales without regulatory permission. Despite the FDA's weak posi-
tion, an FDA staff member successfully delayed in 1961 the transition of a new tranquil-
izer, thalidomide, from clinical tests into full-scale commercial marketing. It became
clear shortly thereafter that sales of the drug in Europe, where thalidomide was ap-
proved and used widely to combat morning sickness, had led to the birth of some 8,000
malformed babies. Thalidomide became a cause celebre, leading the U.S. Congress to
reorient an ongoing investigation of drug prices and profits toward legislation strength-
ening regulatory controls. The Kefauver-Harris Act of 1962 eliminated the 180-day
loophole, required the FDA to certify that new drugs were not only safe, but also effica-
cious, and gave the FDA substantial new powers over drug testing and marketing.
Under regulations subsequently articulated by the FDA, an organization seeking to
test a new chemical entity in human beings must first obtain from the FDA an "Inves-
tigation of New Drug" (IND) authorization. IND applicants must submit evidence of
non-toxicity at anticipated dosage levels from tests in animals such as dogs or mon-
keys and describe in detail their contemplated experimental design, including controls
that will ensure statistical validity. The FDA has 30 days from the IND application's
submission to raise objections. When clinical testing begins, it is typically arrayed over
the following three-phase sequence, with median time spans and attrition rates (condi-
tional upon passing through the prior phase) estimated from 93 clinical trials conducted
between 1970 and 1982:13
Months Attrition
(median) rate
Phase I: Administration to a small number of healthy volunteers to test for absorp- 15.5 25%
tion, metabolism, and (at varying dosages) toxicity.
Phase n: Administration to a few and then dozens of patients with the disease to be 24.3 52%
treated.
Phase III: The drug is administered in double-blind tests to at least two large sam- 36.0 36%
ples of patients with the disease. Long-term toxicity tests are conducted in parallel.
12 The next several paragraphs are drawn in part from Scherer (1996, pp. 346-350).
13 The source is DiMasi et al. (1991). The times and attrition rates ignore possible overlaps between stages.
1310 EM. Scherer
If the drug clears all of these hurdles successfully, its sponsor applies to the FDA for
a new drug approval (NDA). The FDA often requests additional tests and information
before making its decision. On average, for the 93 clinical trials summarized above, 23
percent of the chemical entities that entered Phase I testing emerged with FDA approval.
The median lag for that sample from NDA application (following Phase III) to FDA
approval or disapproval was 30 months. Adjusting for overlapping phases, the median
time span from the start of clinical testing to NDA issuance was 98.9 months, or slightly
more than eight years. The total amount of time required for successful drugs to pass
through these testing and regulatory hurdles has tended to increase in successive decades
since the Kefauver-Harris Act was passed - e.g., from 4.7 years on average in the 1960s,
6.7 years in the 1970s, 8.5 years in the 1980s, and 9.1 years in the mid-1990s [DiMasi
et al. (1994) and Kaitin (1997)].
The increase in clinical testing mandates and the time needed to satisfy them led to
striking increases in the average cost of clinical trials. A study by Mansfield (1970) of
17 new drug development projects before the 1962 law took effect found an average suc-
cess rate of 37 percent. The average development and testing cost per successful NDA,
including the pro-rated costs of failed chemical entities, was $1.05 million (in current
dollars, equivalent to roughly $4 million in 1987 dollars). For the 93 projects studied by
DiMasi et al. (1991) whose histories are summarized above, the average clinical devel-
opment cost per NDA had escalated to $48 million (in 1987 dollars).' 4 Correcting for
general price inflation, the increase in testing costs between the late 1950s and the late
1970s was roughly ten-fold. Grabowski et al. (1978) took advantage of a difference be-
tween U.S. and British testing regulations to illuminate the reasons for increasing costs.
U.K. regulations were strengthened in 1971 from reviewing safety only, as had been
FDA practice up to 1962, to requiring proof of efficacy also. Between 1960-1961 and
1966-1970, average inflation-adjusted drug development costs in Great Britain rose by
a factor of three, while those in the United States increased sixfold. This suggested that
more stringent regulation in the United States was responsible for a twofold increase,
while influences common to both regulatory jurisdictions explained the remaining three-
fold increase. Among those other influences were the recognition by drug companies
that extensive testing was required to avoid repeating the thalidomide disaster, which
imposed huge tort liability losses on European firms; the need to accumulate evidence
used in persuading physicians to favor new drugs over tried-and-true older formulations;
and perhaps also increasing difficulty in finding superior new drugs following the first
wave of successes experienced during the 1940s and 1950s.
14 This estimate is for out-of-pocket clinical testing period costs only. Costs incurred in the pre-clinical dis-
covery phases, when pro-rated over successful NDAs, approximately double the total R&D cost of a marketed
NDA. It has become customary in the literature to take into account the opportunity cost of invested funds by
capitalizing phased R&D costs to the time of market approval, which raises even more the relative weight of
pre-clinical work, farthest in time from the date of a new product's introduction. Capitalized R&D costs as
high as $200 million per successful NCE are commonly reported.
/U i I I I i
0
QA)
cL
60
..... ... .......2....2....!... ......... :
w 50 I--- I .
.......+ All NCEs .
E .... Important
....... _ ......... .............
..........
m 40
0
(a Ju
o
C 20
E
Z 10n
. . .. *. C. ·
n I I -l ,
1940 1950 1960 1970 1980 1990

Year
Figure 1.Trends in U.S. new drug approvals, 1940-1990.
Despite rapidly rising aggregate drug R&D expenditures, the number of new drugs
approved for marketing in the United States declined markedly. This is shown by Fig-
ure 1.15 The solid line, counting the number of new chemical entity approvals, reveals
NDAs rising erratically and then peaking at 65 in 1959, only to plummet and hover
around an average of 18 per year following enactment of the Kefauver-Harris law in
1962.
The Food and Drug Administration was criticized for the drop in new drug appear-
ances and later for "drug lag" - the slow approval of even those drugs that ultimately
did reach the market, causing the United States to fall behind other nations in making
the most up-to-date therapies available to its consumers. See, e.g., Peltzman (1973),
Wardell and Lasagna (1975), and U.S. House of Representatives (1980). Peltzman laid
out inter alia a theoretical schema for FDA regulation, whose principal benefit, he ar-
gued, is to eliminate what might otherwise be information failures in the marketplace. It
is illustrated in Figure 2. Suppose the demand curve reflecting some combination of pa-
tient and prescribing physician preferences without testing information required by the
FDA is BDUNc. With no FDA intervention, consumers will consume OXu units of the
drug per year at a price (assumed fixed over all cases) of OP. If, however, the true full-
information demand curve DT1 lies much more to the left because FDA-mandated test-
ing reveals adverse side effects, consumers with full information would consume only
15 It is drawn from Scherer (1996, p. 351), tapping a variety of sources.

1312 EM. Scherer
0)
au
0
0
CL
IlUlIIULI Ul rIl : .sll4LlUl4 l
Figure 2. Welfare consequences of uncertainly reduction through clinical testing.
OXI units. Over-consumption due to imperfect information causes a consumers' sur-

plus loss measured by near triangle GFDT1, offset to a modest extent by the sacrifice of
consumers' surplus EGP during the period when the drug is not available to consumers
because of FDA-mandated testing. If on the other hand FDA-mandated tests show the
drug to be better than poorly-informed decision-makers' beliefs, the full-information
demand curve will shift to ADT2 and consumption will increase to OX 2, adding con-
sumers' surplus triangle KCF. However, if the tests are so costly that companies choose
not to develop the new drug, consumers will lose trapezoidal surplus AKFP as a result
of FDA intervention. From a statistical analysis of the decline in drug approvals, and
making the key assumption that the decrease evident as early as 1960 would in time
have been reversed, as earlier declines were, Peltzman concludes that the 1962 drug law
caused a substantial loss of consumer benefits.
In response to its critics, the FDA insisted that at least part of the sharp drop in NDAs
was intentional. What it had done by requiring more rigorous and costly testing, its of-
ficials said, was mainly to discourage "me too" variants adding little or nothing to the
therapeutic benefits of already existing drugs. It supported this argument by pointing
to its internal ratings of newly approved drugs according to their therapeutic novelty.
The dotted line in Figure 1 tracks the appearance of drugs considered by the FDA to
offer important therapeutic gains. In the 1950s, there was a surge of important drug
discoveries as the drug-finding revolution took hold, but then, long before 1962, the ap-
pearance of important new chemical entities stabilized in the range of zero to six NDAs
Ch. 25: The PhannaceuticalIndustry 1313
4 n. .... .. ... . ... . . .........................

I UV - - ... - - - --------I --
---, - ---. ........ · .I ...... ....... ... -- ... '' - - - I
v#
a0
90
cn
- -. -. 1111... I - - - - ...
Ca 80
0
o- 70
o 0
.... ........... -
2) 60
a0 50
L
Ca 40
13,
30
(n
........- .....
x~3 3
uJ,
a 10n~2
USA Switzerland France Sweden Japan

Germany UK Italy Netherlands
Nation
Figure 3. Percent of multinational drug company sales outside home market. Source: Thomas (1996).
per year, with a mean of 3.6. What had been weeded out, the FDA continued, were
mainly the drugs yielding little or no therapeutic gain. Critics questioned the accuracy
of the FDA's drug importance evaluations. Industry spokesmen acknowledged that they
had refocused their R&D efforts away from "me-too" drugs and toward the goal, not
always achieved, of developing only therapeutically important molecules.
A retrospective analysis by L.G. Thomas (1996) injects surprising new insights into
this debate. For a while, the United States had the most stringent drug testing require-
ments in the world. The United Kingdom and Germany were the next major nations to
follow the U.S. lead. By setting high standards for new drugs, Thomas argues, the reg-
ulatory agencies in those nations forced their domestic drug firms to target their R&D
on drugs of superior therapeutic efficacy, and when success was attained, the resulting
drugs sold well not only at home but also in foreign markets. The data Thomas com-
piled in support of his argument are summarized in Figure 3, which shows the fraction of
the 1985 sales of companies at home in nine leading drug-developing nations realized
outside the firms' home markets. The smaller a nation's home market, the larger was
the fraction of sales realized away from home. This plus the fact that all three nations
have excellent university systems explains the relatively high external sales ratios for
Switzerland, Sweden, and the Netherlands. Taking into account this population effect,
the stronger the home regulatory regime, the larger companies' sales were outside their
home markets. France, Italy, and Japan, with populations as large as those of (West) Ger-
many and the United Kingdom, had unusually weak domestic testing standards, which
1314 FEM. Scherer
are reflected in poor external market performance. The United States, Germany, and the
United Kingdom, with tough standards, all do well. 1 6
Despite this inadvertent mercantilist policy success, the U.S. Congress continued to
be alarmed over "drug lag" claims and especially the long decision-making lags ex-
perienced when applications for a New Drug Approval reached the FDA. A new law
passed in 1992 and amended in 1997 permitted the FDA to collect fees on new drug
approval filings and also annually from drug-producing plants and from drugs approved
after being in the FDA's review "pipeline" by September 1992 or later. See Shulman
and Kaitin (1996). The proceeds of these fees, totalling roughly $325 million between
1993 and 1997, were used to augment the FDA's technical staff, as a consequence of
which the FDA agreed to a time-phased program of reaching an increasing fraction of
its decisions within one year on conventional NDA applications and even more quickly
for high-priority drugs. An appreciable reduction in decision-making intervals in fact
resulted, from an average of 31 months over 1987-1991 to 16 months in 1997 [PhRMA
(1998, p. 24)]. It and several highly publicized withdrawals of approved products from
the market precipitated criticism that FDA approval standards may have been reduced
in the bargain. The acceleration of decision-making and clearing out of backlogs also
led to 53 new drug approvals in 1996, compared to an average of 23 in 1987-1991.
Regulatory reforms also occurred in other industrialized nations. The most impor-
tant change was the creation, beginning in January 1995, of the European Medicinal
Products Evaluation Agency (EMEA), which provided a uniform forum for regulating
new drug introductions in all member nations of the European common market. See
Heppell (1996). Initially, only medicines developed using new biotechnology methods
were placed within the sole jurisdiction of EMEA. For other products, manufacturers
retained the option of seeking approval either before EMEA or with national regula-
tory authorities, whose decisions would be mutually recognized in all European Union
nations. It is unclear whether this procedure led to convergence of national authorities'
standards - e.g., whether greater stringency was introduced into what critics called the
"French impressionist school of safety regulation" [Thomas (1996, p. 116)]. Beginning
in January 1998, the mutual recognition approach was to be phased out and manufac-
turers seeking to market new products in more than one European nation were required
to have their test results reviewed by EMEA. There has been too little experience with
EMEA's decisions to ascertain whether they differ substantively from those of the more
stringent national jurisdictions. 17
Differences between regulatory regimes point to several more fundamental philo-
sophical questions. Granting or denying permission for full-scale marketing of a new
drug is an exercise in decision-making under uncertainty. Clinical test insights can
16 See also PhRMA (1998, p. 71).

17 In a 1995 seminar at Harvard University, an EMEA official said that his agency was likely to slant its
tradeoffs somewhat differently from those of the U.S. FDA, placing relatively more emphasis on speed of
new product availability and relatively less on conclusive proof of efficacy and safety.
be sharpened through the use of appropriate statistical methodologies, but they can-
not eliminate uncertainty, especially for adverse side effects of very low incidence. 18
Decision-makers must weigh the risk of Type I errors - approving a drug when it is
not truly safe or effective - against the risk of Type II errors - withholding from the
market entities that are truly effective and safe. See, e.g., Scherer (1996, pp. 353-355).
In governmental agencies, there is a natural tendency toward placing more weight on
avoiding Type I errors, since officials who have approved a product that leads to cancer
or malformed babies will be singled out for castigation in public fora. The tradeoff can
be narrowed by increasing the size of clinical trial samples, but that solution increases
costs, possibly discouraging the development of some drugs, and it is likely also to delay
the availability of new drugs.
Several further questions follow. If uncertainty is high but the possibility of life-
saving benefits is also substantial, shouldn't a regulatory agency illuminate the prob-
lematic tradeoff and let individual physicians and/or patients make their own risk-taking
decisions, rather than being restrained by the choice of a bureaucracy? Stung by crit-
icism that its decision-making was denying potentially vital therapies to patients with
life-threatening diseases, the U.S. FDA began during the 1980s to make increasing use
of "compassionate NDA" procedures under which experimental drugs that have not yet
been approved formally are made available to physicians. For HIV drugs, it also waived
the requirement that double-blind tests be conducted, for to assign a patient randomly
to the placebo group could be tantamount to imposing a death sentence.
Recognizing such exceptions, one must ask the further question, why should a reg-
ulatory agency be the ultimate decision-maker on whether any new drug can be used?
To be sure, absent regulatory requirements, drug manufacturers might perform too little
clinical testing to ascertain whether a drug is superior to existing alternatives. Meager
testing was the norm in the pre-thalidomide era. An information market failure may
need correction. But why doesn't the regulator merely require appropriate testing and
disclosure of test data, letting physicians decide from the data whether the drug is safe
and efficacious? If there is an argument for regulation of whether new drugs may be
marketed, it must lie in a further information market failure - e.g., from the possibility
that most physicians are too busy to make well-informed independent decisions.
Carrying the debate one step farther, why should prescriptionsbe required to obtain
drugs? They are not required for over-the-counter drugs or, in many less-developed na-
tions, for any available drug. The prescription system implies that patients are unable
to make well-informed decisions about their own welfare, so physicians must act in
loco parentis. That may be true, but obtaining a prescription imposes costs, and Sam
Peltzman's (1987) statistical analysis suggests that there is no clear indication of higher
18 Withdrawal of approved drugs because of safety problems is fairly rare. Three to four percent of the drugs
approved for marketing between 1974 and 1993 in the United States, the United Kingdom, and Spain were
subsequently withdrawn for safety reasons. The number of withdrawals was 10 in the United States, 16 in
Spain, and 20 in the U.K. [Bakke et al. (1995)].
1316 EM. Scherer
poisoning or mortality rates in middle-income nations where prescriptions are not re-
quired.
Similar perplexities are encountered in the regulation of medical devices. We fo-
cus only on the U.S. experience. Reacting to numerous deaths and maladies attributed
to faulty artificial heart valves, cardiac pacemakers, and intrauterine devices, the U.S.
Congress in 1976 increased the FDA's regulatory authority over the marketing of medi-
cal devices and therapeutic apparatus. See U.S. Office of Technology Assessment (1984,
Chapter 5). All such devices were to be classified by the FDA into three categories,
depending upon safety and effectiveness risks. Devices already on the market were re-
viewed by FDA panels, and those assigned to the highest-risk Class III were required
to submit safety and efficacy data to the FDA, with forced cessation of marketing re-
sulting for devices found to be unsafe. Newly-developed Class III devices must run a
gamut of tests similar to those for new drug chemical entities before receiving mar-
keting approval from the FDA. Lower-risk class devices are subject to manufacturing
and performance standards published by the FDA. As in the history of drug regulation,
there were complaints that FDA test and approval procedures impose excessive costs,
discourage innovation, and delay the availability of potentially beneficial devices. In
1997 an unsuccessful attempt was made in the U.S. Congress to privatize significant
elements of the FDA's device certification functions.
6. Patents and pharmaceutical innovation
Owing to a combination of regulatory mandates and intrinsic difficulty whose exact

weights remain unknown, developing new drug products is a costly, high-stakes game.
It is also a risky game. The risks are shown in part through evidence assembled by
Grabowski and Vernon (1990, 1994), summarized here in Figure 4. They compiled data
on the domestic and foreign sales of 100 new drug chemical entities (NCEs) developed
by U.S. companies and introduced into the U.S. market (following FDA approval) dur-
ing the 1970s. Using standard ratios, they subtracted estimated production, distribution,
and marketing costs and plant investments from product sales, leaving estimates of the
"contribution margin" or Marshallian quasi-rents left to repay research and development
investments and provide a profit. The 100 products' discounted quasi-rents (assuming
1986 price levels) were arrayed into deciles in descending order of absolute magnitude.
What emerges is a highly skew distribution shown by Scherer (1998) to approximate log
normality and to resemble the payoff distributions in other high-technology fields. The
most lucrative ten NCEs contributed 55 percent of total quasi-rents. Products in seven
of the ten deciles did not even repay their capitalized research and development invest-
ments (including the pro-rated cost of projects yielding no marketable product), whose
average value was found to be $81 million per NCE.19 Products in the third decile came
19 A positive but statistically insignificant correlation was found between the magnitude of quasi-rents and
individual product R&D costs.
--
) )UU
n
400
0
z
0 300
0)
r 200
e)
- 100 Average R&D Cost

F- _ ................ I.................................................
o
0
UW
A1
Y v
1 2 3 4 5 6 7 8 9 10
Decile
Figure 4. Distribution of 1970s drug NCE quasi-rents. Source: Grabowski and Vernon (1990).
close to break-even, while those in the second decile provided returns roughly twice av-
erage R&D investments. Summed over 100 new products, the discounted present value
of quasi-rents exceeded the sum of R&D costs by only three percent. Thus, a relatively
few "winners", and especially the "blockbusters" in the first decile, had to repay the
losses on the majority of low-payoff products.
6.1. The unusual importance of drugpatents
Although altruistic motives undoubtedly enter, profit is the principal lure leading drug
makers to invest large sums toward the discovery and development of new drugs. Specif-
ically, company leaders hope to develop products sufficiently important, and sufficiently
well insulated from competition, to repay or more than repay their R&D investments.
Patent protection on marketed new chemical entities is a significant component of their
profit-earning expectations. Indeed, surveys asking research and development managers
what factors permit them to reap the profit benefits from their innovations - including
the competitive advantages of being first on the market with a new product, superior
sales and service efforts, secrecy and complexity of product and production process
technology, and patents - consistently show the pharmaceuticals industry to be one in
which the greatest stress is placed on patent protection [Levin et al. (1987); Cohen et al.
1318 EM. Scherer
(1997)].20 In most industries, imitators must duplicate much or all of the R&D to field a
product competitive with the original innovation, and this plus a head start and the abil-
ity to build brand loyalty in the interim are often sufficient to make R&D investments
profitable. 21 But in pharmaceuticals, much of the R&D investment is incurred to dis-
cover which molecules have medically interesting properties and to test their efficacy in
humans. Absent patent protection or some equivalent barrier,22 imitators could free-ride
on the information created by the innovator's hundred-million-dollar R&D and testing
investment, spend a few hundred thousand dollars on process engineering, and begin
competing with the innovator, eroding its quasi-rents. 2 3 At the same time, drug product
patents, unlike the patents in many other fields of technology, protect a clearly identi-
fied chemical molecule around which the marketing of substitute variants is impossible
without undergoing a complete new array of FDA clinical trials.
The development of significant new drugs is concentrated largely in the nine highly
industrialized nations covered by Figure 3 [Ballance et al. (1992, pp. 86-88)]. Many
other nations, and especially the less-developed countries, have traditionally denied
patent protection to medical and (frequently) food products. In this way, they are able
to avoid the monopoly prices and drain of scarce foreign exchange that would occur
if patent protection were granted and the patented drugs were imported (1992). Even
Switzerland, home to three of the world's leading pharmaceutical companies, provided
no drug product patent protection until 1977. A few nations allowing no drug product
patents - notably, during the 1960s and 1970s, Italy, and more recently, India - evolved
domestic pharmaceutical industries adept at "knocking off" new drugs still patented
in industrialized nations and selling them at bargain prices both at home and (through
exports) in other patent-denying nations. See, e.g., Lanjouw (1997).
6.2. Consequences of the Uruguay Round agreement
Even though the half of the world's population living in underdeveloped nations ac-
count for less than 20 percent of world pharmaceutical consumption, the patent policies
of LDCs were viewed as a thorn in the sides of multinational pharmaceutical produc-
ers. During the 1980s the United States government, urged on by a lobby organized
by pharmaceutical executives, began threatening such nations with international trade
20 The only industry in the second [Cohen et al. (1997)] survey placing more stress on patents than the
pharmaceutical industry was medical equipment.
21 On brand loyalty as a barrier to imitation of innovative drug products, see Bond and Lean (1977).
22 On new drug approval as a regulatory barrier to imitation see Kitch (1973). In the Orphan Drug Act of
1983, the U.S. Congress ordained a seven-year period of exclusivity following FDA approval of drugs treating
symptoms affecting fewer than 200,000 persons, whether or not patent protection co-exists. This orphan drug
privilege serves as a surrogate patent. Between 1983 and 1998, 140 drugs were approved by the FDA under
orphan drug rules.
23 There are exceptions. Some drugs are very difficult to produce. An example is the cephalosporin antibiotic
cefaclor, whose production entails ten complex steps and on which competitive imitation was retarded even
after a product patent and several key process patents expired during the early 1990s.
sanctions under Section 301 of the U.S. trade law unless they conformed to U.S. intellec-
tual property standards. See Santoro (1992) and Ryan (1998). Several nations changed
their laws to comply, and to pave the way for a bilateral trade treaty, Canada amended
its laws in 1987 to end compulsory licensing of drug product patents at quite mod-
est royalties, which had led to relatively low new drug product prices in Canada. See
McFetridge (1997). U.S. pharmaceutical industry leaders extended their lobbying cam-
paign to include compatriots in Europe and Japan and also attracted the collaboration of
motion picture, musical recording, and software interests. Their governments responded
by making harmonization of intellectual property standards a high-priority item on the
agenda of the Uruguay Round trade negotiations. The effort was successful. The re-
sulting Treaty of Marrakech, ratified in April 1994, requires inter alia that all signatory
nations provide full patent protection for pharmaceutical products - for industrialized
nations, by 1999, and for less-developed nations, by the year 2004.
These measures continued to be controversial even after the Marrakech Treaty was
signed. The parliaments of some LDCs balked at enacting the required law changes. For
their nations, the benefits from a modest increase in the number of new drugs likely to be
forthcoming as a result of broader international patent coverage were plausibly believed
to be outweighed by the loss of consumers' and producers' surpluses from having to
import at elevated prices the most up-to-date drugs from multinational companies. See
Deardorff (1992). The balance might change if granting drug product patents led to
the emergence of a local pharmaceutical industry skilled at discovering and developing
innovative new drugs for local production and export. However, the experience of Italy,
whose Supreme Court ordered in 1978 that the government begin issuing drug product
patents, undermining a foundation of Italy's strong "knock-off" industry, suggests that
making the transition from imitative to innovative industry is at least time-consuming
and perhaps even unlikely [Scherer and Weisburst (1995)]. Canada's experience after
it strengthened drug patent rights was more favorable because it was able to extract
from the multinational companies a pledge to move a substantial fraction of their R&D
activity to Canada and because Canada had a surplus of well-trained life scientists upon
which that shift could build. See McFetridge (1997).
7. Pricing
Although companies selling new drug chemical entities commonly enjoy patent pro-
tection for a number of years after their product is introduced, more often than not
they must face competition from chemically differentiated molecules that might be pre-
scribed to treat the same symptoms. The pharmacy benefit management company PAID
published a formulary for 1993 that can be subdivided, applying some judgment, into
141 specific symptom groups. 2 4 The number of drugs per symptom group ranged from
24 PAID Prescriptions Inc. (1993), The PAID National Formulary (Montrale, NJ). Some sub-categories were
combined where it was clear that the same symptoms were being treated in the same general mode.
1320 EM. Scherer
one to 50, with a median of five drugs and mean of 6.04. Thus, the typical market
structure for first-line drugs is differentiated oligopoly. How are prices set under those
circumstances?
7.1. Pricingbranded drugs
The most thorough relevant study is by Lu and Comanor (1998), who analyzed the pric-
ing of 148 new branded chemical entities introduced into the U.S. market between 1978
and 1987.25 All but 13 had at least one fairly close substitute in their principal therapeu-
tic indications. The average number of substitutes was 1.86. The authors distinguished
among other things between drugs that offered an important therapeutic gain, as evalu-
ated by the FDA (10 percent of the sample); those providing modest gains (37 percent);
and those offering little or no gain relative to substances already on the market. Lu and
Comanor found that drugs contributing important therapeutic gains were introduced at
prices 3.2 times the level of substitute products in equivalent dosages; those offering
modest gains were priced at 2.17 times the average for substitutes; those making little
or no gain were priced at roughly the same level as pre-existing substitutes. During the
four years after introduction, prices of the important new drugs (adjusted for general
inflation) tended to decline by about 13 percent on average, while the prices of drugs
making little or no therapeutic contribution rose on average by 22 percent. Introduc-
tory prices tended to be lower by 8 to 10 percent, all else equal, for each additional
competing differentiated substitute available at the time of initial marketing.
7.2. Tort liability risks andprices
Prices are also influenced by the amount of tort liability risk borne by drug manufac-
turers. Such risks vary both with drugs' inherent chemical properties and with national
or state legal rules. Controlling for these two variables with a matched sample of drugs
sold both in Canada and the United States, Manning (1997) found that roughly half the
difference between the prices of identical drugs in those two nations was attributable
to drug-specific tort liability risk, which was generally higher under U.S. legal prece-
dents. Because they inject modified actual disease vectors into the human body, vac-
cines pose particularly high risks of unfavorable outcomes 2 6 - so much so that public
health authorities in the United States experienced difficulty during the early 1980s in
obtaining supplies for immunization programs. See U.S. Office of Technology Assess-
ment (1993, pp. 176-182), Garber (1993), and Manning (1994). As a consequence, the
U.S. Congress passed in 1986 the National Childhood Vaccine Injury Act, which re-
duced companies' risk exposure by adopting a no-fault compensation scheme designed
25 See also Schwartzman (1976, Chapter 12). This paragraph is drawn with minor changes from Scherer
(1996, p. 369).
26 The author's bunk mate died, and the author came close to death, as a result of defective vaccines admin-
istered at a U.S. Army basic training camp in 1954.
to minimize the frequency of large damages awards based upon traditional negligence
theories.
7.3. Generic drug competition
Once patents expire, a new form of competition may emerge - competition from generic
drugs, that is, products with the same active chemical ingredient(s) as the original pio-
neer drug, and which are normally sold with little or no advertising or field sales promo-
tion. The extent to which generic drugs are substituted for original branded drugs and
their impact on prices varies widely from nation to nation and also across therapeutic
categories. A survey by Ballance et al. (1992, p. 210) reports little use of generic drugs
in Belgium, France, and Spain during the early 1990s, but extensive and rising use in
the United States, Denmark, Germany, and the United Kingdom. See also Perry (1996).
We focus initially on the U.S. experience.
As the post-World War II drug revolution gained force, most states in the United
States had laws limiting the ability of pharmacists to dispense anything other than the
specific brand prescribed by a physician. Since physicians typically prescribed by brand
name rather than by (less memorable) chemical name, this precluded the substitution of
a generic for a prescribed branded drug, and generic use was modest. During the 1970s
and early 1980s all states with such anti-substitution laws repealed them and passed
new laws permitting (with wide variation from state to state) some generic substitution.
But significant barriers to substitution remained. As of 1980, generics were dispensed
on only one fourth of the prescriptions for which substitution was feasible [Masson and
Steiner (1985, p. 26)].
There were two main limitations on generic use. For one, remaining legal hurdles, in-
cluding obstacles as trivial as the design of the physician's Rx form, interacted with
weak patient or pharmacist incentives as impediments to substitution [Masson and
Steiner (1985, Chapters 4 and 5)]. These will be explored in a different context shortly.
Second, FDA and judicial interpretations of the 1962 Kefauver-Harris Act required
would-be generic drug producers seeking approval from the FDA after an innovative
drug's patent expired to carry out tests nearly as extensive and costly as those associated
with a new chemical entity. Since the generic products would sell at much lower prices
than the original branded drug and might secure only modest market shares, generic
suppliers were often deterred from undertaking the required effort.
A grand compromise embodied in the Waxman-Hatch Act of 1984 remedied this
problem along with one troubling the research-oriented drug manufacturers. Under the
new law, generic suppliers were obliged only to demonstrate before the FDA that their
drug had the same active ingredient(s), that their formulation was absorbed into the
blood stream at a rate within plus-or-minus 20 percent of the original drug's norm (usu-
ally shown through tests on 24 human subjects), and that they adhered to good man-
ufacturing practices. Moreover, they were allowed to manufacture test samples before
the original drug's patents expired so that tests could be conducted and documenta-
tion submitted to the FDA in advance of patent expiration. Thus, they could hit the
1322 EM. Scherer
ground running - perhaps even on the day of patent expiration. As a quid pro quo for
the branded drug makers, Congress authorized an extension of the patent protection pe-
riod to compensate for the delays caused by FDA-required clinical testing regulations. 2 7
Typically, patents were obtained on promising new molecules at about the time when
clinical testing began, so if the tests and FDA decision-making took eight years, only
17 - 8 = 9 years of patent life remained. 2 8 Under Waxman-Hatch, the period of exclu-
sive protection could be extended by as much as five years, e.g., in the illustration here,
to 14 years. 29 Thus, drug developers would be given a longer period of exclusive sales,
but would have to face tougher competition once the period of exclusivity ended. Both
features strengthened incentives for vigorous new drug development.
7.4. Branded vs. generic drug price competition
The new legal mandates spurred a wave of entry into generic drug manufacturing, saving
U.S. consumers (or their insurers) an estimated $8 to 10 billion in 1994 [U.S. Congres-
sional Budget Office (1998, p. 31)]. Generics' share (by countable units, e.g., tablets) of
U.S. prescription drug sales rose from 18.6 percent in 1984 to 32.9 percent in 1990 and
44.3 percent in 1998 [PhRMA (1998, p. 57)]. Their share of dollar sales at retail was
much lower, e.g., 17.3 percent (vs. 36 percent by number of prescriptions) in 1994,3 ° in
part because the newest and highest-priced drugs were still covered by patent protection
and because, when generic entry began, the price regime for the typical drug bifurcated.
Generics tended to enter the market at whole sale prices 40 to 70 percent of those pre-
vailing before the original drug's patent expired. As additional generic competitors en-
tered a product category, the generic price fell, e.g., to 29 percent of the pre-competition
price with 10 generic rivals and 17 percent with 20 rivals [Caves et. al (1991, p. 118)].
Meanwhile, prices of the original branded drug remained essentially stable, according
to the analysis of Caves et al. (1991) or even rose, according to Grabowski and Vernon
(1992) and Frank and Salkever (1997).
Figure 5 illustrates the not atypical history of pricing after the product patent covering
the cephalosporin antibiotic cephalexin (sold as Keflex by the original patent holder, Eli
Lilly) expired in April 1987. 3 1 Two price trajectories for generic entrants are shown,
one averaging the wholesale prices charged by R&D-oriented firms (dotted line) for
100 250-mg capsules and one (dashed line) the comparable prices of generic specialists.
27 Japan enacted a similar law effective in January 1988, and in June 1992, European Community regulation
1768/92 authorized "supplementary protection certificates" with similar effect. See Ager (1997).
28 At the time, the normal life of a U.S. invention patent was 17 years from the time the patent was issued. As
a consequence of the Marrakech Treaty, the law was changed in 1994 to make the patent life 20 years from
the time an initial application is filed. The average time from application to issue was two to three years, but
with considerable variation about the mean.
29 The average actual extension was 2.3 years [Tufts University (1997)].
30 See U.S. Congressional Budget Office (1998, p. 15).
31 It is drawn from Griliches and Cockburn (1993, Figure 1).
^^
YU
80
Branded Keflex
o, 70
.,.,.,.,,,,,. -- /,....................
-7 "............................. ..........
70
0 60
.......... ............. ......... ..........................
Generics
.........
. . o
1o50
n I i I q I I i I i I i I
19 85 86 87 88 89 90 91
Year
Figure 5. Trends in cephalexin prices with generic entry.
Initial generic entry occurred at prices slightly less than half the branded product's price.
As additional firms began generic sales (ultimately, more than 20), the average generic
price was competed down to as little as 15 percent of the (rising) branded product price.
The reason for this bifurcated pricing behavior is that some physicians retained strong
preferences for the original brand and chose not to permit substitution. Branded drug
manufacturers in effect confronted two markets - one consisting of price-insensitive
consumers willing to pay high prices for the security of a brand name, and another
(growing in relative size as the 1980s gave way to the 1990s) consisting of price-
sensitive consumers willing to shift to generics. See Frank and Salkever (1992) and
Scherer (1996, pp. 376-378). The branded drug suppliers found it more profitable to
serve a minority fraction of their molecule's market (by 1992, 28 percent on average
of total unit sales 3 2) at high prices than to reduce their prices to the low levels required
to match generic competition. Some branded drug manufacturers practiced price dis-
crimination in an attempt to serve both market segments. Price-insensitive consumers
were sold the original brand, but to capture price-sensitive consumers, the patent holder
introduced shortly before patent expiration a "branded generic" - i.e., the same drug
under a different label, priced at lower levels than the original brand but higher than
32 See Grabowski and Vernon (1996, p. 114).

1324 EM. Scherer
no-name generics. In this way they gained a "first mover advantage" in the generic mar-
ket, 33 secured the leading share of generic sales, and perhaps thereby discouraged some
would-be generic suppliers from entering and driving prices even lower. However, most
companies feared adopting such a strategy because it could accelerate the shift from
their high-price branded products to low-priced generics. According to a U.S. Congres-
sional Budget Office study (1998, p. 34), among 112 drugs with generic competition,
the original brand holder sold its own generic product and gained a retail market share
of more than 10 percent in only 13 cases.
Not all of the savings from generic drug price reductions at wholesale are passed on
to consumers in the form of lower retail prices. On average, retail pharmacies retain
higher dollar margins, and hence much higher percentage margins, on the generic drugs
they dispense than on the equivalent, higher-priced, branded drugs. See Masson and
Steiner (1985, p. 36) and Grabowski and Vernon (1996, p. 117). Higher dollar margins
are realized on generics in part to defray inventory holding costs. A more important
reason, however, is that pharmacists must stock the leading branded product, but have
some choice in determining which of several possible generics to stock and dispense.
This appears to be a special case of a more general phenomenon described as the Steiner
effect: retail margins tend to be lower on products with strong consumer pull, e.g., those
that are heavily advertised, than on less-advertised items. See Steiner (1993). Kopp and
Sheffet (1997) report a substantial relative decline in retailers' margins for drugs receiv-
ing direct-to-consumer advertising support between 1988 and 1991. Viagra, the drug
product with perhaps the strongest early consumer awareness in history, was merchan-
dised by retail drug discounters at near-zero margins as a means of attracting patronage
more generally. 3 4
7.5. Stimulating generic substitution
The extent of generic substitution depends upon several variables in addition to the
number and identity of generic suppliers. Generics have been particularly unsuccessful
in replacing branded drugs for which attaining precise diffusion rates within the body
is crucial to effective therapy, e.g., in heart rhythm regulation and the administration
of anticoagulants after a heart attack. Passing over such relatively rare cases, generic
substitution was encouraged during the 1980s and 1990s in the United States by three
phenomena: rapidly rising affiliation of physicians with cost-conscious health care or-
ganizations enforcing strong pro-generic policies; "maximum allowable cost" (MAC)
reimbursement rules under which state Medicaid authorities reimbursed only the cost
33 On first-mover advantages more generally, see Bond and Lean (1977) and Robinson et al. (1994). The race
to be a first mover led some generic producers to falsify test results and bribe FDA employees. See Morton
(1997b).
34 "Kmart, Wal-Mart Compete with Teeny Viagra Prices", Reuters dispatch (obtained on Yahoo Finance),
May 21, 1998.
Ch. 25: The PharnaceuticalIndustry 1325
of the lowest-cost qualified generic substitute for a drug;3 5 and lower co-payments under
private health insurance plans when patients accepted generic as compared to branded
drugs.
The experience of Canada sheds further light on variables affecting generic substitu-
tion. See Gorecki (1986) and McRae and Tapon (1985). Generics were available early
in branded drug life cycles during the 1970s and 1980s as a result of compulsory drug
patent licensing policies. The various Canadian provinces had widely differing rules
with respect to generic substitution, and substitution rates varied as a result in iden-
tifiable ways. Quebec, New Brunswick, and Nova Scotia had especially low substitu-
tion rates because they reimbursed the full cost of any drugs dispensed under Canada's
lower-income and above-65 health insurance programs, whereas provinces with high
substitution rates reimbursed only the cost of the least expensive generic. Designation
of specific generic drugs as interchangeable with branded counterparts on provincial
formularies markedly increased substitution rates. In Quebec, unlike other provinces,
pharmacists were not absolved of malpractice liability in dispensing formulary gener-
ics. This, like Quebec's requirement that pharmacists acquire the patient's consent be-
fore substituting, reduced generic drug usage.
7.6. New institutions;New power relationships
During the 1980s and 1990s, new forces emerged in the United States to countervail
and weaken manufacturers' power to maintain high prices on patented drugs. Crucial
to these changes was the rapid growth of health maintenance organizations (HMOs)
and pharmacy benefit management firms (PBMs). Glied (2000) shows in more detail,
health maintenance organizations affiliate substantial numbers of physicians to provide
comprehensive health care on a prepaid insurance basis. Patient enrollment in HMOs
increased from 9.1 million in 1980 to 33.6 million in 1990 and 46 million in 1995 [U.
S. Bureau of the Census (1996, p. 121)]. An even newer phenomenon was the emer-
gence of PBMs to manage the payment paperwork on out-patient prescription drug in-
surance claims for HMOs, companies with large employee health care insurance plans,
and conventional insurance carriers. By 1993, the six largest PBMs were managing an
estimated 36 percent of all U.S. retail prescriptions [U.S. Congressional Budget Office
(1996, p. 21)1.
To control the rising costs of prescription drugs, many HMOs and traditional hospi-
tals began establishing formularies listing the drugs suitable for use against particular
illnesses. When appropriate generic drugs existed, formularies strongly encouraged af-
filiated HMO or hospital staff to use them in place of higher-priced branded drugs. But
even when no generic substitutes were available, formulary committees began selecting
from the menu of alternative patented drugs those deemed most cost-effective. Physi-
cians were motivated to comply with formulary guidelines by persuasion, paperwork
35 Since reimbursement is to the pharmacist, the pharmacist dispensing a high-priced drug absorbs the cost
difference.
1326 FM. Scherer
burdens justifying non-compliance, and in some cases financial penalties. As the use
of formularies gained acceptance, health care organization drug procurement personnel
realized that they could use the threat of a drug's exclusion from their formulary as a
lever to elicit discounts from pharmaceutical manufacturers. Manufacturers of substi-
tutable patented, branded drugs were played off against each other, and to an increasing
degree they conceded substantial off-list discounts. By the early 1990s, HMOs were
receiving discounts averaging 20 to 25 percent off the average wholesale prices paid by
retail pharmacies [Boston Consulting Group (1993, Figure 1-3)]. A government study
revealed that in 1991, the "best price" offered by a manufacturer to some customer -
usually an HMO or similarly bargaining hospital - carried a discount of 50 percent or
more off the wholesale list price for 32 percent of all single-source (i.e., patented) drugs
[U.S. Congressional Budget Office (1996, p. 28)].
For drugs dispensed on an inpatient basis within an HMO or hospital, there were
several ways the negotiated discount could be realized. The organization could take de-
livery of the drug directly from the manufacturer and pay only the discounted price.
But more frequently, health care organizations preferred to obtain their drug supplies
from local wholesalers, who assumed the burden of maintaining inventories and assured
same-day delivery of orders. In those instances, the wholesaler could bill the HMO at its
standard wholesale price, after which the manufacturer would issue a rebate check to the
HMO for the amount of the negotiated discount. Alternatively, the manufacturer could
notify its wholesalers of its discount arrangements with individual HMOs, whereupon
the wholesaler would deliver drugs to an HMO at the discounted price plus a negoti-
ated wholesaling fee, receiving from the manufacturer a "chargeback" payment to cover
the difference between the discounted price and the price at which the wholesaler had
acquired its inventory from the manufacturer.
PBMs entered the picture for the much larger volume of drugs obtained by patients
from retail pharmacies. HMOs contracted with PBMs to reimburse retailers for insured
prescriptions written by their affiliated physicians. When the HMO negotiated a dis-
count with a manufacturer, the PBM would reimburse the retailer for its drug acquisition
cost plus a negotiated dispensing fee, obtaining from the manufacturer a rebate (passed
on the HMO) to cover the discount negotiated between the HMO and the manufacturer.
(The negotiation of dispensing fees also imposed downward pressure on retail pharma-
cists' traditional 40 percent gross margins. 3 6 ) In addition, PBMs contracted to manage
the retail pharmacy payment paperwork for companies with large numbers of employ-
ees whose health insurance included prescription drugs. To save money for their clients,
36 According to one account, retailers' margins on the sale of ethical drugs had been driven down by 1998
from 40 to 20 percent. "Why drugstore chains are in good health", Business Week, May 4, 1998, p. 170.
See also Scherer (1997, pp. 245-246). In the late 1990s, PBMs and HMOs began concluding contracts under
which pharmacy chains agreed to bill the PBM a fixed total price per prescription, regardless of the cost of the
drug dispensed - an arrangement under which the pharmacies in effect bore the risk of dispensing unusually
many high-cost drugs. Smaller retail pharmacies were wary of accepting such contracts because of the risks,
and therefore lost HMO sales. See "Wrong Rx for their needs", Boston Globe, March 22, 1998, p. C1.
PBMs established their own formularies, negotiated discounts with manufacturers for
formulary inclusion, and assigned staff members to call physicians, urging them to pre-
scribe lower-priced formulary drugs for covered patients when appropriate substitutes
were available.
These hard-ball tactics introduced new and powerful elements of competition into
the pricing of patented drugs. A key lever used in negotiating discounts - exclusion of
non-discounting drugs from formularies - may also have had more subtle adverse side
effects. Before the tactic was widely adopted by HMOs, U.S. state government agencies
frequently excluded new and high-priced drugs from formularies listing drugs available
for Medicaid reimbursement. Statistical analyses reveal that expenditures on drugs were
in fact reduced by such policies, but that hospital admission and similar costs may have
increased by a more than offsetting amount as a consequence of using less effective
drugs. See, e.g., Dranove (1989), Moore and Newman (1993), and, concerning flat limits
on the number of reimbursable prescriptions per patient, Soumerai et al. (1991).
The transfer of substantial control over prescription drug choices from individual
physicians to formulary committees also elicited changes in pharmaceutical manufac-
turers' marketing strategies. Dispatching hordes of "detail persons" to make regular
sales calls on individual physicians became less cost-effective, and so detailing budgets
were cut sharply. To instigate patient pressure on physicians and from them on formu-
lary committees, manufacturers in the United States began devoting substantial sums to
newspaper and television advertising aimed directly at end consumers.
Another unanticipated consequence of the manufacturer - HMO bargaining was an
antitrust law suit that eventually pitted more than 25,000 retail pharmacists against 25
leading U.S. branded drug manufacturers. 3 7 The pharmacists alleged that by granting
discounts to the HMOs who could threaten them with market share losses but not to the
retail pharmacists, the drug manufacturers engaged in illegal price discrimination. They
alleged also that discounts were withheld from the retailers through an illegal conspir-
acy, inferred inter alia from the cooperation between manufacturers and wholesalers to
create a uniform computerized system for keeping wholesalers informed of negotiated
discounts and paying chargebacks to the wholesalers who delivered discounted drugs to
HMOs. Most of the plaintiffs and 13 drug companies agreed in 1996 to a settlement un-
der which the manufacturers would pay damages of $351 million and offer discounts to
retail pharmacy buying groups that could demonstrate "an ability to affect market share"
through their own formulary and physician contact activities. 38 Four other companies
settled for a total of $345 million in July 1998. In December 1998, the presiding judge
37 Inre Brand Name Prescription Drugs Antitrust Litigation, MDL-997, Master file no. 94 C 897 (consoli-
dated before the U.S. Federal Court for the Northern District of Illinois). A symposium in the International
Journal of the Economics of Business, Vol. 4 (November 1997), joins the issues inarticles by eight economists
who had consulted with either plaintiffs or respondents in the litigation.
38 Memorandum opinion of Judge Charles P. Kocoras, MDL-997, June 21, 1996. See also the Seventh Circuit
Court of Appeals decision of August 15, 1997, which rejected many of Judge Kocoras' findings but did not
overturn the settlement.
1328 FM. Scherer
dismissed the remaining conspiracy claims as "far fetched, improbable, and unreason-
able".
A U.S. government initiative seeking to reduce Medicaid drug costs had still another
unanticipated consequence, restricting the size of the discounts manufacturers were
willing to cede to hard-bargaining HMOs and other health care organizations. Under
laws enacted in 1990 and extended thereafter, manufacturers were required to give the
government agencies reimbursing Medicaid prescription drug purchases discounts equal
to the best discount offered any non-governmental purchaser on any given product, but
not less than 15.1 percent. As a consequence, manufacturers confronted with demands
for deep discounts from a strong HMO realized that if they yielded to the HMO, they
would also have to increase their discounts on Medicaid sales, which in 1995 amounted
to 17.6 percent of all out-patient prescription drug sales [PhRMA (1997, p. 35)]. Two
statistical studies have shown that the perverse most-favored customer incentives gen-
erated by the Medicaid discount law substantially reduced the discounts received by
other health care organizations [U.S. Congressional Budget Office (1996, pp. 27-44),
and Morton (1997a)].
8. Profits and price controls

Pharmaceuticals are sold under classic monopolistic competition conditions. Patents
and product differentiation lead to prices that are well above production costs. But com-
panies strive for partial monopoly positions and high margins by introducing new drugs.
To do so, they incur substantial R&D costs and marketing costs, reducing bottom-line
profits. When superior new products emerge from drug R&D, consumers and producers
alike benefit. Whether high expenditures on sales promotion are in the net socially bene-
ficial is more heatedly debated. On one hand, information is disseminated to physicians
and (more recently) consumers, speeding the entry of new drugs into medical practice.
On the other hand, as a governmental commission concluded, "It is doubtful ... that
... detail men ... give invariably unprejudiced and objective advice". [Task Force on
Prescription Drugs (1969, pp. 9-10)]. Compare also Schwartzman (1976, Chapter 9),
Comanor (1986, pp. 1196-1199), Beales (1996), and Kopp and Sheffet (1997).
In the United States, the pharmaceutical industry has for decades appeared at or near
the top of industry rankings by after-tax profit returns on stockholders' equity [Scherer
(1996, p. 342)]. However, profit reports prepared following conventional accounting
practices, including current-year writeoff of research and development expenditures,
tend to overstate true economic profitability, given the growth rates experienced by
pharmaceutical firms. When drug makers' R&D outlays were capitalized and amortized
at plausible rates, the industry's overall rate of return on invested capital in the 1970s
and 1980s was found to exceed all-industry averages by only two to three percentage
points [U.S. Office of Technology Assessment (1993, Chapter 4)].
The perception, correct or incorrect, that pharmaceutical prices and profits have been
excessive, the taxpayer burdens from rising public health care costs, and the belief,
especially in smaller nations, that reducing drug prices and profits will at best have
a minor impact on R&D expenditures by companies oriented toward serving world-

wide markets, have led many governments to impose more or less thoroughgoing price
controls on pharmaceutical products. Among 56 nations whose governmental policies
toward the pharmaceutical industry were surveyed by Ballance et al. (1992, pp. 140-
145 and 166-171), 30 nations (12 industrialized and 18 developing) were characterized
as having "substantial" price controls and 20 (11 industrialized and nine developing) as
having "limited" controls. Only six (all in the developing category) were said to have
no controls.
There are countless variations in the ways governments regulate drug prices. These
can be compacted into five broad groupings - reference pricing, item-by-item negoti-
ation and control, formula pricing, profit or rate of return regulation, and capping or
budgetary constraint controls. See Danzon (1997), Shulman and Lasagna (1994), and
U.S. Office of Technology Assessment (1993, pp. 250-262), on which the next seven
paragraphs are based. Many nations' policies entail a mix of the various methods, with
the mix changing over time, so what follows can provide only selective snapshots.
Under reference pricing, more-or-less comparable drugs are placed into a reference
group, and reimbursement is provided under national or regional health insurance plans
only at the lowest price within the reference group. The U.S. maximum allowable cost
approach to Medicaid drug reimbursement is a relatively innocuous version, placing
generics and branded drugs with identical active ingredients in the same group. A more
drastic approach was taken by Germany beginning in 1989 and Sweden beginning in
1993. Different chemical entities treating the same illness are placed in the same refer-
ence group. The broader the reference group and the more it includes new formulations
along with older drugs, the more likely it is to discourage investments in discovering
and developing superior new drugs.
In France and (until 1993) Italy, the prices of individual outpatient drugs seeking
reimbursement under national' health insurance plans were set in administrative pro-
ceedings taking into account a wide array of criteria, including therapeutic novelty and
contribution to the economy. Drugs produced and developed locally tended to receive
higher prices than imported drugs, which created incentives for local firms to develop
and introduce numerous new drugs of insufficient therapeutic novelty to achieve signif-
icant sales outside the home market. This, along with the low standards imposed by the
agencies regulating new drug introductions, helps explain the relatively modest external
sales of French and Italian drug manufacturers, as shown in Figure 3.
Under the health care reforms proposed in 1993 by U.S. President Clinton but re-
jected by the Congress, a different form of ad hoc regulation was contemplated. An Ad-
visory Council on Breakthrough Drugs was to be charged with reviewing the prices of
new drugs and, in cases where they were considered excessive, implementing measures
ranging from public suasion or "jawboning" to making the drugs ineligible for health
insurance reimbursement. The drugs most likely to be singled out for this regulation
were the "blockbusters" in the first decile of Figure 4. The difficulty with this approach,
obvious from Figure 4, is that curbing significantly the prices and profits of blockbuster
drugs could make it difficult for companies to recover their research and development
1330 EM. Scherer
investments on less successful drugs - i.e., in the fourth through tenth deciles. Severe
impairment of R&D incentives could result.
Many nations, including Italy since 1993 [Fattore (1996)] and Canada since 1987,
relate the reimbursable prices of relatively new drugs to the prices of the same drugs in
other nations. When nations (such as Spain) characterized by generally low prices are
included in the comparison group, this creates incentives for multinational drug manu-
facturers to set prices higher in the comparison group jurisdiction than those they would
otherwise be inclined to charge. 3 9 Or in the case of India, where both low incomes
and regulation limit sustainable prices, multinationals are said sometimes not to market
drugs at all until late in their life cycle so that low prices in India cannot be cited to
reduce prices in other nations making international comparisons. See Lanjouw (1997).
In Japan, the vast majority of drugs are dispensed directly by physicians, who are
then reimbursed by government health authorities on a formula basis for the drugs. New
drugs receive relatively high prices, and after that, their prices are reduced downward
systematically with the drug's age. This system has two important incentive effects. For
one, to encourage the use of their drugs, manufacturers set prices that allow physicians
generous margins between the physician's acquisition cost and the reimbursed price,
leading to the extraordinarily high prescribing rates observed in Japan. Second, because
new drugs command the highest prices, manufacturers have strong incentives, as in
France and Italy, to introduce many new drugs, whether or not they make significant
therapeutic contributions. This in turn is partly responsible for the poor external market
performance of Japanese drug manufacturers evident in Figure 3. See Thomas (1996)
and Ikegami et al. (1998).
The United Kingdom is the only nation known to have a rate of return regulation
system analogous to the way electrical and telephone utilities were regulated in the
United States for many decades. In an annual determination, the assets of individual
companies, including the capitalized value of research and development outlays, are
measured. Each company negotiates with the regulatory authority an allowed before-
tax rate of return on its assets, usually in the range of 17 to 21 percent. Prescription drug
sales revenues are set (or adjusted after-the-fact) so that, after operating, R&D, and
sales promotion costs are deducted, the company is left with profit sufficient to yield the
agreed-upon rate of return on assets. In the cost calculations, promotional expenditures
can be deducted only up to a limit of approximately 9 percent of sales. The U.K. "Price
Regulation Scheme" would appear to reward investments in research and development
39 Wide differences in drug price levels among the member nations of the European Union led to further
complications. Wholesalers in low-price nations such as Spain attempted to ship drugs imported from, e.g.,
Germany back to Germany to arbitrage the higher prices there. When manufacturers sought to curb such
"parallel imports", the Common Market authorities intervened, charging illegal restraint of competition. See
the Commission decision In re Adalat, Case IV/34.279/F3, 1995. For diverse views on policy toward parallel
imports and arguments favoring inter-national price discrimination as a "second best" Ramsey pricing solution
to the problem of recovering fixed R&D costs, see Yarrow (1995), Danzon (1997, Chapter 7), Bangemann
(1997), and Anis and Wen (1998).
Ch. 25: The PhannaceuticalIndustry 1331
and hence to avoid the negative incentive problems in many other nations' regulatory
approaches. However, there is a paradox. If the scheme is implemented mechanically,
large companies with R&D portfolios containing many projects tend to realize substan-
tially higher returns on investment than small companies with few projects. See Scherer
(1995, pp. 36-38). Given the high skewness of drug development project outcomes re-
vealed by Figure 4, companies with many projects can include the substantial R&D
investment from numerous "losers" as well from (the few) blockbusters in their R&D
asset base, and the large investment base will allow the companies to realize most, if
not all, of the profit potential from blockbusters. If a small company is lucky enough
to develop a blockbuster, it will by its very smallness have few losers in its investment
base, so the revenues it is allowed to realize on the blockbuster will be severely lim-
ited by regulation. If on the other hand (with appreciable probability) it achieves no
blockbuster, its returns will be severely limited by market competition.
Germany illustrates the use of aggregate budget constraints and roll-backs. In an at-
tempt to control escalating health care costs, the Federal Health Ministry beginning in
1993 set a tight overall drug budget, requiring inter alia a roll-back from previous spend-
ing levels. The first DM 280 million of spending above that target was to be deducted
from the incomes of physicians. If the budget was exceeded by DM 281-360 million,
the excess was to be deducted from reimbursements to drug manufacturers.4 0 Between
1995 and 1997, German drug budgets were decentralized regionally out to the level
of individual physicians (as is also done in the United Kingdom). An apparent conse-
quence of individual physician spending constraints was that primary care physicians
referred increased numbers of patients to specialists and hospitals, who were subject to
different individual constraints [Sch6ffski (1996)]. In 1998, cost containment empha-
sis in Germany shifted away from drug budget constraints toward increased individual
patient co-payments.
In sum, efforts by national authorities to curb pharmaceutical costs and offset the
demand-increasing effects of generous health care insurance by imposing drug price
controls are found throughout the industrialized and less-developed world. These some-
times succeed in their proximate goal, but cause bulges in other parts of the health
care balloon, bias new drug research and development incentives, and distort interna-
tional trade and investment patterns. Although one may share the underlying cost con-
trol goals, a review of the consequences suggests that the aversion of most economists
to price controls is well-founded.
9. Conclusion
The pharmaceutical industry has made enormous contributions to health care in the half
century since World War II as the drug research and development revolution gained
40 France introduced manufacturer-specific budget targets in 1994.

1332 EM. Scherer
momentum. Progress in biological science and molecular engineering is likely to pro-

vide the basis for further dramatic therapeutic advances in the future. But the conditions
that create strong incentives for investment in pharmaceutical R&D also arouse public
concern over monopoly positions, high prices, and the introduction of products of un-
certain efficacy or safety. From that concern flow regulatory interventions into clinical
testing protocols and pricing that could retard future technological progress. The prob-
lem is complicated by the fact that individual nation states can rationally behave as free
riders or, more accurately, cheap riders, ignoring the consequences of their policies on
drug R&D decisions in other parts of a complex multinational industry. Achieving the
right tradeoff between progress, affordability, and optimal provision of test information
remains an elusive goal.
References
Ager, B. (1997), "Patents, pharmaceuticals and generics", Eurohealth 2:35-36.

Anis, A.H., and Q. Wen (1998), "Price regulation of pharmaceuticals in Canada", Journal of Health Eco-
nomics 17:21-38.
Bakke, O.M., M. Manochia, F. de Abajo, K.I. Kaitin and L. Lasagna (1995), "Drug safety discontinuations in
the United Kingdom, the United States, and Spain from 1974 through 1993", Clinical Pharmacology and
Therapeutics, 58:108-117.
Ballance, R., J. Pogany and H. Forstner (1992), The World's Pharmaceutical Industries: An International Per-
spective on Innovation, Competition and Policy (Edward Elgar for United Nations Industrial Development
Organization, Hants, UK).
Bangemann, M. (1997), "Completing the single pharmaceutical market", Eurohealth 3(1):22-23.
Beales, J.J. (1996), "New uses for old drugs", in: R.B. Helms, ed., Competitive Strategies in the Pharmaceu-
tical Industry (AEI Press, Washington) 281-305.
Berndt, E.R., L. Bui, D.R. Reiley and G.L. Urban (1995), "Information, marketing, and pricing in the U.S.
antiulcer drug market", American Economic Review 85(2):100-105.
Bond, R.S., and D.F. Lean (1977), Sales, Promotion, and Product Differentiation in Two Prescription Drug
Markets (Federal Trade Commission, Washington).
Boston Consulting Group (1993), The Changing Environment for U.S. Pharmaceuticals (Boston).
Caves, R.E., M. Whinston and M. Hurwitz (1991), "Patent expiration, entry, and competition in the U.S.
pharmaceutical industry", Brookings Papers on Economic Activity: Microeconomics 1991:1-48.
Cockbum, I., and R. Henderson (1998), "Absorptive capacity, coauthoring behavior, and the organization of
research in drug discovery", Journal of Industrial Economics 46:157-182.
Cohen, W.M., R.R. Nelson and J. Walsh (1997), "Appropriability conditions and why firms patent and why
they do not in the American manufacturing sector", Working Paper (Carnegie-Mellon University, Pitts-
burgh).
Comanor, W.S. (1986), "The political economy of the pharmaceutical industry", Journal of Economic Liter-
ature 24:1178-1217.
Cooper, J.D. (1970), "The sources of innovation", in: J.D. Cooper, ed., The Economics of Drug Innovation
(American University, Washington) 41-62.
Danzon, P.M. (1997), Pharmaceutical Price Regulation (AEI Press, Washington).
Deardorff, A.V. (1992), "Welfare effects of global patent protection", Economica 59:35-51.
DiMasi, J.A., R.W. Hansen, H.G. Grabowski and L. Lasagna (1991), "Cost of innovation in the pharmaceuti-
cal industry", Journal of Health Economics 10:107-142.
DiMasi, J.A., M. Seibring and L. Lasagna (1994), "New drug development in the United States from 1963 to
1992", Clinical Pharmacology and Therapeutics 55:609-622.
Dorfman, R., and P.O. Steiner (1954), "Optimal advertising and optimal quality", American Economic Review
44:826-836.
Dranove, D. (1989), "Medicaid drug formulary restrictions", Journal of Law and Economics 32:143-162.
Ellison, S.F., I. Cockburn, Z. Griliches and J. Hausman (1997), "Characteristics of demand for pharmaceutical
products: An examination of four cephalosporins", RAND Journal of Economics 28:427-446.
Fattore, G. (1996), "Pharmaceutical policy in Italy 1992-1995", Eurohealth 2:34-35.
Frank, R.G., and D.S. Salkever (1992), "Pricing, patent loss and the market for pharmaceuticals", Southern
Economic Journal 59:165-179.
Frank, R.G., and D.S. Salkever (1997), "Generic entry and the pricing of pharmaceuticals", Journal of Eco-
nomics and Management Strategy 6:75-90.
Gambardella, A. (1995), Science and Innovation: The US Pharmaceutical Industry During the 1980s (Cam-
bridge University Press, Cambridge, UK).
Garber, S. (1993), "Product liability and the economics of pharmaceuticals and medical devices", RAND
Corporation Report R-4285-ICJ (Santa Monica, CA).
Glied, S. (2000), "Managed care", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics
(Elsevier, Amsterdam) Chapter 13.
Gorecki, P.K. (1986), "The importance of being first: The case of prescription drugs in Canada", International
Journal of Industrial Organization 4:371-396.
Grabowski, H.G. (1976), Drug Regulation and Innovation (American Enterprise Institute, Washington).
Grabowski, H.G., and J.M. Vernon (1983), The Regulation of Pharmaceuticals (American Enterprise Institute,
Washington).
Grabowski, H.G., and J.M. Vernon (1990), "A new look at the returns and risks to pharmaceutical R&D",
Management Science 36:804-821.
Grabowski, H.G., and J.M. Vernon (1992) "Brand loyalty, entry, and price competition in pharmaceuticals
after the 1984 drug act", Journal of Law and Economics 35:331-350.
Grabowski, H.G., and J.M. Vernon (1994), "Returns on new drug introductions in the 1980s", Journal of
Grabowski, H.G., and J.M. Vernon (1996), "Longer patents for increased generic competition in the US",
PharmacoEconomics 10(Supplement 2): 110-123.
Grabowski, H.G., J.M. Vernon and L.G. Thomas (1978), "Estimating the effects of regulation on innovation:
An international comparative analysis", Journal of Law and Economics 21:133-163.
Griliches, Z., and I. Cockburn (1993), "Generics and new goods in pharmaceutical price indexes", Working
Paper No. 4272 (National Bureau of Economic Research).
Henderson, R., and I. Cockburn (1996), "Scale, scope, and spillovers: The determinants of research produc-
tivity in drug discovery", RAND Journal of Economics 27:32-59.
Heppell, S. (1996), "The new European system for regulating medicinal products", Eurohealth 2:28-29.
Holmes, J., and J. Dunning (1995), "Factors influencing the location of multinational investment in the phar-
maceutical industry", in: A. Towse, ed., Industrial Policy and the Pharmaceutical Industry (Office of Health
Economics, London) 92-105.
UT Research Institute (1968), Technology in Retrospect and Critical Events in Science (Illinois Institute of
Technology, Chicago).
Ikegami, N., S. Ikeda and H. Kawai (1998), "Why medical care costs in Japan have increased despite declining
prices of pharmaceuticals", PharmacoEconomics 14(Supplement 1):97-105.
Kaitin, K.I. (1997), "The new drug approvals of 1993, 1994, and 1995", American Journal of Therapeutics
4:46-54.
Kitch, E.W. (1973), "The patent system and the new drug application", in: R.L. Landau, ed., Regulating New
Drugs (University of Chicago Center for Policy Study) 81-108.
Kopp, S.W., and M.J. Sheffer (1997), "The effect of direct-to-consumer advertising of prescription drugs on
retail gross margins", Journal of Public Policy and Marketing 16:270-276.
Lanjouw, J.O. (1997), "The introduction of pharmaceutical product patents in India", Discussion Paper
No. 775 (Yale University Economic Growth Center).
1334 FM. Scherer
Leibowitz, A., W.G. Manning and J.P. Newhouse (1985), "The demand for prescription drugs as a function
of cost sharing", Social Science Medicine 21:251-277.
Levin, R.C., A.K. Klevorick, R.R. Nelson and S.G. Winter (1987), "Appropriating the returns from industrial
research and development", Brookings Papers on Economic Activity 3:783-821.
Lichtenberg, FR. (1996), "The effect of pharmaceutical utilization and innovation on hospitalization and
mortality", Working Paper No. 5418 (National Bureau of Economic Research, Cambridge, MA).
Lichtenberg, F.R. (1998), "Pharmaceutical innovation, mortality reduction, and economic growth", Working
Paper No. 6569 (National Bureau of Economic Research, Cambridge, MA).
Lu, .L., and W.S. Comanor (1998), "Strategic pricing of new pharmaceuticals", Review of Economics and
Statistics 80:108-118.
Mann, C.C., and M.L. Plummer (1991), The Aspirin Wars (Harvard Business School Press, Boston).
Manning, R.L. (1994), "Changing rules in tort law and the market for childhood vaccines", Journal of Law
and Economics 37:247-275.
Manning, R.L. (1997), "Products liability and prescription drug prices in Canada and the United States",
Journal of Law and Economics 49:203-244.
Mansfield, E. (1970), "Comment", in: J.D. Cooper, ed., The Economics of Drug Innovation (American Uni-
versity, Washington) 149-154.
Mansfield, E. (1995), "Academic research underlying industrial innovations", Review of Economics and
Masson, A., and R.L. Steiner (1985), "Generic substitution and prescription drug prices", Staff Report (Fed-
eral Trade Commission, Washington).
McFetridge, D.G. (1997), "Intellectual property rights and the location of innovative activity: The Canadian
experience", Working Paper (Carleton University, Ottawa).
McRae, J.J., and F. Tapon (1985), "Some empirical evidence on post-patent barriers to entry in the Canadian
pharmaceutical industry", Journal of Health Economics 4:43-61.
Moore, W.J., and R. Newman (1993), "Drug formulary restrictions as a cost-containment policy in Medicaid
programs", Journal of Law and Economics 36:71-114.
Morton, F.S. (1997a), "The strategic response by pharmaceutical firms to the Medicaid most-favored-
customer rules", RAND Journal of Economics 28:269-290.
Morton, F.S. (1997b), "The objectives of the FDA's Office of Generic Drugs", Working Paper No. 6143
(National Bureau of Economic Research).
Neary, I. (1995), "Japanese industrial policy and the pharmaceutical industry", in: A. Towse, ed., Industrial
Policy and the Pharmaceutical Industry (Office of Health Economics, London) 12-25.
Peltzman, S. (1973), "An evaluation of consumer protection legislation: The 1962 drug amendments", Journal
of Political Economy 81:1049-1091.
Peltzman, S. (1987), "The health effects of mandatory prescriptions", Journal of Law and Economics 30:207-
238.
Perry, G. (1996), "Generic medicines: How to ensure their effective contribution to health care", Eurohealth
2:32-34.
PhRMA (1997), Industry Profile: 1997 (Pharmaceutical Research and Manufacturers of America, Washing-
ton).
PhRMA (1998), Industry Profile: 1998 (Pharmaceutical Research and Manufacturers of America, Washing-
ton).
Rizzo, J.A. (1997), "Advertising and competition in the ethical pharmaceutical industry: The case of antihy-
pertensive drugs", Working Paper (Yale University).
Robinson, W.T., G. Kalyanaram and G. Urban (1994), "First-mover advantages from pioneering new markets:
A survey of the empirical evidence", Review of Industrial Organization 9:1-24.
Ryan, M.P. (1998), Knowledge Diplomacy: Global Competition and the Politics of Intellectual Property
(Brookings, Washington).
Santoro, M. (1992), "Pfizer: Protecting intellectual property in a global marketplace", Case Study N9-392-072
(Harvard Business School, Boston).
Schachtel, B.P. (1994), "Sensible and sensitive clinical trials in the evaluation of over-the-counter drugs", in:
S.R. Shulman, ed., Moving Prescription Drugs to the Over-the-Counter Market (Tufts University Human
Nutrition Center, Boston).
Scherer, EM. (1995), "US industrial Policy in the pharmaceutical industry", in: A. Towse, ed., Industrial
Policy and the Pharmaceutical Industry (Office of Health Economics, London) 26-39.
Scherer, F.M. (1996), Industry Structure, Strategy, and Public Policy (Harper-Collins, New York).
Scherer, EM. (1997), "How U.S. antitrust can go astray", International Journal of the Economics of Business
4:239-256.
Scherer, EM. (1998), "The size distribution of profits from innovation", Annales d'Economie et de Statistique
49/50:496-516.
Scherer, EM., and D.R. Ross (1990), Industrial Market Structure and Economic Performance, 3rd edn.
(Houghton-Mifflin, Boston).
Scherer, EM., and S. Weisburst (1995), "Economic effects of strengthening pharmaceutical patent protection
in Italy", International Review of Industrial Property and Copyright Law 26:1009-1024.
Sch6ffski, 0. (1996), "Consequences of implementing a drug budget for office-based physicians in Germany",
PharmacoEconomics 10(Supplement 2):37-47.
Schwartzman, D. (1976), Innovation in the Pharmaceutical Industry (Johns Hopkins University Press, Balti-
more).
Schwartzman, D. (1997), "Has pharmaceutical research become more scientific?", Review of Industrial Or-
ganization 11:841-851.
Shulman, S.R., and K.I. Kaitin (1996), "The prescription drug user fee act of 1992", PharmacoEconomics
9(2):121-133.
Shulman, S.R., and L. Lasagna, eds. (1994), Pricing and Reimbursement of Pharmaceuticals: An Evaluation
of Cost-Containment Strategies, Symposium proceedings, PharmacoEconomics 6(Supplement 1).
Soumerai, S.B., D. Ross-Degnan, J. Ahom, T.J. McLaughlin and I. Choodnovskiy (1991), "Effects of Medic-
aid drug-payment limits on admission to hospitals and nursing homes", New England Journal of Medicine
325:1072-1077.
Steiner, R.L. (1993), 'The inverse association between the margins of manufacturers and retailers", Review
of Industrial Organization 8:717-740.
Task Force on Prescription Drugs (1968), The Drug Makers and the Drug Distributors (US Department of
Health, Education and Welfare, Washington).
Task Force on Prescription Drugs (1969), Final Report (US Department of Health, Education and Welfare,
Washington).
Temin, P. (1979a), "The origin of compulsory drug prescriptions", Journal of Law and Economics 22:91-105.
Temin, P. (1979b), "Technology, regulation, and market structure in the modem pharmaceutical industry",
Bell Journal of Economics 10:429-446.
Thomas, L.G. (1996), "Industrial policy and international competitiveness in the pharmaceutical industry", in:
R.B. Helms, ed., Competitive Strategies in the Pharmaceutical Industry (AEI Press, Washington) 107-129.
Toole, A.A. (1997), "The impact of federally funded basic research on industrial innovation: Evidence from
the pharmaceutical industry", Working Paper (Christensen Associates, Madison, WI).
Tufts University Center for the Study of Drug Development (1997), "Tufts CSDD testifies before Canadian
Parliamentary Committee about U.S. experience with patent extension", Tufts CSDD Newsletter 22:1-2.
US Bureau of the Census (1992), 1987 Census of Manufactures, Concentration Ratios in Manufacturing,
MC87-S-6 (USGPO, Washington).
US Bureau of the Census (1996), Statistical Abstract of the United States: 1996 (USGPO, Washington).
US Bureau of the Census (1997), 1995 Annual Survey of Manufactures, Value of Product Shipments,
M95(AS)-2 (USGPO, Washington).
US Congressional Budget Office (1996), How the Medicaid Rebate on Prescription Drugs Affects Pricing in
the Pharmaceutical Industry (CBO Papers, Washington).
1336 EM. Scherer
US Congressional Budget Office (1998), How Increased Competition from Generic Drugs Has Affected
Prices and Returns in the Pharmaceutical Industry (Washington).
US Federal Trade Commission (1958), Economic Report on Antibiotics Manufacture (USGPO, Washington).
US Federal Trade Commission (1985), Statistical Report: Annual Line of Business Report: 1977 (Washing-
ton).
US House of Representatives, Committee on Science and Technology (1980), Report, The Food and Drug
Administration's Process for Approving New Drugs (USGPO, Washington).
US National Science Board (1998), Science and Engineering Indicators: 1998 (USGPO, Washington).
US National Science Foundation (1996), Research and Development in Industry: 1993, NSF 96-304 (Wash-
ington).
US Office of Technology Assessment (1984), Federal Policies and the Medical Devices Industry (USGPO,
Washington).
US Office of Technology Assessment (1993), Pharmaceutical R&D: Costs, Risks and Rewards (USGPO,
Washington).
Wardell, W., and L. Lasagna (1975), Regulation and Drug Development (American Enterprise Institute, Wash-
ington).
Weintraub, M. (1994), "OTC drugs: Definitional and methodologic aspects", in: S.R. Shulman, ed., Moving
Prescription Drugs to the Over-the-Counter Market (Tufts University Human Nutrition Research Center,
Boston).
Yarrow, G. (1995), "CED and EC member state industrial policy and the pharmaceutical industry", in:
A. Towse, ed., Industrial Policy and the Pharmaceutical Industry (Office of Health Economics, London).
Zuger, A. (1997), "Drug companies' sales pitch: Ask your doctor", New York Times, August 5, p. C1.
Chapter26
LIABILITY FOR MEDICAL MALPRACTICE
PATRICIA M. DANZON
The Wharton School, University of Pennsylvania
Contents
Abstract 1341
Keywords 1342
2. The theory of tort liability 1345
2.1. Tort liability with perfect information 1345
2.2. Tort liability with imperfect information 1347
2.2.1. Bias in the custom-based standard of care 1347
2.2.2. Uncertain medical and legal standards 1349
2.3. Other quality control mechanisms 1350
3. Empirical evidence on injuries and claims 1351
3.1. Adverse events and negligent injuries 1351
3.1.1. Causes of medical injuries 1353
3.2. Malpractice claims vs. negligent injuries 1354
3.3. Trends in malpractice claims 1355
3.4. The disposition of malpractice claims 1357
4. Malpractice insurance 1359
4.1. Liability insurance contracts: experience rating and co-payments 1360
4.2. Premium levels and availability 1362
5. Effects of liability on medical practice 1364
5.1. Formulating the analysis 1364
5.1.1. Theoretical issues 1364
5.1.2. Empirical issues 1365
5.2. Empirical evidence 1366
5.3. Defensive medicine 1368
6. Overall evaluation of the malpractice system 1369
6.1. Costs of administration and defensive medicine 1369
6.2. Deterrence benefits 1370
1340 P.M. Danzon
7. Traditional tort "reform" 1371

7.1. The structure of awards 1372
7.1.1. Limits on awards for nonmonetary loss 1372
7.1.2. Periodic payments 1374
7.1.3. Collateral source offset 1374
7.2. Reducing litigation costs 1374
7.2.1. Limits on contingent fees 1375
7.2.2. Alternative dispute resolution (ADR) 1376
7.3. Quality and standards of care 1376
7.4. Enterprise liability 1377
8. Radical alternatives 1378
8.1. No-fault programs for iatrogenic injuries 1378
8.2. Private contracting 1381
9. Liability under managed care 1382
9.1. Claims for denial of coverage 1383
9.2. MCO liability for negligent care 1385
9.3. The ERISA pre-emption 1386
10. Non-US experience 1386
10.1. The UK 1386
10.2. The Swedish patient compensation insurance (PCI) 1389
10.3. The New Zealand accident rehabilitation compensation and insurance scheme (ACS) 1392
11. Concluding comments 1395
Appendix 1396
Al. Optimal liability rules 1396
Al .1. First party liability (caveat emptor) 1397
A1.2. Strict third party liability 1397
A1.3. Negligence 1398
A2. Imperfect information 1399
A3. The demand for liability insurance 1400
References 1401
Ch. 26: Liabilityfor Medical Malpractice 1341
Abstract
Physicians are traditionally liable under a negligence rule of liability. Economic analysis
of liability rules, including malpractice, assumes that the primary function of liability is
injury prevention (deterrence). Compensation can be provided more efficiently through
other forms of social or private insurance. In theory, a negligence rule creates incentives
for efficient care, hence there should be no negligence, no claims and no demand for
liability insurance. In practice, the incidence of negligent injury has been estimated at
roughly one per hundred hospital admissions in the US and about one in seven physi-
cians is sued per year.
These discrepancies between the theory and actual operation of the negligence sys-
tem arise primarily because of imperfect information on the part of courts, doctors, pa-
tients, liability insurers and health insurers. Imperfect information and extensive health
insurance lead to biased and uncertain legal standards. Uncertain legal standards create
incentives for physicians to practice defensive medicine and incentives for plaintiffs and
defendants to invest in litigation, leading to high overhead costs, such that compensation
through the malpractice system carries a load of $1.50 per $1.00 of compensation.
Nevertheless, the extreme criticisms of the malpractice system are exaggerated. Mal-
practice premiums are less than 1 percent of total health care costs. There are no com-
prehensive estimates of defensive medicine costs; in any case such costs are likely to
decline with the growth of managed care. Although claim disposition exhibits both Type
1 and Type 2 errors, negligent injuries are much more likely to lead to a claim being filed
and payment to the plaintiff than non-negligent injuries, and awards are strongly related
to loss incurred. The limited empirical evidence of provider response to liability and
the deterrent effect of claims suggests - but cannot prove - that the net benefits of the
malpractice system may plausibly be positive. Nevertheless, reforms designed to reduce
inappropriate compensation and deter excessive litigation and defensive practice would
make the system more cost-effective.
The empirical evidence, based primarily in the US, includes studies of malpractice
injuries; physician response to liability; trends in claim frequency, severity (size), and
claim disposition; and the malpractice insurance market. Analyses of actual and pro-
posed reforms address tort reform, no fault, enterprise liability and optimal liability
under managed care. More limited evidence is available on the negligence regimes in
Canada and the UK, and the quasi no-fault regimes in Sweden and New Zealand.
1342 P.M. Danzon
Keywords
accidents, administrative costs, adverse events, awards, California malpractice study,

caps on awards, Caveat Emptor, claim frequency, claim severity, collateral source
offset, compensation, contingent fees, contractual approach, custom, defensive
medicine, denial of coverage, deterrence, disposition of claims, economic theory of
tort liability, efficient level of care, Employer Retirement Income Security Act
(ERISA), English rule of fee shifting, enterprise liability, experience rating of liability
insurance, fee for service, Harvard study, health care costs, imperfect information,
injuries (iatrogenic), insurance, insurance crisis, learned hand standard, litigation costs,
malpractice premiums, managed care, moral hazard, mutual, physician owned,
negligence rule, New York malpractice study, New Zealand accident compensation, no
fault, overhead costs, pain and suffering (non-monetary loss), periodic payment of
awards, physician response to liability, policy issues, positive analysis, practice
guidelines, prevention of injuries, regulation, scheduled awards, settlement out of
court, sociolegal risk, standard of care, statute of limitations, strict liability,
subrogation, Swedish patient compensation insurance, tort reform, uncertain standards,
verdict, workers' compensation,
Although medical providers have been subject to liability for malpractice for centuries,
malpractice only emerged as a major concern of physicians and of health policy over
the last three decades. Since the late 1960s, malpractice claim frequency (claims per
100 physicians) and claim severity (average amount per paid claim) have risen dramati-
cally but unevenly, leading to so-called "crises" in malpractice insurance markets in the
mid 1970s and mid 1980s. Doctors in the highest risk specialties and regions, such as
neurosurgeons or obstetricians in New York City, face annual premiums of $150,000-
$200,000, although the national average is roughly $16,000 (weighted average over
20 specialties) [Bovbjerg (1995)]. Malpractice premiums account for roughly one per-
cent of total health care spending, hence are not a significant contributor to the level
or growth of health care costs. However, there remain no good empirical measures of
the changes in medical care that are induced by liability - either cost-justified injury
prevention that the system is intended to encourage or defensive practices that are not
cost-justified and that are alleged to contribute significantly to total health care costs. In
the 1970s and again in the 1980s, many states enacted changes in their laws governing
legal liability and liability insurance. Federal legislation has been repeatedly debated but
so far not enacted, reflecting widespread disagreement as to the nature of the problem,
if any, and hence the desirability of change.
The traditional common law of medical malpractice holds health care providers li-
able for iatrogenic (medically-caused) injuries that are attributable to negligence. Un-
der a negligence rule, the plaintiff must show that he or she suffered an injury; that
the injury was caused by medical care; and that the provider's care deviated from due
care. For medical liability, due care is defined as customary practice of practitioners in
good standing or a significant minority of such professionals. Adverse outcomes that
are within the normal risks of customary medical care are to be borne by the patient,
possibly using first party insurance.
The theoretical analysis of medical malpractice liability is an extension of the eco-
nomic analysis of tort liability more generally [Posner (1972), Brown (1973), Shavell
(1980)].1 In that analysis, the primary function of liability is to influence incentives for
care and hence to reduce the rate of inappropriate accidents. Other potential functions of
liability are recognized, in particular, providing compensation to injured patients, jus-
tice and retribution. However, compensation can be provided more efficiently through
other forms of social or private insurance (see below), and economics has little to say
about retribution. Economic analysis has therefore focused on the deterrence function
of liability.
Liability rules are analyzed in terms of their ability to create incentives for first best
efficient levels of care, that is, for investment in precautions to the point where the
marginal cost is equal to the marginal benefit, in terms of reduction in expected injury
I Tort law refers to the set of legal rules and practices that govern wrongful injuries to person or property.
1344 PM. Danzon
costs. Given certain assumptions, this analysis yields the important conclusions that,
first, a negligence rule of liability can provide incentives for potential injurers to take
first best efficient care, and second, a well-functioning negligence system should in-
duce complete compliance, that is, there should be no negligence, no claims alleging
negligence and no demand for liability insurance [Shavell (1982)]. A key necessary
condition is that courts should define negligence as failure to take cost-justified precau-
tions. In that case, it is always cheaper for a potential injurer to prevent an injury that
would be deemed negligent than to insure against it.
The medical malpractice experience is seriously at odds with this prediction, of no
negligent injuries and no claims. The incidence of negligent injury has been estimated
at roughly one per hundred hospital admissions (Weiler, Hiatt et al. 1993). Whereas
before the 1960s, only one in seven physicians had ever been sued, now about one in
seven physicians is sued per year.
Economic theory suggests that the discrepancies between the negligence system in
theory and its operation in practice arise primarily because the decision-makers - courts,
doctors, patients, liability insurers and health insurers - lack the information that is
assumed by the models (Danzon 1991).
Imperfect and asymmetric information can lead to legal standards of care that are sys-
tematically biased and have high variance. Legal standards that are unpredictable and
open to influence can create incentives for physicians to practice defensive medicine and
incentives for plaintiffs and defendants to invest in litigation to influence the outcome,
leading to high overhead costs. Roughly forty cents of the malpractice insurance pre-
mium dollar is spent on litigation, 20 cents is insurance overhead and only forty cents
reaches the plaintiff as compensation. Whether the deterrence benefits of the malprac-
tice system are sufficient to outweigh its high administrative costs plus any net costs of
defensive medicine remains a critical but unresolved question.
The economic literature on medical malpractice falls into two broad categories. The
largest category is primarily empirical and positive in focus, aimed at providing evi-
dence on the actual operation of the malpractice system. It includes studies that attempt
to measure the number of malpractice injuries and claims; physician and patient re-
sponse to liability; determinants of the number and size of claims, claim disposition
and other issues related to the legal system; and the supply and pricing of malpractice
insurance. A second, smaller literature is directly concerned with the policy issues of
malpractice reform, including evaluation of actual and proposed changes. Most of the
literature draws on the US experience, which has provoked the fiercest policy debate and
generated the most reliable data and analysis. However Canada and the UK have also
experienced rising claims and premium costs in the 1980s, giving rise to proposals for
policy changes [Dewees, Coyte et al. (1989), Fenn (1993), Towse and Danzon (1998)].
In the early 1970s, Sweden and New Zealand replaced their traditional negligence sys-
tems with quasi no-fault systems of liability, and this approach has been considered as
a possible model in the US, the UK and Canada.
In this Chapter, Section 2 reviews the standard economic analysis of tort liability and
identifies the problems in applying it to medical care. Section 3 reviews empirical evi-
dence on the incidence of medical injuries, trends in malpractice claims and claim dis-
position. Section 4 discusses malpractice insurance contracts and market performance.
Section 5 reviews the evidence on the effects of liability on medical practice, including
defensive medicine. Section 6 summarizes the evidence on the overall costs and benefits
of the current system, and the case for reform. Sections 7 and 8, respectively, evaluate
the theory and evidence on traditional tort "reforms" (such as caps on awards) and more
radical alternatives such as no-fault plans and contracting out of liability. Section 9 re-
views the emerging issues and challenges posed by managed care. Section 10 reviews
the experience in the UK, Sweden and New Zealand; Section 11 concludes.
2. The theory of tort liability
2.1. Tort liability with perfect information
Accidents are a costly by-product of other beneficial activities, for example, medical
care or driving a car. The probability and size of losses can be reduced at a cost, either by
reducing the rate of risky activities (number of surgeries or miles driven per year) or by
taking costly precautions per unit of activity (care per surgery or per mile driven). Brown
(1973) showed that safety can be viewed as a good like any other. Efficient investment
in safety requires producing both the efficient level of safety and using the lowest cost
mix of inputs. The problem can be viewed most narrowly as minimizing the total social
cost of accidents, defined to include the cost of precautions and the cost of injuries. If
insurance is available and claims disposition is costly, the maximand is expanded to
reflect the utility cost of injuries, net of optimal compensation and transactions cost. An
even broader formulation maximizes a social welfare function that reflects the utility
value of the beneficial activities of which accidents are a by-product, in addition to the
costs of injuries, prevention and overhead (Shavell 1980).
In these economic models, alternative liability rules allocate the costs of accidents in
ways that affect private incentives for investment in safety - the "deterrence" function of
liability. The main rules under consideration are no liability (caveat emptor), negligence
and strict liability. Efficient rules are those that create incentives for socially optimal
care by all participants. Tort liability also provides a source of insurance and compensa-
tion to injury victims. Although the compensation function of liability features promi-
nently in the policy debate, economics has little to say about the equity of different
compensation rules. Economic analysis has been used to evaluate the efficiency of tort
compensation structure and amounts (Section 7) and the efficiency of providing com-
pensation through tort liability vs. either private or public first party insurance. On the
latter point, the conclusion is clear: liability is a much more costly insurance mechanism
than first party insurance (Section 6). The implication is that if liability is to be justified
as an efficient institution for dealing with risk, its deterrence benefits must outweigh its
added costs.
In analyzing the deterrent effects of different legal rules, it is useful to categorize
accidents as unilateral (optimal care is positive for just one party) or bilateral (optimal
1346 PM. Danzon
care is positive for both the injurer(s) and the victim). Medical injuries are generally as-
sumed to be unilateral, due to asymmetric information between providers and patients.
A second useful distinction is whether accidents occur between strangers (for exam-
ple, most automobile accidents) or between parties who are in a market relationship,
such as the patient-physician relationship for medical injuries, the employer-employee
relationship for workplace injuries or the consumer-producer relationship for product-
related injuries. Shavell (1980) provides a systematic analysis of the incentives for care
in these different contexts - unilateral or bilateral accidents, between strangers or mar-
ket participants - under alternative liability rules (including no liability, a negligence
rule and strict liability) and under different assumptions about information asymmetry.
The Appendix applies a similar model to medical malpractice.
For accidents between strangers in a non-market context, an accident is like any ex-
ternality: the injurer has no incentive to invest in any care in the absence of liability or
regulation. Modeling accidents between strangers as a non-cooperative game, Brown
(1973) shows that a negligence liability rule creates incentives for efficient care, pro-
vided that due care is defined by the "incremental Learned Hand standard". According
to this standard, defined by Judge Learned Hand, negligence occurs if "the loss caused
by the accident, multiplied by the probability of the accident's occurring exceeds the
burden of the precautions the defendant might have taken to avert it". 2 In other words,
negligence consists of failure to take precautions if their cost is less than the expected
damages averted. If defined in terms of marginal cost and benefit, this defines negligence
as failure to take efficient precautions.
However, if the potential injurer and victim are in a contractual relationship, as in
the physician-patent relationship in medical care, the value of liability depends on the
extent of information [Spence (1977), Shavell (1980)]. If customers are fully informed,
their valuation of safety is appropriately internalized to producers through the prices
that they are willing to pay for safer products. But if customers misperceive risks, a
producer who invests to make a product safer cannot recoup the investment through a
higher price. In the medical context, if the liability rule is caveat emptor and if patients
underestimate surgical risks in general and cannot monitor the individual surgeon's care,
there would be too many surgeries and too little care per surgery. Conversely, there
would be too little risk-taking if consumers overestimate risks. Since many medical
services are infrequently purchased, it is difficult to become an informed shopper until
it is too late. The existence of a core of informed shoppers will not necessarily assure
appropriate quality for everyone, because the product is patient-specific, not a mass-
produced, homogeneous good.
In theory, a well-functioning negligence rule could, given certain assumptions, create
incentives for optimal care per procedure. To achieve the optimal rate of risky proce-
dures, the definition of negligence should include liability for performing "unnecessary"
procedures, if patients nisperceive average risk. But the simple model of negligence li-
ability as an efficient system of deterrence assumes perfect information on the part of
2 US vs. Carroll Towing 159 Federal Reporter 2d 169 (1947).

courts, potential injurers and, ex post, of victims. The model assumes that courts cost-
lessly enforce the efficient standard of care, providers know the due care standard, and
that patients file a claim if and only if they are injured due to negligence. Under a per-
fectly functioning negligence rule there would be no negligence and no claims, since by
definition it is cheaper to prevent injuries that would be deemed negligent than to pay
for the resulting damages [Shavell (1982)]. There would also be no defensive medicine.
But with perfect information, there would be no need for negligence liability. More real-
istically, consumer information in medical markets is imperfect, which creates a prima
facie case for provider liability. Whether in practice exposing medical providers to tort
liability improves efficiency depends on the information available to decision-makers
under such a rule. If doctors, patients, courts and liability insurers lack good informa-
tion about appropriate medical care and legal rules, then the operation of the negligence
system in practice can diverge significantly from the theoretical ideal.
2.2. Tort liability with imperfect information
2.2.1. Bias in the custom-based standardof care
For professional liability of physicians, attorneys, architects and other professionals,

courts generally define due care as customary practice of practitioners in good standing,
acknowledging their own imperfect information. The explicit and independent use of a
Learned Hand cost-benefit calculus to define negligence is extremely rare. 3 By defer-
ring to custom, courts forego any attempt to correct any systematic bias in customary
care that results from consumer misperceptions in medical markets. Yet the correction
of such distortions is the main justification for liability assumed in the economic lit-
erature. A custom-based standard is likely to be systematically biased, relative to first
best efficient care, in precisely those circumstances where market failure is significant.
At best, a custom-based negligence rule could prevent significant deviations from the
(non-optimal) norms that consumers have come to expect.
The prevalence of comprehensive private or public insurance for medical care creates
further systematic bias in customary care. In the US, over 80 percent of the population
has some form of health insurance, typically with modest co-payment and premiums
unrelated to own use (experience rating). Most other countries have virtually universal
systems of public or quasi-public insurance, with even lower co-payments. Insurance
that reduces the point-of-purchase price of medical care to patients creates moral haz-
ard [Pauly (1968)], in the absence of other provider incentives or controls by insurers.
Norms of quantity and multiple dimensions of quality are likely to be distorted, relative
to the first best optimum.
The nature of the biases in customary care depends on the structure of insurance
coverage, in particular, the financial incentives created by provider reimbursement and
3 Helling vs. Carey 83 Wash. 2d 514, 519 P2d 981, 983 (1974).
1348 PM. Danzon
patient co-payments. Under traditional fee-for-service insurance with low co-payments

for patients, moral hazard leads patients to prefer overuse of costly procedures, rela-
tive to a first best optimum. Providers acting as good agents for patients have every
incentive to comply. On the other hand, strict fee regulation, as tends to occur in pub-
lic systems, creates incentives for physicians to reduce real resource intensity per en-
counter, leading to a high frequency of very short visits - as, for example, in Germany,
Japan and traditional US Medicaid. Thus fee-for-service reimbursement leads to a high
volume of reimbursable encounters and procedures, increasing the likelihood of errors
of commission. Errors of omission are also possible, however, particularly if fees per
encounter/procedure are regulated at low levels, since this creates incentives for physi-
cians to reduce resource input per encounter. Skimping on time and cognitive effort can
lead to errors in diagnosis and treatment.
Capitation forms of payment are increasingly being adopted under managed care in
the US and in some other countries, particularly for primary care physicians, to correct
the incentives for overuse created by fee-for-service reimbursement. Since the capitated
physician faces a positive marginal cost but receives zero marginal revenue per unit of
additional service or effort, capitation may create incentives for suboptimal quantity
and quality of care if patients have imperfect information about quality or face costs
of switching physicians. The deterrence value of liability may therefore in theory be
greater under capitation than fee-for-service [Danzon (1994b)], particularly for deter-
ring errors of omission. Note that defining an appropriate due care standard may be
particularly difficult for courts at a time of transition from fee-for-service to capitation,
as is occurring in the US. If the courts continue to define due care in terms of (pre-
sumably excessive) customary practice that developed under fee-for-service insurance,
liability will undermine the potential efficiency gains from moving to managed care.
Drawing the line between appropriate reductions in excess, which is the promise of
managed care, and inappropriate underprovision, which is its potential downside, may
require resort to an external standard other than customary care [Danzon (1997) and
Section 9 below].
Thus when both patients and insurers lack the information necessary to determine
appropriate care and to monitor the care actually delivered, the safe conclusion is that
customary care is likely to deviate systematically from optimal care, both in quantity
and quality. However, the direction and extent of bias depends on the level and form of
provider reimbursement, professional norms, the effectiveness of surrogate monitoring
of quality and outcomes, and other factors.
An alternative benchmark would recognize that the full-information, first best opti-
mum is not a relevant standard, given the costs of obtaining information and controlling
moral hazard in health insurance markets. The issue is then whether a liability standard
defined in terms of customary care might be second best optimal, given the information
available in insurance markets, However, US markets for health insurance and hence
medical care are distorted by the tax subsidy to employer contributions to private in-
surance and direct subsidies to public insurance Medicare and Medicaid. Subsidies to
health insurance drive a wedge between private and social costs. Medical norms will
Ch. 26: Liability for Medical Malpractice 1349
tend to reflect the distorted private costs, assuming that physicians act as reasonably
good agents due to professional norms and competition in medical markets. Managed
care is no antidote to subsidy-induced distortions since it is equally subsidized.
Thus in the case of medical care, the standard model of potential efficiency gains
from tort liability is at best an approximation. Liability may help enforce that actual
care conforms to customary norms which themselves are non-optimal due to imperfect
information and subsidies to insurance. If courts were to attempt to weigh social costs
and benefits, this would create severe conflict between legal standards and competitive
pressures in medical care markets that are driven by private costs and benefits, at least
in the US. This tension may be less in a system with public provision such as the UK
NHS, if doctors adopted treatment norms based on social rather than private costs and
benefits.
2.2.2. Uncertainmedical and legal standards
The literature on variation in medical practice norms within countries [Phelps (2000)]
concludes that much of this variation reflects differences in doctors' beliefs about appro-
priate care. Courts accommodate some difference in medical opinion (and possibly in
patient preferences) by recognizing "customary practice of professionals in good stand-
ing or a significant minority of such professionals". A legal standard that recognizes a
range of acceptable medical practice is potentially optimal, given uncertainty of medi-
cal opinion and differences in consumers' preferences and willingness to pay for care.
However, this necessarily contributes to variance in judicial decisions. A further impli-
cation of a custom-based standard is that appropriate care in each case must be defined
by medical experts. Although such experts owe a duty to the court, they are paid by
the litigants and are presumably selected to try to influence the outcome, contributing
further to uncertainty.
Craswell and Calfee (1986) show that a negligence rule with an uncertain standard
is likely to create nonoptimal deterrence incentives. The precise effect depends on the
bias and the variance of legal standards relative to efficient care. In this context, there is
some presumption for excessive care. With uncertainty, the physician cannot be sure of
avoiding liability simply by taking the required level of care. Because liability is all-or-
nothing, by incurring a small additional cost his probability of a large penalty may be
significantly reduced. Thus uncertain legal rules may be a significant factor contributing
to incentives for defensive medicine.
Imperfect information about medical and legal standards contributes to errors by pa-
tients and their attorneys in filing claims. The evidence from several studies in the US
indicates that many valid claims are not filed and many invalid claims are filed (Sec-
tion 3). Only 43 percent of claims filed with insurance companies receive any pay-
ment [Bovbjerg (1995) and sources cited therein]; the remainder are either dismissed
or dropped. Although negligent injuries are more likely than nonnegligent injuries to
lead to claims and valid claims are more likely than invalid claims to receive payment
[White (1994)], the system is far from perfect. The more variable are legal rulings, the
1350 PM. Danzon
greater the incentive for plaintiff and defense to invest in legal expense to influence the
outcome.
From the standpoint of physicians, invalid claims and uncertain legal standards gen-
erate a demand for comprehensive liability insurance, including legal defense insurance.
This contradicts the theoretical prediction (which assumes perfect information and ef-
ficient standards), that there should be no demand for liability insurance because it is
always cheaper to be non-negligent. Liability insurance would not interfere with deter-
rence if it were perfectly experience-rated, with premiums adjusted to reflect the actuar-
ial risk implied by the physician's actual level of care. In practice, however, malpractice
insurance premiums are based primarily on location, medical specialty and limits of
coverage; individual experience-rated adjustments are infrequent and rarely based sim-
ply on the number of claims filed or paid. Ellis, Gallup and McGuire (1990) have shown
that experience-rating based on a simple count of claims filed or paid would expose
providers to significant risk of inappropriate surcharges, because of the large number of
false positive claims [see also Sloan, Mergenhagen et al. (1989)].
Nevertheless, the widespread purchase of liability insurance with minimal experience-
rating does not necessarily imply that deterrence incentives are nonexistent or even sub-
optimal, as some have argued. Physicians with consistently bad claims experience, ad-
justed for specialty, may face a surcharge on their premium, restrictions on their scope of
covered practice (for example, no surgery) or be denied coverage by low cost insurers.
However, an important implication of pervasive liability insurance with no formal coin-
surance and minimal experience-rating, is that deterrence probably derives more from
the uninsured time, embarrassment and reputation costs faced by physicians, which
plausibly depend more on claim frequency of claims than on size of awards. This has
implications in evaluating reform proposals, as discussed below.
2.3. Other quality control mechanisms
Liability is only one of several mechanisms that may correct the distortions that re-
sult from asymmetric information in medical markets. Altruism, professional or ethical
concerns may motivate physicians to act as better agents for patients than would be
predicted by models that assume purely self-interested income maximization (Danzon
1994b). In most countries, other regulatory mechanisms such as state licensure and
boards of medical quality assurance provide coarse screens to eliminate persistent in-
competence or misconduct by medical providers, although in practice implementation
may be limited [Sloan, Mergenhagen et al. (1989)]. Indirect market mechanisms such as
hospital credentialing committees, reputation and referral networks may provide some
substitute for informed consumers. Performance measurement, to permit monitoring of
providers by patients, employers and other third party payers through outcomes report-
ing and other quality indicators, is a rapidly growing field in the US, the UK and other
countries [for example, Eddy (1998)]. Since professional ethics, regulation and indirect
market forces all provide some correction for the market failure that results from imper-
fectly informed consumers, liability should only be used if it is cost-justified (benefits
exceed costs) and cost-effective relative to these alternative corrective mechanisms [see
Spence (1977) and Appendix].
3. Empirical evidence on injuries and claims
3.1. Adverse events and negligent injuries
The theoretical assumption, that an accident is a well-defined event, is reasonable for

automobile, product and acute workplace injuries that can be causally related to specific
activities. However, a patient who seeks medical care is already either ill or injured and
medical care cannot guarantee perfect health and immortality. Thus defining a medical
"injury" and a causal relation between an imperfect medical outcome and the medical
treatment is problematic. The amount and quality of medical services shifts the prob-
ability distribution of outcomes, hopefully with more weight for good outcomes but
possibly also increased risk of some bad outcomes. For example, surgery reduces the
risk of death from appendicitis but entails risk of other infections or complications, even
if competently performed. Identifying a medical injury therefore requires an assumption
about the expected distribution of outcomes, which presupposes an assumption, implicit
or explicit, about the expected or appropriate level of medical care. Defining a negligent
injury requires specifying a particular threshold within the range of expected care. Given
the complexity of defining and recognizing medical injuries and negligent injuries, it is
not surprising that comprehensive data on the number of iatrogenic injuries, with or
without negligence, are not routinely collected.
The best available information is from two surveys of medical records of hospitalized
patients, the first in California in 1974 [Mills (1977), hereafter California study] and the
second in New York in 1984 [Weiler et al. (1993), hereafter Harvard Study]. In both
studies, a large sample of hospital records was reviewed by experts in law and medicine
- 20,864 patients in 23 hospitals in California and 31,429 patients in 51 randomly se-
lected hospitals in New York- to identify the incidence of injury due to medical care and
the subset of these injuries caused by negligence on the part of a health care provider.
The California study concluded that 4.65 percent of hospitalized patients suffered an
injury due to medical care; of these, 17 percent (or one in 126 patients) involved neg-
ligent injury. In the New York study, 3.7 percent of patients suffered an injury due to
medical care and of these, 28 percent were attributed to negligence, that is, roughly 1
percent of all hospitalized patients suffered a negligent injury. If extrapolated to all New
York hospital discharges in 1984, this would imply that 98,610 hospitalized patients per
year suffered a medical injury. Of these, 57 percent were minimal with recovery in one
month, and an additional 14 percent lasted less than 6 months. However, over 6,000
(7%) patients suffered permanent disabilities and another 13,400 (14%) died. If extrap-
olated to the US population as a whole, this would imply 150,000 iatrogenic fatalities
annually, more than half of which may be attributed to negligence. This far exceeds the
rate of motor vehicle fatalities (50,000 deaths per year) and occupation related fatalities
1352 PM. Danzon
(6,000 per year). These studies and the resulting estimates omit injuries that occur in
ambulatory settings, unless they resulted in hospitalization.
However, these startling figures on iatrogenic injury rates should be viewed with cau-
tion. As the authors note, a substantial (but undetermined) proportion of the patients
whose deaths were attributed to medical management were seriously ill and many would
have died from their underlying illness in months, days or even hours, whereas most vic-
tims of automobile or workplace injuries are healthy. Second, this apparently high rate
of iatrogenic injury in part reflects the broad definition of injury used in both studies.
The Harvard study defined an iatrogenic injury as "any disability caused by medical
management that prolonged the hospital stay by at least one day or persisted beyond
the patient's release from hospital".4 Thus the count of injuries implicitly reflects the
standard of care considered appropriate by the analysts, including their expectation of
an appropriate length of stay and reasonable medical outcome. Any (negative) deviation
from the expected outcome is considered an injury.
A negligent adverse event was defined as the consequence of treatment that failed
to meet the standard of the average medical practitioner. The Harvard study used a
modified locality standard "similar to that employed by peer review organizations ...
we did not require that a reviewer imagine that all the resources of, say, a tertiary care
teaching hospital were available at small community hospitals" (p. 35). Similarly, the
California study defined negligence in terms of the standards likely to be applied by a
jury. Since neither study attempted to define negligence by weighing marginal costs and
benefits of additional precautions, the resulting count of "negligent injuries" does not
necessarily correspond to the number of economically inappropriate injuries, for which
costs of precautions are less than expected damage costs.
As in the California study, the Harvard study found that the percentage of the injuries
attributed to negligence was higher for serious injuries than for minor injuries - for ex-
ample, 51 percent of deaths were attributed to negligence, compared to 23 percent of
impairments lasting less than 6 months. While it is possible that injuries caused by neg-
ligence have systematically worse outcomes than injuries not involving negligence, this
finding could also be influenced by the definitions. Specifically, the definition of neg-
ligence, which invokes legal standards, may implicitly set a higher threshold of injury
severity than that required for an adverse outcome to count as a nonnegligent injury.
These US-based studies of iatrogenic injury have been used to estimate injury rates
for other countries [for example, Smith (1990)]. Such extrapolations are problematic
because of cross-country differences in standards of care, thresholds of legal negligence
and - presumably - in rates of iatrogenic and negligent injury. The direction of bias is
4 Weiler et al. (1993, p. 35) refer to "those adverse events that were the unintended or unexpected harmful
consequences of medical intervention, and that prolonged the hospitalization beyond the time required by the
underlying illness and/or caused disability at the time of hospital discharge or death". This would exclude
adverse but inevitable consequences of appropriate treatment, such as amputation of a limb to treat bone
cancer.
uncertain a priori. The higher claim frequency and premium cost in the US should cre-
ate stronger deterrence incentives for US doctors and hospitals. Financial and agency
incentives may also differ. More fundamentally, any attempt to compare rates of adverse
events and negligent injuries must specify the standard of care implicit in defining in-
juries and negligence, and whether any observed differences in estimated injury rates in
fact reflect differences in care patterns that are appropriate, given differences in costs,
preferences and other factors, or whether they reflect true differences in rates of inap-
propriate injury, conditional on different standards.
3.1.1. Causes of medical injuries
The causes of malpractice injuries are not well understood but both patient and provider
characteristics play a role. In the Harvard study, age over 45 and Medicaid payer status
were associated with significantly higher rates of adverse events; controlling for these
and other factors, race and gender were not significant.5 Age over 65 was also associ-
ated with significantly higher percent of adverse events due to negligence. Weiler et al.
comment that "although their higher injury rates can be attributed in part to the frail
physical state of older patients, the higher negligent injury rates may also be the result
of the quality of care the elderly receive from health care providers". It may also re-
flect the fact that the definitions of adverse events and negligence are not age-specific.
By contrast, economic definitions of optimal injury rates would take into account the
age/condition-specific distributions of expected outcomes and costs of preventing ad-
verse outcomes. If the costs of preventing mishaps increase with age and severity and/
or the expected benefits to precautions decline, then optimal injury rates would increase
with age and illness severity.
For purposes of prevention, an important issue is whether negligent injuries are
caused largely by occasional inadvertent lapses of many, normally competent providers
or by a minority of incompetent, physicians and low quality hospitals. (Of course this as-
sumes, following standard legal practice, that all providers should be held to a common
standard of care, whereas if costs and prices vary, standards of care would optimally
differ.) Studies consistently show that both factors play a role. For the hospitals in the
Harvard study, the percent of adverse events due to negligence averaged 25 percent,
but ranged across hospitals from 1 percent to 60 percent. Controlling for patient age
and DRG category (a proxy for illness severity), the factors associated with relatively
high rates of adverse events were university teaching status, urban location, and small or
medium size (relative to large). Factors associated with a high proportion of injuries due
to negligence were high proportion of minority discharges (positive), university teach-
ing and proprietary status (negative). Resources, case mix, location and related factors
thus appear to affect outcomes.
5 Low income was significantly negatively related to adverse event rates; however this is presumably pos-
itively correlated with Medicaid payer status which is positively related to adverse event rates [Weiler et al.
(1993, p. 48)].
1354 P.M. Danzon
Similarly, analysis of claims experience of groups of physicians indicate that, after

controlling for medical specialty, the distribution of claim frequency is more concen-
trated than would be expected based on chance alone [Rolph (1981), Nye and Hofflan-
der (1987), Sloan, Mergenhagen et al. (1989), Ellis, Gallup et al. (1990)], as discussed
further in Section 4.1.
In the Harvard study, diagnostic errors accounted for only 8 percent of adverse events,
but 75 percent of these were attributed to negligence. Drug-related events accounted for
19 percent of all adverse events, but these were less frequently attributed to negligence
than most other categories. Note that despite the Harvard study's initial sample of over
30,000 records, there are only 1,278 adverse events and 306 negligent adverse events.
This small sample limits the scope for detailed analysis of causation. Identifying the
medical interventions that are most at risk of resulting in negligent injury would require,
in addition, data on the underlying frequency of treatments, which was not part of the
study.
3.2. Malpractice claims vs. negligent injuries
The simple theoretical models of tort liability assume that a claim is filed if and only
if a negligent injury occurs. In reality, several studies suggest that both false negatives
(failure to file valid claims) and false positives (claims filed without negligent injury)
occur frequently in medical care.
The California study did not directly compare claims to injuries. However, a compari-
son of the number of malpractice claims filed in California relative to the estimated num-
ber of negligent injuries concludes that at most one in ten of negligently injured patients
filed a claim, assuming no false positive claims. Only 40 percent of these claimants
received compensation through the tort system (Danzon 1985a). The Harvard study
directly compared filed claims to identified injuries. The total number of malpractice
claims was only about 15 percent of the number of negligent injuries. 6 However, only
2 percent of patients identified as having sustained an injury due to negligence filed a
claim. This suggests a large number of both false positive and negative claims, at least
relative to the definition of negligence and the data available to this study. The ratio of
claims to negligent injuries was much higher for serious injuries: roughly one claim was
filed for every three such injuries and one in six was paid [Weiler, Hiatt et al. (1993)].
The California data also indicated a lower ratio of claims to injuries for minor com-
pared to major injuries, and lower for persons over 65. These findings are consistent
with a simple economic model of the decision to file a claim, with fixed costs of filing
and a payoff that depends on economic loss, hence increases with injury severity and
decreases with patient age.
6 The definition of a claim includes claims filed by patients with insurance companies, even if no legal suit
was filed. Multiple claims per incident are aggregated. In the Harvard study, there were roughly 1.5 provider
claims per patient incident.
3.3. Trends in malpractice claims
Several studies have examined the determinants of trends in claims over time and per-
sistent differences across states and countries. Although malpractice liability has existed
for centuries, such actions were rare until the late 1960s. In the US from the late 1960s
to the mid-1980s malpractice claim frequency (number of claims per 100 physicians)
increased at more than 10 percent a year. Claim severity (the average payment per claim
paid) increased at roughly twice the rate of general inflation [Danzon (1986)], with some
evidence of disproportionate growth for the highest stakes cases [Shanley and Peterson
(1987)]. Claim frequency reached a peak of about 17 claims per 100 physicians in 1986,
and then stabilized around 15 claims per 100 physicians per year. 7 Failure to anticipate
the surge in claims precipitated "crises" in liability insurance markets in the mid-1970s
and mid-1980s.
A simple economic model views the decision to file a claim as an investment with
an uncertain payoff that depends on the nature of the injury, the legal rules defining
negligence and compensable damages, and the costs of filing [Danzon (1984a)]. This
model implies a simple econometric model in which the frequency of claims per capita,
at the state level, depends on the frequency and characteristics of medical treatment; the
legal rules that affect probability of winning and expected award, and the costs of legal
input.
The empirical evidence confirms that the growth in claim frequency in the 1960s
and early 1970s was significantly related to the increase in surgical interventions which
increased the number of adverse outcomes that could be attributed to medical care. At
the same time, pro-plaintiff shifts in legal doctrine increased the expected payoff to filing
a claim, by increasing the grounds or reducing the cost of a showing of negligence. For
example, the abolition of the locality rule substituted a statewide or national standard
for a local standard of due care, which plausibly increased the number of injuries that
would be deemed negligent. Perhaps more important, the move to a national standard
meant that out-of-state experts could testify as to the standard of care, which allegedly
broke down the "conspiracy of silence" that prevailed when local experts were required
to testify. The abolition of charitable and government immunity exposed voluntary and
government hospitals to suit. The doctrine of respondeatsuperiorextended the liability
of hospitals for actions of their employees and staff physicians; and requirement to
obtain the patient's informed consent to treatment was increasingly defined in terms of
the information that a reasonable patient would want, rather than what was customary
for physicians to provide. But these factors had run their course by the mid-1970s and
cannot explain claim growth in the 1980s.
In response to the malpractice insurance crisis of the mid-1970s and again in the
1980s, many states enacted one or more tort reforms.8 For example, 18 states adopted
7 These figures from Bovbjerg (1995) are based on the experience of The St. Paul Fire and Marine Insurance
Company, which is the largest writer of malpractice insurance, operating in 42-43 states. More recent data
are not maintained on the same basis. The experience of other insurers may differ.
8 Tort law and insurance regulation are traditionally areas of state jurisdiction.
1356 PM. Danzon
caps on awards or collateral source offset before 1985 and 23 states adopted these re-
forms between 1985 and 1990 [Kessler and McClellan (1996)]. 9 Estimating the effects
of these changes is problematic because each state's legislation is unique to a degree.
Moreover, reforms were often enacted in response to "crisis" conditions, which raises
the possibility that estimates of effects of reforms may be biased due to endogeneity
and other, unobserved state characteristics. Nevertheless, the findings from three empir-
ical studies that use data from 1975-1984 are reasonably consistent [Danzon (1984a,
1986), Zuckerman, Bovbjerg et al. (1990)]. Caps and collateral source offset reduced
claim severity by 19-39 percent and 11-50 percent, respectively, relative to what it
would have been in the absence of the reforms. Shorter statutes of limitations reduced
claim frequency - one year off the statute of limitations for adults is estimated to reduce
claim frequency by 8 percent. Collateral source offset has also reduced claim frequency
(by 14 percent) presumably because of the feedback effect from lower expected awards
to reduced incentive to file claims. Other reforms do not seem to have had significant
effects.
Although medical and legal factors account for some of the trends and interstate
differences in claim frequency and severity, much remains unexplained. The similarity
of claims filed per negligent injury in the California and Harvard studies suggests that
claim growth in the intervening years cannot simply be attributed to the "catch-up"
filing of claims for a larger percentage of negligent injuries. Growth in the number of
lawyers per capita is not a statistically significant explanatory factor, after controlling
for other attributes of lawyer-dense areas. Danzon (1984a) found that urban areas tend
to have much higher claim frequency and severity, but that this urban phenomenon
could not be explained by specific observable characteristics of urban areas such as
income, unemployment, welfare recipiency or population turnover rates (intended as
a proxy for the "physician-patient relationship"). The growth in malpractice litigation
coincided with abnormal growth in other areas of tort law, notably product liability,
suggesting a role of common legal or social factors. However, the growth in product
claims can plausibly be attributed to pro-plaintiff changes in product liability rules [see,
for example, Henderson (1988)], whereas no comparable legal changes occurred for
medical cases. It is possible that numerous subtle changes in case law and rules of
evidence made it easier for a plaintiff to establish a cause of action and get to a jury,
which effectively reduces the expected cost and increases incentives to file marginal
claims. Such hypotheses are plausible but very hard to test.
An interesting but unanswered question is whether the increase in claim frequency
is largely a response to higher expected awards, leaving the increase in awards as the
main factor to be explained. Evidence from workers' compensation [for example, Butler
(1983)] shows that claim rates respond positively to benefit levels. A similar supply
9 Under traditional damages rules, the tort award is not reduced by the amount of compensation available to
the plaintiff from other (collateral) sources. Collateral offset laws provide for the tort award to be reduced for
some forms of private and public insurance.
response is plausible for medical malpractice, although much harder to measure because
malpractice payment levels are not statutorily determined; rather the average observed
award is endogenous, reflecting the mix of cases actually filed. In a simple model of the
settlement process, the settlement amount depends on the probability of plaintiff verdict
(p) which depends on the degree of negligence N; the size of verdict (V) if successful
which depends on damages (D) and rules of compensation; the litigation costs of going
to verdict for defense and plaintiff (Cd, Cp); and a bargaining parameter (g):
S = p(N)V(D) + (Cd - Cp) g . (3.1)
Thus the observed mean payment S reflects not only the legal rules of damages
V(D) but also the distribution of damages and plaintiffs' probability of winning, litiga-
tion costs, all of which are endogenous, depending on the distribution of cases filed.
The rate of growth of malpractice claim frequency and severity was as high in Canada
and the United Kingdom in the 1980s as in the United States [Dewees, Coyte et al.
(1989), Danzon (1990a, 1990b)]. But in 1987 physicians in the United States were still
at least 5 times more likely to be sued than physicians in Canada or the UK. The mean
payment was somewhat higher in the US than in Canada. However, the similarity of
mean payments does not permit inferences about cross-country differences in net com-
pensation for the same injury for several reasons. First, US awards generally include
past and future medical expenses, whereas tort awards in the UK and Canada are net
of medical costs that are borne by public health care systems. Second, in the US, the
reported award is gross of the attorney's contingent fee (typically one third), whereas
the UK and most Canadian provinces apply the English rule, that the loser pays costs.
Third, the mean observed payment depends on the actual mix of cases filed and the in-
centives to settle out of court, all of which may differ across countries. In particular, to
the extent that the higher frequency of claims in the US reflects disproportionately more
minor injuries or cases of dubious merit (low p V) which settle for a relatively low pay-
ment, the mean observed payment in the US provides a downward biased estimate for
the expected compensation for the case mix in Canada or the UK, where the absence
of contingent fees is likely to deter more cases with low damages or low probability
of winning. It is interesting that the growth in malpractice claim costs has occurred in
Canada and the UK despite the fact that these countries have limits on awards for pain
and suffering, have lower medical costs and lower rates of growth of medical costs,
and do not permit contingent fees for plaintiff attorneys - all factors commonly cited to
explain the growth of US malpractice claims.
3.4. The disposition of malpractice claims
Several studies have examined the disposition of malpractice claims, to determine how
far the outcome conforms to legal principles and reasons for deviations [Danzon and
Lillard (1983), Farber and White (1991), Sloan and Hoerger (1991)]. Claim disposition
can be modeled as a process in which the litigants form expectations about the outcome
1358 P.M. Danzon
at verdict (probability and size of award) and expected costs of litigating to verdict.
A necessary (but not sufficient) condition for settlement is that the difference between
the expected outcomes is less than the sum of the litigation costs. More information
accrues as time and the disposition process advance. Variants of this model introduce
risk aversion, strategic behavior, etc. An important implication of the simple model
is that the disposition process involves nonrandom selection of cases to close at each
stage of disposition. In particular, cases that go to verdict are disproportionately those
in which the plaintiff or defendant tends to overestimate his chances of winning and in
which the stakes are sufficiently large for the difference in expectations to exceed the
litigation costs.
These various studies of claims disposition show similar patterns. Of the claims filed
with insurance companies, 39.6 percent are closed without filing a legal suit, 53.6 per-
cent are closed after suit but before trial, 1.6 percent are closed during trial, 3.2 percent
are closed at verdict and 2.0 percent are closed on appeal. Overall, only 43 percent of
claimants receive any payment [Bovbjerg (1995) and sources cited therein]. As the dis-
position process proceeds, claimants obtain more information about the likely success
of their claims, which contributes to the high percentage dropped or settled before ver-
dict. Sloan et al. (1991) show that claims with bad news in the patient's chart from the
plaintiff's standpoint are more likely to drop their claims [see also Danzon and Lillard
(1983)].
Negligent injuries are more likely than nonnegligent injuries to lead to claims and
valid claims are more likely than invalid claims to receive payment [Weiler et al. (1993),
White (1994)], but the system is not perfect. Farber and White (1991) examine claim
disposition using data on a small sample of institutions with an evaluation of the cause
of the injuries by independent reviewers. They conclude that negligence was present
in 35 percent of claims, not present in 42 percent, with the remainder uncertain. For
claims with negligence, the probability of receiving compensation was 0.66 and the
mean payment was $205,000; for claims without negligence, the probability of payment
was only 0.16 and the average payment was $41,800. White (1994) reviews data from
several studies and concludes that the probability of a claim is 0.026 per negligent injury,
0.01 per non-negligent injury and 0.001 per noninjury. This much higher probability of
suit for negligent treatment than for non-negligent treatment should provide a significant
deterrent effect, despite the high overall error rate in claiming.
These studies consistently show that the size of awards and settlements is strongly
influenced by the plaintiff's economic loss. Jury awards, which are the publicly visible
component of the malpractice system, are a tiny but self-selected and atypical subset of
claims, including disproportionately claims with high stakes and uncertain merit. Given
the selection process in which cases with strong evidence of negligence tend to settle,
it is not surprising that doctors win over two thirds of the cases taken to verdict but
that awards are very large when the plaintiff prevails. Of cases that receive some pay-
ment at verdict or in settlement, the mean award is over $100,000 but the median is
under $50,000. The distribution of payments is right skewed (approximately log nor-
mal), which partly reflects the underlying distribution of injury severity, with modest
losses for the majority of injuries but a few extremely severe injuries Danzon and Lil-
lard (1983)]. The largest awards are most likely to be reduced either by the judge after
trial or on appeal [Shanley and Peterson (1987)].
These findings suggest that the most extreme criticisms of the tort system as a random
lottery are exaggerated. The legal system appears to be quite effective at eliminating in-
valid claims, paying either zero or relatively small amounts to the majority. Moreover,
the apparent shortfall between claim rates and negligent injury rates overall does not
necessarily imply that deterrence incentives and compensation are too low or that re-
forms should be designed to stimulate more claims. Given the existence of professional
norms and other regulatory and market-driven quality assurance mechanisms, the op-
timal deterrence incentive through the tort system is certainly less than without these
alternative sanctions, although it cannot be determined precisely. Ignoring deterrence
concerns and focusing on the compensation function of tort, compensating small claims
through the tort system is probably not cost-effective, given other social and private
insurance programs that provide appropriate compensation for minor injuries at lower
overhead cost. In the Harvard study, nearly 80 percent of the patients who suffered a
negligent injury but did not sue were either fully recovered within 6 months or were
over 70 when the injury occurred, suggesting relatively small compensable damages.
Although the optimal rate of claims per negligent injury remains an unsettled ques-
tion, given other deterrence and compensation mechanisms, it is clear that a reduction
in number of false positive claims and more speedy elimination of those that do occur
would improve the efficiency of the system.
4. Malpractice insurance
The supply of medical malpractice insurance violates several of the predictions of stan-
dard insurance theory, two of which are addressed here. First, the structure and pricing
of liability insurance contracts makes little use of contractual terms or rating provisions
designed to restrain moral hazard on the part of the insured, although in principle the
risk is within the control of the insured - if not, the deterrence justification of tort liabil-
ity must be rejected. In particular, malpractice insurance rarely requires co-payment in
the form of deductibles or coinsurance, which are common in first party insurance such
as health insurance, and premiums are not generally experience-rated, based on prior
claims experience, as is common in automobile liability insurance. Second, although
theory predicts that insurance should be available at a price equal to the expected loss
plus an expense loading, in fact the price and availability of malpractice insurance has
been extremely erratic. The mid-1970s "crisis" was associated with premium increases
of over 300 percent in some states and total withdrawal of commercial carriers in other
states where regulators denied requested rate increases. The mid-1980s witnessed an-
other so-called "affordability crisis". Availability was less problematic, thanks partly
to the supply-side changes that occurred in response to the 1970s crisis. These include
the formation of provider-owned mutuals, reciprocals, hospital-owned captive compa-
nies and risk retention groups that now write over 50 percent of malpractice insurance
1360 PM. Danzon
coverage, and joint underwriting associations that, like high risk pools in other lines
of insurance, are state-mandated suppliers of last resort. In addition, the risk borne by
insurers was reduced by the switch from occurrence coverage, which covers all claims
arising out of practice in the policy year, to claims made coverage, which pays all claims
filed in the policy year, regardless of the practice year in which the alleged injury oc-
curred, provided that the insured was covered by that insurance company at the time
(see Section 4.2).
4.1. Liability insurance contracts: experience rating and co-payments
Shavell (1982) shows that liability insurance need not interfere with the deterrence in-
centives of liability if premiums are perfectly experience rated, that is, the price of
insurance accurately reflects the insured's expected loss. In practice, malpractice insur-
ance has less experience rating than other insurance lines such as automobile liability
or workers compensation, where premiums are usually automatically adjusted to reflect
adverse claims experience. Malpractice insurance rates are a multiplicative function of
limits of coverage (for example, $1 million per occurrence, $3 million total for the pol-
icy year); medical specialty; and geographic location. For example, the territorial rate
for a base class and basic limits is multiplied by specialty differentials and excess limits
factors to obtain rates for other specialties and higher limits of coverage. Rates are gen-
erally not based on volume of business, except for a crude part time adjustment, and are
not automatically surcharged for claims [Danzon (1985a), Sloan (1991)].
Several studies have confirmed that the distribution of claims, conditional on medical
specialty, is highly skewed, with a small number of physicians accounting for a larger
number of claims than would be expected if the probability of a claim were uniform
and the judicial process entirely random [Rolph (1981), Nye and Hofflander (1987),
Sloan, Mergenhagen et al. (1989)]. These findings are consistent with the hypothesis
that malpractice claims are disproportionately due to minority of "bad apple" providers.
However, at least part of the variation in claims experience could reflect unobserved
variation in case mix or volume. Consistent with this interpretation, Sloan et al. (1989)
find that board certified physicians and physicians who work longer hours have more
claims. The evidence is inconclusive on how far claims are triggered by poor commu-
nication with patients as opposed to inferior clinical care. Hickson et al. (1994) find
that physicians with a prior history of malpractice claims are more often the subject
of subsequent patient dissatisfaction over inadequate time and explanation than physi-
cians who had never been sued. Entman et al. (1994) find that prior claims experience is
unrelated to subsequent technical quality of care. Taken together, these findings might
be interpreted to suggest that poor communication is a more important determinant of
claims than poor clinical care. However, an alternative possible explanation is that de-
terrence is effective for clinical care, such that physicians with prior claims experience
perform better in the future, whereas communication skills are not improved.
The frequency of claims per physician is significantly higher for surgical specialties
than for non-surgical specialties, although it seems unlikely that surgeons are consis-
tently more careless than non-surgical specialists. A plausible explanation is that sur-
gical errors are more likely to be severe and causal connections to treatment are more
obvious. 10 This suggests that courts distinguish imperfectly between adverse outcomes
due to bad luck vs. negligence.
The findings of nonrandom distribution of claims has led some to argue for more
experience rating. But Ellis, Gallup and McGuire (1990) show that rating based on
Bayesian conditional means with five years of experience would move premiums only
modestly towards actuarially fair rates on average, while introducing inequities between
physicians with identical underlying risk and exposing physicians to considerable finan-
cial risk of inappropriate surcharges. They estimate that under such a rating scheme, a
single paid claim would result in a four-fold increase in premiums for most medical
specialties. Similarly, Rolph et al. (1991) find that 5 years' prior claims experience has
only modest predictive power for future claims experience. Thus, if judicial error is sig-
nificant, risk aversion would explain the lack of demand for experience-rated policies.
Companies that do base rates on prior experience often conduct an independent review,
rather than impose an automatic surcharge for all claims [Schwartz and Mendelson
(1989)]. The demand for community-rated policies may thus be viewed as insurance
against the risk of error by claimants, the courts and the settlement process [Danzon
(1985b)].
A related apparent puzzle is the relatively infrequent use of deductibles or other co-
payments commonly used in other lines of first party insurance to control moral hazard.
Several factors may limit the demand for policies with monetary deductibles and co-
payment. First, being sued entails uninsurable costs of time, in addition to anxiety and
threat to reputation; these uninsured costs are probably equivalent to a deductible of
several thousand dollars. Second, the potential for a claim in excess of the policy limit
implies additional uninsured risk. Third, since liability insurance covers legal defense
in addition to indemnity payments, a deductible would reduce the malpractice insurer's
incentives to defend claims that could be settled within the limits of a deductible. Fourth,
the sorting of physicians into companies acts as a crude form of experience rating, if
lower priced insurers with stringent underwriting standards reject physicians with bad
prior claims experience. At the limit, physicians who lose their insurance in the standard
market may obtain coverage from surplus lines carriers, who charge high premiums,
and impose large deductibles and restrictions on coverage [Schwartz and Mendelson
(1989)]. Whether these hidden uninsured costs add up to less or more than the socially
10 Weiler et al. (1993) found significant differences in the adverse event rate across medical treatments-
for example, cardiac surgery had three times the adverse event rate (10.8 percent) of general medicine (3.6
percent). However, the differences inproportion of injuries due to negligence were not statistically significant.
Surgical complications accounted for 47 percent of all adverse events, but the percent of these attributed to
negligence was only 17 percent, compared to 37 percent for non-surgical adverse events. Weiler et al. (p.53)
conclude "Although the total number of negligent injuries inflicted by surgeons was higher than the number of
injuries caused by internists, the difference appears to reflect the complexity and riskiness of the procedures
performed by the two groups, and thence the very different consequences of momentary lapses".
1362 PM. Danzon
optimal degree of co-payment is an important but unanswered question. Imposing more

co-payment or experience rating by law would increase uninsured risk for physicians
and could lead to defensive responses, such as refusal to take high risk patients.
Even if it could be shown that the great majority of physicians take appropriate care
most of the time and that most negligent injuries are the result of occasional mistakes
by otherwise competent physicians, it certainly would not follow that liability serves no
useful deterrence purpose. It is possible that liability has a negligible marginal effect but
the average effect is large, such that if liability were eliminated, more physicians would
be less careful and make more frequent mistakes. A related argument is that liability
is unnecessary because of alternatives such as state professional review mechanisms,
hospital quality assurance programs etc. Although in theory these mechanisms could be
substitutes, in practice they may be complements and depend on the liability system.
For example, casual evidence indicates that hospital and other peer review procedures
have been strengthened in direct response to liability. Sloan (1989) report that fewer
than 10 percent of physicians with adverse claims experience were disciplined in any
manner by professional review boards.
4.2. Premium levels and availability
Malpractice insurance premiums appear to follow a cyclical pattern, with periods of

sharp rate increases and limited availability followed by periods of flat or falling premi-
ums. Although over the long term the increase in premiums is explained by the rise in
claim costs, the short term cycle in premiums is more erratic than the trend in claims.
In the early seventies, premium rates initially lagged behind rising claim costs, neces-
sitating sharp premium increases in the mid-1970s of over 300 percent in some states
and withdrawal of traditional insurance carriers in other. In the late 70s, claim costs sta-
bilized and insurance rates fell in real terms. Premiums increased sharply again in the
mid 1980s. Similar "crises" occurred in the mid 1970s and 1980s in other "long tailed"
lines of liability insurance, such as product liability, municipal and environmental lia-
bility, for which claims may be filed many years after the action that allegedly caused
the injury. There is concern that the stable rates so far in the 1990s may partly reflect
aggressively competitive pricing that may ultimately lead to insurer losses followed by
sharp premium increases.
A considerable literature has analyzed the causes of these apparent cycles and crises
in long-tailed insurance lines [Cummins and Outreville (1987), Priest (1987), Harring-
ton and Litan (1988), Winter (1988), Harrington and Danzon (1994), Cummins and
Danzon (1997)]. Malpractice, product liability and other long-tailed insurance lines
have several characteristics that limit the ability of insurers to eliminate uncertainty
simply by writing a substantial share of the relevant risk pool. First, the loss distribution
per risk is extremely skewed. For malpractice, 5 percent of claims account for over 50
percent of dollars paid [Danzon and Lillard (1983)]. This skewness, combined with the
small number of policyholders in a given risk class in many jurisdictions, implies con-
siderable intertemporal variance in the mean loss per insured. Second, the payout tail
on claims may exceed 10 years because the policy covers all claims arising out of in-
juries that occur in the policy year, but certain long-latent injuries such as cancers may
take many years to emerge. Traditionally, the statute of limitations does not begin to
run until the injury has, or with due diligence should have been, discovered. The decay
of evidence contributes to delay in claim disposition, which may add 5-10 years from
claim filing.
Third, although in principle a claim is governed by the rules in effect at the time
of the alleged injury, in practice changes in social norms and legal rules may simul-
taneously affect the loss distribution on all outstanding claims, spanning several policy
years and possibly multiple lines of insurance. These common factors cannot be diversi-
fied through standard, law-of-large numbers pooling. Changes in social norms and legal
rules apparently contributed significantly to the surge in claims for malpractice and other
lines in the 1970s and again in the 1980s. Insurance companies failed to anticipate these
changes, hence were underreserved and suffered significant shocks to their capital base.
The "capacity crunch" theory [Winter (1988), Cummins and Danzon (1997)] posits that
shocks in insurance capacity lead to contraction in the supply of insurance, sharp pre-
mium increases and possibly lack of availability of coverage for high risk activities or
policyholders. However, total withdrawal of commercial insurers, as occurred in some
states in the 1970s, is more plausibly explained by denial of requested rate increases by
state insurance regulators.
In response to the crises, many malpractice insurers shifted from occurrence cover-
age, which covers all claims arising out of practice in the policy year, to claims made
coverage, which pays for all claims filed in the policy year, regardless of the practice
year in which the alleged injury occurred, provided that the insured was covered by that
insurance company at the time. Claims made coverage gives insurers greater flexibility
to adjust premiums to reflect changes in social and legal norms but, as a result, shifts
this risk back to the policyholder. A second response to high premiums or lack of avail-
ability of commercial coverage was the formation of provider-owned insurers, including
physician-owned mutual and reciprocal companies and hospital-sponsored captive in-
surers, which now write over half of malpractice insurance premium volume.
The formation of provider-owned insurers was prompted initially by high premiums
or withdrawal of commercial carriers. Some were formed with medical society spon-
sorship to provide insurance to all members; others use selective underwriting to attract
better risks. The survival and growth in market share of these mutual carriers, includ-
ing many that do not practice selective underwriting, suggests that they have persistent
advantages relative to stock companies. Provider-owned companies may have certain in-
formation advantages over commercial insurers, which facilitates accurate underwriting
and premium rating, settlement and merit rating of policy holders. However, commer-
cial insurers can and do involve physician-policyholders in these functions. For exam-
ple, some commercial policies have been sponsored by state medical societies which
play a role in designing the coverage.
Probably the more important advantage of provider-owned insurers is in bearing of
the undiversifiable component of risk that derives from socio-legal trends that are com-
1364 PM. Danzon
mon to all policyholders in the pool. In long-tailed lines of liability insurance, the to-
tal risk can be decomposed into the policyholder-specific or idiosyncratic component,
which depends on the probability of injury, and the socio-legal or common compo-
nent, which depends on the resolution of claims, given an injury. Mutuals may have
a comparative advantage in bearing the nondiversifiable component [Danzon (1984b),
Doherty and Dionne (1993)] and can diversify the nonsystematic component of risk
through equity markets, by purchasing reinsurance. For example, mutuals can assess
or pay dividends to their members, depending on the realization of the common risk,
whereas stock companies must hold capital reserves to protect against such uncertainty.
5. Effects of liability on medical practice
5.1. Formulatingthe analysis
5.1.1. Theoretical issues
The effect of liability on physicians' behavior is a critical component of any evaluation

of the costs and benefits of the malpractice system. The theoretical analysis of physi-
cian response to liability depends on several key assumptions. First is the physician's
objective function and form of reimbursement. Economic models of tort law gener-
ally assume profit maximization or cost minimization as the objective of the potential
tortfeasor [Brown (1973), Shavell (1980)]. However models of physician behavior in
other contexts generally assume a utility function that includes income and other ar-
guments such as leisure, ethics or reputation or the utility of a representative patient
[Pauly (1980), Farley (1986)] to reflect agency concerns. Using such an agency model,
Danzon (1994b) simulates physician response to alternative liability rules. The conclu-
sion is that the welfare gains from applying tort liability relative to no liability are much
lower if the physician is a reasonably good agent than if he or she is a selfish income
maximizer. Welfare gains are also greater under capitation reimbursement than fee-for-
service reimbursement, ceteris paribus, because of the incentive effect of capitation to
reduce services. (This assumes that the expected benefit of any excess services induced
by fee-for-service reimbursement is positive, although less than marginal cost.)
A second theoretical issue is the assumed relation between liability insurance and
injury prevention. Ehrlich and Becker (1972) show that insurance and prevention are
simultaneously determined responses to injury risk; they may be complements or sub-
stitutes, depending on whether the price of insurance adjusts to reflect risk reduction
due to prevention. Shavell (1982) demonstrates a similar result in the context of liability
insurance. An important implication is that empirical analysis of physicians' prevention
response to liability should treat the level of malpractice insurance coverage or premium
as a simultaneously determined choice, whereas several studies use the premium as an
exogenous measure of liability risk or "climate". A preferred, exogenous measure is the
premium rate charged for specified limits of coverage in the physician's locality.
Applying a simple prevention/insurance model to medical liability presupposes a

functional relationship between precautions, injuries and a liability loss, and that physi-
cians know this relationship. By contrast, the common view of the legal system pre-
supposes either that the risk of suit is independent of the level of precautions or
that it can be influenced by defensive measures such as extra tests, which reduce
the probability of suit with little or no effect on the risk of injury. Danzon (1990a,
1990b) models the physician's choice of defensive practices designed to reduce the risk
of successful suit as a third alternative to prevention and insurance. Assuming that the
physician's objective function includes the patient's utility, in addition to the physi-
cian's own income and leisure, this model concludes that extensive defensive measures
are unlikely unless physicians place low weight on the patient's utility.
A third issue in analyzing physicians' response to liability is to distinguish between
the cost-justified prevention (deterrence) that liability is intended to encourage and
wasteful "defensive" responses. This is discussed in Section 5.3.
5.1.2. Empirical issues
Accurate measurement of the effect of liability on medical care requires detailed infor-
mation on the services provided (M); a comprehensive measure of liability risk (L); and
other factors that may affect care, in particular, the patient's insurance and the provider's
reimbursement (Z). Measurement of effects on injuries requires, in addition, a compre-
hensive measure of health outcomes (H). The care response equation can be written:
Mit = al + a2Lit + a3Z + uit. (5.1)
Measurement of all these variables is problematic. If the observed data on medical ser-
vices, M, are not comprehensive - for example, ambulatory services only - estimated
effects may be upward or downward biased, depending on whether the measured med-
ical services are substitutes or complements for omitted services. The liability risk L
is multidimensional, but can be proxied by the local price for specified limits of insur-
ance coverage, under the assumption that insurance is accurately rated at the community
level; however, the premium paid by the individual doctor confounds price per dollar
of coverage (which is exogenous, assuming no experience rating) with level of cover-
age purchased, which must be treated as endogenous as discussed earlier. The vector of
other determinants of provider behavior should ideally control for all other relevant fac-
tors. Omission of variables that affect practice and are correlated with L - for example,
other locality-specific insurance characteristics such as prevalence of managed care -
would lead to biased estimates of a2.
Estimates of Equation (5.1) using either cross-section or time-series variation, can at
best measure the marginal response to changes in liability. However, it is the total, all-or-
nothing effect of liability that would be required to evaluate radical policy proposals, for
example, for replacing tort liability with a tax-funded, no-fault compensation scheme.
Moreover, Equation (5.1) alone cannot distinguish cost-justified deterrence from waste-
ful defensive medicine in the absence of data on outcomes.
1366 PM. Danzon
Two types of data have been used to estimate variants of Equation (5.1). The first is
surveys undertaken specifically to ask physicians about their response to liability. For
example, Reynolds (1987) use AMA survey data that asked about several dimensions
of practice that are believed to be most affected by liability, including use of lab test,
X-rays, C-sections, referrals, etc. Reynolds et al. conclude that 14 percent of expenditure
on physicians' services are a defensive response to liability. Although widely cited and
updated [Rubin and Mendelson (1993)], these estimates are subject to several biases.
Most obvious, physicians may exaggerate in ascribing their use of costly procedures to
liability rather than to other factors, such as financial incentives. Second, the response
is fully assigned to defensive medicine, none to cost-justified deterrence. Both of these
factors would imply that the Reynolds et al. results are an upper bound on true defensive
medicine. Third, because the listed services are only a subset, changes in other substi-
tute or complementary inputs are not accounted for. Fourth, reporting the estimated cost
of response as a net increase in expenditure presupposes that this cost is fully passed
forward to patients/payers in higher billings, with no incidence on physicians. This as-
sumption may have been valid in the 1980s [Danzon, Pauly et al. (1990)], but is less so
with the growth of managed care.
Surveys designed to collect information for purposes other than measuring response
to liability are less likely to be subject to reporting bias, but typically contain incomplete
information on patterns of practice. The Physician Practice Cost and Income (PPCI) sur-
veys have been used in several studies, but include information on a limited number of
ambulatory care price and service characteristics. Reimbursement claims data are more
complete but lack detail on other practice characteristics, such as time per encounter.
5.2. Empirical evidence
Danzon et al. (1990) and Danzon (1990a, 1990b) analyze several dimensions of physi-
cian response using the PPCI surveys of 1976, 1978, 1983 and 1986, which surveyed
a different, but nationally representative sample of physicians each year. The liability
climate is measured by the premium rate for basic limits of insurance coverage. These
surveys span the years of the liability insurance crises but predate the widespread growth
in capitation and managed care, which may have changed physicians response and lim-
ited their ability to pass on high premium or service costs through additional billings or
fee increases.
This analysis concludes that physicians increased their expenditure on insurance less
than in proportion to increases in expected liability loss costs. Physicians therefore bore
more uninsured risk in states with high liability costs. At that time, liability insurance
increases were passed along promptly through higher fees and reimbursement by health
insurers. The elasticities of routine office and hospital visit fees with respect to liability
insurance rates are between 0.1 and 0.2. This is more than sufficient to pass on the cost
of increased expenditure on insurance, assuming no change in volume, since on aver-
age physicians spent roughly 4 percent of gross revenues on insurance. Elasticities of
reimbursement paid by health insurers are similar to fee elasticities in the 1970s, but
somewhat lower in the 1980s, consistent with increasingly aggressive attempts at ex-
penditure control by insurers. By 1983, the ratio of Medicaid reimbursement to usual
fees is negatively related to liability, possibly because Medicaid reimbursement lagged
most in urban areas which also tend to have high malpractice costs. This suggests that
if, as is often alleged, liability has made physicians less willing to treat Medicaid pa-
tients, the relatively tighter constraints on cost pass-through to Medicaid is an important
contributing factor. But in general, this pass-through of malpractice costs into higher
fees and reimbursement levels was rapid and direct, without requiring an adjustment in
physician stocks. The number of physicians per capita, by county, and the rate of change
between these years was unrelated to either levels or rates of growth of liability costs.
The excess of the fee elasticities over the level required to fully pass-through the
costs of malpractice insurance may reflect several factors: increased physician time per
patient encounter, which could reflect improved care; a compensating differential for
exposure to uninsured claim costs and/or uninsurable time and non-monetary costs as-
sociated with the risk of suit; and a reduction in volume in response to higher fees. There
is weak evidence that liability induced physicians to spend more time per patient visit -
a possible indicator of more care.l 1 By contrast, the frequency of lab tests or procedures
was significantly negatively related to liability costs; the frequency of X-rays or fluo-
roscopies was positively related to malpractice costs in the 1970s but the relationship
was negative in 1983. Total number of office visits was negatively related to liability
costs, which is consistent with standard constraints on demand and inconsistent with
unlimited ability or willingness of physicians to shift demand for defensive purposes,
or with demand shifting outward in response to perceived improvement in quality of
care or higher expected compensation in the event of injury. This evidence is thus not
consistent with significant defensive ordering extra tests, at least in ambulatory care.
On average, physicians' reported net money incomes were not adversely affected
by liability costs through 1983, which is consistent with the evidence of a rapid pass
through of cost increases to fees and no effects on the geographic distribution of physi-
cians. However, several caveats are in order. First, even if net money incomes were
maintained, physicians' real utility may be lower, because of slightly longer hours of
work and increased exposure to the uninsured risk monetary and non-monetary costs
of being sued. Second, as medical care markets have become more competitive, the
ability to pass through premium increases is probably more limited in the 1990s than
at the time of these surveys. Third, these estimates of mean effects may obscure sig-
nificant distributional effects, with losses to some physicians offset by gains to others.
A combination of these factors may explain why physicians lobbied for changes in the
11 Aliability-induced increase intime per visit is plausibly consistent with cost-justified deterrence, assuming
that fee-for-service reimbursement and a fortiori capitation create incentives for suboptimal physician time
per visit inthe absence of liability. By contrast, fee-for-service reimbursement is likely to lead to excessive
use of tests and procedures, even without liability. If so, an increase in tests and X-rays is more likely to be
wasteful defensive medicine.
1368 P.M. Danzon
malpractice system, even if their fees and net money incomes, on average, rose to keep
pace with malpractice insurance costs.
Weiler et al. (1993) surveyed physicians to obtain information on their perceived risk
of suit and actual suit history, in addition to practice changes. These data permit esti-
mates of whether self-reported practice changes are statistically related to self-reported
perception of liability. They find that physicians with prior claims were significantly
more likely to explain risks to patients, and that those with high perceived risk of suit
were significantly more likely to order more tests or procedures and reduce the num-
ber of patients or procedures (such as GPs dropping minor surgery). This study also
found that physicians incurred significant financial and nonfinancial costs of being sued,
confirming anecdotal evidence that physicians perceive a significant tort threat despite
extensive liability insurance.
5.3. Defensive medicine
Defensive medicine may be defined as liability-induced changes in medical practice that

entail costs in excess of benefits, so would presumably not be desired by an informed
patient, given their insurance coverage, and that are intended to reduce the physician's
risk of being sued. Defensive medicine is thus a form of supplier-induced demand [SID,
see McGuire (2000)]. Distinguishing empirically between liability-induced defensive
medicine and insurance-induced waste (use of medical services with expected benefits
less than social cost because the patient is insured) is problematic, because insured pa-
tients have little incentive to refuse any treatment with non-negative benefit. Physicians
may face little patient resistance to low benefit services that increase their income (under
fee for service) and/or reduce their risk of suit. However, such practices and the pass-
through of associated costs to payers should decline with the growth of managed care
and capitation, to the extent that capitation internalizes the cost of defensive practices
to medical providers.
Several studies have focused on the effect of liability on the use of certain procedures
that are allegedly commonly used for defensive purposes. Using patient level data from
31 hospitals in New York in 1984, controlling for patient and physician risk factors,
Localio et al. (1993) find that use of caesarean section is significantly positively related
to three measures of malpractice risk: premium rates in the locality; the mean perceived
risk of suit in the locality, as reported in surveys of physicians; and the prior claim
history of the hospital and the medical staff in which the physician practices.
In order to control for possible bias due to unobserved omitted variables that is present
in the cross-section studies discussed so far, Kessler and McClellan (1996) apply a
difference-in-differences estimator to data on Medicare patients hospitalized for treat-
ment of acute myocardial infarction (AMI) or ischemic heart disease (IHD) over the
period 1984-1990. They use the enactment of tort reforms as a measure of change in
the state-specific liability climate. They estimate the effect of tort reforms on total hos-
pital expenditures in the year after the AMI (a measure of the intensity of treatment) and
on two measures of outcome, mortality or repeat hospitalization for AMI or heart dis-
ease within one year of the initial illness. A key assumption is that statutory changes in
tort law affect providers' perceptions of liability risk and hence affect practice patterns.
The expenditures and outcomes equations control for patient demographic characteris-
tics, state legal and political characteristics, and state and time fixed effects. The effects
of tort reforms are estimated as the difference in time trends between states that changed
and states that did not change their laws.
Kessler and McClellan find that tort reforms have a significant negative effect on
cost but no significant effect on outcomes, which they interpret as evidence that higher
liability creates incentives for socially excessive care or defensive medicine. This is
an imaginative and careful study, but the possibility remains that the estimated effects
may reflect some unobserved, correlated factor. In particular, it is plausible that states
with aggressive managed care were more likely to adopt tort reform, since managed
care limits the ability of providers to pass through liability-related costs. If so, the cost
savings attributable to tort reform could in fact be due managed care. Consistent with
this, in preliminary findings from work in progress Kessler and McClellan find that
controlling for managed care penetration reduces but does not eliminate the negative
association between tort reform and expenditures.
6. Overall evaluation of the malpractice system
A full evaluation of the medical malpractice system must weigh the costs against the
benefits. Although the available data are insufficient for a definitive analysis, rough
calculations are informative.
6.1. Costs of administrationand defensive medicine
Malpractice insurance premiums are less than 1% of total health care spending; never-
theless, this small percentage is roughly $10 billion [Rubin (1993)]. From the patient's
perspective, if tort liability is solely a form of compulsory insurance system for iatro-
genic injuries, for which premiums are included in the cost of health care, then it is
grossly inefficient relative to alternatives. Roughly 40 cents of the malpractice premium
dollar reaches the patient as compensation. Of the remainder, 40 cents per $1 premium
is spent on litigation, roughly equally divided between plaintiff and defense attorneys
and 20 cents is insurance overhead [Danzon (1985a)]. Compensation through tort lia-
bility thus carries a loading charge of $1.50 per $1.00 of compensation, compared to
less than 10 cents per $1.00 of compensation for large group first party insurance. This
overhead cost is pure waste if and only if the investigation into cause and fault has no
deterrence benefit. Tort also entails additional real but hidden costs, including the time
and anxiety costs of litigation; a relatively long mean delay (several years on average)
from the occurrence of an injury to receipt of payment; and uncertainty as to timing
and amount of compensation, whereas insurance is intended to eliminate variance in
income.
1370 PM. Danzon
The second potential cost of the malpractice system is defensive medicine. As dis-
cussed above, if courts lack good information about the optimal standard of care, physi-
cians may anticipate that they can reduce their probability of being found liable by tak-
ing highly visible but unnecessary precautions, such as ordering tests beyond the level
that is desired by patients given their insurance coverage, with expected benefits less
than cost. Defensive medicine remains an unmeasured deadweight loss of the liability
system. Theory suggests that it is likely to be larger under traditional fee-for-service
reimbursement than managed care.
Given these excess overhead costs of insurance and defensive medicine, the clear
conclusion is that if the sole function of liability is to provide compensation, this can be
done more efficiently through other private and social insurance mechanisms. The criti-
cal question is whether these excess costs are matched by at least equivalent deterrence
benefits, such that overall the benefits outweigh the costs.
6.2. Deterrence benefits
Measuring the deterrence benefits of tort liability requires estimating its effect on the
incidence of negligent injuries, which poses severe measurement and estimation issues.
The econometric challenge is somewhat easier for automobile accidents where the fre-
quency of accidents is more readily observable and changes in liability rules across
states and over time provide clear and measurable differences in the liability regime -
for example, change from third party negligence rules to first party no-fault rules have
occurred in several states in the US and in Quebec in Canada. The evidence from these
studies, summarized in Sloan (1998), confirms that tort liability is a deterrent to unsafe
driving, mediated in part by associated changes in the availability and price of insur-
ance.
The only credible study of deterrence of medical negligence is from Weiler et al.
(1993). This study estimated the relation between proportion of negligent injuries and
claims per negligent injury across 49 hospitals in New York state. 12 An instrumental
variables approach is used, using urbanization and population density as instruments, to
control for potential bias due to the endogeneity of negligent injuries in the denominator
of the deterrence measure, claims per negligent injury. The point estimate is negative
but not statistically significant. Taking this point estimate at face value and extrapolating
would imply that tort liability reduced the rate of negligent injuries per admission by
29 percent (from 1.25 with no liability to 0.89 with the current system) and reduced the
overall rate of medical injuries per admission by 11 percent (from 3.7 to 3.3). The failure
to find significant effects may be influenced by the small sample size (49 hospitals)
and imperfect instruments available. Moreover, at best these data would estimate the
marginal effect of changes in liability over the limited range of variation in the New
12 Using the proportion of negligent injuries as the dependent variable controls for unobserved case mix
across hospitals which might lead to variation in rates of nonnegligent injuries
York sample. Considering these intrinsic limitations that bias against finding significant
effects, together with other evidence that physicians do perceive a significant risk of suit
and change their behavior in response to liability, Weiler et al. conclude that liability
plausibly does have a significant deterrent effect.
This empirical evidence on deterrence benefits is consistent with rough calculations
by Danzon (1985a), that under quite generous assumptions about the costs of defensive
medicine, the malpractice system would pay for itself (yield positive net benefits) if
it reduced negligent injury rates by at least 20 percent, ignoring such intangible bene-
fits as retribution or fairness. As Weiler et al. (1993) note, the point estimate from the
Harvard study, that negligent injuries are reduced by 29 percent, easily passes this test.
This perhaps surprising conclusion is possible, despite the high overhead expense of the
malpractice system, because of the low rate of claims per negligent injury. Since the
high administrative loading is incurred only on the small fraction of injuries that lead to
a claim, a modest percentage reduction in injury rates is sufficient to offset reasonable
estimates of overhead and defensive medicine costs. Even if the benefits of the current
system do outweigh its costs, however, the search for marginal improvements or more
cost-effective alternatives remains an important policy question.
7. Traditional tort "reform"
Since the mid-1970s, most states have adopted one or more changes in their traditional
tort rules for medical malpractice. Follow standard usage, these changes are referred
to here as tort "reforms", without implying any endorsement of such changes as desir-
able.
The economic analysis has shown that the most extreme criticisms of the system
as a costly lottery are unfounded: negligent injuries are more likely to result in a claim
and compensation than non-negligent injuries and awards are significantly related to the
magnitude of loss. The various pieces of evidence suggest that medical practice patterns
are affected and may plausibly provide a sufficient deterrent effect to outweigh the costs.
Thus it is plausible but not proven that the system overall yields a positive net social
benefit. Nevertheless, any reform that reduces the deadweight costs of litigation and
defensive medicine or improves the efficiency of deterrence or compensation without
increasing litigation or overhead costs would improve the efficiency of the system.
The economic criterion for evaluating a proposed reform is thus, Is it likely to re-
duce the deadweight loss of litigation and defensive medicine, or improve the efficiency
of deterrence and compensation, recognizing that the practical choice is between im-
perfect alternatives? By contrast, most actual reform proposals aim primarily to reduce
measurable claim costs and liability insurance premiums or budgetary costs to health
care providers. This budget focus is likely to result, at best, in simply shifting costs
from medical providers to patients and taxpayers; at worst, total social costs may actu-
ally increase if, for example, deterrence incentives are weakened.
1372 P.M. Danzon
7.1. The structure of awards
7.1.1. Limits on awardsfor nonmonetary loss
Roughly half the states have enacted caps on awards; most limit only non-monetary loss
but a few limit the total award. Award caps have been estimated to reduce mean payout
per claim by up to 40 percent [Danzon (1984a), Zuckerman, Bovbjerg et al. (1990),
Harrington and Danzon (1994)] and premiums by somewhat less. Such large effects are
possible, although the caps directly constrain only a small percentage of cases, because
roughly 5 percent of cases account for 50 percent of dollars paid.
Economic evaluation of tort awards concludes that, while some limits are desirable,
single caps are at best a second best approach. A tort award in principle serves two
functions: it provides compensation to the plaintiff and imposes a fine on the negligent
defendant, assuming no liability insurance. The traditional guideline for tort awards is
to "make the plaintiff whole". This full compensation principle is unlikely to be optimal
for either compensation or deterrence, at least in the case of seriously disabling injuries
that result in an "irreplaceable loss" [Cook and Graham (1977), Spence (1977), Danzon
(1984b)]. Since tort compensation is a form of compulsory insurance that is tied to the
purchase of medical care, optimal compensation is the amount that consumers would
choose to purchase voluntarily, given the expense loading of the physician's liability
insurance (see Appendix). Insurance can only transfer money from the healthy to the
disabled state, but money is an imperfect substitute for an irreplaceable faculty or pos-
session, and transferring funds is costly. Optimal insurance with zero load equalizes the
marginal utility of income in the injury and no-injury states. With a state-dependent util-
ity function, optimal compensation for an irreplaceable loss could be more or less than
full compensation, depending on whether the injury raises or lowers the marginal utility
of income. With the high loading on malpractice insurance, optimal compensation is
presumably lower.
If victims incur an uncompensated loss after optimal compensation, then optimal
deterrence may require that the defendant pay a fine in addition to the compensatory
award. This optimal deterrence fine depends on consumers' willingness to pay for risk
reduction, given optimal compensation. To illustrate, a bachelor with no heirs may be
willing to pay large sums to reduce his risk of death even though he might choose not
to buy life insurance. More generally, the optimal deterrent fine, conditional on optimal
compensation, is inversely related to the extent to which market prices for medical care
internalize patients' willingness to pay for risk reduction and to the defendant's unin-
surable costs of suit, such as time and reputation [Spence (1977), Danzon (1985b) and
Appendix]. The fine should be paid to the state and refunded as a subsidy to the risky
activity, in order to preserve appropriate relative prices
Traditional tort rules also permit a two part damage award, consisting of a com-
pensatory award for monetary and nonmonetary loss and a punitive award, in cases of
recklessness, wanton or willful misconduct. However, the two part system in practice
differs from the theoretical ideal in several ways. Punitive awards are based on the de-
fendant's conduct, whereas ideally they should reflect consumers' willingness to pay for
Ch. 26: LiabilityforMedical Malpractice 1373
prevention; punitive awards are paid as additional compensation to victims, rather than
to the state; and compensatory awards aim to provide full compensation of monetary
and nonmonetary loss, regardless of consumers' willingness to pay for insurance.
Theoretical analysis alone cannot determine optimal compensation for an irreplace-
able loss, because the marginal utility of income is unobservable. However, the empir-
ical evidence that consumers do not voluntary buy coverage for noneconomic loss in
any other private or social insurance program suggests that such coverage may not be
worth its cost. The lack of a voluntary market for insurance of nonmonetary losses may
reflect severe ex post moral hazard of exaggeration of such losses which cannot be ob-
jectively measured. Assuming that this moral hazard of loss exaggeration is at least as
severe in the tort system, the evidence from private choices supports the case for limits
on compensation for nonmonetary loss through the tort system.
Several studies have therefore concluded that limits on awards for nonmonetary loss
would improve efficiency of compensation in the US tort system [Danzon (1984b),
Bovbjerg, Sloan et al. (1989)]. Many European countries already have such limits. The
preferred approach is a schedule based on the severity of injury and the plaintiff's life
expectancy, in order to approximate the ideal of equalization of the marginal utility of
income. By contrast, in states that have enacted limits on awards since 1975, the great
majority have adopted single caps on nonmonetary loss for all cases, say $500,000,
rather than scheduled benefits for nonmonetary loss. This may be too low for young,
severely injured plaintiffs, excessive for older patients or minor injuries.' 3 This prefer-
ence for single caps may reflect concern to avoid "injury severity creep" or litigation
over whether the differentials are fair - issues that are ignored in the theoretical analysis
of scheduled benefits.
In addition to providing more optimal compensation, scheduled awards are also ex-
pected to reduce litigation expense, by reducing the marginal payoff to investment in
litigation effort. Limits on compensation for nonmonetary loss are unlikely to under-
mine deterrence, because very high awards are typically not used for rating individual
(as opposed to class) liability premiums, being viewed as random bad luck. If deterrence
derives primarily from the uninsured time, anxiety and reputational costs, these are re-
portedly invariant to the outcome of the claim [Weiler et al. (1993)]. Thus scheduled
limits on awards could improve the efficiency of compensation and reduce litigation
expense, with no effect on deterrence.
13 There may be a further objection to single caps, if actual awards already tend to undercompensate for the
economic loss of severely injured patients more than for minor injuries (Sloan, Githens, et al. 1991). The
evidence on this point is inconclusive. In principle, one would like to compare the patient's compensation at
verdict, net of attorney fees, to the economic loss incurred. In practice, the sample of cases closed at verdict
is too small to permit such estimates. Inference from the much larger sample of out-of-court settlements is
problematic because settlements reflect the expected verdict, discounted to reflect the plaintiff's probability of
winning, net of the differential in litigation costs (see Equation 1). If the plaintiff's probability of winning is
inversely related to economic loss in closed claims data, because of fixed costs of going to court (see Danzon
and Lillard, 1983), this could account for an inverse relation between the compensation/economic loss ratio
and injury severity.
1374 P.M. Danzon
7.1.2. Periodicpayments
Under traditional tort rules, compensation for future damages is paid as a lump sum
equal to the discounted present value of future payments. Several states now permit pe-
riodic payment of compensation for future damages. The intended level of patient com-
pensation may often be provided at lower cost to the defendant through purchase of an
annuity or other financial instrument, if courts tend to be more conservative than finan-
cial markets in estimating interest rates, inflation and life expectancies. The amount of
such future payments should be fixed at the time of claim disposition. Periodic payments
that are contingent on the actual reported loss provide more than optimal insurance and
tend to undermine incentives for rehabilitation [see Rea (1981) for theoretical analysis;
Butler (1983) for empirical evidence from workers compensation].
7.1.3. Collateralsource offset
Under the traditional collateral source rule, tort awards in the US are not reduced by the
amount of compensation that the patient receives from private or public insurance. Such
offset occurs automatically in countries such as Sweden, the UK and Canada, where
medical costs that are covered under national health systems are not compensable in
tort. Since 1975, many states in the US have provided for offset from the tort award of
certain forms of insurance, to avoid a windfall of double compensation to the plaintiff
and reduce malpractice premiums.
With perfect information and costless transacting, the collateral source rule would
be irrelevant because consumers could contract around it [Coase (1960)]. Consumers
could choose first party insurance that makes no payment in the event of a tort award
or with subrogation, whereby the first party insurer assumes the plaintiff's tort claim
for covered expenses. However, with nonzero information and contracting costs, the
collateral source rule matters. Which rule is on balance more efficient is an empiri-
cal question. Subrogation preserves the full internalization of injury costs to the tort
defendant and hence preserves stronger incentives for deterrence. By contrast, collat-
eral source offset undermines deterrence by shifting costs from the tort defendant to
other insurance programs and by reducing the plaintiff's incentive to bring a claim be-
cause of the lower expected award. Empirical evidence confirms that collateral source
offset rules have not only reduced claim severity but also claim frequency, consistent
with the prediction that lower awards reduce the incentive to file [Danzon (1984a,
1986)]. However, because subrogation may entail higher transactions costs than collat-
eral source offset, the optimal mechanism for eliminating double compensation remains
an unresolved empirical question.
7.2. Reducing litigation costs
Litigation expense is at least partly a voluntary investment made by the litigants, given
the costs and expected payoff. A simple model of rational investment in litigation yields
important implications for reform. First, measures that reduce the elasticity of awards
with respect to litigation effort, such as damage caps or scheduled benefits, should re-
duce litigation investment. Second, measures that reduce the cost per unit of litigation
input, such as substituting arbitration for more costly court proceedings, may increase
the number of claims filed and total litigation expense per case could rise or fall; effi-
ciency effects are uncertain, but an increase in outlays is certainly contrary to the intent
of such reforms. Third, measures that reduce litigation inputs will also affect outcomes,
hence a full evaluation must consider effects on compensation and deterrence. In partic-
ular, the optimal amount of litigation depends on the social benefits of injury deterrence
and the private benefits of compensation, and the divergence between private and so-
cial costs of litigation that results because each litigant imposes costs on the opposing
party and on public financing of the courts. Shavell (1997) analyses the implications
for liability reform of this divergence between private and social benefits and costs of
suit.
7.2.1. Limits on contingentfees
In the US, plaintiff attorneys on medical malpractice and other personal injury litigation
are typically paid on a contingent basis, that is, they receive a fee if and only if they win
the case. The most common fee is one third of the award or settlement, with a range
of 25-50 percent. Investing in litigation with uncertain payoff is a risky business. An
attorney with a portfolio of cases can more efficiently bear this risk than an individ-
ual plaintiff, for whom the legal expense if paid as an hourly fee may be a significant
fraction of wealth. Contingent fees therefore provide a potentially more efficient alloca-
tion of this risk than hourly fees. Nevertheless, contingent fees have traditionally been
banned in the UK and Canada.
In the US, several states have adopted sliding scale limits on contingent fees. The al-
legation is that contingent fees stimulate an excessive number of suits and an excessive
willingness to reject settlements and gamble for large jury verdicts. Theoretical analysis
predicts that number of claims filed would be higher with a contingent fee, but appro-
priately so, because risk aversion would deter many plaintiffs from filing valid claims
with an hourly fee. More generally, the effect of contingent vs. hourly fees depends on
risk preferences and on competition and information in the market for legal services
[Danzon (1983)]. If attorneys compete for cases based on the fee percentage and the
expected award, then contingent fees may induce a private first best optimum, whereas
risk averse plaintiffs may bring too few cases and invest suboptimally per case if re-
quired to pay an hourly fee regardless of the outcome. This disincentive is higher under
the English rule that the loser pays all costs; however, it may be mitigated in practice by
legal aid (see Section 10).
The objective of limits on contingent fees is unclear and effects of such limits on
claim frequency and disposition - and a fortiori on efficiency - are uncertain. If the
objective is to reduce large awards, this could be achieved more accurately by direct
limits in the form of scheduled benefits. If the goal is reduce litigation expense, mea-
1376 PM. Danzon
sures to reduce uncertainty and the ability of litigants to influence the outcome would
deter investment by both sides.
If the goal is to reduce frivolous suits, a cost-shifting English rule, that assigns both
sides' legal costs to the losing party, is a more promising approach. There is a concern
that this would eliminate many valid claims if plaintiffs are risk-averse; on the other
hand, if plaintiffs are judgment-proof or if defendants choose not to enforce the rule,
there would be little effect. An alternative is to apply the rule to the plaintiff's attorney,
if paid on a contingent basis, rather than the individual plaintiff. Such a rule would in-
crease the plaintiff attorney's incentive to reject cases with weak evidence of negligence.
Applying the English rule to the plaintiff's attorney would almost certainly lead to an
increase in the equilibrium contingent fee percentage, in order to compensate plaintiff
attorneys for the added risk of paying the other side's costs if they lose.
7.2.2. Alternative dispute resolution (ADR)
Several states have adopted forms of alternative dispute resolution (ADR) that are in-
tended to eliminate frivolous claims, expedite claim resolution and reduce litigation
expense. Screening, mediation panels and nonbinding arbitration use less formal rules
but operate within the traditional court system, whereas binding arbitration replaces the
judge and jury with an arbitration panel selected by the litigants.
The effects of these procedures depends on their effect on the incentives and con-
straints of the litigants. Theory and evidence indicate that mandatory screening, without
significant penalties for appeal and without the panel's findings being admissible evi-
dence in court, may simply add an additional tier of delay and costs. For ADR to reduce
litigation delay and costs, it must create incentives for the parties to substitute the in-
formal process for more costly trial in a large percentage of cases. This implies that
formal arbitration proceedings should be binding. For less formal procedures, the par-
ties should face significant penalties for proceeding to trial against the recommendation
of the panel. For example, the early neutral evaluation (ENE) program that has been
adopted in northern California provides each side with information about the other's
case, through prompt and neutral evaluation [Rosenberg and Folberg (1994)]. If com-
bined with a system of early binding offers and a fee-shifting rule for frivolous rejection
of an offer and continued litigation, the costs and delay of claim disposition could in the-
ory be significantly reduced. An early binding offer system, combined with the English
rule, creates incentives for each party to act on their true information, whereas bluff
and strategic manipulation are penalized. By contrast, screening and mediation, without
significant penalties for strategic post-screening behaviour, simply increase delay and
costs.
7.3. Quality and standards of care
Many states have enacted measures intended to encourage quality of care through peer
review, practice guidelines etc. The federal government has established the National
Practitioner Data Bank, to which insurers must report all claims paid on behalf of prac-
titioners, and hospitals and states must report significant disciplinary actions. By re-
quiring hospitals and other institutions to check a physician's prior experience with the
databank before making a staff appointment, the intent is to prevent miscreant doctors
with a bad record in one state from simply moving to another state. Whether the benefits
of this system outweigh the costs of data collection and risk of misuse of the data by
unauthorized parties remains an unanswered question.
The proliferation of practice guidelines may, in theory, simultaneously serve to im-
prove care and reduce liability errors. Medical guidelines are promulgated by approved
bodies to provide guidance to physicians on best practice, hence may reduce the inci-
dence of negligent injury. In addition, they may serve as a defense against malpractice
claims. Maine is undertaking an experiment in which adherence to promulgated guide-
lines is a full defense to a malpractice claim [GAO (1994)]. The net effect of such an
approach depends on the optimality of the guidelines, whether they can be used for
both defensive and offensive purposes, and whether a significant fraction of care can be
routinized in this fashion.
7.4. Enterpriseliability
Enterprise liability would shift the locus of liability from the individual physician to an
enterprise such as a hospital, HMO or health plan. Proponents of hospital-based enter-
prise liability [Weiler (1991), Abraham and Weiler (1994)] argue that it would improve
deterrence and reduce litigation costs. Improvement in deterrence could occur if hos-
pitals have better information than individual doctors and the authority necessary to
implement systems-based loss control and quality assurance programs. Moreover, the
incentive to adopt such measures might be strengthened if the larger risk pool increases
actuarial credibility and hence permits more accurate experience rating of malpractice
premiums at the enterprise level than is possible at the individual doctor level. Reduc-
tion in litigation cost could occur because the enterprise would be the sole defendant,
whereas currently it is common to sue multiple doctors as well as the hospital, and each
of these defendants may hire separate counsel.
The arguments against replacing individual doctor liability with enterprise liability
of the hospital are several. First, hospitals already have strong incentives to take those
precautions that are within their control, including monitoring of staff, since plaintiffs
already name the hospital as a co-defendant if there is any possibility of involvement.
Moreover, hospitals already frequently arrange for - and sometimes provide through a
captive - the liability insurance for members of their medical staff. It is not clear that
enterprise liability would add significantly to the existing information or incentives for
system-wide loss prevention measures. On the other hand, the deterrence of individual
doctors would be weakened unless hospitals implement increased surveillance measures
sufficient to offset the elimination of the individual deterrence incentive on doctors.
Second, the savings in litigation costs could be small. Individual physicians would
presumably still be required to testify in order to determine what actually occurred. As
1378 P.M. Danzon
long as the liability rule is a negligence rule, showing negligence would require showing
that some member of staff failed to take appropriate precautions, and this would require
individual testimony, for which physicians might continue to retain their own counsel.
Third, for physicians who have affiliations with multiple hospitals but also some am-
bulatory practice, there would be either duplication or ambiguity of coverage for their
ambulatory practice. Fourth, with the decline in the importance of hospitals in the deliv-
ery of care and growth in other institutional arrangement, including integrated systems,
large physician group practices, etc., it is increasingly anachronistic to view the hospital
as the focus of care and hence as the best locus of liability.
These issues can only be resolved empirically. If enterprise liability is potentially
efficient, it could already be adopted by voluntary contract between hospitals and their
medical staff. In fact, such contractual enterprise liability is already the norm in at least
one staff model HMO, in most teaching hospitals and in other contexts where physicians
are salaried hospital employees. However, it has not occurred widely between hospitals
and physicians who are independent contractors or in the looser, increasingly common
network and independent practice HMOs. Plausibly, in such network environments, the
hospital or HMO has neither the information nor the authority to control the practice of
individual providers, hence retaining individual liability is more efficient. As noted, in
such arrangements the hospital, integrated system or HMO may arrange for the purchase
of insurance by participating physicians, in part because of their common interest in loss
control.
The liability insurance products offered on the market are adapting to meet the needs
and risks of these new institutional arrangements. As long as the market for health care
requires providers to compete on cost and quality, providers have incentives to contract
for assignments of liability and insurance that offer the best trade-off in terms of cost
and deterrence incentives. This may include several variants of total or partial enter-
prise liability, depending on other institutional factors. In this environment, enterprise
liability that is assumed voluntarily is to be welcomed. By contrast, a uniform, manda-
tory requirement could distort the natural evolution of the delivery system, distorting
deterrence and possibly care delivery, with no evident benefits.
8. Radical alternatives
8.1. No-fault programsfor iatrogenicinjuries
No-fault programs provide compensation for injuries caused by medical care, without
regard to the fault or negligence of the medical provider. Some proposals would shift
from a negligence rule to strict liability, and shift the locus of liability from the individ-
ual physician to an enterprise such as a hospital or health plan [Weiler (1991), Weiler
et al. (1993)]. An analogy is drawn with the workers' compensation program, in which
employers are strictly liable for work-related injuries, without regard to fault. Other
variants would compensate all iatrogenic injuries from broad-based taxes rather than
premiums paid by medical providers. In addition, most no fault proposals provide for
claims adjudication through an administrative agency rather than the courts; benefits
are usually limited to monetary loss, with collateral source offset and at most modest
scheduled payments for pain and suffering. The intent of all these proposals is to reduce
litigation delay and expense and to provide compensation to more victims of iatrogenic
injury. Virginia and Florida have established no-fault programs of compensation for
severe, birth-related neurological injuries caused by medical care, with some general
funding.
An evaluation of the efficiency effects of a broad-based no-fault scheme for medical
injuries must consider effects on all costs, including the costs of injuries, prevention
and overhead. Since most of the proposed programs change several components simul-
taneously, it is important to identify the marginal effects of the individual components
of a proposed program, distinguishing features that could be adopted within the current
negligence rule and features that are intrinsic to no-fault programs.
Proponents claim three sources of savings from no-fault programs, relative to the sta-
tus quo. First, cost per case would be lower due to lower payments of nonmonetary loss
and collateral source offset. Both of these changes could be adopted without changing
the fault-based rule of liability and both - particularly collateral source offset - simply
shift rather than reduce social costs. Second, no fault is usually combined with enter-
prise liability, with associated advantages and disadvantages discussed earlier.
Third, it is said that eliminating negligence as a condition for compensation would
reduce litigation expense. Proponents point to the lower overhead expense ratio of the
workers' compensation system in the US or the quasi no-fault systems of accident com-
pensation in Sweden and New Zealand. However, none of these systems provides a
good analogy for a no-fault system for medical malpractice, because of differences in
context and structure.
The argument that no-fault would reduce litigation costs rests on the assumption that
it would be simpler, less litigious and less costly to define a compensable event as a
medical injury rather than a negligent medical injury. However, evidence from workers'
compensation is not necessarily persuasive because of the difference in context: work-
ers are generally in good health, hence the occurrence of a work-related injury is easy
to define. This demarcation is less clear and litigation costs are correspondingly higher
for occupational diseases and cumulative trauma than for acute injuries. For iatrogenic
injuries, whether an imperfect medical outcome is an iatrogenic injury rather than an
imperfect cure within the range of normal risk requires an implicit assumption about
appropriate care and the probability distribution of outcomes with and without appro-
priate care. Since an operational definition of medical causation presupposes a standard
of appropriate care, it would often be a minor additional step to determine whether that
standard had in fact been met. Making a related point, Epstein (1978) argues that show-
ing cause would often require showing some "defect" in treatment, which is very similar
to showing negligence. Weiler et al. (1993) recognize that "in cases of medical omission,
the judgement about whether a patient's disability was caused by medical management
actually rests on an implicit identification of fault on the part of some provider". Since
most injuries can be framed as failure to take some precaution that would have reduced
the risk of adverse outcome, this implicit equivalence of a causation test and a fault test
1380 PM. Danzon
potentially applies far more broadly than the set of injuries that might be classified as
errors of omission under the current system.
The medical reviewers in the Harvard study were able to make more reliable judg-
ments about causation (adverse events) than about negligent events [Weiler et al.
(1993)1, which is one factor leading Weiler (1991) to conclude that litigation would
be less costly under strict liability rule than under a negligence rule.14 Such consistency
might be considerably less with lay adjudicators in the adversarial context of actual
litigation than in the clinical environment of the study.
The Swedish and New Zealand systems, also cited as evidence for the low overhead
costs that could be realized by no-fault systems, also provide a misleading analogy.
Although both eliminate the terminology of fault or negligence, medical causation is
a necessary but not a sufficient condition for compensation [see Section 10, and Dan-
zon (1994a, 1994d)]. The Swedish system retains a notion of medical error. More im-
portant, both Sweden and New Zealand owe their low litigation percentages partly to
the fact that compensation is their sole function, with no attempt at deterrence. Since
providers are not financially liable or exposed to sanction through these systems (with
minor exceptions noted later), providers have no reason to oppose - and some reason to
support - compensation for their patients. This is very different from no fault with strict
provider liability for costs proposed in the US. Moreover, if compensation is denied, pa-
tients have much more limited right of appeal than in the US tort system. Thus the low
overhead costs reflect lack of incentives for either party to contest the administrative de-
cision on compensation. Whether total social costs of iatrogenic injury are higher, due
to increased frequency of iatrogenic injury that offsets any savings in litigation expense
remains an important but unanswered question.
Several models for financing no-fault programs have been proposed for the US, with
different expected effects on overhead costs and deterrence. The first - and least likely
- is to impose strict liability on individual physicians with experience-rated premiums.
This would expose individual physicians to unacceptable financial risk, given their rel-
atively small patient load, hence would be inefficient for risk pooling and entail high
uninsurable time costs on physicians of defending the much larger number of claims.
Recall that if all iatrogenic injuries are compensable, regardless of fault, at least a ten-
fold increase in the number of claims might be anticipated, based on the New York and
California data, even under the optimistic assumption of no invalid claims. 15 If courts
err in dismissing false positive claims, providers would have strong incentives to avoid
the sickest patients. Strict liability with collateral source offset would create incentives
14 The high degree of reliability may partly reflect the fact that the medical reviewers were highly trained and
used an elaborate Adverse Event Form, which structured their decisionmaking. Reliability might be consider-
ably less if decisions are resolved through an adversarial process and with lay adjudicators. Even the random
sample of physicians surveyed showed "marked variation ... in their willingness to label certain outcomes as
iatrogenic" [Weiler et al. (1993, p. 125)].
15 The number of claims could be reduced by excluding minor injuries [Weiler (1991)], but this shifts the
cost to the individual or other insurance programs, without reducing costs.
for providers to avoid uninsured patients, for whom the expected liability cost, condi-
tional on an injury, would be higher than for an insured patient.
The second alternative is to place no-fault (strict) enterprise liability on hospitals or
health plans [Weiler (1991), Abraham (1994)]. The case for and against hospital-based
enterprise liability has been discussed. As noted, if enterprise liability is potentially
efficient, it can already be adopted by voluntary contract between hospitals and their
staffs, regardless of whether the liability rule is strict liability or negligence. In fact,
such contractual enterprise liability has only been adopted in limited circumstances,
usually where doctors are in a close and exclusive relationship to a single hospital, as in
a staff model HMO or teaching hospitals.
The third financing alternative for a no-fault program is a broad-based tax on medical
providers, insurance companies, or general revenues, as in Virginia and Florida. Fi-
nancing by a tax on medical providers eliminates individual deterrence, but retains the
internalization of costs to the health care industry (assuming accurate adjudication of
claims). Financing from general revenue taxation eliminates all internalization of costs
and deterrence, hence is pure social insurance. It is arguably neither efficient nor equi-
table to single out victims of medical injury for special compensation, unless there is
a deterrent benefit. Compensation can be provided more cheaply through broad-based
private insurance and social insurance programs, such as Social Security Disability,
Medicare and Medicaid, which provide compensation without regard to cause. Incur-
ring the cost of determining that a particular condition was caused by medical care,
rather than genetic or other factors, is worthwhile only if this information is used to
promote deterrence.
In theory, deterrence could be preserved despite broad-based funding if the program
itself brought suits for negligence against medical providers. Such decoupling of com-
pensation and deterrence could arguably provide prompt compensation for medical in-
juries, regardless of fault, while preserving deterrence. However, in practice this decou-
pling approach could increase overhead costs because two actions would be required,
one on causation and one on negligence. Moreover, significant tax financing would still
be necessary to pay for the nonnegligent injuries. Such compensation is hard to justify
on equity grounds when persons in similar condition from other causes, for example
birth defects, would not be eligible. Excess burdens of tax financing would add to the
real social costs.
8.2. Private contracting
Tort liability establishes a form of mandatory compensation tied to medical care that
may provide rules of liability, compensation and dispute resolution quite different from
those that patients would prefer if given the choice. Since medical injuries occur in a
context in which the parties are in a contractual relation, there is a prima facie case for
permitting the parties to contract out of judicially mandated tort rules [Epstein (1978),
Havighurst (1995)]. Such contracts might specify the circumstances for liability (for
example, gross negligence only), the rules of damages (for example, economic loss
only), and the rules and forum for dispute resolution (for example, arbitration).
1382 PM. Danzon
One objection to private contracting is that patients are poorly informed and are in no
condition to consider such issues when they are in need of medical care - indeed this is
the basic rationale for exposing medical providers to tort liability. Consistent with this
view, courts have generally overturned contracts entered into at the point of care - for
example, a contract providing for arbitration signed when the patient was admitted to a
hospital. However, contracts entered into as part of the health insurance agreement are
not signed under duress and have generally been upheld by the courts.
Although contractual reassignments of liability are rare to date, this may reflect the
difficulty of internalizing benefits to those signing the contract than to lack of potential
interest on the part of patients and providers. Under traditional fee-for-service insurance
with free choice of provider, patients have little incentive to adopt contracts that limit
their tort rights since they cannot realize the full savings from lower premiums unless
all the doctors and patients in the area adopt the same contract. At most, patients might
face a lower co-payment if they chose a provider who had signed a contract with limited
tort rights and hence had lower fees, but the saving would be only a fraction of the total.
Moreover, if the provider tried to target the savings through lower fees to the patients
who had agreed to the liability restriction, this might be viewed as a contract of adhesion
by the courts because it would be patient-specific.
By contrast, if managed care plans can lock in patients to the providers that have
adopted a cost-reducing contractual change, then the full savings can be passed on to
the patients through lower premiums for the insurance, which avoids the legal problem
of contracts at the point of service. Thus managed care offers the potential for more
contractual specificity both with regard to the conditions of compensation for iatrogenic
injury and the coverage of medical care (see Section 9.1). Stipulating such provisions as
part of the health insurance contract would permit consumers to make informed choices
before they need care, which in turn increases the likelihood that courts would uphold
the contracts.
9. Liability under managed care
The development of managed care in the US has led to fundamental change in the na-
ture of health insurance contracts and in the organization of the medical care delivery.
For patients, managed care means accepting restrictions on choice of providers and
covered services, in return for lower premiums, lower co-payment, or broader cover-
age than under traditional fee for service. For providers, managed care establishes risk
sharing forms of reimbursement in place of fee-for-service or cost-based reimburse-
ment. Direct controls such as treatment protocols, utilization review, drug formularies
etc. narrow the scope of covered services, although patients can purchase non-approved
services by paying out-of-pocket. The growth of managed care is a major force driv-
ing the restructuring of the delivery system, including horizontal and vertical mergers,
alliances and integrated delivery systems, in order to better pool risks, control costs
through economies of scale and scope and coordination of care, and compete better for
multistate employer contracts while protecting bargaining power.
The growth of managed care, together with the associated changes in provider rela-
tionships, has led to new grounds for liability claims. Two types of claim in particular
raise issues that are fundamental to the efficiency of the liability and health care systems.
First, claims for refusal to pay for care may be brought against the managed care organi-
zation (MCO), the physician or the utilization review entity. This raises questions about
the appropriateness of changes in treatment norms. Second, in cases alleging negligent
treatment by an individual physician, a claim may also be brought against the MCO in
addition to the physician. This raises questions of whether managed care plans, which
are often insurance entities that contract with health care providers, should be held li-
able for the negligent care of their contracted providers, if they have exercised due care
in screening and selection. These issues are discussed in detail in Havighurst (1995,
1997) and Danzon (1997).
An issue which pervades all cases but is particularly relevant to claims for withhold-
ing care is the definition of the standard of care for patients enrolled in managed care
health plans. As discussed earlier, due care is traditionally defined as customary care.
However, when most patients have comprehensive, fee-for-service insurance, customary
care is likely to exceed the social optimum for quantity and some dimensions of quality
of care, because of the moral hazard created by traditional indemnity insurance. Man-
aged care can be viewed as a competitive response of insurance and medical markets to
the growing demand of consumers and employer/payers for forms of insurance that pro-
vide better value for money than traditional indemnity insurance. But if the purpose of
managed care is in part to eliminate the excesses and distortions of indemnity insurance,
then if courts adjudicate managed care cases using the fee-for-service norms of care, the
ability of managed care to reduce the waste of traditional norms will be undermined.
9.1. Claimsfor denial of coverage

In Wickline v. State, the plaintiff sued MediCal for negligent denial of coverage of
an 8-day hospital stay, claiming that the 4-day stay permitted caused the subsequent
complications that led to the amputation of her leg. In this case the court ruled for
the defense, finding the MediCal restriction was not the cause of her discharge, which
was the responsibility of her physician who did not appeal the decision. The court also
noted that the discharge was consistent with usual standards of medical practice in the
community and hence that MediCal was not culpable of breach of duty. However, the
court left open the possibility that third-party payers could be liable for "defects in the
design or implementation of cost containment mechanisms", holding it "essential that
cost limitation programs not be permitted to corrupt medical judgment". Other courts
have used the standard of "medically necessary" care.
A very different conclusion was reached in Fox v. Healthnet, in which a California
jury awarded approximately $89m., including $77m. in punitive damages, finding that
Healthnet acted in bad faith, breached its contract for care and intentionally inflicted
emotional distress through reckless denial of coverage of a bone marrow transplant for
treatment of Mrs. Fox's breast cancer. Although the award was drastically reduced in
settlement prior to appeal, the potential magnitude and publicity of such awards could
1384 P.M. Danzon
clearly influence HMOs' coverage decisions. Unlike Wickline, Fox did not allege neg-
ligence but relied on contract theories commonly used to challenge insurance coverage
decisions.
These coverage denial cases raise two issues. First and most problematic, What is
the basis for liability for denial of coverage, if any? The courts appear to implicitly as-
sume the existence of an objective and appropriate standard of care defined by medical
judgment, which in turn defines appropriate cost containment mechanisms. However, at
best medical science can tell us the probability distribution of health outcomes and risks
from particular medical treatment. Deciding whether the treatment is worth performing
requires comparing the value of the expected outcome to the costs. Valuation ultimately
depends on consumer preferences and willingness-to-pay. For private programs, this
can be evaluated using willingness-to-pay. For public programs, willingness-to-pay can
be defined to include the altruistic willingness-to-pay for others too poor to pay for
themselves. Based on this analysis, Danzon (1997) concludes that claims for denial of
coverage should be viewed not as negligence claims but as contract disputes, in which
the question is: Would enrollees (or similar consumer groups) be willing-to-pay for
insurance coverage of this service ex ante, given the cost and expected outcomes? In-
evitably, an ex ante willingness-to-pay standard will appear to conflict with the interests
of the individual patient once sick, who would then want coverage of all services that
offer any positive expected benefit. However, to achieve an efficient standard of liability
for coverage disputes, courts must ignore the ex post or patient-specific private opti-
mum and focus on the ex ante or group optimum, which also approximates the social
optimum (ignoring tax distortions). But it is the ex post or private patient optimum that
underlies the traditional norms of indemnity insurance.
This contractual approach to coverage disputes would permit the standard of care to
vary, depending on the type of plan, the premium and explicit and implicit contractual
terms. If instead, courts apply a uniform standard of "medically necessary" care to all
plans, health plans will be constrained in their incentives to compete by developing in-
novative, more cost-effective patterns of care and to differentiate their product offerings
to cater to heterogeneous consumer preferences.
The second issue is, In the event of failure to pay for services in conformity with the
contract, who should be liable - the health plan, the doctor, the UR agency or all three,
on grounds of joint and several liability? Transactions costs considerations indicate that
liability should be placed only on the health plan, not the individual provider or UR
agency, since it is the plan that defines the contract, operates or contracts for the UR
controls, and ultimately bears the financial risk of paying for the contracted services
within the premiums paid. If liability is placed on individual physicians or UR agencies,
they are likely to seek contracts of indemnification from the health plan, hence it is more
efficient to place liability directly on the plan. 16
16 Of course in a provider-sponsored MCO, liability in coverage disputes would be on the provider group in
their role as plan sponsor, not as a provider of care.
9.2. MCO liabilityfor negligent care
Since managed care plans typically restrict enrollees to the network of selected
providers, plans are required to use due care in selecting and monitoring participat-
ing providers and may be held liable for negligence in performing these credential-
ing functions. Similar liability for negligent credentialing already applies to hospitals
with respect to their credentialing of staff physicians, including independent contractor
physicians with admitting privileges.
A separate and far more contentious issue is whether an MCO should be liable, under
theories of vicarious liability or ostensible agency, for the negligence of its contractor
physicians, assuming that the MCO has exercised due care in credentialing. Propo-
nents argue that patients look to HMOs as providers of care, in part as a result of the
HMO's own promotional material. Havighurst (1995, 1997) argues that the default po-
sition should be enterprise liability of health plans (for POS plans, liability would only
extend to torts of affiliated providers). He would offer MCOs the option of contracting
for other allocations of responsibility. His argument is that "MCOs, although in control
of many levers that can affect the quality of care for better or for worse, are not, in the
eyes of the law, routinely answerable for poor quality. Enterprise liability is the logical
legal culmination of the shift to de facto corporate responsibility that is revolutionizing
American medical care" [Havighurst (1997, p. 588)].17
The argument against holding MCOs liable for the negligence of their contracted
physicians is that, for loose networks such as IPA and POS plans, the plan lacks the
information and authority to control the details of care delivery, which remains in the
hands of individual providers. Danzon (1997) argues that liability for negligent perfor-
mance should therefore remain with these individual providers, if the MCO has ade-
quately met its obligation to screen participating providers. To add the MCO as a de-
fendant in claims alleging negligence by the provider, simply adds another deep pocket
defendant, which may distort the outcome of claims, without adding useful deterrence.
Exposing MCOs to liability for the negligence of their contracted providers is likely to
lead them to select more restrictive networks and exercise tighter control, which appears
contrary to the preferences of consumers who increasingly choose plans with broad net-
works and POS options.
Ultimately, the issue is whether it is practical and sensible to distinguish between
coverage decisions and negligent performance, conditional on the coverage decision.
Danzon assumes that this distinction can be made and is important, hence argues for
plan liability for coverage decisions but provider liability for implementation of treat-
ment. By contrast, Havighurst subsumes these different dimensions into a broad notion
of "quality", notes that plans do intervene in some dimensions of this broadly defined
quality, and hence concludes that they should be liable for all aspects, including treat-
ment.
17 He also argues that placing liability on health plans would lead them to take more effort to define the
standard of care in their contracts with patients.
1386 PM. Danzon
9.3. The ERISA pre-emption
The federal Employer Retirement Income Security Act (ERISA) preempts state laws to
the extent that they "relate to" an employee benefit plan that is subject to ERISA, which
includes all self-insured employer plans. This has been interpreted to bar tort claims for
denial of coverage by employees against HMOs where the HMO coverage is sponsored
by a private employer. 8Employees may receive compensation of some money damages
if, for example, they could show an administrator's misconduct, but compensation for
pain and suffering and punitive damages would not be authorized. The extent of the
ERISA preemption has been uncertain, with the trend toward gradual erosion through
court decisions and explicit statutory changes.
The ERISA preemption is consistent with the contractual view of employee benefit
plans. Economic analysis concludes that, in the long run, the costs of employer-provided
health insurance are borne by employees through lower wages and that employers have
incentives to design such plans to maximize the utility of covered employees, since
this minimizes the money wage or other benefits that must be offered to attract a given
workforce. Given this interpretation, it makes no sense to permit individual employ-
ees to sue for denial of coverage, except where contractual commitments have been
breached. Efficiency and equity argue for eliminating the inconsistencies between the li-
ability exposure of ERISA-protected plans and non-ERISA plans, the question is which
rule should prevail. Some current legislative proposals would effectively eliminate the
ERISA preemption by granting patients statutory rights to sue HMOs and possibly their
employers. The analysis outlined above suggests that, in the absence of a well-defined,
contractual approach to defining responsibilities for paying for care, such an extension
of liability could seriously undermine the ability of HMOs to eliminate the waste that
was embodied in customary care.
10. Non-US experience
10.1. The UK
Medical negligence and its costs emerged as a public policy issue in the UK in the 1980s,
following several years of rapidly rising claim costs and premiums. In 1996, medical
negligence is estimated to have cost the NHS in England £235m., with an estimated
rate of increase of 17-25 percent per annum, and is again a major concern [Dobson
(1998)].
The liability of medical providers in the UK derives from the same common law ori-
gins as in the US, but with significant difference in detail. First, in the UK cases are
decided by common law judges rather than juries. Second, tort awards are reduced by
t8 The employer sponsor in Fox was a public school district and hence was not protected by ERISA.
the amount of compensation available from social insurance programs and NHS med-
ical benefits, and awards for pain and suffering are more modest. This full collateral
source offset, which also occurs in Canada, New Zealand, Sweden and most European
countries, is a major factor contributing to the apparent differences in costs of medical
negligence in different countries. Third, lawyers traditionally are paid by the hour and
are not permitted to take contingent fees. The English rule allocates the legal costs of
both sides to the losing party. In theory, this loser-pays-all rule could significantly deter
risk averse plaintiffs from bringing claims. In practice, most medical negligence plain-
tiffs in the UK receive Legal Aid and the English rule has not been applied to Legal Aid;
however, the hourly rates paid by Legal Aid may not fully compensate some lawyers
for the opportunity cost of their time. Whether on balance the incentives of patients and
lawyers to bring claims are too high or too low, relative to the social optimum, remains
an empirical question.
The data on injuries and claims in the UK are very limited. There has been no com-
prehensive study of the incidence of iatrogenic injuries and negligent injuries. Although
some analysts have extrapolated the findings of the Harvard study for New York state
to estimate the number of negligent injuries in the UK [Smith (1990)], such estimates
are tentative at best, because they assume the same rate of iatrogenic injury and negli-
gence in the UK as in New York. As noted earlier, in the New York study the count of
iatrogenic injuries depended on the standard of care implicitly assumed by the review-
ers, case mix and severity of hospital admissions, and the count of negligent injuries
depended on the reviewers beliefs about legal standards. It is a big leap to assume that
rates of adverse events and negligent injuries defined in this context-specific manner
would be the same in other countries that differ from New York in their norms of care,
case and severity mix of hospital admissions, and legal standards.
There are no comprehensive data on the level and trends of negligence claims for
the UK. Based on a careful study that pieced together information from several sources
[Ham, Dingwall et al. (1988)], it appears that from the mid-1970s to the mid-1980s
the rate of increase in number of claims and size of awards was at least as rapid in
the UK and Canada as in the US. Nevertheless, because these countries started from
a lower base, by 1987 doctors in the US were still five or six times more likely to be
sued than doctors in Canada or the UK, and awards for comparable injuries were several
times larger in the US [Danzon (1990a, 1990b)]. This overstates the difference in real
compensation to patients, because the attorney's fee (usually one third) is subtracted
from the award in the US and because medical costs are borne by public health care
systems in the UK and Canada.
Although medical negligence costs to the NHS are reportedly growing at 17-25 per-
cent per annum, the basis for these estimates is unclear. Also unclear is the relative
contribution of increased frequency of claims, increased severity of claims or simply
increased rates of treatment; whether the reported figures refer to costs paid out or costs
accrued in a given year; and how far the apparent cost growth reflects a shift from
pay-as-you-go accounting for hospital trusts to accrual accounting [Towse and Danzon
(1998)].
1388 P.M. Danzon
Recent changes in the locus of liability and insurance responsibility in the NHS have
complicated the tracking of claim trends. Prior to 1988, all NHS doctors were required
to join one of the two medical defense organizations (MDOs). In the mid-1980s, as
claim costs and premiums rose, commercial insurers entered the market offering lower
rates to GPs and other low risk specialties, undermining the traditional community-rated
premiums of the MDOs. 19 The ensuing price war for low risk specialties threatened
significant premium increases for high risk specialties, including orthopaedic surgeons,
obstetricians, etc., particularly in the higher risk urban areas, which in turn threatened
the NHS policy of equal net incomes. The outcome was that the NHS instructed Health
Authorities to assume full responsibility for all new and existing claims against em-
ployed staff, up to a cap of£300,000 (including legal costs), with responsibility for
claims from pre-1990 practice to be met partly by the MDOs. GPs continued to have
their subscriptions (premiums) fully refunded ex post through their expenses. With the
formation of trust hospitals, the responsibility for negligence claims devolved from the
health authorities (HAs) to the trusts.
In effect, the UK thus established a form of enterprise liability, in which either the
HAs or the trusts are responsible for all liability arising out of the practice of their con-
sultants on NHS business (consultants must continue to provide for their own cover
for their private business). This experience is thus of considerable interest to the en-
terprise liability debate in other countries. Since claimants had usually sued the HA as
well as the doctor, on grounds of vicarious liability, the practical effect was to reduce
the doctors' role in claims settlement and their concern over premium rates. The pro-
fession opposed NHS enterprise liability on grounds that it would lead to settlements
that ignored effects on professional reputation and lead to greater NHS management of
clinical activity and interference with medical judgement - arguments often made by
physicians in the US who oppose enterprise liability. In practice, the HAs are probably
too far removed from the delivery of care to exercise well-informed risk management
and loss control. Thus the net effect of shifting liability from the individual consultants
to the HA would probably be to reduce deterrence.
Shifting enterprise liability from HAs to individual trust hospitals offers greater op-
portunity for realizing the potential risk-management benefits of enterprise liability.
However, trusts' incentives to engage in risk management and loss control depend on
the extent to which they bear the costs of negligent injuries. In 1995, the NHS litiga-
tion authority was set up to manage the Clinical Negligence Scheme (CNS) for trusts.
This is a voluntary reinsurance arrangement, in which participating trusts pool their loss
19 The MDOs are technically not insurance companies, hence are not subject to the same reporting, reserving
and solvency requirements as insurance companies and are under no legal obligation to pay claims against
their member doctors. In practice, MDOs perform the same functions as insurance companies. However, their
looser reporting and reserving requirements may enable the MDOs to operate and set prices on a pay-as-you-
go basis rather than an accrued cost basis. The special legal status of MDOs may also reduce the contestability
of the liability insurance market in the UK, Canada and other countries in which MDOs operate. Such entry
barriers reduce competitive pressures for experience rating and permit community rating to survive.
above an excess threshold ranging from £10,000 to £500,000, depending on the trust's
size and activities. All but 6 of the first 226 joiners chose the lowest excess, typically
£25,000. Since there is no participation in losses above the retention and premiums are
apparently not experience rated, this reinsurance arrangement significantly undermines
trusts' incentives for loss control, although the remaining retained risk should leave
some incentive (the mean settlement is around £50,000). However, a further critical
question is the actual incidence of liability costs. If trusts face a very inelastic demand
for their services, because of local market power and/or long term contracts with HAs,
then those trusts that incur relatively high liability costs can simply pass on these costs
as higher charges to payers. Even if the CNS reinsurance premiums were experience
rated, the deterrence incentives that flow from enterprise liability in the NHS are likely
to be weak if competition is weak in the market for hospital services, such that trusts can
simply pass through their liability costs to payers, as higher charges and/or reduced ser-
vices provided for fixed budgets. Thus in the case of the NHS, replacing the liability of
individual consultants by enterprise liability of HAs or trusts has probably undermined
deterrence, by reducing both the monetary and non-monetary incentives of doctors to
take care, and because weak competition in the market for hospital services permits
trusts to pass on rather than internalize their liability costs.
The UK experience with claims for denial of payment for care is also of great in-
terest as the US wrestles with the same issues in the context of managed care. The UK
courts have consistently supported the HAs in cases where treatment has been denied on
grounds of shortage of funds, provided that the decision making process is reasonable
[for example, Ham (1998)]. Thus the UK courts defer to payers to make substantive
resource allocation decisions of their limited budgets and to define the process, and so
far have protected them from liability. It remains to be seen whether US courts will sim-
ilarly recognize that HMOs and other insurers must operate within budget constraints
and respect their autonomy in both the process and the substantive outcomes of the in-
evitable resource allocation decisions. A key difference appears to be that UK courts
believe that money saved in denying care to one patient will be spent on another pa-
tient, whereas US courts may be led to believe that money saved in denying coverage
goes into the pockets of the HMO shareholders. Both naive views are probably too ex-
treme: UK HAs are not necessarily perfect agents for consumers, whereas US HMOs
are under competitive pressure to deliver value for money to consumers. Nevertheless,
the naive views may prevail over these economic views, permitting UK courts to adopt
more appropriate, ex ante social criterion in these coverage cases, rather than the ex
post, individual patient criterion that US courts may be pressured to adopt.
10.2. The Swedish patient compensation insurance (PCI)
The PCI was established in 1975 by voluntary contract between medical providers and
a consortium of insurers, to preempt the threatened statutory expansion of tort liability
1390 P.M. Danzon
in order to make compensation available to a larger number of patients. 20 Although pa-

tients retain the choice to sue in tort under traditional negligence rules, tort claims have
been extremely rare since the PCI was established. Key features of the Swedish model
include the elimination of the concept of negligence or fault and decoupling of patient
compensation from deterrence or any sanctioning of providers. Patient compensation is
provided by the Patient Compensation Insurance (PCI), while the discipline of medical
providers is handled by the Medical Responsibility Board (MRB). No information is
transmitted between them, in order to obtain the doctors' cooperation with the PCI. The
PCI is administered by a consortium of insurers, with appeal to a special advisory panel
and ultimately to arbitration.
The Swedish model has attracted interest in other countries because of its relatively
low budget cost, low overhead rate and widespread acceptance by medical providers.
In the early 1990s, claim frequency stabilized at about 21 per 100 physicians per year,
compared to 13-16 claims per 100 physicians in the US. The higher rate in Sweden is
not surprising, given the lower costs of filing, somewhat broader criteria of compens-
ability and less reason to oppose payment of marginal claims (see below). Roughly
40 percent of these claims receive compensation in both countries. But the PCI costs
roughly $2.38 per capita, or 0.16 percent of health care costs in Sweden, whereas med-
ical malpractice insurance premiums are about 1 percent of higher health expenditures
in the US. Thus the per capita budget cost of the Swedish PCI appears to be roughly one
tenth of US. malpractice premiums. Administrative overhead is 14-18 percent of total
PCI premiums, compared to roughly 60 percent in the US. This low overhead rate is
often cited as evidence of the potential savings from eliminating negligence in favor of
a no-fault (causation-only) rule of compensability for medical injuries [Weiler (1991)].
However these inferences are based on a misunderstanding of the PCI. The low bud-
get cost of the PCI, despite the higher claim frequency, reflects primarily two factors.
First, the collateral offset rule shifts most of the wage loss and medical expense of ia-
trogenic injuries to other social insurance programs. In fact, such cost-shifting does not
reduce and may actually increase the social costs of injuries, by undermining cost in-
ternalization which undermines general and specific deterrence. Because the economic
loss related to iatrogenic injury is largely shifted to other social programs, the PCI bud-
get vastly understates the true cost of compensating iatrogenic injuries in Sweden.
Second, awards for noneconomic loss are below those in most other European coun-
tries and roughly one tenth of those in the US. 21 The PCI can pay low awards without
driving patients to select the tort option because Sweden's tort system offers even lower
20 Only roughly ten patients a year received compensation through the traditional negligence-based tort lia-
bility system. One alleged obstacle was the reluctance of physicians to testify on behalf of plaintiffs, as was
required by the custom-based standard of care. Risk-averse plaintiffs may also have been unwilling to pay the
legal fees in the absence of contingency arrangements or legal aid.
21 In 1987, the mean payment for noneconomic loss under the PCI was $3,800, the maximum was $117,070.
Nevertheless payments for noneconomic loss account for roughly 74 percent of total PCI payments, because
economic loss is heavily covered through collateral sources.
payments than the PCI, with lower probability of success. 2 2 Thus the PCI offers plain-
tiffs an expected payoff that at least matches their expected tort recovery, net of costs,
in order to deflect tort claims. However, other countries that have more generous tort
systems could not adopt a voluntary contractual alternative such as the Swedish model
and expect to realize costs as low as in Sweden, unless they also adopted significant tort
reform. Indeed, if a Swedish-style PCI were offered as a voluntary alternative to tort
in the US, it would not offer significant cost-savings relative to those that can be re-
alized through out-of-court settlement, which already provides a voluntary, contractual
alternative that operates in the shadow of the tort system.
The PCI's low overhead percentage is not the result of using a causation-only test
for compensability. Although the PCI is often called no-fault, this is misleading. From
the patient's perspective, the criteria of compensability are quite similar to a traditional
negligence rule based on customary practice. Under the PCI, an injury is compensable
if (1) it occurred with "substantial probability" as a direct consequence of medical inter-
vention, and (2) either the treatment was not medically justified or the injury could have
been avoided by performing the treatment differently. Thus although the terminology of
fault and negligence have been eliminated, compensation requires some notion of "er-
ror". Adverse outcomes caused by medical care are explicitly excluded, if the treatment
was medically justified.
But from the provider's perspective, the PCI is both no-fault and no-liability. The PCI
eliminates all reference or inquiry into fault or negligence, does not require the patient
to identify a particular provider who failed in a duty of care, and entails neither financial
nor reputational consequences for individual providers. This no-liability scheme bears
no resemblance to the proposed strict enterprise liability of hospitals [Weiler (1991)],
since the strict enterprise liability proposal would place the liability for paying damages
on the defendant hospital, in order to promote deterrence.
The low PCI litigation percentage reflects several features that are unlikely to be
acceptable in the US. These include the elimination of all link between patient com-
pensation and provider liability and deterrence, and the modest level of patient rights,
compared to a US tort plaintiff (although not necessarily compared to a Swedish tort
plaintiff). Under the PCI, because physicians are not liable and have no personal stake
in the outcome, they generally cooperate rather than opposing compensation of an in-
jured patient. Patients have little to gain from appeal to the review panel or to arbitration.
Both are closed to the press and public, and the panel has ruled in favor of the insurers
in 90 percent of cases. 23
Thus the low litigation rates reflect primarily the fact that neither plaintiffs nor
physicians have strong incentives to oppose or appeal the insurers' adjudication of
claims. Other contributing factors are the simple claim filing process; administration
22 Among other obstacles, plaintiffs allegedly have difficulty obtaining the expert testimony required to sup-
port a claim for negligence under the custom-based negligence rule.
23 Since 1992, major decisions of the review panel and all arbitration decisions are published.
1392 P.M. Danzon
by a monopoly consortium of insurers, which eliminates insurers' incentives to vigor-

ously oppose plaintiff claims or to experience rate premiums; and lack of competition
and provider-specific accountability for costs in the health care system, which makes
providers more willing to tolerate flat-rated premiums, despite significant geographic
differences in claims experience. 24
The deterrence function, to the extent that it exists, was assigned to the Medical Re-
sponsibility Board. Patients can file a claim with the Medical Responsibility Board if
they feel that their treatment was negligent or contrary to the statutory code of medical
practice. They bear their own filing costs and receive no compensation. Providers may
receive a reprimand or warning, but this has no financial consequence and probably at
most a minor reputation effect. There are roughly 6 MRB claims per 100 physicians
per year, of which one in 6 receives some sanction. Thus the ratio of MRB sanctions
to paid PCI claims is less than 1 in 10 - a rough measure of the loss in potential de-
terrence that results from decoupling compensation from medical discipline [Danzon
(1994a)]. Although the PCI database on iatrogenic injuries might in theory be used to
identify and control persistent sources of risk, in practice the information collected is
insufficient. Moreover, although clinics and hospitals are informed about their claims
experience, the responsible individuals and sometimes even the nature of the injury are
not identified.
Thus the main lesson from the Swedish PCI experience is that a sufficient and possi-
bly a necessary condition for low overhead costs is to forego all links between compen-
sation and deterrence. Whether this results in more iatrogenic injuries with social costs
that outweigh the reduction in litigation costs is an unanswered empirical question. The
answer would probably differ across countries, depending on the costs and benefits of
their tort systems and on the costs and effectiveness of other mechanisms for quality
control.
10.3. The New Zealand accidentrehabilitationcompensation and insurance

scheme (ACS)
The New Zealand ACS was established in 1972 as a comprehensive no-fault compen-
sation system for victims of "personal injury by accident", including "medical misad-
ventures". The establishment of the ACS followed the elimination of traditional tort
rights for such injuries, in contrast to the Swedish PCI, where tort actions remain an
option. Claims are administered by the ACS, with appeal to a special ACS Authority.
Compensation was set at a relatively high percentage of wage loss for workers, plus
scheduled lump sum payments for noneconomic loss. Medical costs were borne by the
24 The PCI is financed by premiums paid by the county councils, who are responsible for financing and
provision of the public health care system in Sweden, and by private physicians, dentists and other parapro-
fessionals. For each provider category, premiums are assessed on a flat per capita basis, regardless of claims
experience.
National Health Service (NHS), except that the ACS paid directly for services in private
hospitals, co-payments and services not covered by the public system.
Between 1975 and 1989 total expenditures for all types of injury covered by the ACS
grew at a nominal rate of over 20 percent a year, or roughly 6 percent a year after
adjusting for inflation, which exceeds the average rate of increase of US malpractice
premiums over the same period. For 1985-1997, expenditures rose at an annual real
rate of 8 percent [NZBR (1998)]. However, these trends are not strictly comparable to
trends in malpractice costs in other countries because the ACS data include all injuries.
The ACS did not track iatrogenic injuries as a separate category until 1993. Financing is
through payroll and general taxation, with no separate assessment of medical providers.
Concern over the rapid increase in ACS costs, the inequity of the incidence of costs
(low risk employers and drivers subsidizing high risks) and the neglect of injury preven-
tion led to significant changes in the 1992 Accident Rehabilitation and Compensation
Insurance Act. This 1992 Act redefined medical misadventure. The problems under the
original ACS and the changes adopted are instructive.
"Personal injury by accident" was broadly defined by the original ACS statutes to
include "physical and mental damage caused by medical, surgical, dental and first aid
misadventure". The intent was to exclude illness and normal risks of medical care but
to include medical injuries that fall outside the realm of normal risk, including those
caused by negligence or with either very low probability or unexpected severity. Dif-
ficulties in implementing this definition led to various proposals for change over the
years, including use of ICD-9 definitions of injuries, or extending the system to include
all incapacity. The broad range of reform proposals may reflect the conceptual problem
faced by any compensation scheme for iatrogenic injuries that is not focused on de-
terrence, of providing an equitable justification for compensating some victims but not
others in similar condition but from other causes.
The 1992 ACS reform provides a statutory definition of medical misadventure as
"personal injury resulting from medical error or medical mishap". "Medical error" is
"the failure of a registered health professional to observe a standard of care and skill
reasonably to be expected in the circumstances". "Medical mishap" is an adverse con-
sequence of treatment that is both rare and severe. An injury is "rare" if it has less than
a 1% probability of occurring. An injury that would normally be rare may not be rare
given the particular circumstances of the patient, if this greater risk was known to the
patient. An injury is severe if it results in death, hospitalization for more than 14 days,
significant disability lasting more than 28 days or qualifies for ACS's Independence
Allowance. Medical mishap specifically excludes delayed abnormal reactions and com-
plications of procedures. Injuries related to lack of informed consent, misdiagnosis or
treatment omissions are compensable only if they result from negligence.
The 1992 reforms require the ACS to pay for all medical costs incurred by compen-
sated victims, effectively restoring the traditional collateral source rule, in the interests
of accountability and to increase the ACS's incentives for loss control.
The New Zealand ACS has often been acclaimed for its low overhead costs (less than
10 percent of total expenditures) and prompt payment of compensation. Traditionally,
1394 PM. Danzon
this partly reflected the ACS's practice of accepting the majority of claims as filed, re-
lying largely on physicians as gatekeepers to certify that a claim is a "personal injury
by accident", and the simple claims adjudication procedures. Data collection costs were
kept low, with little detailed information on the causes of injuries. Thus like the PCI,
the ACS has maintained low overhead by simply adjudication and data systems and by
severing all links between compensation and deterrence, which eliminates the incen-
tives of those causing injuries to oppose the compensation. Whether such a strategy of
skimping on overhead is "penny wise but pound foolish", leading to higher real social
costs of injuries, is an important but unanswered empirical question.
Under the 1992 reforms a separate Medical Misadventure Unit of the ACS was estab-
lished to handle medical injury claims. It has ultimate authority on all claims, but seeks
advice on complex claims (including all medical error claims) from a three-person ad-
visory committee (a health professional from the relevant specialty, a lawyer and a lay
person) drawn from a pool of non-ACS people. Both parties are given 15 working days
to comment on the preliminary findings of the advisory committee, which may meet
up to 3 times. Roughly 45 percent of claims have been accepted; of these, roughly 86
percent are based on mishap (no fault) and 14 percent are based on error (negligence)
[ACS (1996)]. Claimants may appeal for review of the decision by an independent of-
fice within ACS. Providers may appeal only those medical error claims associated with
negligent failure to obtain informed consent or misdiagnosis.
Although the 1992 legislation provides for the Medical Misadventure Account to be
funded by premiums levied on health care professionals, so far this has not been im-
plemented and the account remains funded out of general and payroll taxes. Imposing
experience-rated premiums on individual providers would expose them to great risk
that is largely beyond their control (since only 14 percent of claims are due to negli-
gence) which would be inequitable and inefficient, particularly since providers are paid
fixed fees that are not risk-adjusted for patient mix. Those compensated claims that are
deemed attributable to negligence are reported, but the ACS has no record of whether
disciplinary proceedings were taken.
The original New Zealand ACS structure illustrates pitfalls to be avoided rather than
a useful prototype that other countries might adopt. The original definition of a com-
pensable event was difficult to implement. The 1992 reforms clarified the definition of
medical mishap, adding explicit reference to medical error and negligence, with associ-
ated possibilities for disciplinary actions. The low reported administrative costs of the
ACS reflect relatively low investments in determining the causes of injuries, limited
data collection and little attempt at deterrence, which may have contributed to the rapid
escalation of total costs. These comments apply to injuries in general, because medical
injuries were not identified prior to 1992. If the proposed introduction of experience-
rated premiums for medical providers were implemented, providers would have more
incentive to oppose claims and litigation costs would likely increase. A more accurate
measure of the true overhead cost of an accident compensation scheme would include
not only the reported overhead but also the unmeasured deadweight loss from unnec-
essary injuries and inappropriately compensated claims. This is not observable, but is
likely to be higher for both the PCI and the ACS than for tort-based systems which incur
higher reported overhead costs with the purpose of deterring inappropriate injuries.
11. Concluding comments
The basic rationale for medical malpractice liability is to improve provider incentives
for safety, assuming that asymmetric information leads to market failure in medical
markets. The evidence of a significant rate of negligent injury, invalid claims and physi-
cians' preference for insurance policies with minimal explicit co-payment or experience
rating indicate that the efficiency of the malpractice system is severely constrained by
imperfect information on the part of courts, doctors, patients and liability insurers with
respect to appropriate care and legal standards. Just as imperfect information under-
mines the efficient functioning of the market, imperfect information undermines the
efficient functioning of the liability system. The fundamental problem is that changing
the liability rule does not correct the information asymmetry.
Nevertheless, the evidence shows that the worst criticisms of the malpractice system
as a random lottery are unfounded. Negligent injuries are more likely to lead to claims
and receive payment than non-negligent injuries, many invalid claims receive zero or
low payments, and payments are strongly related to the loss incurred by the patient.
Malpractice also does not appear to be a major driver of health care costs. Malprac-
tice premiums are only roughly one percent of total health care spending. Defensive
medicine has not been accurately measured or distinguished from insurance-induced
overuse of care. To the extent that managed care internalizes to providers the costs of
unnecessary care, defensive medicine is likely to become a less pressing issue.
The fault-based system of liability for medical injuries may be worth retaining if
the benefits, in terms of injuries deterred, exceed the excess costs of litigating over fault
and defensive medicine. A full cost-benefit evaluation is not possible given the available
data, but rough calculations suggest that a positive net benefit is plausible. Nevertheless,
overhead costs are high. Certain incremental reforms could make the system more cost-
effective for purposes of both deterrence and compensation, in particular, scheduled
awards for non-economic loss; redefinition of the standard of care to reflect managed
care and heterogeneity of health plans; and possibly some forms of ADR. Freedom to
contract out of the system, for alternative rules of liability (no provider liability, liability
only for gross negligence, etc.) for alternative benefit levels or for arbitration of disputes,
could be encouraged as part of the health insurance contract. Freedom for providers to
contract for various forms of enterprise liability could offer savings if adopted voluntar-
ily. For the US, the key issue going forward is to find the best basis for holding managed
care plans to their contractual commitments to care, without undermining their ability
to continue to search for better ways to define and implement appropriate care.
There is no consensus on whether switching from a fault-based rule (possibly more
clearly defined) to a no-fault basis for compensation would offer significant savings in
litigation costs. The evidence from Sweden and New Zealand suggest that other factors
1396 PM. Danzon
contribute to their low overhead costs. If the deterrence benefits are considered too
uncertain to warrant retaining malpractice liability of individual medical providers, then
there is no strong case for a special, tax-financed compensation program for victims of
iatrogenic injury. The adequacy of compensation for iatrogenic injury is simply part of
the broader question of the efficiency and equity of the existing network of private and
social insurance programs.
Appendix
Al. Optimal liability rules
This model is formulated in terms of medical injuries but could apply in any market
context where consumers may be injured by product failure which depends only on the
care taken by producers. State-dependent utility of patients and physicians is assumed,
because personal injury plausibly affects the patient's utility of income and liability
claims impose uninsurable time and reputation costs on physicians.
Assume that expenditure on prevention affects the probability but not size of loss,
and each patient buys just one unit of medical care. Patients can buy first party and/or
physicians can buy liability insurance, with perfect experience rating and a proportion-
ate loading charge. The following notation is used:
V(B) = patient's utility of initial wealth, V' > 0, V" < 0,
p(r) = probability of injury, p' < O0,p" > 0,
r = quality (prevention per unit of service),
s = price of services,
c(r) = production cost per unit, c' > 0, c" < 0,
L = monetary loss to patient if injury occurs,
M = first party insurance coverage bought by patient,
yp = premium rate per dollar of first party coverage, where y ) 1 is the
loading charge,
h(p) = patient's perception of p, h' > 0, h" < 0,
U(A) = physician's utility of initial wealth, U' > 0, U" < 0,
D = damages paid by the physician if a loss occurs,
Q = liability insurance coverage bought by physician, Q < D,
)~p = premium rate per dollar of liability coverage, where A> 1 is the
loading charge,
it = Lagrange multiplier.
Subscript '0' denotes the state in which an injury occurs. Subscript '1' denotes the
state in which no injury occurs. Subscripts 'f', 's', and 'n' denote first party, strict and
negligence liability, respectively. Initially, patients are assumed to be fully informed.
Al.1. Firstparty liability (caveat emptor)
If patients are fully informed and markets are competitive, the physician chooses the
level of safety (r) and product price (s) to maximize expected utility of patients, E(V),
subject to maintaining an opportunity level of utility, UC, determined by the physician's
alternative use of time. 2 5 Patients select first party insurance coverage (M), given the
supply price per dollar of coverage yp. Informed markets thus solve the following opti-
mization problem:
max = (1-p)V1[B-s-ypM]+pVo[B-s -ypM-L+M]

M,s,r
+ A{U[A +s - c(r)] - Uc}. (A.1)
Maximization with respect to M, s and r yields
yV = VO, (A.2)
V' = tU', (A.3)
· c=- pf[V1,V ' oyM]M (A.4)
where V' = (1 - p) V + pVO.

Equation (A.2) may be written
V= l-y
v; -yp <1 asy3l
VO Y - Yp
This is the familiar result, that optimal coverage does not fully equate the marginal
utility of income if the insurance premium contains a proportionate loading charge
(y > 1). Equation (A.4) shows that if injury entails irreplaceable loss (V1 > Vo), op-
timal prevention (rf*) may exceed the optimal level with risk neutrality (c' = -p'L),
even with full insurance of any monetary loss.
A1.2. Strict thirdparty liability
Under a rule of strict third party liability the physician pays for all iatrogenic injury.
An omniscient benevolent dictator would choose the damage award (D), physician's
liability insurance coverage (Q), prevention (r) and product price (s) to maximize
25 If the physician has monopoly power, UC includes some rent but the structure of the problem is not affected.
The model ignores other arguments in a physician's utility function, such as prestige and ethics.
1398 P.M. Danzon
the patient's expected utility, subject to maintaining the opportunity level of utility for
physicians UC:
max = (1-p)Vi[B-s]+pVo[B-s-L+D]
D,Q,s,r
+ /{(1 - p)SU [A + s - c(r) - p Q]

+ pUo[A + s - c(r) - X, Q - (D - Q)] - UC}. (A.5)
Maximization with respect to D, Q, s and r yields
V = U, (A.6)
3U = U, (A.7)
V' = pU', (A.8)
c=-p V + (a ± Q],
+ (A.9)
where U' = ( - p)U; + pU.

Equations (A.6) and (A.8) together imply:
VI, U = _P <1 as X3 1.
The optimal tort damage award, D*, provides the level of insurance patients would
choose to buy voluntarily, but with the loading factor of the physician's liability in-
surance. Thus D* Z M, as A ~ y. The optimal safety level (r,*) implicitly defined
by Equation (A.9) may exceed or fall short of (rf*) under first party liability as
(U - Uo)/U' + XQ M.
If X= = 1 and the physician incurs no uninsurable liability loss, then r = r. In
the more realistic case of X > y and U1 > Uo because of time and reputation costs of
suit, the optimal level of prevention is higher under strict liability.
A1.3. Negligence
Under a negligence rule, the physician is liable only if he fails to meet the due care
standard, r, and the patient is injured. The social welfare function is given by Equa-
tion (A. 1) for r < r*, and by Equation (A.5) for r > r. If either U is state-dependent
or > 1, the physician is not fully insured against the loss. The social welfare function
is discontinuous at r* because of the costs of suit to physicians.
A first best solution can be achieved by setting D at D*, the optimal compensatory
award under strict liability, and r* = rf*. The physician's decision problem is then to
choose Q, s and r to maximize E(V), subject to E(U) > U' and subject to the penalty
D = D* if r < r*. But this private objective function is identical to the social welfare
function, i.e., it is a discontinuous function equal to Equation (A. 1) for r > r* and equal
to Equation (A.5) for r < r*. If either A > y or Uo < U1, the physician's incentive
is to choose r = r,*, i.e., to be non-negligent and hence have no demand for liability
insurance. On the other hand, if 1 = X < y and U is not state-dependent, then physicians
would choose r < r,*, i.e., would choose to be negligent and to insure the resulting
losses. As a practical matter, this case can be ignored. 26
A2. Imperfect information
The analysis so far has shown that if consumers are fully aware of risks, first party lia-
bility and negligence are equally efficient and superior to strict liability, when shifting
liability imposes uninsurable losses on defendants. 27 But if consumers underestimate
risks, under first party liability they buy too little insurance and non-optimal safety.
Spence (1977) shows that under strict liability, a first best solution with respect to com-
pensation and prevention can be achieved by means of a two-part penalty. A compen-
satory award equal to D* is paid to victims. A fine, paid to the state initially but refunded
as a subsidy to the hazardous activity, is set equal to (1 - h')[(VI - Vo)/V'], where h(p)
is the consumer's perception of p and (V1 - Vo)/V' is the dollar measure of loss due to
injury, or the willingness to pay for injury reduction.
With a risk averse defendant and incomplete insurance, provided the standard of care
is correctly set at rf*, the fine necessary to achieve compliance is less under negligence
than under strict liability because of the discontinuity of the payoff function. 2 8 The
physician will choose to meet the standard provided:
- Uo +Q > V - Vo( h ) +YM.

Ur + 3.Q1-
+ I (vP .
Thus if the load on liability insurance is at least as great as the load on first party in-
surance ( > y), a fine over and above the compensatory award paid to victims is not
necessary to induce compliance with a negligence standard, if the compensated cost of
suit to physicians (U1 - Uo)/U' exceeds the distortion in market incentives due to con-
sumer misperceptions [(V1 - Vo)/V'](l - h'/p'). Since the fine-subsidy mechanism is
presumably costly to administer, this is an added attraction of a negligence rule over
strict liability.
26 If A > y > 1, it might be optimal to provide compensation through first party coverage and impose a
liability fine on physicians to achieve optimal deterrence. Enforcement would depend on subrogation actions
by the patient's first party insurer against the physician or his liability insurer.
27 This analysis ignores costs of adjudicating claims, which would probably be highest for strict liability,
lowest for first party liability.
28 Note that there is a range within which the due care standard can fall and still induce compliance.
1400 PM. Danzon
A3. The demand for liability insurance
It has been shown that, under a negligence rule with the standard of care and rule of
damages optimally defined, physicians have incentives to be non-negligent. Hence there
should be no demand for liability insurance.
This argument presupposes that courts enforce an efficient due care standard with
perfect accuracy, and that this is known to physicians and patients. The demand for lia-
bility insurance can arise out of either Type 1 or Type 2 errors by the courts, or penalties
insufficient to offset consumer misperception of risk. If courts set the standard too high,
holding physicians liable for some injuries where the cost of prevention exceeds the ex-
pected benefits (Type 2 errors), it is cheaper for the physician to insure than to prevent
these injuries, and this is socially optimal. But if victims or courts also commit Type 1
errors, failing to file or to find liability in all true instances of negligence, or if liability
payments are too low, then it is cheaper for the physician to insure than to avoid some
instances of negligence, and this is not socially optimal.
These two cases are illustrated in Figure 1. The curves labeled FF and SS show the
marginal social benefits of care per patient encounter, under first party and strict liability
respectively, with fully informed consumers and courts. Under first party liability, the
marginal benefit of additional care (FF) is simply the reduction in probability of injury
to the patient, which is assumed to be subject to diminishing returns. Under strict lia-
usC
t~
Prevention per
patient encounter
Figure 1. Optimal prevention under first party, negligence and strict liability.
bility, there is an additional benefit from care if the physician incurs some uninsurable
loss of time, reputation or inconvenience in responding to claims, so SS > FF.
The curve labeled CC shows the marginal cost of care. The discontinuous heavy line
NN shows the marginal social benefits of care under a negligence rule with the due care
standard, r, optimally set equal to rf*, the level that would be chosen by fully informed
patients. NN is discontinuous at rn* because for levels of care that equal or exceed that
level, the physician meets the due care standard; hence he bears no liability.
Type 1 errors (failure to file or find liability in all instances of negligence) or a sub-
optimal penalty if negligent imply a downward shifting of NN. Provided the vertical
segment of NN intersects CC at r*, physicians will still choose to be non-negligent. But
if the Type 1 errors are sufficiently large, the intersection occurs to the left of rn*. It is
then cheaper for physicians to practice with less than due care and to purchase liability
insurance than to be non-negligent.
Type 2 errors consist of setting the standard to the right of r*. As long as rf < rn* <
rs*, physicians will meet the standard. But if r* > r*, it is cheaper for physicians to
choose r, i.e., to practice below the excessively high standard set by the courts and
to insure against the resulting claims. Patients are thereby better off than if physicians
adhered to the excessively high standard, but are worse off than if the standard were set
at r*, because medical fees rise to cover the additional prevention costs (r* - r*) and
physician's uninsured disutility of suit.2 9
Casual evidence suggests that Type 1 errors dominate the demand for liability insur-
ance. In medical and other lines of professional liability, the courts defer to the custom-
ary practice of professionals "in good standing" as the standard of due care, rather than
apply the Hand cost-benefit calculus in each case. This creates a bulwark against Type 2
errors for above-average physicians but a presumption of Type 2 errors and consequent
incentive to insure for those of below-average competence. Thus, the Type 2 errors hy-
pothesis might predict that a substantial fraction of physicians would not buy insurance,
which is not the case. On the other hand, Type 1 errors, which give all physicians an
incentive to buy insurance, are very common (see Section 3.2).
References
Abraham, K.S. and P.C. Weiler (1994), "Enterprise medical liability and the evolution of the American health
system", Harvard Law Review 108:381-438.
ACS (1996), What is Medical Misadventure (Accident Rehabilitation and Compensation Insurance Corpora-
tion, Wellington, New Zealand).
Bovbjerg, R.R. (1995), Medical Malpractice: Problems and Reforms (The Urban Institute, Washington, DC).
Bovbjerg, R.R., F.A. Sloan, et al. (1989), "Valuing life and limb in tort: Scheduling pain and suffering",
Northwestern University Law Review 83(4):908-976.
Brown, J.P. (1973), "Toward an economic theory of liability", Journal of Legal Studies 2:323-349.
29 In principle, the additional cost of liability insurance is offset by a reduction in first party insurance costs.
1402 PM. Danzon
Butler, R.J., and J.D. Worral (1983), "Workers' compensation: Benefits and injury claim rates in the 1970s",
Review of Economics and Statistics 65:580-589.
Coase, R. (1960), "The problem of social cost", Journal of Law and Economics 3:1-44.
Cook, P.J., and D.A. Graham (1977), "The demand for insurance and protection: The case of irreplaceable
commodities", Quarterly Journal of Economics 91:143-156.
Craswell, R., and J. Calfee (1986), "Deterrence and uncertain legal standards", Journal of Law, Economics
and Organization 2(2):279-303.
Cummins, J.D., and P.M. Danzon (1997), "Quality, price and capital flows in insurance markets", Journal of
Financial Intermediation 6(1):3-38.
Cummins, J.D., and J.F. Outreville (1987), "An international analysis of underwriting cycles in property lia-
bility insurance", Journal of Risk and Insurance 54(2):246-262.
Danzon, P.M. (1983), "Contingent fees for personal injury litigation", Bell Journal of Economics 14:213-230.
Danzon, P.M. (1984a), "The frequency and severity of medical malpractice claims", Journal of Law and
Danzon, P.M. (1984b), "Tort reform and the role of government in private insurance markets", Journal of
Legal Studies 13(3):517-549.
Danzon, P.M. (1985a), Medical Malpractice: Theory, Evidence and Public Policy (Harvard University Press,
Cambridge).
Danzon, P.M. (1985b), "Liability and liability insurance for medical malpractice", Journal of Health Eco-
nomics 4:309-331.
Danzon, P.M. (1986), "New evidence on the frequency and severity of medical malpractice claims", Law and
Contemporary Problems 5(49):57-84.
Danzon, P.M. (1990a), "The "crisis" in medical malpractice: A comparison of trends in the United States,
Canada, the United Kingdom and Australia", Law, Medicine and Health Care 18:48-58.
Danzon, P.M. (1990b), Medical Malpractice: Incidence and Incentive Effects (The Wharton School, Univer-
sity of Pennsylvania, Philadelphia).
Danzon, P.M. (1991), "Liability for medical malpractice", Journal of Economic Perspectives 5:51-69.
Danzon, P.M. (1994a), "The Swedish patient compensation system: Myths and realities", International Re-
view of Law and Economics 14:453-466.
Danzon, P.M. (1994b), "Alternative liability regimes for medical malpractice: Results from simulation analy-
sis", Journal of Risk and Insurance 61(2):219-244.
Danzon, P.M. (1994d), 'Tort reform: The case of medical malpractice", Oxford Review of Economic Policy
10(1):84-98.
Danzon, P.M. (1997), "Tort liability: A minefield for managed care", Journal of Legal Studies 26(2):491-519.
Danzon, P.M., and L. Lillard (1983), "Settlement out of court: The disposition of medical malpractice claims",
Journal of Legal Studies 12:345-377.
Danzon, P.M., M.V. Pauly and R. Kington (1990), "The effects of malpractice litigation on physicians' fees
and incomes", American Economic Review 80(2):122-127.
Dewees, D., P. Coyte, et al. (1989), Canadian Medical Malpractice Liability: An Empirical Analysis of Recent
Trends (University of Toronto).
Dobson, F. (1998), Dobson to Tackle Rising Levels of Litigation in the Health Service (Department of Health,
London).
Doherty, N.A., and G. Dionne (1993), "Insurance with undiversifiable risk: Contact structure and organiza-
tional form of insurance firms", Journal of Risk and Uncertainty 6:187-203.
Eddy, D.M. (1998), "Performance measurement: Problems and solutions", Health Affairs 17(4):7-25.
Ehrlich, I., and G.S. Becker (1972), "Market insurance, self-insurance and self-protection", Journal of Politi-
cal Economy 80:623-648.
Ellis, R.P., C.L. Gallup, et al. (1990), "Should medical professional liability insurance be experience rated?"
Journal of Risk and Insurance 57(l):66-78.
Entman, S., C. Glass, et al. (1994), "The relationship between malpractice claims history and subsequent
obstetric care", Journal of the American Medical Association 272(20):1588-1591.
Epstein, R. (1978), "Medical malpractice: Its cause and cure", in: S. Rottenberg, ed., The Economics of
Medical Malpractice (American Enterprise Institute Press, Washington, DC).
Farber, H.S., and M.J. White (1991), "Medical malpractice: An empirical examination of the litigation pro-
cess", Rand Journal of Economics 22(2):199-217.
Farley, P.J. (1986), "Theories of the price and quantity of physicians' services: A synthesis and critique",
Fenn, P. (1993), "Compensation for medical injury: A review of policy options", in: C. Vincent, M. Ennis and
B. Audley, eds., Medical Accidents (Oxford University Press, Oxford).
General Accounting Office (GAO) (1994), Medical Malpractice: Maine's Use of Practice Guidelines to Re-
duce Costs (US Government Printing Office, Washington, DC).
Ham, C., R. Dingwall, et al. (1988), Medical Negligence: Compensation and Accountability (The Kings Fund,
London).
Ham, C., and S. Pickard (1998), Tragic Choices in Health Care: the Case of Child B (The Kings Fund,
London).
Harrington, S., and P.M. Danzon (1994), "Price cutting in liability insurance markets", Journal of Business
67(4):511-538.
Harrington, S., and R. Litan (1988), "Causes of the Liability Insurance Crisis", Science 239:737-741.
Havighurst, C.C. (1995), Health Care Choices: Private Contracts as Instruments of Health Reform (American
Enterprise Institute Press, Washington, DC).
Havighurst, C.C. (1997), "Making health plans accountable for the quality of care", Georgia Law Review
31(2):587-647.
Henderson, J.A., and T. Eisenberg (1988), "The quiet revolution in products liability: An empirical study of
legal change", UCLA Law Review 37:479-553.
Hickson, G., E. Clayton, et al. (1994), "Obstetricians' prior malpractice experience and patients' satisfaction
with care", Journal of the American Medical Association 272(20): 1583-1587.
Kessler, D., and M. McClellan (1996), "Do doctors practice defensive medicine?", Quarterly Journal of Eco-
nomics 111(2):353-390.
Localio, A.R., A.G. Lawthers, et al. (1993), "Relationship between malpractice claims and caesarian deliv-
ery", Journal of the American Medical Association 269(3):366-373.
McGuire, T.G. (2000), "Physician agency", in: A.J. Culyer and J.P. Newhouse, Handbook of Health Eco-
nomics (Elsevier, Amsterdam) Chapter 9.
Mills, D.H., J.S. Boyden, et al. (1977), Report on the Medical Insurance Feasibility Study (California Medical
Association and California Hospital Association, Sutter Publications, San Francisco, CA).
Nye, B.F., and A.E. Hofflander (1987), "Economics of oligopoly: Medical malpractice insurance as a classic
illustration", Journal of Risk and Insurance 54(3):502-519.
NZBR (1998), Accident Compensation: Options for Reform (New Zealand Business Roundtable, Wellington,
New Zealand).
Pauly, M.V. (1968), "The economics of moral hazard: comment", American Economic Review 58:531-537.
Phelps, C.E. (2000), "Information diffusion and best practice adoption", in: A.J. Culyer and J.P. Newhouse,
Handbook of Health Economics (Elsevier, Amsterdam) Chapter 5.
Posner, R.A. (1972), "A theory of negligence", Journal of Legal Studies 1:28-96.
Priest, G. (1987), "The current insurance crisis and modem tort law", Yale Law Journal 96(7):1521-1590.
Rea, S. (1981), "Lump sum vs. periodic damage awards", Journal of Legal Studies 10:131-154.
Reynolds, R.A., J.A. Rizzo, et al. (1987), "The cost of medical professional liability", Journal of the American
Rolph, J.E. (1981), "Some Statistical Evidence on Merit Rating in Medical Malpractice Insurance", Journal
of Risk and Insurance 48:247-260.
Rolph, J.E., R.L. Kravitz, et al. (1991), "Malpractice claims data as a quality improvement tool: Is targetting
effective?", Journal of the American Medical Association 266(15):2093-2097.
1404 PM. Danzon
Rosenberg, J.D., and H.J. Folberg (1994), "Alternative dispute resolution in a civil justice reform act demon-
stration district: Findings, implications and recommendations", Stanford Law Review 46:1487-1551.
Rubin, R.J., and D.N. Mendelson (1993), Defensive Medicine and Medical Liability Reform: Estimating
Costs and Potential Savings (Lewin-VHI, Fairfax, VA).
Schwartz, W.B., and D.N. Mendelson (1989), "The role of physician-owned insurance companies in the de-
tection and deterrence of negligence", Journal of the American Medical Association 262(10):1342-1346.
Schwartz, W.B., and D.N. Mendelson (1989), "Physicians who have lost their malpractice insurance: Their
demographic characteristics and the surplus lines companies that insure them", Journal of the American
Shanley, M.G., and M.A. Peterson (1987), Posttrial Adjustments to Jury Awards (The RAND Corporation,
Santa Monica, CA).
Shavell, S. (1980), "Strict liability vs. negligence", Journal of Legal Studies 9:1-25.
Shavell, S. (1982), "On liability and insurance", Bell Journal of Economics 13:120-132.
Shavell, S. (1997), "The fundamental divergence between the private and the social motive to use the legal
system", Journal of Legal Studies 26(2, Pt. 2):575-612.
Sloan, EA. (1998), "Automobile accidents", in: P. Newman, ed., The New Palgrave Dictionary of Economics
and the Law (Macmillan, London).
Sloan, EA., and T. Hoerger (1991), "Uncertainty, information and the resolution of malpractice disputes",
Journal of Risk and Uncertainty 4(4):343-363.
Sloan, EA., P.B. Githens, et al. (1993), Suing for Medical Malpractice (University of Chicago Press, Chicago,
IL).
Sloan, EA., P.M. Mergenhagen, et al. (1989), "Medical malpractice experience of physicians: Predictable or
haphazard?", Journal of the American Medical Association 262(23):3291-3297.
Smith, R. (1990), "The epidemiology of malpractice: Adverse events common, negligent cases rare", British
Medical Journal 301:621.
Spence, M. (1977), "Consumer misperceptions, product failure and product liability", Review of Economic
Studies 64:561-572.
Towse, A., and P.M. Danzon (1998), An Economic Analysis Of Clinical Negligence: The Options For The
NHS (Office of Health Economics, London).
Weiler, P.C. (1991), Medical Malpractice on Trial (Harvard University Press, Cambridge).
Weiler, P.C., H.H. Hiatt, et al. (1993), A Measure of Malpractice: Medical Injury, Malpractice Litigation and
Patient Compensation (Harvard University Press, Cambridge).
White, M.J. (1994), "The value of liability in medical malpractice", Health Affairs 13(4):75-87.
Winter, R.A. (1988), '"The liability crisis and the dynamics of competitive insurance markets", Yale Journal
of Regulation 5(2):455-500.
Zuckerman, S., R. Bovbjerg, et al. (1990), "Effects of tort reforms and other factors on medical malpractice
insurance premiums", Inquiry 27(2):167-182.
Chapter 27
ANTITRUST AND COMPETITION IN HEALTH CARE

MARKETS*
MARTIN GAYNOR and WILLIAM B. VOGT

H. John Heinz III School of Policy and Management, Carnegie Mellon University, Pittsburgh, PA
Contents
Abstract 1406
2. Background 1408
3. Is the health care industry different? 1409
3.1. Differentiated product 1410
3.2. Imperfect information 1413
3.2.1. Moral hazard in health insurance 1413
3.2.2. Adverse selection in health insurance markets 1414
3.2.3. Agency problems in health care markets 1415
3.3. Government regulation 1417
3.4. Not-for-profit firms 1418
4. Hospital mergers 1419
4.1. Enforcement 1419
4.1.1. Process 1419
4.1.2. Framework for analysis 1420
4.1.3. Outcomes to date 1421
4.2. Market definition 1426
4.2.1. Background and issues 1426
4.2.2. A proposed method 1429
4.2.3. Managed care and market definition 1435
4.3. Mergers, conversions, and ownership status 1436
4.3.1. Implications of the model: conversions 1439
4.3.2. Implications of the model: mergers 1440
4.3.3. Identifying the model 1443
*We gratefully acknowledge support from the Robert Wood Johnson Foundation and from the National Bu-
reau of Economic Research. We wish to acknowledge Deborah Haas-Wilson for many extensive conversations
and discussions that were instrumental in helping to frame much of the material in this chapter. Thanks also
go to Bob Berenson, Bob Bloch, David Dranove, Winand Emons, Roger Feldman, Ted Frech, Joe Newhouse,
Mark Pauly, Tom Philipson, Carol Propper, Bill Sage, Mark Satterthwaite, Mike Scherer, and Dennis Yao.
The usual caveat applies.
Handbook of HealthEconomics, Volume 1, Edited by A.J. Culyer and J.P Newhouse

( 2000 Elsevier Science B. V All rights reserved
1406 M. Gaynor and W.B. Vogt
4.4. Empirical evidence on hospital competition and mergers 1445

4.4.1. Structure-conduct-performance: price 1448
4.4.2. Structure-conduct-performance: non-price 1450
4.4.3. MAR, scale economies, and the efficiencies defense 1451
4.4.4. Event studies 1453
4.4.5. Studies of entry 1455
5. Monopsony and foreclosure 1456
5.2. Monopsony power 1457
5.2.1. Monopsony and antitrust enforcement 1457
5.2.2. Monopsony and welfare 1459
5.2.3. Detecting monopsony power 1461
5.2.4. Studies of monopsony in health care markets 1463
5.3. Bilateral market power 1466
5.3.1. Identifying bilateral market power 1466
5.3.2. Welfare implications 1468
5.3.3. Studies of relative bargaining power 1468
5.4. Foreclosure 1470
5.4.1. Efficiencies 1471
5.4.2. Anti-competitive concerns 1473
5.4.3. Health economics research on vertical restraints 1476
6. Conclusions 1478
References 1478
Abstract
In this chapter we review issues relating to antitrust and competition in health care
markets. The chapter begins with a brief review of antitrust legislation. We then discuss
whether and how health care is different from other industries in ways that might affect
the optimality of competition. The chapter then focuses on the main areas in which
antitrust has been applied to health care: hospital mergers, monopsony, and foreclosure.
In each of these sections we review the relevant antitrust cases, discuss the issues that
have arisen in those cases, and then review the relevant economics literature and suggest
some new methods for analyzing these issues.
JEL classification:Ill, L40, L44, L12, L13, L31, D21

Ch. 27: Antitrust and Competition in Health Care Markets 1407
1. Introduction
The U.S. health care system is organized around markets. l There has, however, been
ongoing concern about the functioning of these markets, so much so that some have
despaired of these markets working at all. The policy response to this concern has been
disjointed. Health care markets are subject to many regulations and interventions. Some
of these policies have attempted to substitute regulation for competition, regulating en-
try and investment (certificate of need laws, health planning) or price (all-payer regula-
tion). At the same time, health care markets have been subject to antitrust enforcement.
Recent years have seen a shift away from regulatory policies and toward competition.
Antitrust policy towards health care markets has become much more vigorous since the
early 1970s. Antitrust is intended to ensure the efficient functioning of these markets.
As a consequence, competition and antitrust policy have become prominent issues in
U.S. health policy.
Economic research is vital to addressing issues of competition and antitrust in health
care. This includes issues of market definition and detecting anti-competitive conduct.
In rule of reason cases, where benefits are weighed against costs, it includes measuring
the loss of consumer welfare resulting from a particular practice against any gains in
consumer welfare resulting from it.
In this chapter we consider research issues in the analysis of competition and the
application of antitrust to health care markets. 2 We outline our views on the analytical
issues and review the relevant literatures both from health economics and industrial
organization and antitrust generally. Our focus is mainly on hospitals and interactions
between hospitals and insurers. This is due, in part, to where there has been antitrust
activity. Physician markets have been for the most part very unconcentrated, and as
such have not lent themselves to the kinds of anti-competitive conduct the antitrust
laws prohibit. 3 Although issues of competition and antitrust in pharmaceutical markets
are fascinating and important, they differ in some fundamental ways from markets for
health care services, and as such, we exclude them from this chapter. 4
In what follows we first provide some background on antitrust and health care. We
then discuss distinctive features of the health care industry, then proceed to horizontal
and vertical issues in antitrust. Finally, we present a set of recommendations for future
research.
i While many health care systems outside the U.S. do not depend to the same degree on markets, competition
is nonetheless important. In particular, where there is competitive bidding or any system that uses cross firm
comparisons for contracting, competition will be important to ensure efficiency. We do not discuss such issues
in this chapter. See Chalkley and Malcomson (2000) on government purchasing of health services.
2 See also Dranove and Satterthwaite (2000) on the industrial organization of health care.
3 See McGuire (2000), Gaynor (1994), or Frech (1996) for a review of issues concerning physician markets.
4 See Scherer (2000).
1408 M. Gaynor and WB. Vogt
2. Background
The major antitrust statutes of the United States are the Sherman Act (1890), the Clayton
Act (1914), and the Federal Trade Commission Act (1914).5 The Sherman Act prohibits
attempts to restrain trade and attempts to monopolize. Sections 2 and 3 of the Clayton
Act prohibit price discrimination, tying, or exclusive dealing that substantially lessen
competition or create a monopoly. Section 7 prohibits mergers or other combinations
that could reasonably be expected to reduce competition or create a monopoly. The
Federal Trade Commission Act created the Federal Trade Commission and prohibits
unfair methods of competition which affect interstate commerce. An important law di-
rectly relevant to health care is the McCarran-Ferguson Act (1948), which exempts the
business of insurance from the antitrust laws if regulated by the state (excepting actions
such as boycott, intimidation, or coercion).
Antitrust issues in health care have only been an issue where markets are relied upon
to a large degree to determine prices and quantities (and other characteristics) of health
care services. Thus antitrust enforcement has been relevant only in the United States
for the most part, since most other countries have relied mainly on the public sector for
price or quantity determination. 6 Active application of the antitrust laws to the health
care sector in the U.S., however, can only be dated to the late 1970s and early 1980s.
Up until 1975 the courts fluctuated on the application of antitrust to the professions. 7
The debate was whether the "learned professions", such as law and medicine, were
subject to the antitrust laws. For example, in FTC v. Raladam Co., 283 U.S. 643 (1931)
it was suggested that medical practitioners "follow a profession and not a trade". On the
other hand, the Supreme Court ruled in 1943 that the American Medical Association
had violated the antitrust laws by helping the District of Columbia Medical Society
to deny staff privileges to physicians in the Group Health Association of Washington,
D.C., an early HMO [American Medical Ass'n. v. United States, 317 U.S. 519 (1943)].
In a landmark case in 1975, the Supreme Court decided in Goldfarb v. Virginia State
Bar [Goldfarb v. Virginia State Bar, 421 U.S. 773 (1975)] that there was no "learned
professions" exemption from the antitrust laws. Since that time there has been exten-
sive antitrust enforcement in health care. Sixteen health care cases have been decided
by the Supreme Court in the period following Goldfarb. Many more cases have been
decided by lower courts. A number of these have struck down activities by medical as-
sociations, such as parts of the ethical codes of the American Medical Association and
other professional associations which ban advertising, 8 associations setting fee sched-
ules for their members, 9 associations prohibiting members from certain practice forms
5 For a fuller description see Miles (1998), Chapter 1.

6 Note that competition policy issues have emerged in a number of countries as reforms have attempted to
create market incentives in the health system.
7 See Chapter 1, "General background" in Miles (1998).
8 For example, American Medical Association, 94 F.T.C. 701 (1980), enforced, 638 F.2d 443 (2d Cir. 1980),
aff'd by equally divided court, 455 U.S. 676 5 (1982).
9 Goldfarb v. Virgina State Bar, op. cit., Arizona v. Maricopa County Medical Society, 457 U.S. 332 (1982).
(notably HMOs), 1° and associations imposing restraints related to standards setting.ll

The courts have also seen cases involving mergers, denial of staff privileges, exclu-
sive dealing, most-favored-nations contracts, market allocation, and joint ventures [see
Miles (1998) Volume 2 for extensive detail].
In the rest of this chapter we focus on the types of behavior that have both been most
relevant in health care antitrust and, in our judgment, are most amenable to economic
analysis. We begin in the next section by discussing whether the health care industry
is different with regard to implications for the optimality of competition. In Section 4
we consider hospital mergers. Section 5 covers monopsony power and foreclosure, and
Section 6 contains our conclusions.
3. Is the health care industry different?
The economic justification for antitrust enforcement is that competition maximizes so-
cial welfare. While in most industries the link between competition and social welfare
is (or seems to be) direct, it is not obviously so in health care. Health care markets are
usually thought to differ from textbook competitive markets in a number of important
ways. Health care markets are characterized by multiple imperfections, in large part de-
riving from the uncertainty and asymmetry of information between buyers and sellers
that are inherent in the nature of health and medical care [Arrow (1963), Pauly (1978),
Gaynor (1994)]. As a consequence, questions of social welfare must be decided in the
context of the second-best. This implies not only that competition may not be second-
best, but that as a consequence, antitrust laws possibly should not be enforced in health
care markets, or enforced differently in health care than in other markets. Indeed, a com-
mon claim has been that health care markets are so fundamentally different from other
markets that meaningful competition is either an impossibility or counterproductive. As
a consequence, adherents of such views have been in favor of extensive government reg-
ulation or production in health care and hostile to the application of antitrust to health
care.12 It should, however, be pointed out that there are few real world markets that
satisfy the requirements for a textbook perfectly competitive market. Most markets are
characterized by some imperfections, and in many markets the deviations from the re-
quirements for perfect competition are substantial. The antitrust view is that all markets
are special, and that while competition in health care markets should be examined in
light of its special character, so should competition in any other market.
10 American Medical Association v. United States, op. cit., American Society of Anesthesiologists, 93 F.T.C.
101 (1979) (consent order).
11 Wilk v. American Medical Association, 671 E Supp. 1465 (N.D. III. 1987), aff'd, 895 E2d 352 (7th Cir.)
cert. denied, 496 U.S. 927 (1990).
12 A government regulator or purchaser may still want competition among providers in order to provide
incentives for efficiency and quality.
In what follows we consider the differences between health care markets and other
markets and discuss what is known about the implications of these differences for the
optimality of competition in health care markets.
Health care markets differ from textbook markets in the following major ways: the
product is differentiated, information is imperfect, there is extensive government regu-
lation, and many firms are not-for-profit. We discuss the general issues associated with
these differences and review the health economics literature on these specific imperfec-
tions and their implications for the desirability of competition in health care markets.
3.1. Differentiatedproduct
Health care is a differentiated product. Health care is a service and all services are
by their nature inherently heterogeneous and non-retradable. A haircut, an auto re-
pair, a will preparation, are never the same thing twice. 13 Further, these products
can only be purchased by direct interchange with the seller, thus they cannot be re-
traded among consumers. This is even more true of made to order, or customized, ser-
vices, e.g., a fashion makeover, music lessons, catering, investment advice, or repre-
sentation in divorce proceedings. This certainly applies to health care services, from
those which are routine, such as check-ups, to those which are specific to the pa-
tient (i.e., made to order, or "customized"). Further, preferences are heterogeneous.
Some patients will prefer extensive explanation and discussion of their case, others
would prefer the bare minimum. Some will prefer "Cadillac" or "gold-plated" treat-
ment, whereas others will only wish to pay for only that which is adequate or neces-
sary, the "Ford" or "Chevrolet" treatment. This may also be the case with immutable
seller characteristics (or those which cannot be changed ex post, like "putty-clay" in-
vestment, e.g., a physician's board certification, medical school degree, or a hospital's
religious affiliation or neighborhood location). Individuals may prefer physicians of a
particular age, sex, race, or ethnic background, or with a particular appearance (tall,
short, brown hair, blue eyes, conservatively dressed, fashionably dressed, hip, etc.).
Some may prefer a general practitioner for their primary care, while others may fa-
vor a board-certified internist. Similarly, some may prefer a Catholic or Jewish hospi-
tal, a hospital with extensive landscaping or modern architecture. Satterthwaite (1979,
1985) has stressed the idiosyncratic nature of preferences in the physician services mar-
ket in particular.
13 Strictly speaking, this is true of tangible products as well. Steel of a certain grade produced at the same
foundry varies between runs. No two cars are identical, even if they are produced on the same assembly line
by the same workers. Even simple tap water varies in its composition. This is simply the nature of the physical
world. Regardless, it is commonly agreed that services are inherently more heterogeneous than manufactured
or mined goods.
It is this combination of a heterogeneous product with heterogeneous preferences

which is key.l4 As Satterthwaite has argued persuasively, this bestows the seller with
market power. Patients choose sellers who produce the type of services and have char-
acteristics which best match their preferences. The fact that patients choose sellers who
give them the highest utility gives sellers market power, since switching to another
seller will reduce a patient's utility. The less substitutable are sellers for one another,
the greater the degree of market power, i.e., the lower is the elasticity of demand a
seller faces. In the case of physicians, since the vast majority of physicians are located
in urban markets with many other physicians, the market can be characterized as mo-
nopolistically competitive. 15, 16 In the case of hospitals, since most hospital markets are
characterized by a relatively small number of hospitals interacting over a long period,
this implies a differentiated product oligopoly. These are powerful paradigms which
can help to explain observed patterns in these markets. In particular, this implies that
we should expect sellers to possess power over price in these markets, even absent any
form of anti-competitive behavior. Health care markets do not seem to be fundamentally
different in this regard from other markets with substantial product differentiation.
Competition may or may not be optimal in markets with product differentiation. The
theoretical literature on monopoly and product quality and monopoly and product va-
riety does not provide strong guidance. In general, anything can happen. Competition
(monopoly) can result in too little quality or variety, too much, or just the right amount
[Spence (1976), Dixit and Stiglitz (1977), Salop (1979), Anderson et al. (1992)]. The
intuition is as follows. Competition may underprovide product variety, since competi-
tive firms cannot capture the consumer surplus from additional variety. It thus may not
be profitable for competitive firms to offer a product which has positive social value. A
monopolist, however, may do so, due to its ability to extract consumer surplus. On the
other hand, competition may overprovide product variety. In a competitive market, part
of the profit from offering a new variety will come from "stealing demand" from other
firms. A firm deciding whether to offer a new variety will not take into account this
external effect and thus too much product variety may be offered in a competitive equi-
14 It would probably be more accurate to say that it is preference heterogeneity which is critical, since this
should call forth heterogeneity in production by producers. This is especially likely to be true for services,
since the cost of altering the product to suit the customer (e.g., a barber chatting more with his garrulous
customers and less with his laconic ones) is likely to be low.
15 Over 88 percent of all physicians were located in metropolitan areas in 1989 [Roback et al. (1990)].
16 Even if physicians are very similar, information problems can cause patients to value physicians they
know about (in terms of quality or characteristics they value) more highly. This can bestow market power
on physicians and lead to a monopolistically competitive market structure even if physicians are essentially
homogeneous. Information is discussed in the next section.
librium. Whether product variety/quality is optimal, insufficient, or excessive depends

on whether and which of these two effects dominates. 17 18
Further, if price is fixed by regulation, then competition can lead to excessive prod-
uct differentiation, as in models of non-price competition in the (price) regulated airline
industry [e.g., Douglas and Miller (1974), White (1972), Van der Weide and Zalkind
(1981)]. Some have claimed that, due to insurance, hospitals do not (or did not) com-
pete on price to attract patients, but rather compete solely on quality or facilities to
attract patients (or doctors, who then bring patients with them). This has been called
the "medical arms race" [Robinson and Luft (1985)].19 This has some plausibility for
hospitals prior to the 1980s (although the empirical evidence is weak), since most hos-
pitals were reimbursed on the basis of costs and patients were both heavily insured and
unrestricted in their choice of health care provider. This notion of competition leading
to excessive product differentiation and higher prices through a medical arms race has
been influential in at least one antitrust case, in which the judge declared "[H]ospital
rates are lower, the fewer the number of hospitals in an area". 20
What all of this means is that economic theory does not give general guidance to
appropriate antitrust policy in markets with significant product differentiation. In this
regard health care is no different than many other markets. It is clear that price fixing
reduces social welfare. There is little else in the domain of antitrust policy that is clear
in general from economic theory. Mergers resulting in increased market power, price
discrimination, attempts to deter entry, and attempts to reduce competition may either
reduce or increase social welfare. While it may be possible to agree that attempts to
reduce entry or eliminate competitors such as limit or predatory pricing 2 1 are "unfair"
and should not be allowed, the welfare effects of such practices will depend on the
specifics in each case.
It is worth pointing out, however, that the antitrust enforcement agencies only con-
cern themselves with short-term anti-competitive impacts. The 1984 Merger Guidelines
[U.S. Department of Justice (1984)] specifically state that anti-competitive impacts are
of concern for only two years. If product characteristics are fixed for the short term, then
the only competitive concern is price competition. Price competition is clearly optimal
when product characteristics are fixed and no entry or exit occurs.
17 More definite results can be derived in specific models. For example, Deneckere and Rothschild (1992)
show for a class of models that competition is approximately optimal when firms' fixed costs go to zero
relative to the size of the market.
18 In an intriguing paper, Benrry and Waldfogel (1997) present empirical evidence indicating that there is
excessive entry in radio broadcasting. To our knowledge, this is the only empirical evidence on this issue.
19 It is not clear that the results for a price regulated industry apply directly to health care. Since price is
controlled by health care firms, it is not clear that the results would carry through in a carefully specified
model.
20 United States v. Carilion Health System (892 F 2d 1042).
21 Suspending for a moment the question of whether such strategies are rational or successful.
3.2. Imperfect information
It has long been understood that the structure of information is critical for understand-
ing health care markets [Arrow (1963)]. What is critical is uncertainty and, more im-
portantly for the issues here, asymmetries in information. The important asymmetries
lead to moral hazard in health insurance, adverse selection in health insurance markets,
and agency problems in health care markets. What is critical here is the extent to which
these information problems either impair the workings of health care markets so that
competition is ineffective, or are exacerbated by increases in competition. We review
these issues as associated with each of these imperfections, in turn.
3.2.1. Moral hazard in health insurance
Consumer uncertainty about illness and its associated losses leads to a demand for health
insurance. However, since it is difficult to verify the exact nature of the illness and the
appropriate treatment, there is an asymmetry of information between the consumer and
the insurer leading to moral hazard.
One of most important imperfections in health care markets is the moral hazard ef-
fect on consumption due to health insurance [Arrow (1963), Pauly (1968), Zeckhauser
(1970)]. The vast bulk of health care bought and sold is covered by insurance. This
means that the consumer of the service himself pays only a small fraction of the cost,
while the majority is reimbursed by third-party payers. Insuring individuals against risk
reduces the price they face for medical care, thereby inducing excessive consumption.
Since moral hazard induces excessive consumption, it might seem that market power
on the part of firms selling health care services could improve matters by restricting
output [Crew (1969), Frech (1996), Pauly (1998)]. Indeed, for a fixed insurance policy
for which there is excess consumption of health care goods and services at the compet-
itive (marginal cost) price, raising the price of health care goods and services above the
competitive level increases welfare. 2 2
It turns out, however, that if the insurance market is competitive, then setting price at
marginal cost in the health care product market is always second-best optimal [Gaynor,
Haas-Wilson, and Vogt (forthcoming)]. The insurance market will sell insurance poli-
cies that maximize consumer welfare conditional on prices in the health care product
market.
Another form of moral hazard is the decreased incentive for consumers to search for
the lowest price [Newhouse (1978)]. In general, however, insurance may lead consumers
to search less, or more, in equilibrium [Dionne (1984)].23 The presence of insurance
22 This is essentially the intuition in Crew (1969). Gaynor. Haas-Wilson, and Vogt (forthcoming) show this
formally.
23 He assumes a homogeneous product. It is not clear what impact introducing product heterogeneity might
have on the results. The papers which consider patient search with a heterogeneous product, e.g., Dranove
and Satterthwaite (1992), do not consider the effects of insurance.
will tend to lead to greater dispersion in prices. It is possible that consumers may search
more because the increased dispersion in the price distribution means that there is a
bigger potential payoff to search, since the expected gap between the current price and
the lowest price is larger. On the other hand, since consumers pay only a small fraction
of the expense, the payoff to search may still not exceed the costs. The emergence of
highly price responsive buyers of health care in the form of managed care plans can be
thought of as an institutional innovation in response to this problem.
Therefore in health care markets insurance may lead to quality's being the salient
competitive factor (as in the medical arms race story). In this case, search may focus
entirely on quality. Dranove and Satterthwaite (1992) show that if there is imperfect
information about both price and quality then in the case where only information about
quality is improved (beyond the level required for first-best quality), both price and
quality increase to supra-optimal levels. This result is suggestive of what might happen
in health care markets, but not definitive, since there is no insurance in the Dranove and
Satterthwaite formulation.
Dranove and Satterthwaite (1992) also show that it is possible for increased price
information to either decrease or increase welfare. If only information about price is
improved (beyond the point necessary for first-best quality), it is possible that con-
sumer welfare will decrease. Price will fall, benefiting consumers, but firms' margins
fall enough that quality decreases, causing a net decrease in consumer welfare. It is
also possible for the welfare effect of the price decrease to outweigh that of the quality
decline, thus resulting in an increase in welfare.
This result occurs in a setting where consumers pay the full price for their purchases.
It is not clear that improved information about price will lead to a significant price de-
crease in health care markets where consumers are heavily insured. Hibbard and Weeks
(1989) provide evidence that increased information on prices has little effect on con-
sumers' decisions, in large part due to insurance coverage. Nonetheless, the Dranove
and Satterthwaite result may occur in the presence of highly price-responsive managed
care plans. Indeed, much of the concern over managed care seems to be due to a per-
ceived emphasis on price at the expense of quality.
3.2.2. Adverse selection in health insurance markets
Another form of information asymmetry between consumers and insurers is when con-
sumers are of different risk types known to themselves but not to insurers, leading to
adverse selection in health insurance markets. This is another well-known imperfection
in health care markets. The classic paper by Rothschild and Stiglitz (1976) shows that
competitive insurance markets may fail in the face of adverse selection. When individ-
uals know their risk types but insurers don't, an equilibrium in which both types are
pooled in a single insurance contract does not exist. A separating equilibrium may exist
in which high risk individuals purchase complete insurance, while low risk individuals
purchase incomplete insurance. This equilibrium will not exist if there are "sufficiently
few" high risk types. Stiglitz (1977) analyzes the equilibrium insurance contracts of-
fered by a monopolist in the face of adverse selection. He shows that a pooling equi-
librium does not exist in the case of monopoly either. In the separating equilibrium the
high risk individuals purchase complete insurance. Low risk individuals either purchase
partial insurance or no insurance. The terms of low risk individuals' contracts are such
that they are indifferent between purchasing insurance and having no insurance.
The welfare comparison between competition and monopoly in insurance markets
based on the Rothschild-Stiglitz analysis is unclear. 24 Competition is suboptimal, but
monopoly may be no better. Further, there are other theoretical analyses in which com-
petition does not pose a problem for existence [Spence (1978), Miyazaki (1977), Ri-
ley (1979), Wilson (1977), Cave (1986)]. Another issue is that it is possible that the
Rothschild-Stiglitz assumption of an immutable risk type known to the individual is not
plausible. Some recent evidence indicates that, while there is some persistence among
individuals in their health expenditures, this only lasts at most for 4 or 5 years [Eichner
et al. (1997)]. In consequence, the normative implications for competition in real world
health insurance markets is unclear.
The empirical evidence on the functioning of health insurance markets is mixed. First,
it appears that insurers do not price differentially by risk, either because of transactions
costs or regulations banning such practices. As a consequence, most analysts are more
concerned with favorable selection by insurers (this is often referred to as "cream skim-
ming" or "cherry picking") seeking out good risks and avoiding bad risks, than with
adverse selection by individuals [Pauly (1986), Newhouse (1996)]. Second, while it
appears there is risk selection in terms of enrollment in HMOs versus conventional in-
surance plans [Hellinger (1995), Newhouse (1996)], there is also evidence pointing to
consumer persistence in their choice of insurance plans [Neipp and Zeckhauser (1986),
Madrian (1994), Royalty and Solomon (1997)].
3.2.3. Agency problems in health care markets
The product in this market is not a typical service in that sellers (physicians) do not
provide only medical care, or treatment, but also diagnosis and advice. The physician
is hired to make a diagnosis and provide advice on the appropriate course(s) of action.
This, in and of itself, is not unique to health care markets. Repair services generally
have this characteristic. An auto mechanic diagnoses a car problem and provides a rec-
ommendation for action.2 5 Appliance repairmen operate in the same way. Stockbrokers
or financial advisers diagnose an individual's financial situation and advise them on in-
vestment strategies and on particular investments. Taxi drivers are told the destination of
24 Monopoly does have the advantage that the insurer knows the total quantity of a consumer's insurance
purchases. Rothschild-Stiglitz assume that this is known in a competitive market, but it may be difficult for
such a market to set up this exchange of information [see Jaynes (1978) and Hellwig (1998)]. If information
cannot be exchanged by firms in the competitive market then competition results in a 3rd best, and is thus
inferior to monopoly.
25 Hubbard (1998) investigates the functioning of the market for automobile repairs. He finds evidence in-
consistent with significant agency problems in this market.
the passenger and allowed/requested to choose the route. The purchase of such services
has been analyzed under the literature known as "games of persuasion".26
Such games have the following information structure. The consumer observes a
symptom (engine knocking, pain in the side). The consumer observes only the symp-
tom, but does not know the nature of the problem. There is some probability, known
to all, that the problem is a serious problem, with an expensive treatment/repair and
some probability it is minor (inexpensive treatment). The expert can observe (let's say
with certainty) whether the problem is serious or minor, and can perform both the ex-
pensive and the inexpensive treatments.2 7 The consumer prefers the appropriate treat-
ment/repair, but cannot observe the problem. Thus, the consumer's choice can only be
based on a report that has been sent strategically by the doctor. There is an asymme-
try of information between the buyer and seller, and thus an agency relationship. This
creates the possibility for the seller to misrepresent the nature of the consumer's prob-
lem to him. An incentive to do so exists when the more expensive repair is also more
profitable.
This problem has been referred to in the health economics literature as "induced de-
mand" and is another frequently mentioned imperfection in health care markets. While
deductively it is plausible that physicians exploit their role as agents to induce some
demand from their patients, it is not clear whether this is an empirically significant phe-
nomenon. 28 The empirical literature on this topic suffers from such severe methodolog-
ical flaws that it does not provide useful evidence on either the existence or magnitude
of this effect [see McGuire (2000) or Gaynor (1994) or Frech (1996, Chapter 5) for
discussion].
The relevant issue here is whether competition leads to more or less inducement. One
of the most common attempts to test for demand inducement by physicians takes the
form of regressing price against the number of physicians per capita (and controls). The
notion is that if the market is competitive, the presence of more physicians implies a
greater supply and hence a lower price. Demand inducement, by contrast, is hypothe-
sized to occur in the following way. Entry of more physicians implies a reduction in
demand per physician on average. Physicians will thus induce demand to increase their
incomes. While this "model" and its empirical testing suffer from severe flaws, what
concerns us here is the relationship of inducement to competition.
A number of models with profit maximizing physician firms produce the result that
competition reduces or eliminates inducement. Stano (1987) constructs a model in
26 See, e.g., Darby and Karmi (1973), Milgrom (1981), Wolinsky (1993), Plott and Wilde (1980), Crawford
and Sobel (1982), Pitchik and Schotter (1987), Green and Stokey (1980), Nitzan and Tsur (1989) generally,
and Pauly (1980), Dranove (1988), and Glazer and McGuire (1992) for discussions of this in health care.
27 A recent paper by Ma and McGuire (1994) analyzes the case where the accuracy of the diagnosis is in-
creasing in physician effort. Physician effort is noncontractible, and this noncontractability causes physicians
to shirk and undersupply effort.
28 See, however, Emons (1997) for a result in which no inducement occurs in equilibrium, even though sellers
have the ability to induce.
which profit maximizing physician firms choose prices and inducement levels. Stano
shows that, so long as entry increases the price elasticity of demand facing the physician
firm and marginal inducement costs are constant, entry reduces inducement. 2 9 Stano's
model does not allow physician firms to leave the local market and move to another one
in response to entry. Since exit from a local market does not imply (necessarily) exit
from the industry, the only losses from relocation to a new market are the sunk costs
of entering a local market, which for physicians are the costs of building a practice.
If exit from the local market is a viable strategic choice, exit may further reduce the
attractiveness of inducement as a response to entry.
Consider also the following simple competitive model. Physician firms maximize
profits. They have constant marginal costs and fixed capacities. The firms have different
marginal costs. They enter a market when their variable profits exceed the (common)
fixed cost of entry. Price is set at the intersection of supply and demand schedules,
keeping in mind that supply becomes vertical at industry capacity. Then as demand
in a market grows firms will enter in (inverse) order of their marginal costs. Even if
firms price at marginal cost, then markets with more physicians will have higher prices.
This leads to a positive correlation in equilibrium between price and entry, but implies
nothing about demand inducement.
If physician firms are not profit maximizers, but utility maximizers, then it may be
possible that income effects can lead to increased inducement in response to entry. How-
ever, to our knowledge there are no papers that have analyzed precisely this situation.
McGuire and Pauly (1991) find that a monopolist only engages in inducement in re-
sponse to a (regulated) price decrease if income effects outweigh substitution effects.
There is some weak evidence for the existence of income effects in physician labor
supply [see Sloan (1975), Hurdle and Pope (1989), Lee and Mroz (1991), Rizzo and
Blumenthal (1994)]; however, the results of Rizzo and Blumenthal (1994) show that the
substitution effects of a wage change outweigh the income effects by a wide margin.
Thus, even if physicians maximize utility (rather than profit) the available evidence is
not consistent with an increase in inducement in response to entry and price decreases.
While it seems likely that some inducement exists due to asymmetric information and
agency problems [Darby and Karni (1973)], neither theory nor current evidence indi-
cate that competition is likely to increase distortions from this market imperfection, but
rather just the opposite.
3.3. Government regulation30
Health care markets are characterized by extensive regulation. Regulations cover mar-
ket entry (physician licensure, hospital certificate-of-need regulation), pricing (hospital
29 If marginal inducement costs are increasing, the results are indeterminate, because entry decreases the
marginal revenue from inducement but also reduces marginal cost.
30 See Salkever (2000) for more extensive discussion.
all-payer regulation, insurance premium regulation), product offerings (insurance man-

dates), and the activities of not-for profit firms (community benefit requirements). Some
regulations are not intended to replace competition, but rather to augment it. Regulations
like physician licensure or requirements for full information disclosure on insurance
coverage are intended to protect consumers who may be poorly informed. Regulations
such as these do not affect the desirability of competition.
Other regulations, such as all-payer regulation, are intended to replace market deter-
mination of price. Whether competition is beneficial in the presence of such regulations
may depend on what particular constellation of regulations are present. For example,
when hospitals' prices are controlled by regulation (e.g., all-payer regulation), then
competition may occur over services and could well result in supra-optimal levels of
service adoption. Certificate-of-need regulation may be an optimal response to this sit-
uation (the evidence, however, is that certificate-of-need regulation has not been effec-
tive). Another example is mandated community rating of premiums for insurance firms,
i.e., insurers must charge a single premium to all consumers based on expected losses in
the community. It is certainly possible that in this case competition among insurers may
take the form of competing for the good risks, i.e., favorable selection, as discussed
previously. Such competition is inimical to welfare. The solution may be regulations
designed to combat this phenomenon, or removal of the original price regulation. Situ-
ations like these, however, rarely become the concern of antitrust enforcement. Usually
state government does such extensive regulation and this usually supersedes the antitrust
laws under a doctrine known as "state action immunity".
3.4. Not-for-profitfirms
A distinctive institutional characteristic of the health care sector in the United States
is the presence of not-for-profit firms. In particular, the hospital sector is dominated
by not-for-profit firms. This has led to two issues about the desirability of competition
in hospital markets. One question is whether not-for-profit hospitals exercise market
power. The alternative that has been advanced is that not-for-profit hospitals seek to
maximize the welfare of the community in which they are located, and thus do not ex-
ercise market power if given the opportunity. This argument was very influential in a
recent court decision to allow the merger of the two largest hospitals in Grand Rapids,
Michigan. 3 1 The second issue is the impact of increased competition on charity care
provided by hospitals. Hospitals, particularly not-for-profit hospitals, provide a lot of
charity care to indigent patients. To the extent that such care is financed out of profits,
increased competition may reduce charity care. While it is true that financing charity
care via monopoly profits is inefficient, charity care which is lost due to increased com-
petition may not be replaced, due to the vagaries of politics. We take these issues up in
the next section.
31 Federal Trade Commission v. Butterworth Health Corporation and Blodgett Memorial Medical Center
(1996, 2 Trade Cases 71,571); 1996 QL 570479 (W.D. Mich. September 26, 1996).
4. Hospital mergers
We now turn to a discussion of hospital mergers. Hospital mergers are both one of the
most prominent areas of antitrust enforcement in health care and an area where eco-
nomic analysis plays a critical role. We first provide an overview of the enforcement
process and outcomes in Section 4.1. In Section 4.2 we discuss market definition. Sec-
tions 4.3 through 4.5 contain a discussion of merger analysis, focusing in particular
on the effect of not-for-profit status. In Section 4.6 we review the empirical evidence
relevant to hospital mergers.
4.1. Enforcement
4.1.1. Process
Hospital mergers are covered by the same antitrust laws as are mergers in other indus-
tries. In particular, Section 1 of the Sherman Act and Section 7 of the Clayton Act forbid
mergers which, in the words of the Clayton Act, "may be substantially to lessen com-
petition, or to tend to create a monopoly". Typically enforcement of the antitrust laws
with respect to hospitals works as follows. Section 7A of the Clayton Act requires, in
the case of large acquisitions (and hospital mergers are normally large for these pur-
poses), that the merging parties inform the Department of Justice (DOJ) and the Federal
Trade Commission (FTC) of their intent to merge. The firms must then wait thirty days,
during which time the DOJ and FTC investigate the proposed merger.
During this thirty day period, the agencies may issue a request for additional infor-
mation to the merging firms, requiring the firms to produce information relevant to the
competitive impact of the proposed merger. At this time also, the agencies decide which
of them is to pursue the matter. The merging firms are then required to wait twenty days
after they have complied with the request for additional information before they may
consummate the merger.
During this time, the relevant agency may seek a preliminary injunction in federal
court to prevent the merger. If the agency is the DOJ, it will seek a preliminary and
permanent injunction, typically; whereas, if the agency is the FTC, it will seek a prelim-
inary injunction, pending a trial-like administrative proceeding within the FTC before
an administrative law judge. If the agencies fail to obtain a preliminary injunction, the
merger can proceed. If the agencies fail to obtain a preliminary injunction, they may
still pursue a trial in federal court (DOJ) or the administrative procedure (FTC) in order
to force the firms to divest; however, this is not typically done. A decision at the district
court stage may be appealed to the circuit court and to the Supreme Court. A decision
in the administrative proceeding may be appealed to the full Federal Trade Commission
and then to circuit court and to the Supreme Court.
This process may end in a number of ways. Most commonly, the relevant agency takes
no action. The agency may also offer to allow the merger, provided that the merging
firms agree to some conditions (for example divestiture of some assets). If the merging
firms agree, a consent decree or consent order is the outcome. If the agency decides to
try to stop the merger, it must go either to court or to an administrative proceeding, to
obtain an order either to enjoin the merger or to force the merged firms to divest. We
review (briefly) the decisions in the eleven cases that we know of in which the process
has run all the way to a decision in this last stage.
4.1.2. Frameworkfor analysis
In deciding whether or not a merger is illegal under the Clayton or Sherman Acts, the
courts and the enforcement agencies go through a fairly routinized sequence of steps.
We follow Miles (1998) in describing the sequence as a six-step process. 3 2 The steps
are:
1. Definition of the relevant product market.
2. Definition of the relevant geographic market.
3. Identification of the competitors in the relevant product/geographic market.
4. Calculation of market shares of the competitors and Herfindahl index of concentra-
tion.
5. Calculation of merging firms' post-merger market share and the post-merger
Herfindahl, and determination of the likely competitive effect of the merger.
6. Consideration of any factors which mitigate or exacerbate anticompetitive effects. 3 3
In the definition of the relevant market, both product and geographic, the idea is to
find the smallest group of products for which there are not close substitutes. A product
is to be included in the relevant market if buyers could easily substitute to it in response
to a price increase of the merging firms. In addition, a supplier is to be included in
the relevant market either if it does produce the same product, or does not produce the
relevant product currently, but could easily come to produce it. The Merger Guidelines
articulate a test of market definition which, at least in principle, can be implemented.
The Guidelines say that the relevant market, both product and geographic, is the smallest
market in which the sellers, acting as a cartel, could profitably implement a "small but
significant and nontransitory price increase". They later go on to say that, most often,
this means a 5% increase for the period of at least one year.
Once market definition is completed, market shares and concentration indices are
straightforward to calculate. There is ambiguity in the hospital industry regarding the
proper variable for calculating market shares. Proposals include total revenues, inpatient
revenues, beds, admissions, and patient-days. Choosing among these alternatives does
not seem to make a substantial difference in the outcome of the calculations. Once
the various indices have been calculated, the analyst looks at the post-merger market
32 The DOJ/FTC Joint Merger Guidelines [U.S. Department of Justice and Federal Trade Commission
(1997)] describe it as a five step process. The organization is somewhat different, but consists of essentially
the same factors.
33 These include the potential for entry and any likely impact, potential for efficiency gains, and the potential
that one of the merging firms will fail without the merger.
share, the pre and post merger Herfindahl-Hirschmann Index (HHI),3 4 and the change
in the HHI, as well as other evidence (testimony of competitors and third party payers,
for example) in order to determine the likely competitive impact of the merger. No
firm standard appears to exist for a concentration index threshold. However, there are
guidelines. The Supreme Court [U.S. v. Philadelphia National Bank et al. (1963, 374
U.S. 321)] has found that a post-merger market share of 30% is high enough to make a
merger presumptively illegal. The Merger Guidelines say that a merger which increases
the HHI by 50 or more points and results in an HHI of more than 1800 is likely to
be anticompetitive and that a merger which increases HHI by 100 or more points and
results in an HHI of more than 1000 is also likely to be anticompetitive.
If this evidence and the testimony of competitors and third party payers is strong
enough, the merger is presumptively illegal, and the merging firms must present evi-
dence that there are other (non-concentration related) reasons that the merger will have
pro-competitive effects. The most common such defense is the efficiency defense. Merg-
ing firms argue that they will achieve cost savings by the merger, which will be passed
along to consumers in the form of lower prices. Another defense is the failing firm de-
fense, in which the merging firms argue that, without the merger, one of the firms will
fail, so that the merger itself will not result in fewer firms (that there will be fewer firms
anyway). To make this defense the firms must establish not just that one firm is failing
but that there are no other potential buyers for the failing firm. In addition, firms may
argue that entry into the market is fast and easy, so that an attempt at supra-competitive
pricing will be defeated by entry of new firms.
In hospital merger cases, other defenses have been attempted. Hospitals have argued
that they are in a closely regulated industry, and that the government will not permit
them to behave in an anti-competitive way. Hospitals have claimed that, since they are
not-for-profit firms, they will not behave in an anti-competitive way. Hospitals have
claimed that, since there is no price competition in the first place, there is no competition
to be protected. Hospitals have also argued that competition in the hospital industry is
a bad thing, and that getting rid of it is therefore beneficial to consumers. The analyst
must then weight these considerations against the potential harms associated with less
competition to decide whether the merger will benefit consumers or not.
4.1.3. Outcomes to date
Table 1 presents a list of cases. We denote the cases by the name of the city in which
the merging hospitals were located.3 5 For the cases the government lost, we provide
our reading of the primary reason for the decision. For the sake of brevity, in what
34 The HHI is defined as the sum of squared market shares for all firms in the market. The antitrust authorities
use percentage of the market (as opposed to proportion) for market share, thus the maximum value for the
HHI is 10,000 and the minimum is n(100/n)2 = 10,000/n, where n is the number of firms.
35 Citations for the decisions are as follows: Poplar Bluff [FTC v. Tenet Healthcare (1998, U.S. Dist. Lexis
11849)], Long Island [U.S. v. Long Island Jewish Medical Center (1997-2 Trade Cases [71,960)], Grand
1422 M. Gaynor and WilB. Vogt
Table 1
Hospital merger cases
Year Case Winner Primary reason for decision
1998 Poplar Bluff FTC

1997 Long Island Hospitals Product market definition
1996 Grand Rapids Hospitals Not-for-profit organization,
Efficiencies
1995 Dubuque Hospitals Geographic market definition
1995 Joplin Hospitals Geographic market definition,
Not-for-profit organization
1994 Ukiah Hospitals Geographic market definition
1991 Augusta (District) Hospitalsl Not-for-profit organization
1991 Augusta (Circuit) FTC
1989 Rockford DOJ
1989 Roanoke 2 Hospitals Market definition
1985 Chattanooga FTC
1984 San Luis Obispo FTC
l This decision was vacated and remanded by the circuit court, see the next line in the
table.
2 This case was a jury trial. Both the court and the jury agreed that the merger should
be allowed; however, the findings of the two were slightly different, so we present the
court's opinion.
follows we generically say "court" where we could mean court, or administrative law
judge, or full Federal Trade Commission, since different cases are decided by different
entities, depending on the circumstances, as discussed in the preceding section. When
the government has prevailed, it has managed to convince the court on steps one through
six above. 3 6 As can be seen from the table, the government has prevailed in five of the
eleven cases, but, before its 1998 victory in Poplar Bluff, had not prevailed in a case
since 1991.
In order to gain some insight into the key questions in these cases, with an eye to-
wards identifying where economic research can provide valuable input, we will proceed
through the six steps outlined above, summarizing as we go the typical method that the
court uses in its analysis.
Rapids [FTC v. Butterworth Health Corporation (1996, 946 F. Supp. 1285)], Dubuque (902 F. Supp. 968,
1995), Joplin [FTC v. Freeman Hospital (1995, 911 F. Supp. 1213)], Ukiah [ Adventist Health System/West
(1994, 117 FTC 223)], Augusta [District Court: FTC v. University Health (1991-1 Trade Cases 69,444).
Circuit Court: FTC v. University Health (11th Circuit, 1991, 938 F.2d 1206)], Rockford [U.S. v. Rockford
Memorial Corporation (1989, 717 F. Supp. 1251)], Roanoke [U.S. v. Carilion Health System (707 F. Supp.
840)], Chattanooga [Hospital Corporation of America (1985, 106 FTC 361)], San Luis Obispo [American
Medical International (1984, 104 FTC 1)].
36 In fact, the court rarely accepts the government's (or the hospitals') case in its entirety, but for the govern-
ment to prevail, the court must accept, in large measure, each of the government's points.
In defining the product market in which hospitals compete, courts have typically used
a "cluster market" approach. So, a typical product market definition is "general acute
care hospital services". However, outpatient care was included in the product market
in the Roanoke decision. Later, in the calculation of market shares, this product mar-
ket definition means that inpatient revenue, beds, or inpatient days for all services are
treated as a "single product" for the purpose of measuring market share.
As is discussed, for example, by Baker (1988), this cluster of services approach is
likely to overstate the size of the relevant market if the component parts of the clus-
ter are not easily substitutable in demand or supply and if the various hospitals in the
relevant geographic market produce the various component services in differing pro-
portions. For example, consider two hospitals, one of which specializes in cardiac care
and another in oncology, and suppose that resources are not easily and quickly fungible
between these two uses on the supply side. These two hospitals will appear to be com-
petitors in "general acute care hospital services"; whereas they are really operating in
totally separate product markets. This is acknowledged to some extent by the courts, as
there are some variations in the product market definitions. Some courts have divided
the product market into primary/secondary and tertiary inpatient care (Roanoke, Long
Island, Poplar Bluff) or into primary and general inpatient care (Grand Rapids).
Geographic market definition has most often relied heavily upon the Elzinga and
Hogarty (1973) technique of analyzing shipments data in order to infer the boundaries of
geographic markets [Morrisey et al. (1988)]. Using this technique, a putative market is
expanded until it encompasses a geographic area large enough so that sales from sellers
outside the area to buyers inside the area are small and sales from sellers inside the
area to buyers outside the area are also small. In other words, the geographic market is
expanded until "imports" and "exports" are a small proportion of total sales. In hospital
cases this is typically measured by flows of patients (inpatient discharges is the actual
measure used) out of and into the area. There are no universally applied cutoffs for these
criteria, although the courts follow Elzinga and Hogarty in calling a cutoff of 10 percent
a "strong market" and a cutoff of 25 percent a "weak market". This exercise is regarded
by the courts as a starting point in market definition. Additional evidence which the
courts typically consider are the admitting and referral patterns of physicians, overlap
in physician admitting privileges among hospitals, claims by executives of hospitals on
whether they consider themselves to be competitors, planning documents of merging
and non-merging hospitals, and claims by third party payers about their ability to defeat
potential price increases by shifting their patients over geographic space.
Notwithstanding the similarity in the courts' approaches and what appear to be fairly
similar facts, there is a great deal of heterogeneity in the outcomes of the market defi-
nition process. Geographic markets have been found to be as small as a single county,
with a single city as a relevant submarket (San Luis Obispo) to as large as 19 counties
(Roanoke) or to include hospitals as far away as 100 miles from the merging hospitals
(Dubuque). Several issues come up repeatedly in the disputes over market definition.
First, it is typically found that patients flow from outlying areas into a central city.
The hospitals then argue that this militates for including hospitals in rural areas in the
relevant market. The government argues that these flows do not establish demand sub-
stitutability between rural and urban hospitals, since people flowing from rural to urban
areas are seeking different care (are in a different product market, essentially) than are
the people staying in the rural area. Alternatively and relatedly, the government argues
that the hospitals in the urban area are differentiated from the rural hospitals by repu-
tation, quality, breadth of services, etc. (again, essentially an argument that they are in
a different product market). Sometimes the courts accept this argument (Rockford, San
Luis Obispo, Grand Rapids, Poplar Bluff) and sometimes they do not (Joplin, Dubuque,
Roanoke, Long Island, Ukiah).
Second, the relative importance of managed care and physician loyalty in steering
patients to providers is frequently disputed. The government typically claims that since
physicians admit to few hospitals (near their offices) and since patients are very loyal to
their physicians, geographic product differentiation is extremely important and, there-
fore, it would be very difficult for managed care plans to channel demand away from
local hospitals in response to price increases. The hospitals argue that managed care
has great power to channel patients and would do so in response to a price increase. 37
Again, sometimes the courts find that patients are tightly tied to their local doctor and
hospital (San Luis Obispo, Chattanooga, Rockford, Poplar Bluff) and sometimes the
courts find that they are easily directed to distant hospitals by financial incentives and
selective contracting (Joplin, Dubuque, Ukiah, Long Island). 38
The calculation of market shares and HHI proceeds once the relevant market has been
decided. The shares are usually based upon such things as beds, admissions, revenue,
or patient-days, and differing choices among these measures do not produce noticeably
different results, at least in the decisions where more than one of them are presented.
Table 2 presents values for post-merger share, post-merger HHI, and change in HHI
for mergers in which these figures could be gleaned from the decision. In all cases the
market definition settled upon by the court was used.
With few exceptions these markets are very concentrated, the merging firms would
command a large market share post-merger, and the HHI would increase a great deal as
a result of the merger. 3 9 For the most part, the government theorizes that anticompetitive
37 Distance is an extremely important observed determinant of patient choice of hospital. Demand for a
managed care plan's product depends on the hospitals in its network. Therefore a managed care plan has a
derived demand for hospitals for its network based in part on hospitals' distances from consumers. Managed
care demand also depends on price, which in turn is affected by hospitals' prices. Which of these two positions
is correct depends on the tradeoff between price and distance consumers make when choosing insurance.
38 While how far consumers are willing to travel for hospital care is not a settled matter, some empirical
evidence indicates that managed care has almost no impact on the observed distance consumers travel for
hospital services [Mobley and Frech III (1997)].
39 All relative to prior decisions in other industries and to the Horizontal Merger Guidelines [U.S. Depart-
ment of Justice and Federal Trade Commission (1997)]. The Guidelines state that a market with an HHI
below 1,000 is considered unconcentrated, a market with an HHI between 1,000 and 1,800 is considered
moderately concentrated, and a market with an HHI above 1,800 is considered highly concentrated. Mergers
in moderately concentrated markets that result in an HHI increase of 100 points or more are considered to
Table 2
Merger effects on concentration
Location Post merger share Post merger HHI Change in HHI
Poplar Bluff 84% 6,000-7,000 2,700-3,200

Grand Rapids 47-65% general 2767-4521 general 1064-1889 general
65-70% primary 4506-5079 primary 1675-2001 primary
Joplin 24% 1624 222
Augusta 43% 3200 >630
Rockford 72% 5647 2621
Roanoke about 20 hospitals
Chattanooga 18.2% 2495 151
San Luis Obispo 87% city 7775 city 3405 city
76% county 6025 county 2207 county
effects will take two forms. First, transaction prices paid to managed care organizations
will rise with the increase in market power caused by the merger. Second, quality of
care will fall, as hospitals gain market power in their competition for patients whose
payers pay a fixed, non-negotiable fee (Medicare and Medicaid). With the exception of
Roanoke, it is reasonably clear that in all of the mergers in the table the government
met its prima facie case for anticompetitive effects, using either the market share test or
the standards in the merger guidelines. Of the cases in the table, the government lost in
Grand Rapids, Joplin, and Roanoke.
Hospitals in merger cases argue that there will be efficiencies from the merger. These
claimed efficiencies typically are to result from consolidated operations, savings on ad-
ministrative expenses, improved management practices, elimination of duplicative facil-
ities and equipment, and, sometimes, via the elimination of cost-increasing competition
("medical arms racing"). In hospital merger cases, as in antitrust cases in general, the
efficiency defense is very hard to make successfully. Cost savings are usually specula-
tive and disputed. Neither the realization of the cost savings nor the passing along of the
savings to consumers is easily enforceable. Courts usually regard the efficiency defense
with skepticism.
In hospital merger cases, courts have sometimes found the efficiency claims plausi-
ble. In the Grand Rapids case, the District Court's decision in Augusta, the Roanoke
case, the Long Island case, and the Ukiah case, the courts found that the mergers would
generate significant efficiencies. These courts found both that there would be operat-
ing efficiencies from the consolidation of laundry, administration, and similar functions
and that there would be savings associated with reducing non-price competition. In the
potentially raise significant competitive concerns. In highly concentrated markets mergers that raise the HHI
by more than 50 points but less than 100 points are considered to potentially raise significant competitive
concerns. Mergers in highly concentrated markets that increase the HHI by 100 points or more are presumed
likely to create, enhance, or facilitate the exercise of market power.
Roanoke case, the court found that "hospital rates are lower, the fewer the number of
hospitals"; in the Grand Rapids case, the court found that the merger would serve to end
the "medical arms race" which was previously occurring; in the Augusta case the court
found that the merger would eliminate wasteful duplication of services caused by "a de-
sire among those who administer hospitals to have expensive equipment because other
hospitals have it". In the Ukiah case, the administrative law judge found that hospital
costs were lower in one hospital than in two hospital towns and that this was indicative
of wasteful non-price competition.
In cases involving the merger of not-for-profit hospitals, the merging parties often
argue that their not-for-profit status makes it unlikely that they will exercise any market
power which they gain. Again, there is heterogeneity in courts' willingness to accept this
argument. The decisions range from the District Court in Augusta saying "The board
of University Hospital is quite simply above collusion", to Judge Posner in HCA vs.
FTC (1986, 807 F2d 1381) saying "no one has shown that [nonprofit status] makes the
enterprise unwilling to cooperate in reducing competition ... which most enterprises
dislike and which nonprofit enterprises may dislike on ideological as well as selfish
grounds" and the court in Rockford which found that the two nonprofit hospitals in that
case had, in fact, colluded in the past. The court in the Grand Rapids case found the not-
for-profit defense quite persuasive. It referred repeatedly in its opinion to Lynk (1995a)
and to an analysis done by Lynk on behalf of the defendants [see Lynk and Neumann
(1999) for similar analysis using the same data], which find that not-for-profit hospitals
do not mark up their prices in concentrated markets, and remarked repeatedly on the
unchallenged nature of the findings. It is worth noting as well that the government's only
success in recent years has come in Poplar Bluff, where both hospitals in the proposed
merger were for-profit, so that this defense did not arise.
We now turn to a discussion of what we see as the principal points on which economic
scholarship would be useful in improving the analysis of merger cases. We will consider
in turn, market definition, the effects of merger on competition, the effect of ownership
status on competition, and the efficiency defense.
4.2. Market definition
4.2.1. Background and issues
In hospital merger cases, it seems apparent that market definition is the most important
issue. When the government has succeeded in convincing courts to adopt its preferred
product and geographic market, it has typically won (the leading exception being the
Grand Rapids case). Market definition is always an important component of antitrust
cases, with the party seeking to block a proposed merger, often the government, arguing
for a narrow market and therefore fewer competitors and the merging parties arguing for
a broad market with correspondingly many competitors. In the 1997 Horizontal Merger
Guidelines [Department of Justice and Federal Trade Commission (1997)] the FTC and
DOJ have articulated a test for market definition. 4 0 The Guidelines are explicit that mar-
ket definition considers only demand substitution and not supply substitution. A market
is the smallest group of competing firms such that they could, acting in coordination
(the guidelines say a "hypothetical monopolist"), raise their prices by more than 5%
profitably 41 for at least one year [DOJ and FTC (1997)]. This market definition may, in
principle, be implemented by an analyst in possession of the full demand system faced
by the relevant universe of hospitals [Scheffman and Spiller (1987), see also Froeb and
Werden (1992)].42 Many methods for market definition have been proposed. Here we
describe those that have been employed in analyzing hospital markets. 43 In the next
section we describe a proposed method for hospital market definition using commonly
available data.
There are two important dimensions of market definition in hospital merger cases,
product market and geographic market. As we discussed above, the product market is
typically taken to be "general acute care inpatient hospital services" or some functional
equivalent. Hospitals are thought to have little market power, in practice, in outpatient
care, since there are a large number of actual competitors and since entry into this mar-
ket is relatively easy. By definition, only hospitals are in the market for inpatient care.
As was discussed previously, this definition is problematic. Economic product market
definition depends on both demand and supply substitutability. There does appear to be
some substitutability in demand between inpatient and outpatient care, as, for example,
one effect of Medicare PPS is thought to have been the substitution of outpatient and
home health for inpatient care [Eldenburg and Kakapur (1997), Kenney (1991), ProPAC
(1995)]. 44 On the supply side, treating aggregate inpatient services as a product is not
likely to be problematic if there is easy substitution on the supply side among different
categories of inpatient care or if hospitals provide inpatient care of different types in
relatively fixed proportions.4 5 Both theoretical and empirical research is needed in this
area.
The determination of the geographic market has been subject to more research. A
common method of market definition in antitrust cases has been the analysis of "ship-
ments data". In many states, there are detailed discharge data detailing the residential
40 The 1997 Guidelines for horizontal mergers are a revision of the 1992 Guidelines with regard to effi-
ciencies. They are unchanged with regard to market definition. They do differ on market definition from the
Guidelines previously issued by the DOJ in 1984 and the FTC in 1982.
41 Actually, the guidelines say "small but significant and nontransitory" price increase and later go on to say
that, most often, this means 5%for the period of at least one year.
42 We note that the 1997 Guidelines specify market definition in terms of a price increase, holding the prices
of all other goods constant. The 1984 DOJ Guidelines do not specify that all other prices are held constant.
This difference gives rise to different implementation methods, as we discuss below.
43 For general discussion of market definition, see Scherer and Ross (1990, pp. 176-184), Geroski (1998).
44 Courts have not usually recognized this substitution as important, however. They appear to rely on the
testimony of physicians that many inpatients could not easily substitute to outpatient care. This reasoning
appears to be faulty, since the fact that some people are not price sensitive at current prices is irrelevant,
unless hospitals somehow may price discriminate among patients who are and are not sensitive.
45 Fixed proportions are not supported empirically. See Farley and Hogan (1990).
location and hospital chosen by each hospital inpatient. It is possible, through the use
of such data, to define markets using the method of Elzinga and Hogarty (1973).46 The
market is defined by increasing the size of the potential market until some threshold for
both exports (termed LOFI)
LOFI "Little out from inside" =1 - patient inflows

patients treated in area
and imports (termed LIFO)
LIFO "Little in from outside" = 1 - patient outflows

patients treated in area
is crossed. No threshold is universally applied, but 75% and 90% are focal numbers in
court decisions, since they appear in Elzinga and Hogarty (1973, 1978).
The Elzinga-Hogarty market definition method is attractive since it can be calculated
in a fairly straightforward way with commonly available discharge data; however, this
method suffers from theoretical shortcomings. As has been detailed by Werden (1990,
1981), an Elzinga-Hogarty market may overstate or understate the true size of the mar-
ket. The producers in two markets may be very close substitutes for one another, though
no product flows between them, if, for example, their products are very close substi-
tutes and are priced similarly. In this case, no product will flow between the putative
markets (since consumers economize on transport costs). As the putative markets are
not truly distinct, were the producers in one putative market to increase their prices,
product would flow between them. Conversely, producers in two different markets may
be selling widely differentiated products, so that there are considerable flows between
the markets caused by consumers seeking out their more preferred products. However,
this does not imply that the producers are in the same market, since their products are
widely differentiated and therefore the price of one product in one market does not ef-
fectively constrain the price of the other product in the other market. This issue arises
often in hospital markets when a large "tertiary care" hospital providing a wide range
of services draws patients from outlying areas, which are served by much smaller "pri-
mary care" hospitals. Though there is flow between the two markets, there may not be
much in the way of substitution possibilities between the products produced in the two
markets.
In the scholarly literature, a wide variety of market definition techniques have been
employed, with considerable controversy concerning appropriate methodology. These
market definitions have included the Elzinga-Hogarty method discussed above. Another
method is the "fixed-radius" technique [Robinson and Luft (1985), Gruber (1994)] in
which a hospital's market is defined to be an area 5 or 10 or 15 miles in radius around
46 Morrisey et al. (1988) advocate the use of this method for hospitals and illustrate its application.
the hospital. Still another type is the "variable radius" technique [Melnick et al. (1992),
Gruber (1994)] in which a hospital's market is, in essence, a weighted average of fixed
radius markets, where the weights come from observed or predicted market shares in zip
codes. The most common definitions are political or census divisions such as counties,
Metropolitan Statistical Areas (MSA), Health Services Areas (HSA), urbanized areas,
etc. [Romeo et al. (1984), Lynk (1995a), Dranove et al. (1992)].
There has been relatively little work in providing either theoretical or empirical justi-
fication for these choices. There are exceptions, however. Phibbs and Robinson (1993)
attempt to validate fixed radius limits with variable radius measures. Makuc et al. (1991)
validate their aggregates of political subdivisions with Elzinga-Hogarty measures. Dra-
nove and Shanley (1989) criticize both fixed radius limits and political subdivisions,
showing that even relatively carefully defined political division measures suffer from
problems. All of these methods also are subject to Werden's critique, since they do not
take account of price and substitution possibilities.
4.2.2. A proposedmethod
We now briefly outline what one would like to do in order to measure geographic hos-
pital markets, and then discuss why this procedure has not been feasible to date.
Suppose there are three hospitals pricing, as the Guidelines suggest, Bertrand 4 7
at P, P2, P3 and producing quantities, Q, Q2, Q3 at constant marginal costs,
cl, C2, c3. We wish to know if firms 1 and 2 are in the same market, as defined by
the Guidelines. The firms price as:
Pi = c i (4.1)
aQilaPi
Here, a Qi / aPi is both demand and residual demand, by the Bertrand assumption. Now,
following the Guidelines, imagine a hypothetical cartel of firms 1 and 2. Pricing by the
cartelized firms is at:
______ aQ 2 /aP 1
P1 = cl -(P 2 - c 2) ,Q1/P (4.2)
Q2 QPllaP2
P2 = C2 - (P1 - cl) (4.3)
aQ2 /aP2 aQ2 /aP 2
The above two equations, the structural demand equations Qi (P1 , P2, P3), and the as-
sumption of constant P3 suffice to solve for the new, cartelized P, P2 and Qc, Qc.
47 Actually, the guidelines only use the Bertrand assumption to assess the profitable price rise after the hypo-
thetical merger or cartelization.
Also, the average price increase from the hypothetical cartelization can be calculated
as:
(P cQC + P2CQ)/(Qc + Q2) - (P 1Ql + P2 Q2 )/(Ql + Q2)

(4.4)
(PI QI + P2 Q2 )/(QI + Q2)
The Guidelines suggest that if the price increase is at least 5%, then the two firms are in
the same market. 48 If not, then a 3rd firm is added, and so on. Notice that this method
(the method of the guidelines) is subject to the "reverse cellophane fallacy" critique
detailed by Froeb and Werden (1992). 4 9
This method requires that the researcher be in possession of the full demand structure
facing firms. This typically requires knowledge of a very large number of cross-price
elasticities of demand, and absent some method for reducing this curse of dimensional-
ity, market definition along these lines is likely to be problematic. As a simple example,
consider the following constant elasticity market demand curve for product j,
ln(qj) = aj + Ze 1jk ln(k) + Sj, (4.5)

k
where qljk is the elasticity of good j with respect to the price of good k. If there are
N goods there will be N 2 elasticity parameters to estimate. 50 This can potentially be a
very large number. There are, however, promising avenues for addressing this curse in
hospital markets. In what follows we describe a method for estimating demand using
data on individual patient choices that allows us to overcome this curse of dimensional-
ity and use demand estimates to define markets.
48 This basic method is due to Scheffman and Spiller (1987). They, however, propose using estimates of resid-
ual demand. This method takes, e.g., firm 3's reaction to firm 1 and 2's actions into account and substitutes
this reaction function out, so the resulting demand is only in terms of the actions of firm I and firm 2. Using
residual demand, rather than structural demand, as we propose, has the advantage of allowing for other firms'
reactions to those of the "cartel" firms. There are two disadvantages, however. First, the 1997 (and 1992)
Guidelines explicitly state that market definition takes all other firms' prices as given. This cannot be done
using residual demand. Second, since residual demand is only in terms of the actions of the firms in the hy-
pothetical cartel, separate residual demands must be estimated for every possible combination of firms. This
may be an onerous or impossible task in markets with anything other than a small number of firms.
49 The cellophane fallacy refers to a case in which the court concluded that cellophane was in the same prod-
uct market as other flexible wrapping materials, based on a high cross-price elasticity of demand [U.S. v. E.I.
duPont de Nemours and Co., 351 U.S. 377 (1956)]. A rational monopolist, however, will set its price so that
its product is a substitute for alternatives. The reverse cellophane fallacy is drawing the false conclusion that
a firm with a low cross-price elasticity of demand enjoys market power. The problem with respect to market
definition is that the estimated price increase will be based on demand elasticities estimated at prevailing
prices. If elasticity is not constant and there is not sufficient variation to estimate it over the range relevant for
the simulated merger then a low elasticity may lead one to falsely conclude two firms are in the same market
when in fact they are not.
50 This example is from Berry (1994).
Since the choice of which hospital to patronize is discrete, it is natural to structure

the demand for hospital services using discrete choice models [McFadden (1974), Berry
(1994), Berry et al. (1995)]. The utility to patient i of consuming care at hospital j is:
Uij = j - lpj - ydij + Eij. (4.6)
In this notation, j is the mean utility ("quality") of hospital j,5; pj is the price of
hospital j's services, dij is the distance from patient i 's residence to hospital j, and eij is
the idiosyncratic part of consumer i's evaluation of hospital j. It is well documented that
the physical distance from a patient's residence to a hospital is a major determinant of
choice of hospital. In fact, there is a moderately large literature on the subject [see Burns
and Wholey (1992), Garnick et al. (1989), and references therein].
A consumer chooses hospital j if the utility derived from going there is higher than
any other alternative (we ignore outside goods, although it would not be hard to include
them). The parameters of the utility function in Equation (4.6) can be estimated using
data on individual consumer choice of hospital. Market share for hospital j is the prob-
ability of an average consumer choosing hospital j. Market demand for product j is
the product of its market share and the number of consumers in the market. Cross-price
elasticities can then be calculated from market demand. Note that if we assume that
the £ij are distributed Type I extreme value (logit) this requires estimating only a sin-
gle parameter on price, and thus avoids the dimensionality problem. This approach to
addressing the curse of dimensionality is discussed, for example, by Berry et al. (1995).
In practice hospital choice is a complex combination of the consumer's choice of
health plan, the health plan's choice of providers to contract with, the consumer's choice
of physician, and the consumer-physician-health plan choice of whether and where to
admit the consumer. The choice model above implicitly combines all of these. Modeling
all of these choices simultaneously is probably too demanding, but it would be useful
in practice to interact whatever health plan or physician variables are available with the
above characteristics.
In additive random utility models such as the one above, in which the Eij are in-
dependent across consumers, implausible substitution patterns among products at the
consumer level are generated. In particular, the slope of demand in both own and rival's
price are dependent only upon choice probabilities. That is, two products with identi-
cal choice probabilities will have identical own-price slopes and identical cross-price
slopes with all other products. In the case of logit demand, the slope of the demand for
product j in its own price is -asj (1 - sj), where sj is product j's choice probability.
Furthermore, the slope of j's demand curve in the price of another product, k, is asksj,
where k is product k's choice probability. Clearly if another product, , has the same
market share as j then it will have identical own price and cross-price effects.
51 Notice that our use of j is slightly non-standard in this literature in that we break price out separately.
These undesirable properties of the individual level demand curves carry over to the
market level demand curves faced by firms if consumers are identical (except for real-
izations of Eij). However, if there is consumer heterogeneity and there are interactions
between consumer and firm characteristics, then the undesirable properties need not
carry over to market demand curves. In this case, dij, the distance from the consumer to
the firm, is such an interaction. It arises from the interaction of a consumer characteristic
(location) and a firm characteristic (location). The consequence of this is that hospitals
which are physically closer to one another will be better substitutes than are hospitals
further apart. For more complete discussion of these issues, see Berry (1994) and Berry
et al. (1995).
It is easy to see, in a simple example, that the tie between market shares and sub-
stitution patterns is broken by including the interaction. Suppose that there are three
locations, 1, 2, 3. Hospital A is in location 1 as are a share I of consumers. Hospital B
is in location 3 as are a share )*3 of consumers. The remaining share ) 2 of consumers
are located between the two hospitals at location 2. The locations are arranged on a line
with one mile separating location 1 from location 2 and one mile separating location 2
from location 3:
A B
1 2 3
If the ij are distributed type I extreme value, y = 1 and A = 8B = 0 (so that choice
only depends on distance and price), then the unconditional choice probabilities of hos-
pitals A and B are:
SA = ILSIA + 2S2A + 3S3A, (4.7)
SB = lSlB + 2S2B +3S3B, (4.8)
where SIA is the conditional probability of A being chosen by consumers in location ,

and so on for the other conditional probability terms.
The conditional probabilities SlA, 2A, s3A have the usual multinomial logit form, so
that the derivatives are:
OSA _ -a[ISIA(1 - SIA) + ).2S2A(1- S2A) + X*3S3A(S3A)],

A= u(X(AIS B + 2S2AS2B +A3S3AS3u).(4.9)
aPB
These two expressions are equal in absolute value (only because there are only two
hospital choices, however, i.e., 1 - slA = siB, and so on).
As discussed by Berry (1992) and Berry et al. (1995) the problems in the market level
substitution patters arise in additive random utility models because all price effects are
determined by a single price parameter, a, and the vector of choice probabilities (or,
equivalently, the product-specific mean utility levels, ). Our point is that the above
slopes are not dependent only upon the choice probabilities, SA, SB. They are also de-
pendent upon A. Evaluating at PA = PB, S2AS2B is greater than either SI AS1B or S3AS3B,
so that any re-allocation of consumers from location 2 to locations 1 and 3 would de-
crease the slope of hospital A's demand curve with respect to either PA or PB. Thus,
with geographically distributed consumers and hospitals, the particulars of the distribu-
tion of consumers and firms affects substitution elasticities directly (not only via market
shares), so that the one-to-one correspondence between market shares and substitution
elasticities is broken.
Since dij is an interaction effect, the distribution of the dij in the population is rel-
evant to the slopes of the demand curves, especially since dij will not be independent
across choices. Obviously, with more hospitals and locations, one would see more in-
teresting effects of the dij as hospitals closer to one another in geographic space would
be better substitutes than would hospitals further apart.
Assuming that a consistently estimated model of the above type is available, all of
the relevant demand elasticities, including own and cross-price elasticities may be cal-
culated. This will allow us to perform the test described above in order to establish
the size of the market. Additionally, if we possess cost information and are willing to
make assumptions about (or have measurements on) price-setting in the market, we can
simulate the effects of merger. Indeed, there is a recent literature concerning the use
of "unilateral" theories of market power in differentiated products industries [Werden
and Froeb (1994), Baker (1997), Werden (1997)]. To evaluate product market aspects of
market definition (say primary vs. secondary vs. tertiary) additional interactions, either
explicitly on consumer characteristics or implicitly via random coefficient modeling,
would be needed.
These techniques, potentially at least, offer the ability to assist with product as well
as geographic market definition. By including hospitals' characteristics, both measured
and unmeasured, in interaction with consumer characteristics, both measured and un-
measured, in Uij, it is possible to evaluate the "locations" in both geographic and prod-
uct space of various hospitals. This would allow for substitution among providers to
differ based upon (for example) "tertiaryness" or "quality" and based upon the patient
population under examination. Using such a model, it would be possible to evaluate
whether or not patient flows from rural to urban areas truly represent substitution pos-
sibilities or just variation in the perceived product market location of hospitals.
The analysis described above requires that the demand system for the hospitals be
estimated correctly. In particular, the parameter on price must be estimated correctly.
Demand systems of roughly this type have been estimated in the hospital choice litera-
ture [see Bums and Wholey (1992), and Garnick et al. (1989), and references therein].
The chief difficulty with using the estimates from these studies for the purposes de-
scribed above is that there are significant econometric problems with the estimation of
the coefficient on price, leading most researchers to regard these coefficient estimates
as highly suspect. There are two main sources of econometric difficulties.
1434 M. Gaynor and WVB. Vogt
First, the price data typically used in such estimations are list prices, or "charges".
These prices are not the prices faced by the relevant decision-makers. To the extent con-
sumers are the relevant decision-makers, they are insulated from prices by insurance,
which often pays all or most of the costs of a hospital stay. To the extent that health
insurance plans are the relevant decision-makers (through their choice of provider net-
work), charges mismeasure the prices they face, since managed care plans typically
negotiate substantial discounts from list charges, or negotiate reimbursement schemes
with hospitals which are not directly based upon charges. At best, these two facts mean
that there is substantial measurement error in the price variable, and (intuitively, at least)
this tends to bias a toward zero. 52
Second, standard discrete choice techniques encounter significant problems when
there is substantial unmeasured quality, specifically if there is correlation across con-
sumers in the idiosyncratic portion of consumers' evaluations of providers. If in addi-
tion to being unobserved to the analyst these variations in quality are (at least partially)
observable to patients (e.g., through experience, the experience of friends, or the evalu-
ations of their physicians), then there is good reason to believe that unmeasured (by the
analyst) quality enters systematically into consumer decision-making.i 3 Since hospitals
will take into account consumers' perceptions of quality in their price-setting behavior,
hospitals' prices will be correlated with unmeasured quality [see Berry et al. (1995) for
this argument developed fully]. Again, reasoning intuitively, one would expect high un-
observed quality hospitals to be able to charge higher prices and to, nevertheless, receive
"unexplainably" high demand. This is likely to lead to a negatively biased estimate of a.
Both of these influences on a are in the same direction (assuming that a is positive);
thus, we should expect conventional discrete choice techniques to underestimate the
(absolute value of) elasticity of demand facing hospitals. In the hospital choice liter-
ature, measures of price a are frequently found to be negative or to have implausibly
small positive values. Luft et al. (1991) estimated structural demand elasticities facing
California hospital for Coronary Artery Bypass Graft surgery, using a hospital choice
model like the one described above. They found own-price elasticities of demand for
this service in the neighborhood of -1. This estimated elasticity leads to very high pre-
dicted markups for hospital care for any standard equilibrium model of price-setting.
52 If the coinsurance rate and/or discount is equal across decision-makers, no bias in the elasticity of demand
would arise; however, we do not believe that this is the case.
53 There is a widespread belief among researchers in health economics that there is substantial variation in
quality among hospitals and that variation is difficult for third party analysts to measure. Indeed, there is a very
large literature in health services research whose stated purpose is to develop measures of hospital quality.
Since this literature has not met with great success and since there are so many different quality measures,
there is a paucity of easily available evaluations of the relative quality of different hospitals; although, in
recent years several states have released evaluations of hospital mortality and, for a time, the U.S. Health Care
Financing Administration did the same. As a consequence, the issues raised here with respect to unobservable
quality are relevant, although future developments in quality measurement may reduce the scope for these
sorts of problems.
Given the low level of accounting profits generated by hospitals, such markups seem
implausible.5 4
The solution to both of these problems is instrumental variables estimation. The issue,
as always, is finding suitable instruments. In the Section 4.3 we develop a model of
hospital price and quality setting. Identification of a hospital pricing equation based on
the theoretical model is discussed in Section 4.3.3.
4.2.3. Managed care and market definition
As mentioned in Section 4.1, the impact of managed care on geographic markets has
emerged as a critical issue in a number of hospital merger cases. The standard intuition
has been that managed care will expand the size of hospital geographic markets as a re-
sult of their aggressively shopping on price. An alternative story is that since managed
care plans compete for consumers based in part on their provider networks, competitive
pressures will lead to their networks being more inclusive that imagined by the stan-
dard intuition. Thus there may be little or no impact of managed care on the size of
the hospital geographic market. As was mentioned previously, research by Mobley and
Frech (1997) that shows no significant impact of managed care coverage on observed
consumer travel for hospital services in California is presumably consistent with this
story.
Consider the following structure for examining this issue. Each insurer faces a de-
mand for its services as a function of its price (the premium), benefits (coverages, cost-
sharing, rationing mechanisms, etc.), provider network, and other factors that shift de-
mand, such as the distribution of consumer distances to hospitals, other hospital charac-
teristics, consumer risk aversion, etc. 55 Their costs will depend on wages and other fac-
tor prices, the characteristics of the insurance product, and the prices they pay providers.
Insurers will choose premia, benefits, and hospitals for their networks to maximize
profits. This will result in a derived demand for each hospital by each insurer which
is a function of the number and characteristics of the covered lives of the insurer, the
probability that the hospital is in the insurer's network, and the probability that vari-
ous competing hospitals are in the insurer's network. This will yield a derived demand
system for hospitals in a managed care environment.
Note that all insurers do not have to be managed care plans. An insurer's network
can include all hospitals, as in conventional insurance. Assuming it is feasible to do so,
estimating this system will provide own and cross-price elasticities of hospital demand,
which can then be employed for the purposes of market definition as described in the
54 Of course, these facts can be rationalized by extreme scale economies or extreme errors arising from the
distinction between accounting and economic profits.
55 We ignore strategic interactions with other insurers for purposes of simplicity in exposition. This clearly
would have to be part of a complete analysis.
preceding section. 56 Examining the effect of managed care on hospital market definition
would amount to simulating the effects that changes in managed care's market share will
have on market boundaries.
This description is not intended as a blueprint for analysis in this area; it simply pro-
vides a starting point in terms of a framework for thinking about the issue. In particular,
identification must be derived from a fully specified structural model of the insurance
and hospital markets.
4.3. Mergers, conversions, and ownership status
In the next few sections we outline a general structure for thinking about hospital com-
petition, ownership, mergers, and conversions. The structure is used to analyze parts of
the hospital competition literature and its relevance to antitrust policy.
Hospitals sell products that are differentiated on a number of dimensions. Hospitals
have their physical plant in distinct geographic locations, and consumers are known to
value hospitals that are close to their homes. Hospitals have different religious affil-
iations. They are differentiated in the breadth of product line they offer, in the tech-
nological sophistication of their services, in the quality of the "hotel" services they
offer, in their use and deployment of staffing, in their mortality rates, and probably in
other dimensions as well. It seems reasonable, therefore, to model hospital competition
and hospital mergers using models of differentiated oligopoly [see, for example, Haus-
man and Leonard (1997), Werden (1997), Anderson et al. (1992), Baker and Bresnahan
(1985)].
An added complexity in the case of hospitals is that many hospitals are not-for-profit
organizations. A literature has grown up around the idea that hospitals, unlike other
firms, do not maximize profits, but rather some utility function, possibly reflecting the
preferences of the board of trustees, the administrators, the employees more generally,
or the physician staff [Newhouse (1970), Pauly and Redisch (1973), Lee (1971), Sloan
and Becker (1981), Lakdawalla and Philipson (1998)].
We present a simple theoretical model of hospital behavior in order to focus our
discussion of hospital mergers. This model is not fully general, and one should keep this
in mind in reading it. Each firm owns and operates a single hospital. Hospitals maximize
a utility function, U(7T, Q, a). 7r are the profits earned by the hospital, Q is the quantity
of output produced by the hospital, and a are characteristics of the output produced by
the hospital. We are intentionally vague about the interpretation of a. It may be quality,
charity care, technological sophistication, educational activities, "status", or some other
attribute of hospital services hospital administrators care about. 5 7 To ease exposition, we
will simply refer to a as quality throughout, however. In pursuit of their goals, hospitals
56 We have ignored consumer choice of hospital conditional on insurance plan. While this is important in the
insurer's decision, since they must take into account expected hospital utilization, consumer choice will not
be affected by hospital prices since they are fully insured under managed care.
57 We proceed as if a is a scalar quantity. Making it a vector changes the analysis very little.
choose the quantity of their output and the characteristics of their product, taking into
account that they are acting in a market in which other firms are present.
Hospitals are assumed to produce products differentiated (potentially) by a and by
location. Demand facing hospital i is represented as P = Di(Q, a', aQ-i, a-). Us-
ing a characterization of the equilibrium of the game and solving out for other firms
actions [see Baker and Bresnahan (1988), for example], we may write firm i's residual
demand, D1, as
pi = Di(Qiai )
i i i
= D (Q ,a , Q-'(Q, ai), a-(Qi, ai)). (4.10)
The elasticity of residual demand is a function of demand elasticity, cross-demand

elasticities, and the elasticities of the response functions of rivals, as perceived by the
firm:
aD i aD i aQJ aD aa J
Ii ji
aD i a a
= a ' QJ i + E .
D~ = aBaiL
a OQJa aaa
-B. i aal
aaJ Ba'
Following the conjectural variations empirical literature, is sometimes convenient to

rewrite these as follows, so that rQ is the firm's "conjectural variation" arising from a
quantity change and ra is the conjectural variation arising from a quality change 58:
oDi (aD (aD'aQj)(aQ/ Qi (BDi/ ai)( a a i aQi)

aQ
B +ff (aDi/aQi) I (aDi/aQi)
aDi
=Q(l+rQ),
-' aDI' 1 (aDi/aQi)(aQJ/aai) (aDi/aai)(Bai/ai))

D2= aai i
(aDi/a ) +E (aDi/aai )
i/i
BDi
= ai(1+ ra). (4.12)
58 Although conjectural variations are a convenient way to think about market power, firms' expectations are
not measured empirically by r. What is measured empirically is the behavior of firms, so that firms behave
"as if" their C.V. was r. See Breshnahan (1989) on this point. [If preferred, one can think of 1 + r as a conduct
parameter, as in Breshanan (1989).]
Hospitals solve:
max U(r, Q, a),

Q,a
= pQ - C(Q, a), (4.13)
s.t. r r L ,
where we assume that rL is a lower bound on hospital profits and that the functions U
and C are typical well behaved utility and cost functions.
The first-order conditions for a maximum can be easily manipulated to yield:
P = C1 U2 Q-
U
1 ±. (4.14)
U3
C2 = QD 2 + .
Ul +
Here the subscript on C 1 denotes a partial derivative with respect to its first argument,
quantity. . is the LaGrange multiplier on profits. Assuming that C22 is positive, we may
rewrite these as:
U2
P = C U2 Qb,
3 +2.Q) (4.15)
a= f (QD 2 + U 1 + '3,Q)'
where f is increasing in its first argument.

Economists typically have strong priors on the form of U in for-profit firms. In par-
ticular, setting U (r, Q, a) = Or (and dropping the profit constraint) yields familiar first-
order conditions:
P = C 1 -QDi, a = f(QD2 , Q). (4.16)
Notice that the difference between the NFP and FP first-order conditions for price is
simply the term -U 2/(U 1 + .). Assuming that U2 is positive (NFPs derive positive
utility from quantity), then we can think of NFPs as FPs with "lower costs", i.e., "NFP
marginal costs" are C 1 = C 1 - U2 /(U1 + .) [see Lakdawalla and Philipson (1998)
for a fuller development]. Thus, ignoring quality, we expect lower prices and higher
quantities for NFPs than for FPs, due to the positive utility NFPs derive from quantity.
We do, in fact, observe that NFP prices are lower on average than FP prices.
For the most part, we will drop the LaGrange multiplier on profits (assuming that
the profit constraint does not bind). Furthermore, it will be convenient frequently to
"linearize" the problem by imposing strong functional form assumptions. When conve-
nient, we will treat the "slopes" of demand and residual demand curves, QD 1, QD 2,
and the "conjectures" rQ, ra as if they were parameters (possibly varying across firms
and markets with covariates), rather than functions of firms' choice variables.
4.3.1. Implications of the model: conversions
Conversions are not directly relevant for antitrust purposes; however, we consider them
since we believe that studying them can shed light on the alleged differences between
for-profit and not-for-profit institutions and these differences can be relevant for antitrust
purposes.
To understand the likely effects of conversion from not-for-profit to for-profit status,
it is useful to look in detail at the components of the first-order conditions determining
price and quality for not-for-profit (NFP) and for-profit (FP) firms:
NFU2
NFP: P = c NFP _-1 QD I (r(4.17)FP
Q
a= fNFP QD2(1 rNFP) + U +' ('
NF(o:(rt~8") U 1+ rj)
FP: P C -QD1 (1 r P),
(4.18)
a = fFP(QD2 (l + rP), Q).
The change in price and quality caused by the conversion breaks out into several
components. To keep the discussion simple, we approximate changes in price as below.
We suppress discussion of quality, since it is similar.
AP (C nP cNFP)
rFP)
+ U2
-CFP
QD1 (r -rFP) (4.19)
The price change resulting from the conversion has three components. The first com-
ponent, (CFP _ CNFP), captures any efficiency gained in the conversion. If for-profit en-
terprises have lower (marginal) costs, say because of administrative scale economies re-
lated to large for-profit chains, economies related to "better" management, or to changes
in the quality or quantity of service provided (i.e., decrease in a or Q), then this ef-
fect will tend to reduce the price charged by the enterprise. The second component,
U2/ U1, reflects the differences in goals between the for-profit and not-for-profit enter-
prises. The for-profit values output only to the extent to which output increases profit;
whereas the not-for-profit firm values outputper se. Since the NFP values output per se,
it has added incentive to lower price in order to increase output. The third component,
QDi (r P - rNFP), captures any differences which may exist in the reactions of firms in
the market to a for-profit or not-for-profit firm. If other firms in the market, for example,
find it easier to collude with for-profit firms for whatever reason, this will be reflected
in rFP - rNFP > 0.
Q Q
4.3.2. Implicationsof the model: mergers
As we discussed above, courts have often found that the not-for-profit status of hospitals
involved in mergers weighs in favor of the merger, and some courts have given this
factor very great weight (Grand Rapids, and the District Court in Augusta). In addition,
scholarly commentators have put forward the proposition that mergers between not-for-
profits should be treated differently [Kopit and McCann (1988), Lynk (1994, 1995a)]
than those between for-profits.
We examine the effects of a merger between two hospitals in the same market using
a residual demand curve approach. Key issues in this industry revolve around asymme-
tries in the behavior of FP and NFP firms and how these asymmetries affect the price
and quality of product after the merger. Here, we assume that the two firms (each with
its single hospital) merge but keep both hospitals open. They may realize cost savings,
and they set prices and qualities cooperatively after merging.
Consider firms 1 and 2 and write their demand curves to emphasize their roles:
P = D'(Q1 ,al, Q2, a2 , Q-1,2, a-1 2 ),

(4.20)
p2 = D2a(QI a 1, Q a 2 , Q-2, a-1,2). (4.20)
As before, using whatever is the relevant solution concept for the market, solve out for
all other firms' choices as a function of the choices of firms 1 and 2:
p = Dl(Q,al, Q2, a 2)
= Di(Q ,a 1, Q2 a2 , Q-',2(Q1 ,a 1, Q2a2),a-,' 1 2 (Q al,aQ 2, a 2 )), (4.21
p2 = D2(Ql,a, Q2, a2)
= D2(Q,l al, Q2 a2, Q-1,2(Q al, Q2, a2),a-l,2(Qlal Q2,a2)).
The residual demand curves of the two firms are thus Dl(Ql,a, Q 2, a2) and
b 2 (Q , a', Q2 , a 2). Pre-merger prices are given by:
p = Cl U2_
2Q
' =C-
e D t(4.22)
p2 = C12 U2 2(4.23)
Ul
Presumably the post-merger firm will also seek to maximize some objective function
which contains variables other than profits. However, the theoretical literature of which
we are aware provides little guidance as to how the merged entity should evaluate and
trade off its objectives. In the sequel, we assume that the merged firm's utility function
is identical to the separate firms', but that its arguments are the sum of the individual
firm's arguments.5 9 This appears to be sensible for profits and quantity, but (given the
vagueness with which it is interpreted) may or may not be sensible for a.60 In addition,
although we allow economies by permitting the merged firm's cost functions to differ
from the separate firms', production in the two facilities is assumed to be non-joint. The
new, post-merger firm maximizes:
t,
max
2 2
U(7rw + r2, Q1 + Q2,a 1 +a2),
Q Q ,a',a
r1 + r2 = p Q 1 + p 2 Q 2 _ C1 (Q 1 , a)-C 2
(Q 2 a2 ), (4.24)
s.t. + 22 >
>L
In the sequel, we will use U to denote the utility function evaluated at the new, merged,
profit, quantity, and quality. The post-merger cost function at hospital i is denoted Ci.
Post-merger prices and qualities are given by:
I U22 QD21 Q2 ,
U1
I = Cl -QD()2j)2 _ 3 +
a = f 2(Q
+2 + (4.25)
p2 =2
PA-=Ci- ~ -Q l1Db 3 -Q 2nD3 ,
U2
a2 = !(Ql +Q2b2 +3 Q1+ Q2)
For FPs, the same equations may be written, but with the terms involving utility func-
tions omitted in both the pre- and post-merger price and quality equations.
The question of interest is whether one should expect NFPs to have substantially
different post-merger behavior than FPs regarding price and quality changes. Assuming
that the degree of collusion among the other firms and the merging firms does not change
with the merger, the price changes for FP and NFP hospitals are:
NFP: AP'
NFP: =
= C' -C I - + -2 Q1 I,-D _ Q 1,
(4.26)
Ap2
Ap2 = 522 _ UU2
__2-__2 2
+ ~-1
U2
U _ Q2( 2_ ) _ Q1i
U1 U3 3
59 Note that we have assumed identical preferences for both firms. If that were not the case, some method for
aggregating the firms' preferences would have to be assumed.
60 If a represents charity care, this seems a reasonable procedure. If it represents an index of technologi-
cal sophistication, it may also be a reasonable procedure. It seems least reasonable when a is interpreted
as "process" quality [Donabedian (1980)]. In this case, perhaps "average" a in some sense should be used
instead.
1442 M. GaCynor and W. Vogt
FP: AP = C -C -Q(DI -D) _ Q2D 2

2
I I P1 1~~~~~ ~(4.27)
A p2 = 2 _ C2 Q2(/ _ D2) _ Q I D.
For FP hospitals, the price change for firm 1 consists of an "efficiency" compo-
nent (potentially tending to reduce price), Cl - Cl, reflecting any savings in marginal
costs realized by the merger and a "competitive" component (tending to increase price),
-Q D D - - 2 DI, IQreflecting the fact that the merged firm internalizes the effects
of each hospital's output decision on the output of the other hospital.
As in the case of conversions, if there is a difference in FP's and NFP's ability to
realize reductions in marginal costs due to merger, then this generates differences in
the price changes which mergers bring about. Setting this point aside, the principal
difference in the price change terms is U2 /U 1 - U 2 /U 1 . To assess the impact of this
term on the competitive impact of merger, we consider several cases.
The simplest case is one in which the NFPs have linear utility functions:
U(zr, Q, a) = t1, + 2 Q + a3a.
In this case, U2 /Ui - U2/Ui = 0 and the price increases one would expect from FP
merger and NFP merger are equal. So, the fact that hospitals place value on dimensions
other than profit is not a compelling reason, in and of itself, for expecting FP and NFP
pricing to react differently to merger.
What if the NFPs do not have linear utility functions? In this case, one should expect
NFP firm 's price increase to be higher (lower), than that for FPs if U2 / U 1 is lower
(higher) at (r I + 7r2, Q + Q2, a1 + a 2 ) than it is at (rn', Q1 , a ).61 This condition
is, roughly, that the marginal utility of output should fall more slowly than the marginal
utility of profits as the scale of output and profit rises. What precise shape a NFP's utility
function has is obviously an empirical issue (which we address below), but, so long as
profit constraints are not binding, a claim that there is a difference in price increases be-
tween a FP and NFP merger amounts to a claim about complicated curvature properties
of the NFP's utility function and/or residual demand curve. The important issue, then,
to be addressed when thinking about NFP mergers is whether or not NFPs' marginal
61 Note that in this analysis, we are implicitly assuming that the hospitals' residual demand curve has the
functional form p = Bo + Pi In Q + f(a). This simplifying assumption allows us to ignore the effect of
changing the point at which demand is evaluated on marginal revenue, much as the linearity assumption on
utility allowed us to ignore the effect of point of evaluation on the "marginal cost" attributable to the utility
component. Relaxing the demand assumption also introduces additional complexity. In the case of linear
utility functions as described in the text, a constant-elasticity demand will yield smaller price increases for
NFPs, while with constant-slope demand, NFPs will have larger price increases. We are grateful to Tom
Philipson for explaining the importance of the shape of the demand curve to us here.
utility of profit falls more sharply than do their marginal utility of quantity 62 and the
curvature properties of the demand curve.6 3
4.3.3. Identifying the model
From the previous sections we know that (ignoring any differences in FP and NFP
production technology and in the responses of competitors to FP and NFP actions) the
behavior of NFPs with respect to mergers will differ from the behavior of FPs only if the
ratios U2 /U 1, U3/U1 are non-constant. It is, therefore, of interest to ask whether one
should expect these ratios and their variance with differing levels of price and quality to
be identified.
The NFP's pricing and quality equations are:
P = CNFP _-U NFP

pC --- 122 QDI(1+r
e 1'
(4.28)
3 Q
a = fNFP QD2(l+rNFP) + )
To facilitate discussion, we linearize C N FP and U2/ U1 and drop the quality equation
(take a as exogenous):
CNFP = 8 NFP + NFP Q + 8NFPa ' (4.29)

(4.29)
U2
U.= 0+ lr+
lt 2Q + 11 3a,
U1
where , are parameter vectors and W is a vector of input prices. Substituting into
the pricing equation we get:
p = 8,NFP + NFP + NFP+aNFP' W
- (o + ±127
+Q 8L3a) - QDI (1 + r FP),
0) NFP -
(4.30)
P = (NFP _- + ( - A2) Q + (82NF - 3)a
+ NFP' W _i -rQD(1 r NFP)
Let's assume that we have an estimated demand system (this would require W to have
at least two elements), so that we may view D 1, D2 as data.
62 It is worth pointing out here that a NFP with a constant marginal utility of output, a zero marginal utility
of profit (hence a binding profit constraint), acts like a NFP with a more sharply declining marginal utility of
profit than marginal utility of output - post-merger, the shadow price of profit will likely fall, since profit is
more easily gotten, so that U2 /(U1 + k)(= U2 /X) will fall post-merger.
63 For a related discussion, see Froeb et al. (forthcoming).
In the pricing equation, the separate identification of the scale effect (SNFP _- 2) and
the market-power effect (1 + r F ') has been treated elsewhere extensively [Bresnahan
(1989)]. This application introduces three new identification issues. First, firm profits
appear in the pricing equation, second, quality appears in the pricing equation, and,
third, the intercept and the slopes on Q and a are mixtures of cost function and utility
function parameters.
Strategies for the identification of all of these will obviously depend upon data. There
are fairly rich data sources available for hospitals, collected by both state and federal
governments. These data sources permit the estimation of demand systems for hospital
services (as discussed previously), cost functions for hospitals (see the discussion below
under efficiencies), and (accounting) profit functions [Hoerger (1991)]. Furthermore, as
in many other industries, imperfect measures of price, based upon average revenue or
"list" prices, can be constructed (see the discussion below in the structure-conduct-
perfonnance section). So, we proceed below as if demand, cost, price, and profit es-
timates are readily at hand. Depending upon how quality is interpreted in a particular
application, it is either readily measured or measured with difficulty and noise (see the
discussion on this issue in Section 4.4.2). Again, we proceed as if estimates are readily
in hand.
We can speak in general about the identification of the pricing equation. Later, we
will take up the separation of utility and cost function parameters. The constant and
coefficients on W are obviously identified. The coefficients on Q and D 1 Q are identified
as long as there is a shifter of demand and a (price) rotator of demand [Bresnahan
(1989)]. Similarly, the coefficient on a will be identified by a (quality) rotator.
The most problematic variable is r. Two difficulties exist here. First, as in other
industries, measures of economic profits are rarely available, and the measures of ac-
counting profits may be poor proxies [Fisher and McGowan (1983)]. Second, it is not
immediately obvious that rr is identified separately from Q. Profit is determined by:
= PQ - C(Q,a) (4.31)
If the profit function meets the usual conditions for a neoclassical profit function, then
absent arbitrary functional form restrictions, all shifters of C (input prices, fixed factors)
are also present in the marginal cost functions. Furthermore, Q is already in the pricing
equation and P is affected by Q and a, both already in the pricing equation. So, the
coefficient on profits would seem not to be identified. There are a few possible strategies.
Obviously, functional form restrictions are one. Another is to introduce fixed costs with
(at least one) separate shifter into the cost function. A third possibility is to introduce
an additional source of revenue, say returns from an invested endowment, into the profit
function to provide exogenous variation in profit. The second and third possibilities are
potentially viable strategies. Fixed costs are a very significant component of hospital
costs. Likely candidates for separate fixed cost shifters are factors affecting the cost
of capital, like interest rates, bond ratings, etc., construction costs, and certificate of
need regulations. Data are available on all of these factors and have been used in health
economics research, although not in this context. Again, these revenues are usually
documented in hospital databases.
Second, we turn to the separation of cost and utility function parameters. It is of
evident and obvious scholarly interest to separate these two effects, but it is not of
evident and obvious regulatory interest to do so. If there is sufficient information in the
pricing equation, as estimated, to simulate the effects of a merger, then there is little
reason for a regulator interested primarily in the effects of the merger on price to take
up the separation of cost and utility parameters. The question of whether NFP's reduce
prices less because they have lesser scale diseconomies or because they have marginal
utility of output which falls more slowly than their marginal utility of profit is irrelevant
if price is the variable of interest to the regulator.
As is immediately apparent by inspecting the pricing equation, and as is defini-
tionally true by the lack of separate identification, the separation of utility and cost
function parameters is irrelevant to the question of pricing. To see this, consider
two NFP firms, identical in all respects but that firm 1 has utility and cost param-
eters so, 6', 81, 81, /, 1 , it while firm 2 has utility and cost parameters
AL,
( + 0), ( + E1), ( + 2), 8, ( + ), /l (I + 1), ( + 2). Obviously,
these two firms will choose identical prices, pre and post merger. So, a regulator inter-
ested in price and quality alone would disregard the separate information even were it
to be provided.6 4
However, there are several strategies for separate identification. The most apparent
is to estimate the cost function along with the pricing equation. In this way, the cost
function parameters would be identified by the cost function estimation and the utility
function parameters by the pricing equation. The other apparent way would be to as-
sume that the cost functions of FPs and NFPs are identical. Under this assumption, the
cost function parameters would be identified by the pricing equations of the FPs (since
the utility function parameters drop out), and the utility function parameters of the NFPs
would be identified from differences in pricing behavior between FPs and NFPs.
4.4. Empiricalevidence on hospitalcompetition and mergers
There have been a large number of studies relevant to hospital merger analysis and
hospital competition generally. There are excellent reviews elsewhere [Dranove and
White (1994), Frech (1996)]. We present a selective summary of some relevant research
findings.
64 A separate and interesting question is whether a regulator should be interested in only the price and (ob-
servable) quality effects of mergers. Presumably, the NFP form has evolved to predominate in the hospital
industry for some reason, possibly related to information asymmetry [c.f. Frank and Salkever (1994)], and,
in this case, the fact that NFPs have different objectives may mitigate the extent to which they exploit these
asymmetries. Discovering that NFPs have different objectives from FPs might lead a regulator to adopt dif-
ferent attitudes towards them for reasons we do not model.
In practice, researchers have sought to identify the pricing and quality equations using
structure-conduct-performance (SCP) regressions. The idea in these studies is to use
some measure of market structure, say the Herfindahl-Hirschmann index (HHI), as a
proxy for some of the terms in the pricing equation, as:
p = ( FP _ 9o) + (N FP - 2)Q + (6FP - 3)a + NFP'W

FP )
- l - QD1 (1 + r (4.32)
DlrFP= DlrNFP(HHI),
= (HH).
In practice, reduced forms of these equations are usually estimated, and the price and
quality equations are not usually estimated together. The equations usually have approx-
imately the following appearance:
P = o + l Q + 2XD + 3W + 4HHI,
(4.33)
a = YO+YIQ+y2XD+y3W+y4HHI,
where XD represents demand shifters.

Sometimes there is an interaction between HHI and a dummy variable for ownership
status. Presumably, this dummy represents differential effects of HHI on r for FP and
NFP hospitals, the fact that r(HHI) appears in the NFP and not the FP equation, and
the fact that the constant, Q-slope, and a-slopes are different.
Using HHI in this way is problematic. First, the use of HHI in a pricing equation
is known to be justified in the homogeneous-goods Cournot model of competition, and
other indicia are known to be appropriate in other models of conduct [Schmalensee
(1989) and citations therein]. In the Cournot model, however, the coefficient on HHI
provides information on the shape of the demand curve, not on the conduct of pricing
(which is already assumed to be Cournot). Since the issue one is trying to address with a
SCP regression is the relationship between structure and conduct, running a regression
which prejudges conduct and provides information on demand properties conditional on
that prejudgment is not ideal. Second, the HHI is endogenous to conduct, both because
high prices attract entry and because differences in demand and cost conditions (often
not fully accounted for in the pricing specification) affect both prices and the number
of firms that will "fit" in a market [Evans et al. (1993)]. Third, the HHI is affected by
the distribution of market share among firms. A firm with unusually low costs (thus, in
many models high market share) will cause the HHI to be large and is likely to cause P
to be lower [Bresnahan (1989), Schmalensee (1989) and, relatedly, Demsetz (1974) and
Peltzman (1977)]. Fourth, even a true relationship between HHI and price need not
reflect differential conduct. If market average MC varies with HHI, then price could
vary even as conduct remains the same. This relationship could arise in a number of
ways. If bigger markets have more firms and these firms operate at different scales in
bigger markets and there are scale economies, a relationship may exist between HHI
and price which has nothing to do with conduct. If bigger markets have more firms
(with more "draws" on the unobservable components of marginal cost) and low-cost
firms have a disproportionate impact on prices, then price and HHI could be related
without an HHI-conduct linkage [Bresnahan (1989)].
In short, the use of the Herfindahl index to assess the likely impact of mergers imposes
implicit assumptions about the nature of the product, the nature of the competitors, the
nature of conduct, and the nature of entry/exit decisions. When these assumptions are
satisfied, of course, the technique potentially provides information about the effect of
mergers on prices. When they are not, the technique does not. This is true of every tech-
nique, of course, so that it is not the imposition of assumptions which is troublesome,
but the simultaneous implicitness and inflexibility of the assumptions which is problem-
atic. In practice, since the assumptions are implicit, their applicability to the market at
hand is not often discussed nor are they readily discussed. Furthermore, it is not appar-
ent how to incorporate industry and market-specific knowledge from outside the data or
how to modify the approach when any particular implicit assumption is false.
The primary competing paradigm to SCP at present is that of the "New Empirical
Industrial Organization". This technique differs in a number of ways from SCP [see
discussion in Bresnahan (1989)]. NEIO studies typically involve making explicit as-
sumptions about production technologies, demand, and conduct. The assumptions typ-
ically leave enough parameters free so that the data identify such things as important
cost function, demand function and conduct parameters. Furthermore, the assumptions
differ from study to study and are tailored to the industry at hand. The strengths of the
approach are, first, the explicitness of the economic assumptions, and, second, the abil-
ity to tailor assumptions to particular industries. This facilitates evaluation both of how
relevant the analysis is to the chosen industry and how failure of the assumptions to hold
is likely to affect the conclusions drawn.
A weakness of the approach is that conduct is typically measured holding industry
structure constant (literally, not via covariates). This has two downside implications.
First, evaluating the effect of a change in structure, say by a merger, typically requires
additional assumptions. For example, Baker and Bresnahan (1985) evaluate the effect
of mergers among brewers by assuming that the conduct of the non-merging parties is
not affected by the merger, and only the two merging parties take account of the change
in market structure. Werden (1997) evaluates the effect of mergers by assuming that
his markets are characterized by differentiated product Bertrand competition. Second,
it is difficult to integrate cross-sectional variation across markets in structure into the
analysis. For some attempts to remediate this problem see Bresnahan and Reiss (1991),
Sutton (1991), and Berry (1992). There are also rhetorical weaknesses of this family of
techniques. First, explicit assumptions are more readily criticized, so that it may be in a
researcher's private interest to avoid them. Second, estimation techniques are frequently
complex, making their explication, especially to nonspecialists, more difficult.
4.4.1. Structure-conduct-performance:price
The first set of findings we present concerns hospital price-concentration studies. In

this branch of the literature, some measure of price is regressed upon some measure of
market concentration. The studies are differentiated in their operationalization of price
(average revenue per discharge or per inpatient day is probably the mode), in their op-
erationalization of concentration (some kind of Herfindahl or adjusted Herfindahl is
probably the mode), in their study time period, in their study population, in their mar-
ket definition, and in their control variables. However, it is not a completely misleading
summary to say that the studies each regress price on a Herfindahl index, market covari-
ates, hospital covariates including output, and dummy variables for ownership type.
Here, we present a tabular summary of the findings of a number of studies. In each
of these studies, we used the results to analyze a "standard merger case" (SMC). In
the SMC, there is a market consisting of five identical hospitals, each with a 20% mar-
ket share, each not-for-profit. We consider a merger of two of the hospitals, and assume
that in the new market configuration the new hospital will have a market share of 40%,
will have twice the output that it had before, and that the other three hospitals will con-
tinue to have market shares of 20%. Thus it is straightforward to calculate the effect on
the merging hospitals' prices, given a coefficient on the Herfindahl index and a coef-
ficient on (log) output. For studies which use beds as a control variable, we consider
two 218-bed hospitals merging and consider both the change in the Herfindahl and the
change in bed size in the calculation [see Keeler et al. (1999)]. For many studies we
have had to take some liberties in order to present the results in this standard format.
For example, some studies use number of hospitals rather than HHI to operationalize
structure: in those cases, we consider a decrease from five to four hospitals.
Several things are apparent from Table 3. First, the empirical basis for drawing in-
ferences about hospital mergers from SCP studies is quite narrow. Five of the eight
studies we examine use California data, and only one covers the entire United States.
The other interesting pattern is that the estimated effect of the merger is increasing in
the recency of the data. The correlation between the estimated effect size and the year
of the data is 0.22 (it is 0.68 without the DL and LN studies). Finally, the California
studies display stronger merger effects, in general, than do the non-California studies.
Since California has had more rapid and complete penetration of managed care than
have other states, one might wonder whether managed care penetration is associated
with increases in the strength of the association between price and concentration. There
is suggestive evidence in several papers that high HMO penetration tends to decrease
prices and to increase the association between the HHI and price [Simpson and Shin
(1997), Dranove et al. (1993), Melnick et al. (1992)].
In addition to this work, there is a small body of work examining the association
of concentration and price, differentiated by the ownership of the hospital. The the-
ory has been advanced that not-for-profit organizations will not exercise any market
power which they might come to posses, by dint of their different objectives [Kopit and
McCann (1988), Lynk (1995a, 1994), FTC v. Butterworth (1996, 946 F. supp. 1285),
Table 3
Effects of standard merger on price from price-concentration studies
Study1 Price effect Data - Year Data - Place
DSW +5%2 1988 CA

MZBP +2% 1987 CA
Lynk -1% 1989 CA
LN -3% 1995 MI
KMZ +6% 1994 CA
SS +10% 1993 CA
DL +17% 1989 CA
BDW +2% 1988-1992 US
SUD +2% 1983 IN
Noether -1% 1977, 1978 US
l The abbreviations are: DSW - Dranove, Shanley and White (1993); MZBP -
Melnick, Zwanziger, Bamezai and Pattison (1992); Lynk - Lynk (1995a); LN -
Lynk and Neumann (1999); KMZ - Keeler, Melnick and Zwanziger (1999); SS -
Simpson and Shin (1997); DL - Dranove and Ludwick (1999); BDW - Brooks,
Dor and Wong (1997); SUD - Staten, Umbeck and Dunkelberg (1987), Noether -
Noether (1988).
2 The authors use a dummy variable to capture scale effects. The dummy variable
big equals 1 if the hospital in question is large enough. Here (to capture the effect
of merger on output) we assume that the authors' dummy variable BIG goes from 0
to 1 but that the dummy variable HIGH TECH does not change.
Table 4
Effects of standard merger on price, by ownership form
Study FP NFP Government
Lynk +17% -1% +7%

LN NA -3% NA
KMZ +16% +6% +9%
SS +9% +10% 8-9%
DL NA +17% NA
FTC v. University Health (1991-1 Trade Cases [69,444)]. As discussed previously, the
courts have, in some cases, accepted this reasoning. There are five papers that we know
of which specifically address this point. Again, with some loss of information in the in-
terests of greater clarity, we present the results of these studies in tabular form (Table 4).
In this table, we consider the effects of the SMC, this time differentiating by ownership
classification.
All of these studies with the exception of LN use California data. There are a number
of differences among these studies, however. The studies use different years of data.
The Lynk and DL results are based upon 1989 data, while KMZ use several years (the
results above are for 1994), SS use 1993, and LN use 1995 data. The studies also differ
on market definition, the product markets examined, hospitals included in the analysis,
and whether the LHS variable is logged. DL, SS, and KMZ provide discussions of their
differences with Lynk. LN discuss differences between Lynk, DL, and KMZ. The mes-
sage of these discussions is that each of the differences in technique and data accounts
for some of the differences in measured effects. As we mentioned above, the absence of
explicit assumptions mapping empirical technique, economic assumptions, and outside
knowledge makes it quite difficult to evaluate the relative merits of the various empirical
procedures.
4.4.2. Structure-conduct-performance:non-price
The empirical literature in the area of non-price competition is also largely of the SCP
variety. This research falls into a number of categories, depending on what non-price
attribute is measured and what normative spin is imparted to the results.
The medical arms racing (MAR) branch of the literature generally regresses some
measure of input use or of costs on some measure of concentration. 6 5 The idea is that
less concentrated markets induce hospitals to compete more aggressively for patients,
possibly via competition for physician affiliations or managed care network inclusion.
This competition takes the form of actions by the hospital to increase the perceived at-
tractiveness of their products, for example by adopting advanced medical technology,
acquiring helicopters, employing more and/or more skilled nursing staff, etc. These
actions are costly. The MAR is to be detected by looking for a correlation between
concentration and the input use in question (regressing the presence of technology on
HHI, regressing nurses per patient on HHI, etc.). A negative relationship is taken as
evidence of (presumptively welfare reducing) non-price competition. Indexes of input
use often used are the presence of particular technologies [Dranove et al. (1992), Luft
et al. (1986), Robinson et al. (1987)], length of stay [Robinson and Luft (1985), Her-
sch (1984)], staffing levels and/or mixes [Robinson (1988), Hersch (1984)], and reserve
capacity [Joskow (1980)]. A typical finding in this literature is that higher levels of con-
centration lead to lower levels of input use. So, for example, Dranove et al. (1992) find
that more highly concentrated California hospital markets have less sophisticated med-
ical equipment; Robinson and Luft (1985) find that length of stay is declining in con-
centration; Joskow (1980) finds that more concentrated hospital markets hold a lower
reserve capacity of beds.
Another set of MAR papers models hospital costs as a function of concentration.
Sometimes the studies estimate what are easily recognizable as neoclassical cost func-
tions, but with Herfindahl index "dropped" in. In others, functional forms and right-
hand-side variables are more ad hoc. The reasoning in these papers is roughly as fol-
lows. If non-price competition is more aggressive in less concentrated markets and if it
65 Again, for a more comprehensive review, see Dranove and White (1994) or Dranove and Satterthwaite
(2000).
is also cost-increasing, then one should expect to find lower costs in more concentrated
markets. A number of papers in this stream of the literature do find that hospitals in
more concentrated markets have lower costs [Robinson and Luft (1985), Hersch (1984),
Zwanziger and Melnick (1988)]. In addition, studies using more recent data in Califor-
nia [Melnick and Zwanziger (1988), Zwanziger and Melnick (1988)] find that costs are
increasing more quickly in high concentration markets, with the inference drawn that
managed care caused hospitals to compete more on price than on quality.
Although most of the non-price competition literature falls into the above categories,
there are papers outside of them. Shortell and Hughes (1988) examine the associa-
tion between in-hospital mortality among Medicare patients in 1983 and concentration.
There was no significant association between the two, and the point estimate of the
impact of concentration upon mortality was small.
Another dimension which might be affected by competition is the provision of un-
compensated care. Gruber (1994) studies the effect of competition on uncompensated
care in California. He finds that an increase in Herfindahl by 0.12 leads to an increase
in uncompensated care of 10% and an increase in the number of emergency rooms (dis-
proportionately used by uninsured individuals) by 4%.
More recently, Kessler and McClellan (1998) find a negative relationship between
heart attack mortality and the Herfindahl index, suggesting that perhaps hospitals may
take advantage of market power by skimping on quality. Hamilton and Ho (1998), how-
ever, find no impact of hospital mergers on heart attack mortality. Volpp and Waldfogel
(1998) find that heart attack mortality in New Jersey increased following hospital rate
deregulation, implying that more active price competition was accompanied by a de-
crease in quality competition.
4.4.3. MAR, scale economies, and the efficiencies defense
In antitrust cases generally, and in hospital merger cases in particular, firms often mount
an "efficiencies defense". An efficiencies defense is a claim by the firms desiring to
merge that the merger will create cost savings which will be passed along to consumers.
These cost savings are claimed to be large enough to offset any anticompetitive ef-
fects of the merger. In the case of hospital mergers, two types of efficiencies are cited.
First, economies of scale in the usual economic sense are often claimed. Indeed, in
their merger guidelines, DOJ and FTC recognize that there may be substantial scale
economies realized by mergers among hospitals with fewer than 100 beds. They have
created a safety zone for such mergers in their statements of antitrust enforcement in
health care [DOJ and FTC (1996)]. In addition, hospitals, relying often on the MAR
hypothesis, claim that merging will enable them to rationalize the production of high-
technology services, which they allege to have been wastefully duplicated due to the
perverse effects of competition between hospitals. (See the discussion above under out-
comes to date.)
There is a large literature estimating hospital costs; however, it is fair to say that few
firm conclusions can be drawn from it. To the extent that any consensus view of scale
economies exists, it is that there are some scale economies for small hospitals, but that
these economies are exhausted above about two hundred beds. Cowing et al. (1983)
contains a review of the early literature on which that consensus is based. This early
literature is, for the most part, based upon ad hoc regression specifications. Typically
some measure of average costs is regressed upon beds and beds squared along with a
variety of other control variables. The typical finding is a shallow U-shaped average
cost curve with a minimum around 200 beds.
More recent contributions usually use specifications which are either neoclassical
cost functions (using "flexible functional forms") or close relatives of neoclassical cost
functions. These papers have produced mixed results, but our reading is that their results
are roughly consistent with the earlier consensus. Vita (1990) and Granneman et al.
(1986) find no scale economies for inpatient care for hospitals in their sample (in fact
they find weak scale diseconomies at all levels of output). Fournier and Mitchell (1992)
find substantial scale economies at point estimates; however, they are unable to reject
constant returns to scale at conventional significance levels, since their parameters are
imprecisely estimated. Cowing and Holtman (1983) find scale economies; however,
Vita (1990) recalculates Cowing and Holtman's scale economy measures, shifting from
Cowing and Holtman's short run cost function to a long run cost function, and finds
none. Dor and Farley (1996) also find scale economies, but do not provide measures,
confidence intervals, or hypothesis tests for them. Finally, Vitaliano (1987) finds scale
economies; however, his output measure is beds, which is likely better thought of as
a proxy for capital stock, so that it is unclear what to make of his estimates. Dranove
(1998) estimates "cost functions" for cost centers (accounting sub-units of hospitals)
within a sample of California hospitals and finds that scale economies are typically
exhausted by hospitals as large as 200 beds.
Two related problems have prevented a firm consensus from emerging on either the
proper methods for estimating hospital costs or on the facts of the extent and size of
scale economies in production in the hospital industry. First, the "casemix" of the hos-
pital, both in terms of which diseases are treated and in terms of the severity of illness,
has not been well controlled in studies to date. To the extent that large hospitals treat
more complex (more costly) cases, measures of scale economies will be biased down-
ward. Given the "primary", "secondary", "tertiary" and the "community" and "regional
referral center" distinctions so common in the rhetoric of health planners and practition-
ers, it is at least plausible that this effect is large.
Second, and similarly, output is invariably captured via very aggregate measures.
"Discharges" or "patient-days" are frequently used aggregate output measures. In some
studies, five outputs are used, but even this level of disaggregation is not common.
Since there are over 400 DRGs (an output classification) for inpatient hospital care,
and since even DRG's are aggregations of more fundamental output distinctions, there
is face validity to this objection as well. Furthermore, larger hospitals usually have a
broader range of services available and provide more specialized treatments (often cost-
ing more). Since this (badly measured) scope of output covaries positively with the scale
of output and since the marginally added outputs are likely to be more costly than av-
erage, this effect works much like the casemix effect above to depress measurements of
true scale economies for large hospitals.
Finally, there is a strain of this literature [Friedman and Pauly (1981), Gaynor and
Anderson (1995)] which demonstrates that consistent estimates of scale economies can
not be obtained without explicitly modeling hospitals' problem of maintaining sufficient
standby productive capacity in the face of uncertain demand. Lynk (1995b) argues that
there may be substantial scale economies arising from this inventory management prob-
lem. To the extent that the hospital has a stochastic demand and relatively fixed capacity
for various outputs, inventory models suggest that there can be substantial economies re-
alized by consolidating the operations of several hospitals. Given the existence of these
inventory related economies and given the already mentioned measurement problems,
the argument goes, the findings of no scale economies in the cost function literature
lack plausibility [Lynk (1995b)]. These arguments depend upon hospital capacity being
truly fixed in the short run, and any scale economies which would result from inventory
effects would have to be balanced against any diseconomies from other operations.
4.4.4. Event studies
In addition to the SCP-type cross sectional work described above, there are several re-
cent papers using a longitudinal approach to measuring the effects of mergers in hospital
markets. Event studies of mergers attempt to infer the effects of mergers by examining
movements in the prices of a firm's stock arising from information related to mergers.
Since most hospitals are not-for-profit, the direct use of stock prices has not been partic-
ularly popular. Most of this literature, then, has used left-hand-side variables other than
stock price, including output, output price, and cost to discern the effect of mergers.
However, one study does use stock price data of several for-profit, publicly traded
hospital firms. Woolley (1989) examines the effects of mergers on the stock prices of
"rival" hospital companies - that is, hospital companies not involved in the merger. The
idea is that, if a merger is anti-competitive and results in a higher price (and profits),
these higher profits will be shared not only by the merging firms but by other market
participants. Conversely, if the merger has no effect on price and only results in cost
savings, the profits of rivals (hence the stock price of rivals) will not be affected by the
merger. Woolley analyzed the movements in the prices of 9 for-profit hospital chains
over the period 1969 to 1985. For each merger or anti-merger occurring in this period,
he examined the effect on the stock price of the firms uninvolved in the merger. For the
most part, the mergers were positively associated with movements in the rivals' stock
prices and the anti-mergers with negative movements. Woolley's interpretation that the
results indicate the exercise of oligopoly power were criticized by Vita and Schumann
(1991), reply Woolley (1991), who argue that: (1) Woolley's definition of the market is
excessively broad, (2) that many of Woolly's mergers occurred in markets sufficiently
unconcentrated that it is implausible that oligopoly power was increased by the merger,
and (3) that the for-profit companies in Woolley's sample are so large, relative to the
number of hospitals affected by each merger, that an effect could not be discernible
in the company's aggregate profits, whence stock price. The question of which view
is most convincing appears to turn on details of market definition and the relationship
between structure and conduct, and these details are not presently known.
Two recent papers [Connor et al. (1997), Connor and Feldman (1997)] examine
122 hospital mergers which occurred between 1986 and 1994. The papers use a dataset
in which the authors follow 3500 U.S. hospitals. Hospital mergers are identified from
the AHA annual survey, and 122 are identified in this period. These papers use what is
essentially a difference-in-differences approach to assess the effects of merger on the ex-
ercise of market power. In Connor et al. (1997), markets are divided into those in which
a merger occurred (treatment) over the time period and those in which one did not (con-
trol). The difference in price (average revenue per admission) and costs are examined
for hospitals in each type of market. The effect of mergers is the difference between the
increase in prices (costs) by the merging firms in the merger markets and the difference
in prices (costs) by firms in the non-merger markets. Overall, merging hospitals had
lower growth in prices and costs than did hospitals in markets without mergers. How-
ever, mergers which occurred in markets which had a higher degree of concentration to
start with generated slightly higher price increases than were seen in the control group,
and generated smaller cost savings than did the mergers in the unconcentrated markets.
Also, mergers that occurred among hospitals with higher capacity utilization, fewer du-
plicative services, and lower managed care penetration generated more anti-competitive
effects. Connor et al. also found that the hospitals involved in mergers were different
in several ways from the control hospitals in non-merger markets. They were bigger,
more urban, more likely for-profit, more likely part of a system, more profitable, and
had higher prices all before the merger occurred.
In Connor and Feldman (1997), the same data are used in a slightly different way to
assess whether or not the mergers led to the exercise of oligopoly power. Using reason-
ing similar to Woolley's event study, the authors used the same difference-in-difference
method to examine the difference between increases in price and costs in non-merging
hospitals in merger markets (treatment) and the increase in price and costs between hos-
pitals in non-merging markets (control). In this analysis, the authors find effects which
are both very small in magnitude (<1%) and not significant. These results were not
changed in a substantive way by stratifying the markets by concentration or by strati-
fying the hospitals by capacity utilization. The authors conclude that there is not strong
evidence that mergers create increases in oligopoly power by non-merging hospitals in
the same market with merging hospitals.
Another recent paper is Alexander et al. (1996). This paper examines the effects of
92 mergers that occurred between 1982 and 1989 on a variety of operating characteris-
tics of the merging hospitals. The authors examine changes in these characteristics from
three years prior to the merger up to the merger and from the merger up to three years
after the merger. These authors also use what amounts to a difference-in-differences es-
timator. For their comparison group, they draw 276 hospitals at random from the AHA
survey. The authors seek to test, in essence, the efficiency-raising properties of merg-
ers, so they look at costs and various correlates of cost, including staffing levels and
beds. Overall, they find that in merging hospitals, the trend rate of growth of costs was
lower for merging hospitals after the merger than it was before the merger, and that
declines in occupancy continued at a slower rate after the merger. In addition, for the
mergers that occurred after 1987, there were reductions in the trend rate of growth of
costs and staffing levels and there was an increase in the trend rate of growth of oc-
cupancy rate post-merger. Like Connor et al., however, Alexander et al. find that there
were substantial differences between merger and non-merger markets before the merg-
ers occurred. Markets in which mergers occurred were declining in both admissions and
beds faster than were non-merger. Their costs were increasing more slowly than were
the non-merger markets, again before the merger occurred.
In a recent, unpublished dissertation, Krishnan (1998) uses a difference-in-differences
technique to assess the effects of hospital mergers for a sample of Ohio hospitals. Her
treatment group consists of lines of business (DRGs) for the merging hospitals in which
the merging hospitals are likely to have gained market power (DRGs where market share
increased by more than 20% because of the merger). Her control group consists of lines
of business for the merging hospitals in which the merging hospitals are unlikely to have
gained market power (change in market share less than 5%). The average difference in
these price changes (pre- and post-merger) was 4%, so that, in lines of business where
merging hospitals are likely to have gained market power, prices rose 4% more, post-
merger, than in lines of business where merging hospitals are unlikely to have gained
market power.
4.4.5. Studies of entry
Two recent studies have employed the method developed by Bresnahan and Reiss
(1991) for inferring the degree of competition in a market by observing entry behavior.
Firms enter a market only if it is profitable to do so. For any given fixed cost of entry, the
lower average variable profits are, the greater the market size required to break even is.
Market size is assumed to be proportional to population. Since competition lowers aver-
age variable profits, this increases the population size required to support entry. Thus, if
entry increases competition, more population should be required to support successive
entrants. The population required to support successive entrants will stabilize when the
prospective entrant's profit margin (i.e., competition) stabilizes.
The advantage of this approach is that it avoids any potential problems associated
with using price or cost data, as discussed previously. There are a number of disad-
vantages, however. The approach relies on the population per firm required to support
entry. This will be increasing as price competition reduces average variable profits. The
problem is that a number of other factors can cause this to increase. Non-price compe-
tition can reduce average variable profits by increasing average variable costs without
affecting price. Increases in fixed costs can also cause the population per firm required
to support entry to increase. If later entrants get a lesser share of the market than do
earlier ones then the population required to support entry can also be increasing in the
number of firms.
Abraham, Gaynor and Vogt (1999) use this approach to look at competition in local
hospital markets. These are basically markets in rural areas of the U.S. They find that
the incremental population needed to support additional hospitals increases with the
number of hospitals in the market, at least through market structures of up to four firms.
If the changing market size increments reflect only increased competition, then there are
large increases in competition with additional firms, at least up through the fourth firm.
While it is not possible to directly distinguish between price and non-price competition
in this framework, they also find evidence that the presence of more HMOs reduces
entry, consistent with HMOs increasing price competition.
Chernew et al. (1999) estimate hospital entry into "markets" for coronary artery by-
pass graft (CABG) surgery by payer type (Medicare, Medicaid, HMO, Fee-for-service,
and Other). They find evidence that fee-for-service insurers are paying rates that al-
low for profitable entry. Medicare payment rates were generous in the mid 1980s, but
allowed for zero profits by the mid 1990s. HMOs consistently paid rates that covered
variable, but not fixed, costs, Medicaid rates are below variable costs.
5. Monopsony and foreclosure
5.1. Introduction
While a great deal of government antitrust enforcement in health care has addressed
horizontal concerns (hospital mergers, physician boycotts, professional society adver-
tising bans), there have also been important antitrust cases addressed to vertical issues. 66
The two major vertical concerns that have arisen in health care markets are monopsony
power and foreclosure. Concerns about monopsony power have been confined to the
exercise of monopsony power by insurers against hospitals or physicians. Foreclosure
concerns have arisen in the context of vertical integration, exclusive dealing, or most-
favored nation (MFN) clauses between various combinations of insurers, hospitals, and
physicians.
While these issues have for the most part not been viewed as anti-competitive by
the courts in the past, they have received renewed attention in recent years. This is due
to two factors: the recent growth to dominance of managed care in the U.S., and the
evolution of economic and antitrust thinking on vertical restraints. As managed care
firms have assumed large market shares in the insurance market, their bargaining power
with hospitals and physicians has grown. This has led to allegations of the exercise of
monopsony power by managed care firms. At the same time, the health care industry has
undergone a great deal of reorganization, including mergers and acquisitions between
66 The most numerous kind of antitrust suits in health care are vertical cases. These are private suits brought
by physicians denied hospital staff privileges, which are invariably rejected by the courts. We do not discuss
these since they have clearly been judged not to be relevant to competition.
insurers, hospitals, and physicians and a great variety of contractual arrangements be-
tween these entities, including exclusive contracts, long term contracts, and MFN con-
tracts. These contractual or integrated arrangements between upstream and downstream
firms have raised increased concerns about the possibility of these arrangements lead-
ing to foreclosure of competitors and thus increasing the market power of incumbents in
these markets. The "Chicago School" thinking on vertical restraints, which overturned
previous antitrust doctrine in this area, is that vertical restraints are pro- rather than anti-
competitive. More recent developments in economic theory have led to the evolution
of the "Post-Chicago School", which holds that vertical restraints have anti-competitive
potential in certain circumstances. In what follows, we take up monopsony issues in
Section 5.2, bilateral market power in Section 5.3, and foreclosure in Section 5.4.
5.2. Monopsony power
5.2.1. Monopsony and antitrust enforcement
The U.S. antitrust statutes can be interpreted as prohibiting monopsonistic as well as

monopolistic practices (e.g., predatory pricing, price fixing, market division, bid rig-
ging, refusal to deal, tying, merger for monopsony, price discrimination, etc.). While
there have been a number of cases in which the exercise of monopsony power was found
to be an antitrust violation [see Blair and Harrison (1993, Chapter 2) for a review], for
the most part the exercise of monopsony power has not been of major concern to en-
forcement authorities. In part this may be due to the view that monopsony is a rare
phenomenon. While this may or may not be true in most markets in the U.S., questions
about the exercise of monopsony power are definitely germane in health care markets,
as we argue below. Another reason for this lack of concern may have to do with the
specifics of antitrust law. U.S. antitrust law does not prohibit price setting by a seller
that takes advantage of monopoly power. As an extension it would seem that the law
would allow monopsonistic price setting as well. 6 7 Most (but not all) of the cases that
have been brought regarding monopsonistic price setting have been decided in favor of
the defendant. Another reason has to do with the antitrust concept of "standing". Stand-
ing is roughly whether the party(ies) involved have the right to sue under the antitrust
laws. The courts have often regarded sellers to an alleged monopsony as being without
standing, since their complaint is that monopsony led to reduced prices. The view is that
the sellers did not suffer any antitrust injury, since the antitrust laws are not intended to
allow a seller to raise prices. In addition, the courts have regarded monopsony power as
leading to reduced prices to consumers (although monopsony does not necessarily lead
67 For example, Ball Memorial Hospital, Inc., v. Mutual Hospital Insurance, Inc. (7th Circuit, 1986, 784 F.2d
1325) and Kartell v. Blue Shield of Massachusetts, Inc. (st Circuit, 1984, 749 F.2d 922) cert. Denied (1985,
471 U.S. 1029).
to reduced prices to consumers, as we discuss below). Since this is beneficial to con-

sumers, they have been loath to consider it an antitrust violation, even if social welfare
is reduced.
In health care most antitrust cases involving complaints about monopsony have in-
volved complaints about unilateral price reductions by a monopsony buyer. In these
cases this has been an insurer buying from a provider of health care services. As indi-
cated previously, the courts have been reluctant for the most part to regard the unilateral
exercise of market power as an antitrust violation and hence have mostly decided for the
defendants in these cases. In the one case involving a complaint of collusion to obtain
monopsony power the court found in favor of the plaintiffs. At present there have been
no health care cases brought involving other sorts of potential monopsony violations.
However, as consolidation continues in the insurance industry it is likely that cases will
be brought concerning mergers between insurers. A possible source of complaint in
those cases is merger for monopsony. We may also see cases alleging predatory pric-
ing (in the case of monopsony this would involve paying high prices to sellers to deter
entry), or collusion, especially as insurance markets become concentrated.
Table 5 contains a list of selected antitrust cases involving monopsony power in health
care. 68 In Kartell v. Blue Shield of Massachusetts, Ball Memorial Hospital v. Mutual
Table 5
Selected health care monopsony cases
Year decided Case (parties) Winner Primary reason
1985 Kartell (Physicians vs. Insurer Not illegal to exercise market

Insurer)1 power to obtain lower prices (via
balance billing ban)
1986 Ball Memorial (Hospi- Insurer Not illegal to exercise market
2
tal vs. Insurer) power to obtain lower prices (via
a PPO)
1990 Ocean State (Physicians Insurer Not illegal to exercise market
3
vs. Insurer) power to obtain lower prices (via
MFN clause)
1982 St. Bernard Hospital Hospital Price fixing by collusive monop-
4
(Hospital vs. Insurer) sony is illegal. Insurer was con-
trolled by participating hospitals,
therefore constituted illegal com-
bination
i Kartell v. Blue Shield of Massachusetts (Ist Circuit, 1984, 749 E2d 922) cert. Denied (1985, 471 U.S.
1029).
2 Ball Memorial Hospital, Inc. v. Mutual Hospital Insurance, Inc. (7th Circuit, 1986, 784 F.2d 1327).
3 Ocean State Physicians Health Plan, Inc., et al. v. Blue Cross and Blue Shield of Rhode Island (Ist Circuit,
1989, 883 F.2d 1101) cert. Denied (1990, 494 U.S. 1027).
4 St. Bernard Hospital v. Hospital Service Association of New Orleans (5th Circuit, 1982, 712 F.2d 978).
68 See Miles (1998, Chapter 15, Section 22) for a complete listing of cases.
Hospital Insurance, and Ocean State Physicians Health Plan v. Blue Cross and Blue
Shield and many other cases, the courts have made it clear that a health insurance plan
can bargain as hard as it wants to get lower prices from providers, providing it is not
predatory (attempting to force prices below marginal cost or average variable cost would
be judged predatory). The exception (St. Bernard Hospital v. Hospital Service Associa-
tion) is a case where the insurer was controlled by nine participating providers and the
monopsony power was thus judged to be illegally obtained via a combination.
In what follows we provide a brief description of the effects of monopsony power
in general, including how to estimate monopsony conduct. Section 5.2.2 contains a
discussion of monopsony and welfare. In Section 5.2.3 we describe a general approach
to estimating monopsony conduct. We then turn to monopsony power in health care in
Section 5.2.4.
5.2.2. Monopsony and welfare
It is well known that monopsony power leads to a welfare loss for sellers of the product.
The exercise of monopsony power in an input market implies that the quantity and price
of the input sold will both be less than their competitive levels. Thus the owners of the
input suffer a loss of surplus under monopsony. What is not commonly understood is
the implication for consumer welfare. Since the monopsonist obtains a reduced price
for the input it might seem that this will result in a lower marginal cost of production,
which might be passed on to consumers.
Consider the following simple model, based in part on Blair and Harrison (1993,
Chapter 3), but adapted to represent the health care institutional setting. A health in-
surance firm produces output (q), with two inputs: purchased hospital output (x) and
another input (y), according to the following (well-behaved) production function,
q = f(x, y). (5.1)
The hospital produces its output (x) with costs described by the (continuous, twice dif-
ferentiable) cost function c(x). Assume, as is usual, that cx > 0 and cxx > 0. Hospitals
are price takers. Then hospitals' marginal cost curves are their supply curves and the
insurer regards the hospital's marginal cost of production, cx, as the unit price of hos-
pital services. Since the insurer possesses monopsony power its marginal factor cost of
hospital services is the derivative of the total factor cost cx (x)x. This is cx + cxx. If
the output market is competitive, the firm's profit function is
J = pf(x, y) - cx(x)x - pyy, (5.2)

1460 M. Ganor and W. Vogt
69
where p is the output price and py is the price of the other input. The profit maximiz-
ing amount of hospital output is where
= Pfx - cx - cxxx = 0. (5.3)

ax
Clearly this implies that less hospital output is purchased under monopsony than in a
competitive hospital market, the price of hospital output is lower (recall that cxx > 0),
and hospitals are worse off.
It is also true, however, that marginal costs are higher under monopsony than with a
competitive factor market. Let the insurance firm's total costs be
TC = c (x)x + pyy. (5.4)
The cost minimizing demands for hospital output and for the other input will be x
70
x(py, q) and y = y(py, q), respectively, thus the cost function is
C(q, py) = cx (x(py, q))x(py, q) + pyy(py, q). (5.5)
Marginal cost is then
aC
a- = CxxXqX + CxXq + Pyyq. (5.6)
aq
Note that the first term does not appear when the insurance firm is a competitive buyer
in the hospital market. Since the first term is positive, the marginal cost of production
is higher under monopsony.71 The general point can be established directly by noting
that monopsony violates the conditions for efficient hiring of inputs, thus costs must be
higher.
What are the implications of this for consumer welfare? Consider the following sce-
nario. Assume that output is sold in a regional or national, competitive, market. The
monopsonized factor, however, is purchased in local markets. Assume that this factor is
72
immobile and that every firm is a monopsonist in its local factor market. The result is
69 See Scherer and Ross (1990, Chapter 14) for an exposition of the case where the buyer possesses monopoly
power in the market for its output. The case when the buyer is a monopoly seller of its output is not qualita-
tively different from the case when it sells in a competitive market. Since the latter is simpler expositionally,
we present that here. Nonetheless, it is straightforward to adapt this analysis to reflect selling market power
on the part of the buyer.
70 Note that since the price of hospital output, c, is a function of an endogenous variable, x, the factor
demands are functions of the only two exogenous variables, py and q.
71 See also Pauly (1998) on this point.
72 Lumber markets are like this. Trees are immobile and logs are heavy and difficult to transport. Boards,
however, are much easier to transport. Hence markets for logs are usually local and often dominated by a
single buyer or a small number of buyers, while markets for lumber are national, with many sellers.
that every producer sells at the competitive market price, but operates at higher marginal
costs. Thus, aggregate output is decreased relative to what would be produced if factor
markets were competitive. This results in a welfare loss for consumers. If, however, only
a small number of sellers possess monopsony power there will be (essentially) no loss
for consumers. In a competitive market an individual seller is infinitesimal relative to the
total, thus the above described output description only leads to a welfare loss in a com-
petitive output market if a large number of sellers are also monopsonists. 73 If a seller
possesses monopoly power, then a monopsony distortion due to higher marginal costs
exacerbates the output restriction due to monopoly. Consumers pay both higher prices
and consume less output from a monopolist-monopsonist than from a monopolist.
5.2.3. Detecting monopsony power
In principle, detecting the exercise of monopsony power via econometric methods is not
much different than is the same exercise for monopoly power.74' 75 Let the question of
interest be whether insurers possess monopsony power in the hospital market. Assume
that the following are observable: the price and quantity of hospital output, the price or
the quantity of insurance, factors that shift the marginal productivity of hospitals, and
factors that shift the supply of hospital output.
Rearranging the equation above for the profit maximizing amount of hospital out-
put yields the equation for hospital output price setting (or quantity of hospital output
setting) by a monopsonist,
Px = cx = fx - clxx, (5.7)
where px is the price of hospital output. Note that if the second term equals zero this
corresponds to the wage setting condition in a perfectly competitive labor market. Fol-
lowing Bresnahan (1989), we can rewrite this to parameterize the extent of monopsony
power,
P = fx(x, , , Ef) - Cxx(X,Z, Y, c)X0MS, 76 (5.8)
73 Strictly speaking, if the market is perfectly competitive, any seller who can price below others will capture
the entire market (i.e., Bertrand equilibrium). It is not clear in such a case that an equilibrium exists with
monopsony power being exercised.
74 For a somewhat different (but related) approach see Just and Chern (1980), who examine oligopsony power
in the market for tomatoes. Sullivan (1989) examines monopsony power in the market for nurses using a very
similar approach to what we describe here.
75 The exercise of market definition for monopsony can in principle be followed using the Merger Guidelines,
as described in Section 4.2.2, with the exercise calculating whether buyers acting as a cartel could decrease
price by more than 5%.
76 Notice that this can be re-expressed as p, -Pfxlpx = (-cxx/px)OMS = OMS/s, where ns is the elas-
ticity of hospital supply. Thus the percentage gap between the factor price and the value of the marginal
product is a function of the degree of monopsony power and the inverse elasticity of supply.
where M is a vector of insurer marginal productivity shifters, including perhaps man-

agement or technology, 3 is a parameter vector, Ef and ec are the econometric error
terms for the marginal productivity and marginal supply functions, Z are hospital sup-
ply shifters, y is a parameter vector, and 0 MS is a non-negative parameter indexing the
degree of monopsony power. As 0 M moves farther from zero, the firm's conduct moves
farther away from that of a perfectly competitive purchaser.
If observations on the quantity of insurance output are available a production function
may be estimated jointly with the price setting equation. If not, however, the parame-
ters i can still be estimated by making an explicit assumption about the form of the
function f.
If the output market is perfectly competitive, output price, p, can be taken as given.
If the firm has market power in the output market then p may be endogenous. In this
case instruments for p may be output demand shifters and marginal cost shifters. 7 7
Estimation of the monopsony conduct parameter requires data on hospital transaction
prices, marginal costs, and output. These data requirements are essentially the same as
those described in Section 4.4 for estimating the exercise of monopoly power. As in-
dicated there, such data are commonly available in state hospital discharge databases.
Analogous to the estimation of monopoly power by a seller, identification of the pa-
rameter 0 MS requires the presence of a supply rotator 78 (which changes the elasticity
of supply) as an instrument for marginal supply. The basic idea is as follows. If the
market is competitive, rotation of supply around the competitive equilibrium point will
not change price or quantity. If the market if monopsonistic, however, the equilibrium
price and quantity will change. Candidates for hospital supply rotators are factor prices
in other industries (in the case of hospitals, prices paid by Medicare or Medicaid), non-
operating income (e.g., from endowments or other sources), or, changes in the tech-
nology of producing hospital output. Changes in the technology of producing hospital
services seem ideal for this task, although measuring them in a meaningful way may be
difficult. Hospital non-operating income also seems a good candidate for a supply ro-
tator, since not-for-profit hospitals' behavior may be subject to income effects. We are
not aware of any studies of monopsony power in health care that utilize this approach.
As indicated in the section that follows, existing studies have used a structure-conduct-
performance approach to estimating monopsony power. As indicated previously in Sec-
tion 4.4, while such an approach can yield some interesting and useful correlations, it
suffers from serious limitations. Applying more structural approaches to estimation may
lead to significant advances in our understanding of the exercise of monopsony power
in these markets.
77 Presumably output price is correlated with the error in the marginal productivity function. The marginal
cost of production will fluctuate with marginal productivity, causing fluctuations in price to be correlated with
the marginal productivity error.
78 It is a demand rotator that is required in the case of monopoly power.
5.2.4. Studies of monopsony in health care markets
There have been two areas in health care in which the question of monopsony power
has arisen. These areas are insurers negotiating prices for health care services and the
purchase of nursing labor by hospitals. We only discuss studies attempting to detect
insurer exercise of monopsony power. See Sullivan (1989) for a state of the art test of
monopsony power in the market for nurses and a survey of the literature in this area.
The question with regard to insurer negotiation of prices with providers is whether
the insurer possesses monopsony power that allows them to artificially depress price
and what the welfare impact of this exercise of monopsony power might be. This issue
has been discussed in a number of articles [Feldman and Greenberg (1981), Adamache
and Sloan (1983), Staten et al. (1987), Pauly (1987), Staten et al. (1987), Pauly (1988a,
1988b, 1988c), Frech (1988), Melnick et al. (1992), Foreman et al. (1996)] which have
focused on discounts obtained from providers by Blue Cross/Blue Shield plans. These
studies have focused solely on the unilateral exercise of monopsony power.
While monopsony power is normally defined as the ability to price below marginal
factor cost, it is clear that this ability is predicated on the purchaser's ability to buy
elsewhere. If insurers have no power to control the providers from which their patients
obtain care, they cannot possibly exercise monopsony power. Frech (1996) discusses
how Blue Cross and Blue Shield plans were able to direct patients via their use of "par-
ticipating provider" programs. 79 Adamache and Sloan (1983), Feldman and Greenberg
(1981), Frech (1988), and Foreman et al. (1996) find a positive relationship between
Blue Cross market share (in the insurance market) and the size of hospital discounts
from list prices (or simply lower hospital prices). Melnick et al. (1992) find a significant
negative relationship between Blue Cross market share in the hospital and the hospital's
negotiated prices. Foreman et al. find a significant negative relationship between Blue
Cross market share in the insurance market and Blue Cross payments per beneficiary.
Staten et al. (1987), however, find no significant relationship between the discounts Blue
Cross receives from hospitals and Blue Cross market share in that hospital, although the
point estimates are consistent with the earlier studies (Blue Cross market share posi-
tively related to hospital discounts). Frech (1988) finds a positive relationship between
Blue Cross discount and per capita health expenditures. While not conclusive, this is
consistent with monopsony power leading to higher marginal costs and higher output
prices (if the monopsonist also possesses monopoly power in the output market). Over-
all, these findings indicate a negative relationship between insurer market share and
hospital prices.
While these results are interesting, there are a number of methodological issues re-
garding whether one can infer monopsony power from these findings. First, strictly
79 Participating providers agree to accept an insurer's fee as payment in full and not bill patients covered by
that insurer for the difference between their price and the payment. Participating providers are identified as
such by the insurer.
speaking, the monopsonistic structure of the market for hospital services is not nec-
essarily related to the structure of the market for health insurance [Pauly (988b,
1998)]. As a consequence, Blue Cross market share (or any other measure of health
insurance market structure) is not the conceptually appropriate measure of the struc-
ture of the market for selling hospital services. It is, however, likely that an insurer
with monopsony power in a hospital market will also possess monopoly power in the
health insurance market, especially if the geographic markets for both products are very
similar.
Neither is Blue Cross share of a specific hospital's revenues an appropriate measure
of monopsonistic structure. Monopsony concentration is how concentrated sales are
among buyers. Thus, for example, a monopsony Herfindahl index would use buyer's
market shares of output purchased for the entire market, not the share of a particular
buyer for a particular seller.
The second point is that even if this is the case, it does not follow that an insurer with
monopoly power will possess monopsony power. A monopoly health insurer may face a
perfectly elastic supply of hospital services. Even though it is a single buyer, it will pay
the competitive price for hospital services and purchase the competitive quantity. Third,
it is not clear that regressions of price or price differentials against market share identify
tests of hypotheses concerning market power. The negative relationship between premi-
ums and market share may simply reflect movement along a demand curve, as opposed
to identifying the exercise of market power.
Fourth, although the Feldman--Greenberg, Adamache-Sloan, and Foreman et al. stud-
ies treat Blue Cross market share as endogenous, it is not clear that there is sufficient
power in their data to identify causal effects.80 Melnick et al. recognize the endogeneity
of their Blue Cross share measure, but do not address it.
Fifth, as Pauly (1987) points out, the difference in Blue Cross and commercial insurer
hospital payments isn't necessarily a measure of monopsony power. Monopsony power
is properly measured by the price paid for the input or the quantity chosen of the input
deviating from their competitive levels. Differential hospital prices for different insurers
may reflect price discrimination by hospitals (i.e., hospital monopoly power) as opposed
to insurer monopsony power. 8 1
Sixth, the variables used by these studies do not properly measure transactions prices.
Staten et al. use the difference between total payments for Blue Cross and commercial
insurers, while Foreman et al. use total Blue Cross payments per Blue Cross beneficiary.
This undoubtedly includes quantity differences, thus it is difficult to know whether the
estimated effects measure price differences or quantity differences due to differences in
benefits or differences in enrollee characteristics between Blue Cross and other plans.
80 Feldman and Greenberg and Adamache and Sloan both have 60+ observations corresponding to individual
Blue Cross plan areas. Foreman et al. have quarterly observations on 47 Blue Cross/Blue Shield plans for two
years.
81 For more detailed criticism of the Staten (1987) paper, see Pauly (1987). See also Staten et al. (1988)
response and Pauly's further response (1988a).
Similarly, payments per member also include quantity effects. Feldman-Greenberg and
Adamache-Sloan use (one minus) the ratio of the fraction of hospital charges paid by
Blue Cross to the fraction of hospital charges paid by commercial insurers. This un-
doubtedly includes bad debts and payment refusals for uncovered services, as opposed
to actual transaction prices. The best measure is that of Melnick et al., who have an
index of prices actually paid to a hospital by the California Blue Cross PPO relative to
the average of this index across all hospitals in the PPO network.
In the context of managed care, Pauly (1998) has noted that lower prices in hospital
markets could result either from increased buyer sensitivity to price (i.e., increased com-
petition), or from the exercise of monopsony power. Hence he concludes that merely
observing lower prices in hospital markets doesn't provide sufficient information to
identify the exercise of monopsony power. Pauly suggests that observing quantity will
provide that information. The idea is that quantity will expand with lower prices under
competition, while quantity will decrease with lower prices under monopsony. Pauly,
however, assumes that managed care results in the elimination of moral hazard effects
ex ante. However, if managed care does not eliminate moral hazard ex ante, competition
can result in both a lower price and a reduction in quantity. A benevolent, omniscient,
insurer will ration care ex post and thereby improve consumers' welfare. A lower price
will lead to more rationing, hence under managed care competition can be associated
with a reduction in quantity [see Gaynor, Haas-Wilson, and Vogt (forthcoming)]. The
consequence is that simple reduced form estimation does not suffice to identify the ex-
ercise of monopsony power.
One important difference between health care markets and most other markets is that
health insurers do not directly choose the quantity of the intermediate good purchased
(hospital services). Since consumers choose the quantity of hospital services, insurers
do not perfectly control the quantity of the intermediate good, even if they "manage"
care. Further, since consumers bear only part of the cost of hospital services they con-
sume, they will demand more hospital services than will an insurer. 8 2 Consumers' de-
mand will depend on the price of hospital care, the health insurance premium, and the
coinsurance rate. In the "normal" case of monopsony, the buyer chooses a quantity that
is off its "demand" curve (strictly speaking a buyer's demand curve doesn't exist in
monopsony, just as a supply curve doesn't exist in monopoly), but on the seller's supply
curve. In this case, however, consumers will demand more than the quantity the insurer
wants to buy and the hospitals want to sell at the monopsony price, due to moral hazard.
However, an insurer will still have an expected quantity of hospital services that con-
sumers will choose at each hospital price, premium, and coinsurance rate. The task of
the insurer is then to choose the hospital price, insurance premium, and coinsurance rate
to maximize profit, subject to the constraint that consumers' demand equals supply. If
82 Moral hazard will lead consumers to demand more than would an insurer who maximizes social welfare.
This will be true even for an insurer with market power, since it will wish to restrict output even further below
the socially optimal level.
there is not a profit maximizing coinsurance rate that balances consumer demand with
supply at the monopsony price, then the insurer will not fully exercise its monopsony
power. It is possible that there will be no exercise of monopsony power in this situation
at all. Pauly (1988b) is the only paper of which we are aware that recognizes this issue,
although Pauly does not impose the constraint that demand must equal supply. Formal
modeling of the optimal choice of hospital price and insurance policy by an insurer pos-
sessing monopsony power remains to be done. The normative consequence is that if an
insurer doesn't fully exercise its monopsony power the usual welfare consequences of
monopsony may not be present in health care markets.
Overall, while the bulk of the empirical work in this area has been consistent with
the exercise of monopsony power by health insurers, existing studies have not directly
tested the monopsony power hypothesis. An agenda for future empirical research is to
employ structural methods, as reviewed in Section 5.2.3, to identify this conduct.
As stated at the outset of this section, the courts have uniformly rejected attempts
to prosecute the unilateral exercise of monopsony power. Nonetheless, as insurance
and health care markets continue to evolve it is very likely that cases will arise that
involve allegations of merger for monopsony or practices intended to create or extend
monopsony power. Advancing economic research in this area will provide the basis for
the economics analysis that will be necessary to support good legal decisions.
5.3. Bilateral market power
5.3.1. Identifying bilateralmarket power
While the discussions in the preceding two sections have focused on monopoly power
and monopsony power in isolation, it is reasonable to suppose that there are health
care markets in which both are present. While bilateral monopoly is a classic textbook
example, it is rarely considered explicitly in most studies of organized markets. Health
care may be a case, however, in which bilateral possession of market power is relevant
[Pauly (1998)]. Markets for hospital services are frequently characterized by both a
small number of hospitals and a small number of insurers (buyers). Recent consolidation
among insurers, hospitals, and physicians has increased the number of markets where
this is likely to be true.
Identification of market conduct becomes difficult in this case. In particular, neither
movements in prices nor quantities serve to identify monopoly or monopsony power
when either is possible. To illustrate, consider an observed price decrease and corre-
sponding quantity increase for a hospital. Such a change could be caused by an increase
in the price elasticity of demand facing a hospital, i.e., a decrease in monopoly power.
However it could just as well be caused by an increase in the elasticity of hospital supply
(marginal cost) facing a monopsony insurer. This situation requires separate identifica-
tion of monopoly and monopsony power effects.
Consider the following framework. Suppose that the price for hospital services (p,)
is determined as a weighted average of the monopoly price and the monopsony price.
We remain agnostic about the process by which this occurs, although one possibility is
that this is the Nash bargaining outcome, in which case the weights represent relative
bargaining power. 83 Price can then be represented by the following equation,
P = C, - PX)coOMp] + (1 - o)[Cx + CxxXOMS]

= Cx- PxXWc MP + CxxX( - )MS, (5.9)
where co is the weight on monopoly versus monopsony, c is the cost of production for
x, Px is the first derivative of inverse demand with respect to x, Mp is the monopoly
conduct parameter, and MS is the monopsony conduct parameter. The first terms in
square brackets represent the monopoly outcome and the second set of terms in brackets
represent the monopsony outcome. While it may not be possible to identify o)separately
from OMp and OMS, it will be possible to identify w0p separately from (1 - ow)MS by
the use of a demand rotator and a supply/marginal cost rotator as instruments, respec-
tively. Separate identification of o may be possible via functional form assumptions
(i.e., nonlinearities) or by the use of instruments that affect relative bargaining strength,
but not demand, marginal cost, or marginal factor cost. 84 Regulations, such as those
which govern selective contracting by insurers, are potential candidates.
One thing this analysis points out is that estimates of monopoly or monopsony con-
duct which assume the absence of one will underestimate the true value of the conduct
parameter, since what is identified is monopoly relative to monopsony power, not the
absolute values of either. Consider for the moment monopoly power (the same points
will also hold for monopsony). If it is assumed that there is no monopsony power when
there actually is, then the degree of monopoly power will be underestimated. The pa-
rameter recovered from the data will be an estimate of c0MP, not of OMP. This issue
has been ignored in most NEIO studies of market power. The reason may be that it ap-
pears to be reasonable to assume no monopsony power in most other markets studied by
industrial organization economists, such as automobiles, steel, railroads, etc. This may
not be the case, however, as a number of studies have found a negative impact of buyer
concentration on seller price-cost margins. 8 5 In health care, however, bilateral market
power is definitely an issue which should not be ignored.
83 This is intended as an heuristic, rather than a formal model. An important task for any future research in
this area is to derive such a pricing function directly from a well defined theoretical model. See Brooks et al.
(1997) for a model of price determination as a Nash bargaining game between a hospital and an insurer. They
do not, however, model competition. Their model is a model of bilateral monopoly, while we are suggesting
what is needed is a model of price determination in bilateral oligopoly.
84 Further work is necessary to determine if such nonlinearities flow from (reasonable) assumptions about the
structure of the model. If not, identification via functional form would be strictly ad hoc and hard to defend.
Svejnar (1986) uses structural functional form assumptions in a model of bilateral monopoly to identify the
bargaining power parameter along with other parameters of the model via nonlinearities in the econometric
model.
85 See Scherer and Ross (1990, pp. 533-535) for a survey of these results.
1468 1. Gaynor and W.B. Vogt
5.3.2. Welfare implications
The presence of market power on both sides of a market changes welfare analysis from
the case where there is market power only on one side (monopoly or monopsony only).
The theory of the second-best applies, and if one market distortion is fixed, adding an-
other may improve welfare. For example, if we take monopsony power as given, an
increase in monopoly power can be welfare improving. This will not, however, be true
in general. If supply is more elastic than demand, then the profit-maximizing monop-
sony output will be greater than the profit-maximizing monopoly output. The reverse is
true when supply is less elastic than demand. Thus the monopolist's preferred outcome
may either improve or worsen social welfare. It price is determined by bargaining and
output is determined by the lessen of supply or demand, what price and output will pre-
wail cannot be generally determined, but outcomes that reduce or improve welfare are
possible. Thus, while monopoly in the face of monopsony (or vice versa) may be used
for a "countervailing market power defense" [Blair and Harrison (1993)], any resulting
efficiencies are not automatic and hence the efficiency of the countervailing power must
be established. 86 The one case in which monopsony is definitely welfare improving is
when supply is perfectly elastic. 8 7 In that case marginal factor cost is identical with
marginal cost and the monopsony outcome is the same as the perfectly competitive out-
come, so the result of monopsony power is to completely offset monopoly power and
restore the competitive equilibrium outcome.
While it is possible for bilateral monopoly to improve welfare over simple monopoly
or monopsony, it is never preferable to perfect competition. For example, consider a hos-
pital merger which would lead to a substantial increase in monopoly power in a market
dominated by a single insurer. While the merger could be allowed on the efficiency
grounds previously described, a superior policy would be to maintain competition in
the hospital market and remove constraints to competition in the insurance market.
Pure bilateral monopoly will rarely, if ever, be observed in practice. The most likely
situation is that of bilateral oligopoly. Since there are no general results on the welfare
effects of oligopoly, we cannot say in general what the welfare impacts of countervailing
power might be when there is oligopoly (oligopsony) power on both sides of the market.
5.3.3. Studies of relative bargainingpower
Empirical work on this issue has been rare. There are, however, two studies which have
addressed themselves to this issue. Melnick et al. (1992) examine the impact of Blue
86 There have been attempts at such countervailing market power defense in hospital antitrust cases. This
argument has also been advanced to allow physician "unions" to collectively set prices for independent physi-
cian firms without threat of antitrust prosecution.
87 Consider the monopsony condition for a factor x, pfi = w + (aw/al)l. This can be transformed to pfi =
w(l + /es), where Ssis the elasticity of supply. Clearly this condition goes toward the competitive condition
(value of marginal product of labor equals its wage) as the supply curve of labor becomes perfectly elastic.
Cross' share of a hospital's inpatient days, a hospital's share of Blue Cross' total hos-
pital days, and the Herfindahl index in the hospital's market on the hospital prices paid
by the California Blue Cross PPO. They find that Blue Cross' share of the hospital's
days has a negative impact on price, a hospital's share of Blue Cross' days has a posi-
tive impact on price, concentration in the hospital market is positively related to price,
and the positive impact hospital share of Blue Cross days on price is greater in more
concentrated hospital markets. These results demonstrate very interesting associations,
but as Melnick et al. emphasize, they should not be regarded as definitive. As mentioned
in the preceding section, there are issues with endogeneity, market definition, and mea-
surement of market share. In addition, the Blue Cross PPO had at most an 8.9% share
of patient days in California hospitals, while the mean was 2.6%. As such, the results
seem somewhat strong for such small shares of hospitals' output.
Brooks et al. (1997) specify and estimate a Nash bargaining model of hospital-insurer
bargaining over prices. The paper represents a significant advance over other work by
writing down a carefully specified theoretical model of bargaining. The model, how-
ever, is one of bilateral monopoly, while most real world situations are likely to involve
bilateral oligopoly. Brooks et al. estimate the relative bargaining power parameter us-
ing measures of actual transactions prices for a specific condition, appendectomy. The
focus on a specific condition eliminates concerns about proper comparisons between
aggregate payments or price indexes across hospitals. The choice of appendectomy is
particularly felicitous, since this is a condition for which treatment is likely to be rela-
tively homogeneous compared to many others. Further, they could control for severity
or complication differences by differences in diagnosis.
The estimating equation which is derived directly from the structural model is
P - PL = Y(PT - PL) + , (5.10)
where P is the transaction price, PL is the hospital's reservation price, PT is the insurer's
reservation price, y is the bargaining power parameter, and s is an error term. In order
to explore whether relative bargaining power is influenced by characteristics such as
market structure, Brooks et al. also estimate
P - PL = (a + -3Z)(PT- PL) + , (5.11)
where Z is a vector of exogenous influences of bargaining power. Brooks et al. report a

point estimate of 0.686 for y. This indicates that hospitals have relatively more bargain-
ing power than insurers, which is not consistent with insurers possessing monopsony
power. They find that hospitals in more concentrated markets for appendectomies have
greater bargaining power. They did not include any measures of concentration in the
insurance market, although the proportion of the population enrolled in HMOs is es-
timated to have a positive impact on hospital bargaining power. This may proxy for
concentration in the market for buyers of hospital services.
Estimation of the bargaining power parameter requires knowledge of each party's
disagreement, or threat, points. In this case the threat points are the reservation prices
for the hospital and for the insurer. The reservation price for the insurer is assumed to
be the hospital's list price. Brooks et al. did not have actual list prices, so they used
a regression model to estimate a predicted list charge. The reservation price is then
assumed to be the greater of the predicted list charge or actual price. The reservation
price for the hospital is assumed to be either the cost of a unit of output or the next best
price paid by an alternative seller. Again, these are not observed, so Brooks et al. use
an involved procedure to estimate this. PL is taken to be the minimum of an estimate of
what Medicare reimbursement would be, the predicted list charge, or the actual price.
There are some potential problems with this approach. First, it isn't obvious that a list
price represents a buyer's reservation price when buyers have alternative sellers from
which to choose. If buyers choose sellers to negotiate with based at least in part on list
prices, then list prices will likely be set below a buyer's reservation price. Analogously,
it is not clear that alternative sellers' prices represent the reservation price. Second, there
is the question of correcting for sampling error in the reservation price estimates, and
separately, about whether these are endogenous. Third, if the predicted Medicare price
and predicted list price are both greater than the transaction price, then the estimates of
the hospital's and the insurer's reservation prices are both equal to the transaction price,
and hence there is no gain from bargaining.
The factors Z affecting relative bargaining power are assumed to be represented by
a Herfindahl index calculated for appendectomies, the number of hospitals, the pro-
portion of the population enrolled in HMOs, state dummy variables, and a number of
hospital characteristics. As discussed above, the results seem reasonable, although this
regression resembles a price-concentration regression typical of the studies of hospital
competition discussed in Section 4.4. As such, it shares their virtues and limitations.
5.4. Foreclosure88
The other major form of vertical behavior that has been of concern in health care ao-
titrust have been vertical restraints that tend to reduce competition in the of the mar-
kets involved. These include vertical integration, exclusive dealing, and most-favored-
nations contracts.89 A commonly used term for this effect is "foreclosure". The reason
for concern is obvious. Consider a situation with a health insurance duopoly and a hos-
pital monopoly. If one of the insurers integrates with the hospital or engages it in an
exclusive contract, it will have the ability to foreclose the other insurer from the market,
thereby gaining monopoly power. This is, however, an unsettled area, both in economics
and in antitrust. As the Chicago School pointed out, in situations like that in the preced-
ing example, foreclosure will not necessarily occur, since it requires that foreclosure is
88 This section draws in part on joint work with Deborah Haas-Wilson and with Albert Ma, some of which
appears in Gaynor and Haas-Wilson (1998) and Gaynor and Ma (1996).
89 We only list the forms of vertical restraints that are relevant to health care markets. For example, we do not
list resale price maintenance, although this has been a major issue in the literature on vertical restraints, since
it is not possible to resell a physician visit or an appendectomy.
profitable both for the insurer and for the hospital. Further, there are potential efficiency
gains associated with vertical restraints. The Post-Chicago School acknowledges this
and analyzes the circumstances under which foreclosure can occur as an equilibrium
phenomenon. Since vertical restraints both involve potential anti-competitive effects
and efficiencies, antitrust cases involving (non-price) 9 0 vertical restraints are judged on
a rule of reason basis. This makes economic analysis of effects on competition and
efficiencies essential in such cases.
Table 6 contains a listing of some of these cases.9 1 These cases have concerned
vertical integration, exclusive dealing, and most-favored-nation contracts. The courts
for the most part have found for the defendants, finding insufficient evidence of anti-
competitive effects. As indicated previously, the vast bulk of exclusive dealing cases
or vertical integration cases have been rejected by the courts (e.g., Hyde, Vicksburg,
Marshfield). The Oltz case represents an exception. In this case a group of anesthe-
siologists obtained an exclusive contract from a hospital after threatening to leave if a
nurse anesthetist was not fired. The anesthesiologists' annual incomes increased by over
40 percent after the nurse anesthetist was fired, from which the court concluded that the
exclusive deal had an anti-competitive effect [Frech and Danger (1998)]. In some recent
cases involving vertical integration (Marshfield, Vicksburg) the courts did consider the
possibility of anti-competitive effects of vertical integration, although they concluded
that such effects were not present in those situations. The courts have also been unsym-
pathetic to claims of anti-competitive effects from most-favored-nation (MFN) contracts
(e.g., Marshfield, Ocean State). In a recent case breaking with this precedent (Delta Den-
tal), the court found for the plaintiff, banning the use of most-favored-nation clauses in
a consent decree. The court also found for the plaintiff in the Reazin case.
There has not been a great deal of research in health economics on these issues,
although there has been some recent increase in research in this area. We expect that this
will increase as these issues increase in importance with the spread of managed care and
these forms of vertical restraint [see Gaynor and Haas-Wilson (1998)]. In what follows
we first review the general literature on vertical restraints, discussing its relevance to
health care, and then turn to the health economics literature.
5.4.1. Efficiencies
One line of investigation in the economics literature has been concerned with under-
standing the potential efficiencies associated with vertical integration and vertical re-
straints, such as exclusive dealing agreements. Both vertical integration and vertical
90 Vertical restraints involving price, i.e., resale price maintenance, is per se illegal.
91 Vertical cases are by far and away the most numerous type of antitrust case brought in health care. How-
ever, the vast bulk of these are private antitrust suits brought by physicians denied hospital privileges (these
often take the form of challenges to exclusive contracts). The courts have invariably rejected such claims. As
a consequence we only list a selected subset of cases involving exclusive dealing. Cases involving vertical in-
tegration or most-favored-nations clauses have been far less numerous. See Miles (1998) for a comprehensive
listing.
E .f 0 h
_
5o t)0 0 0
o 0:
5;
cr
ig 0,
0i 0½
0
ON
'
a h
>, Xn
D7 00OJ 5 te X ,,
SQ°
g $ t 0 o a
-9 00
0 0 0" 9 o 0
9>
O )t L C X o _ X,
at
o
00 -O 00 0. - I
0 , No 0 '0· a
:. h D OD3 E: 5 80 EzD5>Zt GC
Q
0 .0 o o ,5 N
* a) r .-
,·00,,
s0 EI r~a~0.
e~,·~
- ~ ~
a,
- 0* l · 0 0.
00
0 0o
0
t 0 .E2 Ž00
E ~ ~ ~ -
00'"
> j
C0 0 0 a)
0f
o'
,
0,
'0
a ri O,
cs
cc ~~~~~~
0 u
e a~~~~~~~d ~ .t
a
, a) a) O
v
0 m
,-
E ar)
o 5 m-O
c©
0 0 ; SA zX <, , ,
tD i t t a tX O O a )
-O E 0'
a)
C~~ -- e U 3 ON
O
0 5a ~dO~0 0
- i 0 0
3
a m s v~
't
5 · O : : Z C
0
0
a)
0E
,E ~z. 00 ) F 0
~~~~~~~~~i
0 c ,
¢ A
4.
j
a)
a. 0, 0 *0 0O
a)
C
0
,I O
D
'= *. 5 =Ž 0
-P >oo
0
tZ $ : -5'Up
wm
vo
0 a
F ts
I .Or,
·
N' 0 3ON 0
a)B00 0
00,
0
1~_
'iN ~ *0 0,N
- tDiON
0 cd
I .t
5r .~~~I
0
~0Z~
0 h-4$~~~a
> 0 ' O )' e~
'a 0' .0'0
'2E
0 o a) '0 '0a)1 0,· ,40 1f)
01
7 '50~0, 5 0, 0
0~
n ool
00a~~j
~
-0 _i a
.2
PIY-'_
~.2 S I
. 12
0 0 Z
N t , 0 ON ON O
.9 S
C0 t ON ON O O ON
v
restraints can be efficiency enhancing by leading to more efficient input pricing, lower-
ing transactions costs, assuring supply of an input, improving coordination between the
firms, and improving monitoring [Coase (1937), Spengler (1950), Klein et al. (1978),
Williamson (1975, 1989), Carlton and Perloff (1994), Riordan and Salop (1995)]. There
is some empirical evidence indicating efficiency gains from vertical restraints [La-
Fontaine (1992), Masten (1984), Spiller (1985), Crocker and Reynolds (1993), Lieber-
man (1991)].
5.4.2. Anti-competitive concerns
The economics literature on the anti-competitive impacts of vertical restraints has been
extremely unsettled. Early court decisions were extremely restrictive in their treatment
of vertical restraints due to concern that these restraints could lead to market foreclo-
sure and consequently be harmful to competition.92 Strong criticism of case law by
the Chicago School led to a new perspective in which vertical restraints were viewed
as competitively neutral or procompetitive [Bork (1978)]. Current thinking on vertical
restraints builds upon the Chicago School critique by applying modern industrial or-
ganization theory to the analysis of more realistic market structures and conduct. This
"Post-Chicago School" literature [Riordan and Salop (1995)] identifies situations in
which vertical restraints can raise concerns about competition. 9 3 In what follows we
discuss the literature on vertical integration and exclusive dealing together, since they
are analytically similar, and then discuss the literature on most-favored-nations (MFN)
contracts.
5.4.2.1. Vertical integration and exclusive dealing. The Chicago School critique of
antitrust policy toward vertical integration and exclusive dealing consists of three main
points. First, vertical integration or exclusive dealing between a downstream firm and an
upstream supplier does not necessarily imply that the net supply of inputs to rival down-
stream firms is reduced. 94 For example, with multiple upstream firms, rival downstream
firms may now have access to other suppliers previously utilized by the firm involved in
the merger or the exclusive deal. Just because access to one supplier is foreclosed does
not mean that the net supply of inputs has been foreclosed. In addition, entry can occur
in the upstream market, either by independent entrants, or by the excluded downstream
92 For example, U.S. v. Aluminum Co. of America (2nd Circuit, 1945, 148 F.2d 416); Lorain Journal v.
U.S. (1951, 342 U.S. 143), United Shoe Machinery Corp. v. U.S. (1922, 258 U.S. 451, 458) for exclusionary
practices, and Brown Shoe Co., Inc., v. U.S. (1962, 370 U.S. 294), A.G. Spalding & Bros., Inc. (1960, 56
F.T.C. 1125), Kennecott Copper Corp. v. U.S. (1965, 381 U.S. 414), Ford Motor Co. v. U.S. (1965, 381
U.S. 414), Ford Motor Co. v. U.S. (1972, 405 U.S. 562) for vertical integration.
93 This literature has been criticized in turn for only demonstrating the possible existence of anticompetitive
effects, but not considering efficiency enhancing impacts [see Klass and Sattinger (1995)].
94 This applies similarly to the consideration of upstream firms' access to a buyer if an upstream firm merges
or engages in an exclusive deal with a downstream firm.
firms themselves. So even if there is an upstream monopoly, vertical integration or ex-

clusive dealing does not necessarily imply foreclosure. Second, the Chicago School
claims that a monopolist cannot enhance its monopoly power by the use of exclusive
dealing or other vertical restraints. There is a "single monopoly profit", regardless of
the nature of vertical relations [Bork (1978, p. 229)]. In the case of the upstream mo-
nopolist above, the upstream firm has nothing to gain from such a merger, since it is
already maximizing its profits by selling to the entire downstream industry as a monop-
olist. Third, the Chicago School criticizes the case law on forward integration on the
grounds that upstream firms could not induce downstream firms to accept exclusion-
ary contracts if such contracts actually hurt the downstream firms [Director and Levi
(1956)]. In the case with multiple upstream firms above, if the unintegrated upstream
firms gain market power as a result of the vertical merger, the upstream firm approached
by the downstream firm has an incentive to remain unintegrated. The downstream firm
must compensate an upstream partner for lost profits, potentially rendering the deal un-
profitable. Posner (1976, p. 212) and Bork (1978, p. 309) conclude that exclusionary
contracts cannot cause any harm and therefore ought to be disregarded by antitrust law.
The Chicago School conclusion is that since there are no anti-competitive impacts of
vertical integration or exclusive dealing, these vertical restraints will only exist where
they result in efficiencies, hence they must be beneficial.
The Chicago School view is based on economic models that employ strong assump-
tions. In particular, the result of a single monopoly profit depends critically on the as-
sumptions of a fixed proportions technology, perfect competition in the upstream or
input market, and monopoly in the downstream or output market with prohibitive bar-
riers to entry.9 5 The results are critically sensitive to these assumptions [Riordan and
Salop (1995), Westfield (1981), Vernon and Graham (1971)]. These assumptions do not
hold in most markets and certainly do not hold in health care markets.
The Post-Chicago School literature argues that vertical restraints can enhance effi-
ciency, but also that vertical restraints can be anti-competitive under certain conditions.
The anti-competitive impacts of vertical restraints come from: (1) the potential for a
vertically related firm to raise rivals' costs and/or foreclose rivals' access to a necessary
market and (2) the potential for vertical relations to confer market power by facilitating
horizontal coordination or collusion.
A number of papers [Salinger (1988), Ordover et al. (1990), Hart and Tirole (1990),
Bolton and Whinston (1993), Riordan (1998), Choi and Yi (1996)] demonstrate that
vertical integration sometimes can lead to foreclosure in equilibrium, concluding that
vertical integration may be anti-competitive. 96
95 Note that the single monopoly profit result, and its criticisms, apply as well to the case of forward integra-
tion in which there is a monopoly in the input market and perfect competition in the output market.
96 The results can be somewhat fragile. For example, the results of Ordover et al. (1990) depend critically
upon an assumption that the integrated firm can commit to sell to the unintegrated downstream firm at a higher
price than the internal transfer price within the integrated firm [Reiffen (1992)]. Some more recent papers are
less sensitive to assumptions [Riordan (1998), Choi and Yi (1996)].
Another concern with vertical restraints is that they may act as facilitating devices,
leading to horizontal collusion. For example, if a downstream firm is integrated with an
upstream input supplier but continues to buy inputs from other upstream firms then it
will receive price quotes and other information from these firms. It can also transmit
information from the integrated upstream firm to the other upstream firms. This infor-
mation can be used to facilitate collusive practices in the upstream market [Riordan and
Salop (1995)]. In addition, an increase in the market power of a firm through a vertical
agreement may provide it with sufficient power to initiate or enforce collusive horizontal
behavior.
The concern over exclusive dealing agreements is the same as that with verti-
cal integration: they can lead to market foreclosure or facilitate horizontal collusion.
Whether there is foreclosure critically depends on whether an exclusive deal reduces
the net supply of inputs to rival firms in the downstream market. If access is re-
duced, then exclusive dealing agreements will raise the cost of doing business to
rival firms [Krattenmaker and Salop (1986a, 1986b), Salop and Scheffman (1983,
1987)]. This will put rival downstream firms at a cost disadvantage, increase the market
power and the profits of the downstream firm with the exclusive deal, and reduce so-
cial welfare [Katz (1989), Comanor and Frech (1985), Mathewson and Winter (1987),
Schwartz (1987)]. Not only can this deter entry into the downstream market or increase
the unilateral market power of the downstream firm with exclusive contracts, it can make
collusive agreements among downstream firms more likely. An increase in the market
power of the firm with exclusive contracts can provide it with the necessary power to
implement and enforce collusive agreements [Riordan and Salop (1995)]. A number
of papers show that an exclusionary equilibrium can exist if it is not possible for the
upstream firms to coordinate with each other [Aghion and Bolton (1987), Katz (1989),
Rasmusen et al. (1991)].
While the vast bulk of the literature on the anti-competitive impacts of vertical inte-
gration and exclusive dealing is theoretical, there have been some attempts to empiri-
cally test whether there are strategic motives for vertical integration. Lieberman (1991)
and Spiller (1985) attempt to distinguish between efficiencies and anti-competitive ex-
planations for vertical mergers. Neither study finds any support for the anti-competitive
hypothesis.
5.4.2.2. Mostfavored nation clauses. Most-favored-nation (MFN) clauses, also called

most-favored-customer clauses, are vertical contractual agreements in which the seller
(for example, a hospital or physician group/network) agrees not to charge the buyer (for
example, an insurer) more than the lowest price it charges any other buyer. If the seller
offers another buyer a lower price, then the seller must offer the same lower price to the
buyer with the MFN clause. MFN clauses have appeared in contracts between health
care providers and HMOs, Blue Cross and Blue Shield plans, managed dental plans,
pharmacy networks, and vision care insurers. 9 7
97 As Frech (1996) points out, the traditional Blue Cross physician reimbursement policy of paying usual,
customary, and reasonablefees contains, in essence an MFN clause due to the usual stipulation. Many Blue
Recent work on MFN contracts also shows that such contracts can have anti-
competitive effects in some situations; however, Chicago School analysts have inter-
preted MFN clauses as simply representing efforts by firms to obtain low prices and
therefore as promoting competition [Posner (1976)]. Indeed, the fact that firms are will-
ing to sign such contracts is interpreted as evidence that such contracts must not be
harmful.
Modem theoretical analyses find, however, both that MFN clauses have the potential
to be anti-competitive [Salop (1986), Cooper (1986), Png and Hirshleifer (1987), Baker
(1996)] and efficiency enhancing [Butz (1990), Png (1991), Cooper and Fries (1991)1.98
MFN clauses may impair horizontal competition in at least two ways. First, MFN
clauses may facilitate tacit coordination [Salop (1986)]. In the case of health care, MFN
clauses may decrease competition among health care providers by reducing providers'
incentives to offer lower prices to insurers. Since any discount the provider grants to
another insurer means granting the discount to its partner in the MFN contract, the
provider must grant the discount on all the inframarginal units it sells to its partner.
This makes price reductions very costly and may be a way to signal a commitment
to collusive pricing. Further, MFN clauses may reduce insurers' incentives to bargain
with providers for lower prices, since rival insurers with MFN clauses would reap the
benefits of the lower prices too. If MFN clauses facilitate coordination among providers
and discourage selective discounting, then MFN clauses will lead to higher prices for
health care services and/or lower quality health care services. The limiting factor on
this behavior is the ease of entry or expansion into the market. If potential competitors
or existing rivals can enter or expand easily, then the anti-competitive impacts of MFN
contracts will be limited.
Second, MFN clauses may increase rival insurers' costs and thereby deter entry.
When an insurer with a large market share signs a contract including a MFN clause
with a large hospital or physician group, that insurer has effectively increased its own
costs and the costs of rival insurers and potential rivals in the insurance market. The
MFN clause assures that the hospital or physician group will not offer to provide ser-
vices at lower fees to rival insurers or potential entrants. Thus, a large insurer getting
most-favored-customer treatment may be able to charge prices above the competitive
level or lower quality below the competitive level.
There has been very little empirical research on the impacts of MFN clauses. One
study is by Crocker and Lyon (1994), who study the use of MFN clauses in natural gas
contracts. They conclude that the impact is to improve efficiency, not decrease compe-
tition.
5.4.3. Health economics research on vertical restraints
There has not been very extensive research in health economics on vertical restraints.
Three recent papers consider the impacts of exclusive dealing between insurers and
Cross plans, however, do not enforce the usual clause.

98 Corts (1996), however, finds that MFN clauses can be competitive and welfare enhancing.
providers on competition in the insurance market. Gaynor and Ma (1996) consider dif-
ferentiated health care providers and homogeneous insurers. Thus providers possess
some market power, but the insurance market is perfectly competitive. Individuals, how-
ever, do not know which provider they will prefer should they fall ill and they are risk
neutral with regard to this uncertainty, so, in essence, health care providers are homoge-
neous ex ante. Since exclusive dealing means that both the insurer and the provider must
be chosen ex ante, this effectively eliminates the differentiation between the providers.
There is no anti-competitive effect in equilibrium, since the insurance market is per-
fectly competitive. Further, there is no ability to commit. The result is that exclusive
dealing cannot create market power where there was none. Gal-Or (1996) considers the
same problem, but with differentiated insurers. With differentiated insurers foreclosure
can occur in equilibrium. In this case, a provider who agrees to an exclusive deal with an
insurer will likely accept a lower payment rate in return for a larger volume of patients.
If both insurers sign exclusive deals with different providers, this benefits insurers by
reducing the outside options of the providers and thus reducing their payment rates. En-
cinosa (1996) considers exclusive deals between HMOs and physician groups. There is
an incumbent HMO which has a cost advantage over a rival, but must invest in order
to serve the entire market. When the incumbent HMO is risk averse, it may engage in
an exclusive deal with the single provider. This will result in foreclosure and is socially
inefficient. At present, however, exclusive contracts per se appear to be relatively rare
between insurers and health care providers. Long term services contracts are common,
and may confer a degree of exclusivity on an insurer who is a large buyer.
Exclusive contracts between hospitals and physicians are common [Frech and Dan-
ger (1998)]. In antitrust cases in this area, plaintiffs have argued that the exclusive con-
tract(s) between a hospital and a group of physicians amounts to tying, since consumers
who wish to buy from the hospital must also buy from the particular physician group.
The counterargument is that tying does not increase hospital profits [Lynk (1984)]. It
has additionally been contended that exclusive deals between hospitals and physicians
are harmful to physician competition. If hospitals are controlled to a significant degree
by their physician staffs, they may agree to an exclusive deal that reduces competition
in the physician market and increases the staff's profits, even if it harms hospital profits
[Frech and Danger (1998)]. Alternatively, exclusive dealing between a hospital and a
physician group may increase quality or efficiency [Blackstone and Fuhr (1984), Lynk
and Morrisey (1987)].
Lynk and Morrisey (1987) empirically examine the impact of hospital market power
on the probability of an exclusive contract between a hospital and physicians. Lynk and
Morrisey use county level hospital market share and the Herfindahl index as measures
of market structure. They find a negative relationship between hospital market share
and the probability of an exclusive contract, which they interpret as rejecting the hy-
pothesis of exclusive contracts being anti-competitive. Danger and Frech III (1997),
however, point out that Lynk and Morrisey erred in calculating the total effect of market
share, taking into account the effect on the Herfindahl index. Their corrected estimates
indicate a positive impact of hospital market share on the probability of an exclusive
contract. Neither of these efforts, however, directly tests the main hypothesis of an anti-
competitive effect of exclusive dealing between hospitals and physicians: that it will
reduce competition and increase price in the physician services market. Hospital mar-
ket power seems to be a necessary, but not sufficient condition for this effect.
Some recent studies have considered possible gains from vertical integration between
hospitals, physicians, and insurers [Conrad and Shortell (1996), Walston, Kimberley,
and Burns (1996)]. These studies have found very little evidence of increased efficiency
or quality due to integration. We are not aware of any empirical attempts to test whether
such integration affects competition.
While the use of MFN clauses has been fairly common in health care through Blue
Cross reimbursement contracting with physicians, we are aware of no studies that di-
rectly examine this issue. Scott Morton (1997) is a recent study of the impact of the
adoption by the Medicaid program of an MFN clause in purchasing pharmaceuticals.
Scott Morton finds that prices of brand name drugs facing competition from generics
increased about 4 percent after the Medicaid MFN policy took effect. She finds no im-
pact on the prices of patented drugs. In addition, the prices of generic drugs rose more
in concentrated markets following the introduction of the policy. These results suggest
that the MFN clause may have softened price competition.
6. Conclusions
In this chapter we have reviewed the relevant research issues on the economics of com-
petition and antitrust in health care markets. We have attempted to summarize the cur-
rent literature, indicating what we know and what we don't know on the basis of these
research findings. Where appropriate we have attempted to suggest methodological ap-
proaches for some of the major research issues in this area: in particular hospital compe-
tition and insurer monopsony or bilateral market power between insurers and hospitals.
Although there has been a significant amount of research in health economics on the
competitive workings of health care markets, in some sense research in this area is still
in its infancy. Since only relatively recently has vigorous competition emerged in health
care markets, concomitant with the vigorous application of antitrust, much remains to
be understood about competition and antitrust in health care. In addition, the rich set of
behaviors and data from health care markets may be of use to industrial organization
economists and antitrust analysts wishing to understand the impacts of some facilitating
practices generally.
References
Abraham, J., M. Gaynor and W.B. Vogt (1999), "Entry and competition in local hospital markets", unpub-
lished manuscript (Carnegie Mellon University).
Adamache, K.W., and EA. Sloan (1983), "Competition between non-profit and for-profit health insurers",
Aghion, P., and P. Bolton (1987), "Contracts as a barrier to entry", American Economic Review 77:388-401.
Alexander, J.A., M.T. Halpern and S.D. Lee (1996), "The short-term effects of merger on hospital operations",
Anderson, S.P., A. DePalma and J.E Thisse (1992), Discrete Choice Theory of Product Differentiation (MIT
Press, Cambridge, MA).
53(December):941-973.
Baker, J.B. (1988), "The antitrust analysis of hospital mergers and the transformation of the hospital industry",
Law and Contemporary Problems 51(2):96-164.
Baker, J.B. (1996), "Vertical restraints with horizontal consequences: Competitive effects of 'most-favored-
customer' clauses", Antitrust Law Journal 64:517-534.
Baker, J.B. (1997), "Unilateral competitive effects theories in merger analysis", Antitrust (Spring).
Baker, J.B., and T.E Bresnahan (1985), "The gains from merger or collusion in product-differentiated indus-
tries", Journal of Industrial Economics 33:427-444.
Baker, J.B., and T.E Bresnahan (1988), "Estimating the residual demand curve facing a single firm", Interna-
tional Journal of Industrial Organization 6(3):283-300.
Becker, E.R., and EA. Sloan (1982), "Hospital ownership and performance", Economic Inquiry 23(1):21-36.
Berry, S. (1992), "Estimation of a model of entry in the airline industry", Econometrica 60:889-917.
Berry, S. (1994), "Estimating discrete-choice models of product differentiation", Rand Journal of Economics
25:242-262.
Berry, S., J. Levinsohn and A. Pakes (1995), "Automobile prices in market equilibrium", Econometrica
63:841-890.
Berry, S., and J. Waldfogel (1997), "Free entry and social inefficiency in radio broadcasting", unpublished
manuscript (University of Pennsylvania).
Blair, R.D., and J.L. Harrison (1993), Monopsony: Antitrust Law and Economics (Princeton University Press,
Princeton, NJ).
Blackstone, E.A., and J.P. Fuhr (1984), "Antitrust and exclusive agreements: The case of hospital-based physi-
cians", The Antitrust Bulletin 29:739-758.
Bolton, P., and M.D. Whinston (1993), "Incomplete contracts, vertical integration, and supply assurance",
Review of Economic Studies 60:121-148.
Bork, R. (1978), The Antitrust Paradox: A Policy at War with Itself (Basic Books, New York).
Bresnahan, T.E (1989), "Empirical studies of industries with market power", in: R. Schmalensee and
R.D. Willig, eds., Handbook of Industrial Organization, Vol. 2 (North-Holland, Amsterdam) 1011-1057.
Bresnahan, T.E, and P. Reiss (1991), "Entry and competition in concentrated markets", Journal of Political
Economy 99:977-1009.
Brooks, J.M., A. Dor and H.S. Wong (1997), "Hospital-insurer bargaining: An empirical investigation of
appendectomy pricing", Journal of Health Economics 16:417-434.
Burns, L.R., and D.R. Wholey (1992), "The impact of physician characteristics in conditional choice models
for hospital care", Journal of Health Economics 11:43-62.
Butz, D.A. (1990), "Durable good monopoly and best-price provisions", American Economic Review
80:1062-1076.
Carlton, D.W., and J.M. Perloff (1994), Modem Industrial Organization, 2nd edition (Harper Collins College
Publishers, New York).
Cave, J.A.K. (1986), "Subsidy equilibrium and multiple-option insurance markets", Advances in Health Eco-
nomics and Health Services Research 6:27-35.
Chalkley, M., and J. Malcomson (2000), "Government purchasing of health services", in: A.J. Culyer and
J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 15.
Chernew, M., G. Gowrisankaran and A. Fendrick (1999), "Payer type and the return to bypass surgery: Evi-
dence from hospital entry behavior", unpublished manuscript (University of Michigan).
Choi, J.P., and S. Yi (1996), "Equilibrium vertical foreclosure with investment", unpublished manuscript
(Columbia University).
Coase, R. (1937), "The nature of the firm", Economica 4:386-405.

Comanor, W.S., and H.E. Frech III (1985), "The competitive effects of vertical agreements", American Eco-
Connor, R.A., and R.D. Feldman (1997), "Horizontal hospital mergers and their effects on non-merging hos-
pitals in the same market area", Conference Paper, Managed Care and Changing Health Care Markets,
American Enterprise Institute, April 10, 1997.
Connor, R.A., R.D. Feldman, B.E. Dowd and T.A. Radcliff (1997), "Which types of horizontal hospital merg-
ers produce savings and which do not?", unpublished manuscript (University of Minnesota).
Conrad, D., and S. Shortell (1996), "Integrated health systems: Promise and performance", Frontiers of Health
Services Management 13(1):3-40.
Cooper, T. (1986), "Most-favored-customer pricing and tacit collusion", Rand Journal of Economics
17(3):377-388.
Cooper, T., and T. Fries (1991), "The most-favored-nation pricing policy and negotiated prices", International
Journal of Industrial Organization 9:209-223.
Corts, K.S. (1996), "On the competitive effects of price-matching policies", International Journal of Industrial
Organization 15:283-299.
Cowing, T.G., A.G. Holtman and S. Powers (1983), "Hospital cost analysis: A survey and evaluation of recent
studies", Advances in Health Economics and Health Services Research 4:257-303.
Cowing, T.G., and A.G. Holtman (1983), "Multiproduct short-run hospital cost functions: Empirical evidence
and policy implications from cross-section data", 49(3):637-653.
Crawford, V., and J. Sobel (1982), "Strategic information transmission", Econometrica 50:1431-1451.
Crew, M. (1969), "Coinsurance and the welfare economics of medical care", American Economic Review
59(5):906-908.
Crocker, K.J., and T.P. Lyon (1994), "What do 'facilitating practices' facilitate? An empirical examination of
most-favored nation clauses in natural gas contracts", Journal of Law and Economics 37:297-322.
Crocker, K.J., and K.J. Reynolds (1993), "The efficiency of incomplete contracts: An empirical analysis of
Air Force engine procurement", RAND Journal of Economics 24:126-146.
Danger, K.L., and H.E. Frech III (1997), "Exclusive Contracts Between Hospitals and Physicians", Working
Paper 14-97 (University of California, Santa Barbara, CA).
Darby, M.R., and E. Karni (1973), "Free competition and the optimal amount of fraud", Journal of Law and
Economics 16:67-88.
Demsetz, H. (1974), "Two systems of belief about monopoly", in: H.J. Goldschmid, H.M. Mann and J.F. We-
ston, eds., Industrial Concentration: The New Learning (Little Brown, Boston, MA).
Deneckere, R., and M. Rothschild (1992), "Monopolistic competition and preference diversity", Review of
Economic Studies 59:361-373.
Dionne, G. (1984), "Search and insurance", International Economic Review 25:357-367.
Director, A., and E.H. Levi (1956), "Law and the future: trade regulation", Northwestern University Law
Review 51:281-296.
Dixit, A., and J.E. Stiglitz (1977), "Monopolistic competition and optimum product diversity", American
Donabedian, A. (1980), The Definition of Quality and Approaches to its Assessment (Health Administration
Press, Ann Arbor, MI).
Dor, A., and D.E. Farley (1996), "Payment source and the cost of hospital care: Evidence from a multiproduct
cost function with multiple payers", Journal of Health Economics 15:1-21.
Douglas, G., and J. Miller (1974), "Quality competition, industry equilibrium, and efficiency in the price-
constrained airline market", American Economic Review 64:657-669.
Dranove, D. (1988), "Demand inducement and the physician/patient relationship", Economic Inquiry 26:281-
298.
Dranove, D. (1998), "Economies of scale in non-revenue-producing cost centers: Implications for hospital
mergers", Journal of Health Economics 17(1):69-83.
Dranove, D., and M.A. Satterthwaite (1992), "Monopolistic competition when price and quality are imper-
fectly observable", Rand Journal of Economics 23:518-534.
Dranove, D., and M. Shanley (1989), "A note on the relational aspects of hospital market definition", Journal
Dranove, D., M. Shanley and C.J. Simon (1992), "Is hospital competition wasteful?", Rand Journal of Eco-
nomics 23:247-262.
Dranove, D., M. Shanley and W.D. White (1993), "Price and concentration in local hospital markets: The
switch from patient-driven to payer-driven competition", Journal of Law and Economics 36:179-204.
Dranove, D., and W.D. White (1994), "Recent theory and evidence on competition in hospital markets",
Journal of Economics and Management Strategy 3(1):169-209.
Dranove, D., and R. Ludwick (1999), "Competition and pricing by nonprofit hospitals: A reassessment of
Lynk's analysis", Journal of Health Economics 18:87-98.
Dranove, D., and M.A. Satterthwaite (2000), "The industrial organization of health care markets", in: A.J. Cu-
lyer and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 20.
Eichner, M., M. McClellan and D. Wise (1997), "Health expenditure persistence and the feasibility of medical
savings accounts", Tax Policy and the Economy 11:91-128.
Eldenburg, L., and S. Kallapur (1997), "Changes in hospital service mix and cost allocations in response to
changes in medicare reimbursement schemes", Journal of Accounting and Economics 23:31-51.
Elzinga, K.G., and TF. Hogarty (1973), "The problem of geographic market definition in antimerger suits",
Antitrust Bulletin 18:45-81.
Elzinga, K.G., and T.F. Hogarty (1978), "The problem of geographic market definition revisited: The case of
coal", Antitrust Bulletin 23:1-18.
Emons, W. (1997), "Credence goods and fraudulent experts", Rand Journal of Economics 28(1):107-119.
Encinosa, W.E. (1996), "Exclusive contracting in health care markets", unpublished manuscript (University
of Michigan).
Evans, W., L. Froeb and G. Werden (1993), "Endogeneity in the concentration-price relationship: Causes,
consequences, and cures", Journal of Industrial Economics 41:431-438.
Farley, D.E., and C. Hogan (1990), "Case-mix specialization in the market for hospital services", Health
Federal Trade Commission (1996), Revised Federal Trade Commission, Department of Justice Policy State-
ments on Health Care Antitrust Enforcement (US Department of Justice, Washington, DC).
Feldman, R., and W. Greenberg (1981), '"The relation between the Blue Cross share and the Blue Cross
'discount' on hospital charges", Journal of Risk and Insurance 48:235-246.
Fisher, F., and J. McGowan (1983), "On the misuse of accounting rates of return to infermonopoly profits",
American Economic Review 73:82-97.
Foreman, S.E., J.A. Wilson and R.M. Scheffler (1996), "Monopoly, monopsony, and contestability in health
insurance: A study of Blue Cross plans", Economic Inquiry 34:662-677.
Fourmier, G.M., and J.M. Mitchell (1992), "Hospital costs and competition for services, a multiproduct anal-
ysis", Review of Economics and Statistics 74(4):627-634.
Frank, R.G., and D.S. Salkever (1994), "Nonprofit organizations in the health sector", Journal of Economic
Perspectives 8(4):129-144.
Frech III, H.E. (1988), "Monopoly in health insurance: The economics of Kartell v. Blue Shield of Mas-
sachusetts", in: H.E. Frech III, ed., Health Care in America: The Political Economy of Hospitals and Health
Insurance (Pacific Research Institute for Public Policy, San Francisco, CA) 353-372.
Frech III, H.E. (1996), Competition and Monopoly in Medical Care (AEI Press, Washington, DC).
Frech III, H.E., and K.L. Danger (1998), "Exclusive contracts between hospitals and physicians: The antitrust
issues", Health Economics 7:175-178.
Friedman, B., and M.V. Pauly (1981), "Cost functions for a service firm with variable quality and stochastic
demand '", Review of Economics and Statistics 63:620-624.
Froeb, L.M., and G.J. Werden (1992), "The Reverse Cellophane Fallacy in Market Delineation", Review of
Industrial Organizations 7(2):241-247.
Froeb, L., S. Tschantz, P. Crooke and G. Werden (1999), "The effects of assumed demand form on simulated
post merger equilibria", Review of Industrial Organization, forthcoming.
Gal-Or, E. (1996), "Exclusionary equilibria in health care markets", Journal of Economics and Management
Strategy 6(1):5-43.
Gamick, D.W., E. Lichtenberg, C.S. Phibbs, H.S. Luft, D.J. Peltzman and S.J. McPhee (1989), "The sensi-
tivity of conditional choice models for hospital care to estimation technique", Journal Health Economics
8:377-397.
Gaynor, M. (1994), '"ssues in the industrial organization of the market for physician services", Journal of
Economics and Management Strategy 3:211-255.
Gaynor, M., and G.E Anderson (1995), "Uncertain demand, the structure of hospital costs, and the cost of
empty hospital beds", Journal of Health Economics 14(3):291-317.
Gaynor, M., and D. Haas-Wilson (1998), "Vertical relations in health care markets", in: M. Morrisey, ed.,
Managed Care and Changing Health Care Markets (American Enterprise Institute Press, Washington, DC).
Gaynor, M., D. Haas-Wilson and W.B. Vogt (forthcoming), "Are invisible hands good hands? Moral hazard,
competition, and the 2nd best in health care markets", Journal of Political Economy.
Gaynor, M., and C.-t.A. Ma (1996), "Insurance, vertical restraints, and competition", unpublished manuscript
(Carnegie Mellon University).
Geroski, P. (1998), "Thinking creatively about markets", International Journal of Industrial Organization
16:677-695.
Glazer, J., and T.G. McGuire (1992), "The economics of referrals", unpublished manuscript (Boston Univer-
sity).
Granneman, T.W., R.S. Brown and M. Pauly (1986), "Estimating hospital costs, a multiple output analysis",
Green, J., and N. Stokey (1980), "A two-person game of information transmission", HIER Discussion Pa-
per 751 (Harvard University).
Gruber, J. (1994), "The effect of price shopping in medical markets: Hospital responses to PPOs in Califor-
nia", Journal of Health Economics 38:183-212.
Hamilton, B., and V. Ho (1998), "Hospital mergers and acquisitions: Does market consolidation harm pa-
tients?", unpublished manuscript (Washington University).
Hart, O., and J. Tirole (1990), "Vertical integration and market foreclosure", Brookings Papers on Economic
Activity: Microeconomics 205-276.
Hausman, J., and G. Leonard (1997), "Economic analysis of differentiated products mergers using real world
data", George Mason Law Review 5.
Hellinger, F. (1995), "Selection bias in HMOs and PPOs: A review of the evidence", Inquiry 32:135-142.
Hellwig, M. (1998), "A note on the specification of interfirm communication in insurance markets with ad-
verse election", Journal of Economic Theory 46:154-163.
Hersch, P.L. (1984), "Competition and the performance of hospital markets", Review of Industrial Organiza-
tion 1(4):324-340.
Hibbard, J.A., and E. Weeks (1989), "Does the dissemination of comparative data on physician fees affect
consumer use of services?", Medical Care 17:1167-1174.
Hoerger, T.J. (1991), "Profit variability in for-profit and not-for-profit hospitals", Journal of Health Economics
10:259-289.
Hubbard, T. (1998), "An empirical examination of moral hazard in the vehicle inspection market", Rand
Hurdle, S., and G.C. Pope (1989), "Physician productivity: Trends and determinants", Inquiry 26:100-115.
Jaynes, G. (1978), "Equilibria in monopolistically competitive insurance markets", Journal of Economic The-
ory 19:394-422.
Joskow, P. (1980), "The effects of competition and regulation on hospital bed supply and the reservation
quality of the hospital", Bell Journal of Economics 11:421-447.
Just, R., and W. Chern (1980), "Tomatoes, technology, and oligopsony", Bell Journal of Economics 11:584-
602.
Katz, M.L. (1989), "Vertical contractual relations", in: R. Schmalensee and R.D. Willig, eds., Handbook of
Industrial Organization (North-Holland, Amsterdam) 655-721.
Keeler, E.B., G. Melnick and J. Zwanziger (1999), "The changing effects of competition on non-profit and
for-profit hospital pricing behavior", Journal of Health Economics 18:69-86.
Kenney, G.M. (1991), "Understanding the effects of PPS on Medicare home health use", Inquiry 28:129-139.
Kessler, D., and M. McClellan (1998), "Is hospital competition socially wasteful?", unpublished manuscript
(Stanford University).
Klass, M.W., and M.A. Sattinger (1995), "Do new theories of vertical foreclosure provide sound guidance for
consent agreements in vertical merger cases?", Antitrust Bulletin 40:667-698.
Klein, B., R.G. Crawford and A.A. Alchian (1978), "Vertical integration, appropriable rents, and the compet-
itive contracting process", Journal of Law and Economics 21:297-326.
Kopit, W.G., and R.W. McCann (1988), '"Toward a definitive antitrust standard for nonprofit hospital mergers",
Journal of Health Politics Policy and Law 13(4):635-662.
Krattenmaker, T.G., and S.C. Salop (1986a), "Competition and cooperation in the market for exclusionary
rights", American Economic Review, Papers and Proceedings 76:109-113.
Krattenmaker, T.G., and S.C. Salop (1986b), "Anticompetitive exclusion: Raising rivals' costs to achieve
power over price", Yale Law Journal 96:209-293.
Krishnan, R. (1998), "Market restructuring and pricing in the hospital industry", unpublished doctoral disser-
tation (Katz Graduate School of Business, University of Pittsburgh).
LaFontaine, F. (1992), "Agency theory and franchising: Some empirical results", Rand Journal of Economics
23:263-283.
Lakdawalla, D., and T. Philipson (1998), "Nonprofit production and competition", Working Paper No. 6377
(National Bureau of Economic Research).
Lee, M.L. (1971), "A conspicuous production theory of hospital behavior", Southern Economic Journal
38(1):48-58.
Lee, R.H., and T.A. Mroz (1991), "Family structure and physicians' hours in large, multispecialty groups",
Inquiry 28:366-374.
Lieberman, M.B. (1991), "Determinants of vertical integration: An empirical test", Journal of Industrial Eco-
nomics 39:451-466.
Luft, H.S., D.W. Garnick, C.S. Phibbs, D.J. Peltzman, E. Lichtenberg and S.J. McPhee (1991), "Modeling the
effect of hospital charges and quality on choice", Journal of Health Care Marketing 11(3):2-11.
Luft, H.S., C.S. Phibbs, D.W. Garnick and J.C. Robinson (1988), "Rejoinder to Dranove and Shanley", Journal
Luft, H.S., J.C. Robinson, D. Garnick, S. Maerki and S. McPhee (1986), "The role of specialized clinical
services in competition among hospitals", Inquiry 23:83-94.
Lynk, W.J. (1984), "Tying and exclusive dealing: Jefferson Parish Hospital v. Hyde", in: J.E. Kwoka and
L.J. White, eds., The Antitrust Revolution: The Role of Economics (Harper-Collins Publishers, New York).
Lynk, W.J. (1994), "Property rights and the presumptions of merger analysis", Antitrust Bulletin 39(2):363-
383.
Lynk, W.J. (1995a), "Nonprofit hospital mergers and the exercise of market power", Journal of Law and
Lynk, W.J. (1995b), "The creation of economic efficiencies in hospital mergers", Journal of Health Economics
14:507-530.
Lynk, W.J., and L.R. Neumann (1999), "Price and profit", Journal of Health Economics 18:99-105.
Lynk, W.J., and M.A. Morrisey (1987), "The economic basis of Hyde: Are market power and exclusive con-
tracts related?", Journal of Law and Economics 30(2):399-421.
Ma, C.-t.A., and T.G. McGuire (1994), "Noncontractible inputs and health insurance-payment systems", un-
published manuscript (Boston University).
Madrian, B. (1994), "Employment-based health insurance and job mobility: Is there evidence of job-lock?",
Quarterly Journal of Economics 109:27-54.
Makuc, D., B. Haglund, D. Ingram and J. Kleinman (1991), "Health Service Areas for the United States",
Vital and Health Statistics, Series 2, No. 112 (National Center for Health Statistics).
Masten, S.E. (1984), 'The organization of production: Evidence from the aerospace industry", Journal of Law
and Economics 27:403-417.
Mathewson, G.F., and R.A. Winter (1987), "The competitive effects of vertical agreements: comment", Amer-
ican Economic Review 77:1057-1068.
McFadden, D. (1974), "Conditional logit analysis of qualitative choice behavior", in: P. Zarembka, ed., Fron-
tiers in Econometrics (Academic Press, New York).
McGuire, T.G. (2000), "Physician agency", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health
McGuire, T.G., and M.V. Pauly (1991), "Physician response to fee changes with multiple payers", Journal of
Melnick, G.A., and J. Zwanziger (1988), "Hospital behavior under competition and cost containment policies,
the California experience, 1980-1985", Journal of the American Medical Association 260(18):2669-2675.
Melnick, G.A., J. Zwanziger, A. Bamezai and R. Pattison (1992), "The effect of market structure and bargain-
ing position on hospital prices", Journal of Health Economics 11:217-233.
Miles, J.J. (1998), Health Care and Antitrust Law (West Group, St. Paul, MN).
Milgrom, P. (1981), "Good news and bad news: Representation theorems and applications", Bell Journal of
Miyazaki, H. (1977), "The rat race and internal labor markets", Bell Journal of Economics 8:394-418.
Mobley, L.R., and H.E. Frech HI (1997), "Managed care, distance traveled and hospital market definition: An
exploratory analysis", unpublished paper (University of California, Santa Barbara, CA).
Morrisey, M.A., F.A. Sloan and J. Valvona (1988), "Defining geographic markets for hospital care", Law and
Contemporary Problems 51:165-194.
Neipp, J., and R. Zeckhauser (1986), "Persistence in the choice of health plans", Advances in Health Eco-
nomics and Health Services Research 6:47-72.
Newhouse, J.P. (1970), "Toward a theory of nonprofit institutions: An economic model of a hospital", Amer-
ican Economics Review 60(1):64-74.
Newhouse, J.P. (1978), 'The structure of health insurance and the erosion of the medical marketplace", in:
Warren Greenberg, ed., Competition in the Health Care Sector: Past, Present, and Future (Aspen Publishers,
Germantown, MD).
Newhouse, J.P. (1996), "Reimbursing health plans and providers: Selection versus efficiency in production",
Journal of Economic Literature 34(3): 1236-1263.
Nitzan, S., and J. Tzur (1989), "Price and quality of professional services and codes of ethics", European
Ordover, J.A., G. Saloner and S.C. Salop (1990), "Equilibrium vertical foreclosure", American Economic
Review 80:127-142,
Pauly, M.V. (1968), 'The economics of moral hazard: comment", American Economic Review 58:531-537.
Pauly, M.V. (1978), "Is medical care different?", in: L. Goldberg and W. Greenberg, eds., Competition in the
Health Care Sector: Past, Present, and Future (Federal Trade Commission, Washington, DC).
Pauly, M.V. (1986), "Taxation, health insurance, and market failure", Journal of Economic Literature 24:629-
675.
Pauly, M.V. (1987), "Monopsony power in health insurance: thinking straight while standing on your head",
Pauly, M.V. (1988a), "Reply", Journal of Health Economics 7:85-87.
Pauly, M.V. (1988b), "Market power, monopsony, and health insurance markets", Journal of Health Eco-
nomics 7:111-128.
Pauly, M.V. (1988c), "Competition in health insurance markets", Law and Contemporary Problems 51:237-
271.
Pauly, M.V. (1998), "Managed care, markets, and monopsony", unpublished manuscript (University of Penn-
sylvania, PA).
Pauly, M.V., and M. Redisch (1973), "The not-for-profit hospital as a physicians' cooperative", American
Peltzman, S. (1977), "The gains and losses from industrial concentration", Journal of Law and Economics
20:229-263.
Phibbs, C.S., and J.C. Robinson (1993), "A variable-radius measure of local hospital market structure", Health
Pitchik, C., and A. Schotter (1987), "Honesty in a model of strategic information transmission", American
Plott, C.R., and L.L. Wilde (1980), "Professional diagnosis versus self-diagnosis: An experimental examina-
tion of some special features of markets with uncertainty", Social Science Working Paper 269 (California
Institute of Technology).
Png, I.PL. (1991), "Most favored customer protection versus price discrimination when customer demand is
uncertain", Journal of Political Economy 99:1010-1028.
Png, I.P.L., and D. Hirshleifer (1987), "Price discrimination through offers to match price", Journal of Busi-
ness 60:365-383.
Posner, R.A. (1976), Antitrust Law: An Economic Perspective (University of Chicago Press, Chicago).
Prospective Payment Assessment Commission (ProPAC) (1995), "Medicare and the American health care
system", Report to Congress (Commerce Clearing House, Chicago).
Rasmusen, E., J. Wiley and M. Ramseyer (1991), "Naked exclusion", American Economic Review 81:1137-
1145.
Reiffen, D. (1992), "Equilibrium vertical foreclosure: comment", American Economic Review 82:694-697.
Riley, J.G. (1979), "Informational equilibrium", Econometrica 47:331-359.
Riordan, M.H. (1998), "Anticompetitive vertical integration by a dominant firm", American Economic Review
88:1232-1248.
Riordan, M.H., and S.C. Salop (1995), "Evaluating vertical mergers: A post-Chicago approach", Antitrust
Law Journal 63:513-568.
Rizzo, J.A., and D. Blumenthal (1994), "Physician labor supply: Do income effects matter?", Journal of
Roback, G., L. Randolph and B. Seidman (1990), Physician Characteristics and Distribution in the U.S.
(American Medical Association, Chicago).
Robinson, J.C. (1988), "Market structure, employment, and skill mix in the hospital industry", Southern
Economic Journal 55(2):315-325.
Robinson, J.C., D.W. Garnick and S.J. McPhee (1987), "Market and regulatory influences on the availability
of coronary angioplasty and bypass surgery in US hospitals", New England Journal of Medicine 317(2):85-
90.
Robinson, J.C., and H. Luft (1985), "The impact of hospital market structure on patient volume, average
length of stay, and the cost of care", Journal of Health Economics 4:333-356.
Romeo, A.A., J.L. Wagner and R.H. Lee (1984), "Prospective reimbursement and the diffusion of new tech-
nologies in hospitals", Journal of Health Economics 3:1-24.
Rothschild, M., and J.E. Stiglitz (1976), "Equilibrium in competitive insurance markets: An essay on the
economics of imperfect information", Quarterly Journal of Economics 90:629-650.
Royalty, A.B., and N. Solomon (1997), "Health plan choice: Price elasticities in a managed competition
setting", unpublished manuscript (Stanford University, Stanford, CA).
Salinger, M.A. (1988), "Vertical mergers and market foreclosure", Quarterly Journal of Economics 103:335-
356.
Salkever, D.S. (2000), "Regulation of prices and investment in hospitals in the United States", in: A.J. Culyer
and J.P. Newhouse, eds., Handbook of Heath Economics (Elsevier, Amsterdam) Chapter 28.
Salop, S.C. (1979), "Monopolistic competition with outside goods", Bell Journal of Economics 10: 141-156.
Salop, S.C., and D.T. Scheffman (1983), "Raising rivals' costs", American Economic Review, Papers and
Proceedings 73(2):267-271.
Salop, S.C., and D.T. Scheffman (1987), "Cost-raising strategies", Journal of Industrial Economics 36:19-34.
Salop, S.C. (1986), "Practices that (credibly) facilitate oligopoly coordination", in: J. Stiglitz and G.F Math-
ewson, eds., New Developments in the Analysis of Market Structure (The MIT Press, Cambridge, MA).
Satterthwaite, M.A. (1979), "Consumer information, equilibrium industry price, and the number of sellers",
Bell Journal of Economics 10:483-502.
Satterthwaite, M.A. (1985), "Competition and equilibrium as a driving force in the health services sector", in:
R. Inman, ed., Managing the Service Economy (Cambridge University Press, Cambridge, MA).
Scheffman, D.T., and PT. Spiller (1987), "Geographic market definition under the US Department of Justice
merger guidelines", Journal of Law and Economics 30:123-147.
Scherer, FM. (2000), "The pharmaceutical industry", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of
Scherer, F.M., and D. Ross (1990), Industrial Market Structure and Economic Performance, 3rd edn.
(Houghton Mifflin, Boston).
Schmalensee, R. (1989), "Inter-industry studies of structure and performance", in: R. Schmalensee and
R. Willig, eds., Handbook of Industrial Organization (North-Holland, Amsterdam).
Scott Morton, F.M. (1997), "The strategic response by pharmaceutical firms to the Medicaid most-favored-
customer rules", Rand Journal of Economics 28:269-290.
Shortell, S.M., and E.F.X. Hughes (1988), "The effects of regulation, competition, and ownership on mortality
rates among hospital inpatients", New England Journal of Medicine 318:1100-1107.
Schwartz, M. (1987), "The competitive effects of vertical agreements: comment", American Economic Re-
view 77:1063-1068.
Simpson, J., and R. Shin (1997), "Do non-profit hospitals exercise market power?", unpublished paper (Fed-
eral Trade Commission).
Sloan, F.A. (1975), "Physician supply behavior in the short run", Industrial and Labor Relations Review
28:549-569.
Sloan, F., and E. Becker (1981), "The internal organization of hospitals and hospital costs", Inquiry 18:224-
239.
Spence, A.M. (1976), "Product selection, fixed costs, and monopolistic competition", Review of Economic
Studies 43:217-235.
Spence, A.M. (1978), "Product differentiation and performance in insurance markets", Journal of Public
Spengler, J.J. (1950), "Vertical integration and antitrust policy", Journal of Political Economy 53:347-352.
Spiller, P.T. (1985), "On vertical merger", Journal of Law, Economics, and Organization 1:285-312.
Stano, M. (1987), "A further analysis of the physician inducement controversy", Journal of Health Economics
6(3):227-238.
Staten, M., W. Dunkelberg and J. Umbeck (1987), "Market share and the illusion of power: Can Blue Cross
force hospitals to discount?", Journal of Health Economics 6:43-58.
Staten, M., J. Umbeck and W. Dunkelberg (1988), "Market share/market power revisited, a new test for an
old theory", Journal of Health Economics 7:73-83.
Stiglitz, J.E. (1977), "Monopoly, non-linear pricing and imperfect information: The insurance market", Re-
view of Economic Studies 44:407-430.
Sullivan, D. (1989), "Monopsony power in the market for nurses", Journal of Law and Economics 32:S135-
S178.
Sutton, J. (1991), Sunk Costs and Market Structure: Price Competition, Advertising, and the Evolution of
Concentration (MIT Press, Cambridge, MA).
Svejnar, J. (1986), "Bargaining power, fear of disagreement, and wage settlements: Theory and evidence from
US industry", Econometrica 54:1055-1078.
US Department of Justice (1984), Merger Guidelines (US Department of Justice, Washington, DC).
US Department of Justice and Federal Trade Commission (1997), Horizontal Merger Guidelines, issued
April 2, 1992, revised April 8, 1997 (US Department of Justice, Federal Trade Commission, Washington,
DC).
van der Weide, J., and J. Zalkind (1981), "Deregulation and oligopolis price-quality rivalry", American Eco-
Vernon, J.M., and D.A. Graham (1971), "Profitability of monopolization by vertical integration", Journal of
Vita, M.G. (1990), "Exploring hospital production relationships with flexible functional forms", Journal of
Vita, M.G., and L. Schumann (1991), "The competitive effects of horizontal mergers in the hospital industry:
A closer look", Journal of Health Economics 10:359-372.
Vitaliano, D.F. (1987), "On the estimation of hospital cost functions", Journal of Health Economics 6:305-
318.
Volpp, K., and J. Waldfogel (1998), "Competition and the quality of hospital care: Heart attack mortality after
the onset of price competition in New Jersey", unpublished manuscript (University of Pennsylvania, PA).
Walston, S.L., J.R. Kimberley and L.R. Burns (1996), "Owned vertical integration and health care: Promise
and performance", Health Care Management Review 21(1):83-92.
Werden, G.J. (1981), "On the use and misuse of shipments data in defining geographic markets", Antitrust
Bulletin 26:719-737.
Werden, G.J. (1990), "The limited relevance of patient migration data in market delineation for hospital
merger cases", Journal of Health Economics 8:363-376.
Werden, G.J. (1997), "Simulating unilateral competitive effects from differentiated products mergers", An-
titrust (Spring).
Werden, G.J., and L.M. Froeb (1994), 'The effect of mergers in differentiated products industries: Logit
demand and merger policy", Journal of Law, Economics and Organization 10(2):407-426.
Westfield, F.M. (1981), "Vertical integration: Does product price rise or fall?", American Economic Review
71(3):334-346.
White, L. (1972), "Quality variation when prices are regulated", Bell Journal of Economics 3:524-436.
Williamson, O.E. (1975), Markets and Hierarchies: Analysis and Antitrust Implications (Free Press, New
York).
Williamson, O.E. (1989), "Transaction cost economics", in: R. Schmalensee and R. Willig, eds., Handbook
of Industrial Organization (North-Holland, Amsterdam).
Wilson, C. (1977), "A model of insurance markets with incomplete information", Journal of Economic Theory
16:167-207.
Wolinsky, A. (1993), "Competition in a market for informed experts' services", Rand Journal of Economics
24:380-398.
Woolley, J.M. (1989), "The competitive effects of horizontal mergers in the hospital industry", Journal of
Woolley, J.M. (1991), "The competitive effects of horizontal mergers in the hospital industry: An even closer
look", Journal of Health Economics 10:373-378.
Zeckhauser, R. (1970), "Medical insurance: A case study of the tradeoff between risk spreading and appro-
priate incentives", Journal of Economic Theory 2:10-26.
Zwanziger, J., and G. Melnick (1988), "The effects of hospital competition and the Medicare PPS program
on hospital cost behavior in California", Journal of Health Economics 7:301-320.
Chapter 28
REGULATION OF PRICES AND INVESTMENT

IN HOSPITALS IN THE UNITED STATES*
DAVID S. SALKEVER
The Johns Hopkins University
Contents
Abstract 1490
Keywords 1490
Introduction 1491
1. The demand for regulation - a priori rationales and conjectures 1491
2. Scope and nature of regulation 1493
2.1. Certificate of need controls 1493
2.2. Rate regulation 1495
2.3. Federal regulatory controls 1497
2.4. Medicare prospective payment and the changing policy environment 1498
2.5. Hypotheses about regulatory structure, incentives and effectiveness 1499
3. The demand for regulation: empirical models 1503
4. Effects of regulation 1504
4.1. Dependent variables 1504
4.2. Modeling frameworks 1506
4.3. Regulatory impacts on hospital costs 1507
4.3.1. Empirical research on cost impacts of regulation 1507
4.3.2. Analyses of intra-program variations 1515
4.4. Regulatory impacts on the quantity of hospital services used (utilization) 1516
4.5. Regulation impacts on hospital investment 1519
4.5.1. Studies based on summary capital stock measures 1519
4.5.2. Regulation impacts on diffusion/availability of specialized services 1521
4.6. Regulation impacts on reserve margins or excess capacity 1523
4.7. Overview of results 1524
4.8. A review of selected research on PPS impacts on hospital costs 1527
5. Concluding observations 1530
5.1. On the noncovergence of empirical results 1530
5.2. Future research directions 1531
References 1532
*I gratefully acknowledge the assistance of Jim Collins and Judy Shinogle in the preparation of this chapter.
I also wish to thank Ken Warner and Joe Newhouse for helpful comments on an earlier draft. Responsibility
for any remaining errors and inadequacies is mine alone.
Handbook ofHealth Economics, Volume 1, Edited by A.J. Culyer and J.P. Newhouse
1490 D.S. Salkever
Abstract
With the spread of cost-based hospital payment systems in the United States in the
1960s, and the implementation of the Medicare and Medicaid programs in 1966, rapidly
rising hospital costs imposed unexpected pressures on Federal and state budgets and
generated a demand for regulatory interventions. Large numbers of states responded
with regulatory controls on hospital investment and a significant minority of states en-
acted hospital price regulation (rate-setting) laws. As strong hospital sector inflation
continued into the 1970s, Federal efforts to regulate prices (through the Economic Sta-
bilization Program) and to encourage additional state regulation were also enacted. This
chapter reviews the economic research on the impact of these regulatory interventions,
focusing on econometric studies in particular. Several conclusions emerge from this re-
view. First, studies of adoption of regulation show that pressure on state budgets and
pro-regulation political views were more influential than "provider capture" or "rent-
seeking" factors. Second, cost-containment impacts of state rate-setting programs var-
ied over time, with changes in the national health care economy and major Federal
policy thrusts. Third, there is little evidence that investment controls reduced the rate
of cost growth though inconsistent reports of constraining effects on numbers of beds
and diffusion of some specialized services did appear. Fourth, econometric studies of
the Medicare Prospective Payment System (PPS) supported the presumption that PPS
would constrain the growth in cost per case, but concomitant increases in case-mix in-
tensity and declines in admissions raised questions about (1) the extent to which PPS
truly induced efficiency gains and (2) the adequacy of our analytic models of hospital
behavior. Fifth, as cost-based payment was replaced by prospective payment in Medi-
care, Medicaid and the private sector, and as managed care encouraged price compe-
tition, the evidence of regulatory cost savings dwindled and rate regulation virtually
disappeared. While investment regulation is still widespread, its role and effect in the
new hospital marketplace is still unclear.
Keywords
rate setting, rate regulation, investment regulation, certificate of need, CON,

prospective payment, PPS, economic stabilization program, ESP, demand for
regulation
JEL classification: 11t, 118

Ch. 28: Regulation of Prices and Investment in Hospitals in the U.S. 1491
Introduction
One of the prominent features of the post-Medicare health economy in the United States
was the implementation of direct government regulation on pricing, entry, and expan-
sion decisions of private institutional health care providers, and in particular hospitals.
The primary modes of this regulation were price controls, and certificate-of-need (CON)
controls over investment and the offering of new services. Prior to this time, Federal and
state governments impacted on health care markets through their roles as purchasers,
through regulation of payers (i.e., health insurance plans) and tax policy, and through
facility and personnel licensure regulations. With the major increase in the government
role as purchaser under Medicare and Medicaid, however, direct price and CON controls
began to be deployed.
As direct price and CON controls became more widespread and as experience with
these regulatory programs accumulated, many economists and policy analysts under-
took studies to evaluate their economic effects. A range of research strategies was ap-
plied, including "black-box" studies that sought to estimate such regulatory "outcomes"
as reduction in unit or per capita costs of hospital services, studies that sought to mea-
sure regulatory influences on hospitals' decisions regarding input choices and product
characteristics, studies that modeled political decision-making in establishing regula-
tory programs, and descriptive studies of the operations of these programs. The purpose
of this chapter is to provide an overview of this literature of econometric studies of
hospital costs, volume and investment.
Our review begins by considering the various rationales that have been offered for
adopting these regulatory programs. We also discuss the chronology of their rise, dif-
fusion, and decline. We then consider the empirical evidence on the demand for these
programs and the consistency of this evidence with the rationales offered by regulatory
proponents. This is followed by the major portion of our review, dealing with the lit-
erature on the economic effects of regulation. After a brief digression to survey some
of the recent research on the Medicare Prospective Payment system (PPS), we provide
concluding comments and a distillation of the major research findings from this litera-
ture.
1. The demand for regulation - a priori rationales and conjectures
The major arguments for price and investment controls in health care that were ad-
vanced during the late 1960s and early 1970s drew upon the literature in health
economics which explored in detail the special structural characteristics of health
care markets and their differences from the characteristics of well-functioning, com-
petitive markets. Early contributions to this literature described these special struc-
tural characteristics [Mushkin (1958)], and interpreted them in the context of physi-
cian profit maximization [Kessel (1958)]. Arrow's (1963) seminal analysis explored
the role of uncertainty, limited information, and incomplete markets for risk bear-
ing as explanations for these unusual market characteristics. Klarman (1965) and
1492 D.S. Salkever
Fuchs (1972) expanded upon these earlier discussions, identifying increasing returns
as a potentially important problem; Fuchs also discussed at length (albeit some-
what skeptically) the implications of externalities for resource allocation. Pauly (1968,
1971) extended Arrow's analysis by focusing on the special form that insurance took
in medical care as a result of limited information. Feldstein (1971) explored the im-
plications of this form of insurance for technical change and resource allocation in the
hospital sector.
These early discussions developed a number of the ideas that would be marshaled,
after the enactment of Medicaid and Medicare in 1966, to argue that medical spending
was excessive: examples are moral hazard effects of insurance which encouraged "over-
utilization", inefficiency in production due to the attenuated role of profit incentives and
the absence of managerial rewards for efficiency under insurers' cost-based reimburse-
ment systems, and supra-competitive prices as a result of entry barriers. Direct controls
on price and entry were proposed as remedies that would constrain spending to a level
that was presumably more consistent with (1) the preferences of well-informed con-
sumers and (2) economic efficiency in the production process.
This optimistic view of regulatory controls is often referred to, applying Posner's
(1974) terminology, as the "public interest" theory of regulation. Pessimistic observers
drew upon experience with regulation in other sectors to question this theory, suggest-
ing alternative theories of regulatory "capture" and "producer protection" to explain the
demand for regulation [see, for example, Noll (1975)]. Policy alternatives to regulation
were also offered from various points on the political spectrum, from the promotion
of market competition among health plans, to the expansion of government purchas-
ing to bring all services under a "single-payer" national health system. This pessimism
notwithstanding, direct regulatory controls were in fact the favored initial response, at
both Federal and state levels, to the increased rate of growth of medical spending in the
post-Medicare era.
A second rationale for regulation also focused on the need to constrain the growth
in expenditures but this argument focused specifically on excessive expenditures under
publicly financed programs. One variant of this argument was that the budgetary costs of
publicly funded entitlement programs such as Medicare or Medicaid were exceeding the
amount that taxpayers were willing to spend for these programs. While these budgetary
costs might be controlled by benefit reductions or by reducing payments to providers
serving program beneficiaries, the first of these remedies was not politically appealing
and the second ran the risk of discouraging providers from participating and thereby
creating "access" problems for beneficiaries. Broader-based controls on total hospital
costs, through price and investment regulation, were seen as offering budgetary relief
without potential "access" problems because these controls would apply to all private
patients as well as to program beneficiaries. Moreover, in the case of the Medicaid
program, Federal mandates limited other state options for controlling program costs.
For example, eligibility levels could not be reduced unless the state was also willing
Ch. 28: Regulation of Pricesand Investment in Hospitals in the U.S. 1493
to reduce eligibility levels for cash assistance programs (since Medicaid eligibility was
tied to cash assistance).l
A third line of argument for direct regulation of hospitals parallels the "taxation-
by-regulation" or consumer cross-subsidy rationales explained by Posner (1971), Noll
(1975) and others. In the case of regulation of hospital prices, this argument has been in-
voked in recent years with reference to the funding of charity or "uncompensated" care. 2
According to this rationale, as managed care plans promote price competition in mar-
kets for hospital services, hospital profit margins will be squeezed and the willingness
of hospitals to supply charity care will diminish. Therefore, price regulation schemes,
which afford some protection to hospitals from the pressures of unfettered price com-
petition, can preserve the supply of charity care. Even before the advent of the managed
care revolution, a similar argument was cited by Cohen (1978) in explaining the support
for rate regulation among hospitals in the state of Maryland. In particular, Cohen noted
that hospitals supported regulation as a means to require the major private insurer (Blue
Cross) pay a portion of the hospitals' uncompensated care costs.
Taxation-by-regulation has also been invoked in the case of CON controls. Here the
approval by regulators of proposed entry or expansion plans of a hospital is made con-
tingent, through formal conditions or through informal negotiation, on the willingness
of the hospital to supply services that are perceived by the regulators as in the public in-
terest, such as charity care or outpatient clinic services in poor "under-served" localities
or neighborhoods.
2. Scope and nature of regulation
2.1. Certificate of need controls
Of the two types of regulatory controls we are considering, capital expenditure controls
on entry and expansion were the first to appear in the U.S. and became much more
widespread. New York State enacted the first CON law in 1964, and four additional
states (California, Connecticut, Maryland and Rhode Island) enacted similar laws in
1968 and 1969, shortly after the implementation of the Medicare and Medicaid pro-
grams. The most rapid spread of this form of regulation occurred during the 1970s, with
25 states implementing CON programs in the years 1970-1975 [Russell (1979, p. 39)].
The Federal government encouraged adoption of CON laws during this period by
passage of Title XV of the Public Health Service Act (P.L. 93-641) in 1974. This act
required states to establish CON programs as an eligibility condition for receipt of var-
ious Federal grant monies. By August 31, 1977, 36 states and the District of Columbia
I A similar argument explaining the desire of states to limit the supply of nursing home beds and thereby
constrain Medicaid costs was detailed by Scanlon (1980).
2 Uncompensated care includes charity care (where no bill is presented to the patient because they are
presumed unable to pay) and bad debts.
1494 D.S. Salkever
had adopted CON laws for hospital investment that remained in effect [Chayet and Son-
nenreich (1978)]. 3 In all, 22 states enacted CON laws in 1975 or later and by 1979 only
3 states did not have such laws [Sloan (1981)].
A related mechanism for controlling investment was also made available to states
through the passage of Section 1122 of the Social Security Act in 1972. This section pro-
vides for the establishment, in agreements between the Federal government and states,
of designated Section 1122 review programs. Under these programs, designated state
agencies reviewed and approved proposed hospital investment projects. Providers who
undertook investment projects that were not approved were subject to denial of reim-
bursements under the Medicare and Medicaid programs for depreciation, interest and
other costs associated with these investments. As of August 1977, 37 states had Section
1122 programs in operation; three others had implemented but then terminated such
programs [Chayet and Sonnenreich (1978)]. By January 31, 1978, all but one of the
states (Missouri), as well as the District of Columbia, had either a CON or Section 1122
program in operation. After adopting CON laws, however, some states came to view
their Section 1122 programs as superfluous. As a result, even though all but 9 states
adopted Section 1122 programs at some time during the 1970s, by late 1979, 15 states
terminated or were in the process of terminating their programs [Sloan (1981)].
State CON laws specified the types of investments or changes in programs requiring
regulatory approval, the types of health care providers subject to the law, the agencies
responsible for the review process, and applicable sanctions for providers undertaking
non-approved investments. Three types of investment or changes were typically cov-
ered: investments in buildings and facilities, investments in equipment, and changes in
services offered. 4 (In some states, service changes requiring CON approval included
both additions of new services and discontinuation of old services.)
To eliminate the need to approve very small investment expenditures, a dollar thresh-
old was specified for exempt investments. In the mid-1970s, plant and equipment thresh-
olds were typically about $100,000 though they ranged from $15,000 to $350,000
[Salkever and Bice (1978)]. These thresholds have, of course, increased over time with
inflation and with recent state legislative moves to reduce the scope of CON regulation.
At the present time, thresholds are usually in excess of $1,000,000 [Moore (1997)].
Most CON laws that covered new services did not have threshold exemption provisions
for such changes so typically all new service offerings are covered.
CON reviews were typically carried out in a two-step or three-step process. The ini-
tial review would be performed by a local area health-planning agency while the final
review and approval was the responsibility of a state health department or a state health
3 One state, North Carolina, adopted CON but then had their law declared by a state court to be in conflict
with their state constitution. Also note that one other state, Oklahoma, limited its CON law to nursing homes
so hospital investment was not regulated.
4 Changes in services would cover, for example, new services that did not require substantial plant or equip-
ment expenditures, such as home care services offered by a hospital, or conversion of hospital inpatient beds
from one type of service to another.
planning commission or agency. Hospitals proceeding with unapproved investments

were usually subject to the threat of license revocation or denial, or to court-ordered
sanctions (e.g., fines). In several states, reimbursement for Medicaid costs related to
disapproved projects could be withheld.
While our focus in this chapter is on hospital regulation, it is relevant to note that state
CON laws covered other institutional providers as well, specifically nursing homes.
Some providers of ambulatory care, such as kidney dialysis centers and ambulatory
surgery centers, were also usually covered by the CON laws, while ambulatory care
clinics, medical office buildings or private physician offices were generally exempt from
CON review.
As noted above, the decade of the 1970s witnessed the rapid diffusion of CON laws.
With the lapse of related Federal programs and incentives (i.e., the Section 1122 pro-
gram and the financial incentives created in P.L. 93-641), and the emergence of alterna-
tive cost-containment mechanisms (e.g., the Medicare PPS and increased market pen-
etration of managed care and capitation contracts), a number of states began reducing
or terminating their CON programs in the 1980s. This trend away from CON controls
has, however, been far short of wholesale abandonment. As of August 1997, 37 states
and the District of Columbia still maintained CON programs [Moore (1997)]. General
acute care hospital services were still subject to CON regulation in 29 states, while
long-term care facilities (i.e., nursing homes) were still under CON regulation in all ju-
risdictions with a CON program. Specialized acute care services were also commonly
subject to CON controls, such as open-heart surgery (in 29 states) and cardiac catheter-
ization laboratories (in 27 states). It also appears that more emphasis in the political
debates is now being placed on "taxation by regulation" rather than general cost control
concerns as a rationale for continuing CON programs. In particular, more emphasis is
being placed on conditions for CON approvals (such as requiring provision of minimum
levels of charity care) and protecting community hospitals that currently supply charity
care from competition by new entrants seeking to "cherry pick" profitable services such
as rehabilitation and cardiac surgery [Moore (1997)].
2.2. Rate regulation
Legally mandated rate regulation for hospitals could be viewed as evolving out of
the experience of large private insurers (specifically, Blue Cross plans) with so-called
"prospective reimbursement" programs. As in the case of CON regulation, New York
was the first state to enact a mandatory rate-setting law that took effect in 1969. 5 By
5 Mandatory rate-setting programs are those which required hospitals to participate in the programs and also
required them to actually comply with the rates that were set by the state agencies. A number of other states
adopted either voluntary participation-mandatory compliance programs or mandatory participation-voluntary
compliance programs. For the most part, rate-setting in this review will refer to programs requiring mandatory
participation and compliance that have jurisdiction over private-sector rates. Thus we exclude Medicaid-only
or Medicare-only arrangements as well as programs that are not fully mandatory.
1496 D.S. Salkever
the end of 1975, seven additional states had joined New York: Connecticut, Mary-
land, Massachusetts, New Jersey, Rhode Island, Washington, and Wisconsin. Two other
states, Colorado and Illinois, passed laws providing for the establishment of manda-
tory rate-setting but the lives of their programs were very brief. The Illinois program
was officially in existence by 1978 but its legislation required that all payers, including
Medicare, agree to its rates for compliance by hospitals to be mandatory. Since the Fed-
eral Medicare program never reached the necessary waiver agreement with the State of
Illinois, the Illinois program was automatically repealed in 1982 without ever having the
authority to set binding rates. Colorado established a rate-setting program in October of
1977 but in May of 1979 legislation was signed terminating the program as of March 1,
1980.
In retrospect, the experiences in these two states seem consistent with a broader na-
tional trend away from rate setting (including the rejection, at the Federal level, of the
Carter administration's proposals discussed below). During the decade of the 1980s,
only one additional state (West Virginia) enacted a mandatory rate-setting program.
Rate setting programs had legal authority for approving the rates that hospitals could
charge. In some states, this authority included all "private" patients as well as all Med-
icaid (and other state-supported) patients. In others, Medicaid patients or specific cat-
egories of private patients were not included under the law. Medicare patients were
generally not subject to regulated rates but in some instances the Federal government
granted states a waiver to cover Medicare patients; in those instances, essentially all
patients were covered by rate setting and hence they were referred to as "all-payer" rate
regulation states. (The only states who had such waivers at any time were Maryland,
Massachusetts, New Jersey and New York.)
The actual process for setting rates varied considerably from state to state. Most
states used a detailed budget review process where hospitals submitted proposed bud-
gets along with projections of service volumes. Regulators reviewed the budgets and
volume projections, approved or amended these figures, and set rates based on the vol-
ume projections so that the costs which they deemed to be "allowable" or "reasonable"
would be covered by total revenues. A variant on this process was to conduct the de-
tailed budget review in one year, apply an inflation factor in several subsequent years
to trend rates forward, and then conduct another detailed budget review (say) four or
five years later. The advantages of this process were that it reduced demands on the reg-
ulators and it increased profit incentives by building a regulatory lag into the process.
Finally, one state (New York) did not conduct budget reviews and relied instead on the
application of "formulas" to set rates. These formulas included lagged values of each
hospital's average costs, an inflation "trend" factor, and other formulaic provisions such
as penalties for low occupancy rates.
The units for which rates were set varied across states and over time. In the early to
mid-1970s, rates were either set on an all-inclusive charge per day basis or on the basis
of individual services (e.g., charge per "relative value unit" for pathology lab services,
charge per minute for operating room use, etc.). By the late 1970s, Maryland began im-
plementation of a per case system where the maximum allowable charge was computed
on a case-mix adjusted per case basis. New Jersey also phased in a per case system over
the 1980-1982 period. The Federal Prospective Payment System (PPS) for Medicare,
implemented in 1983, used a similar per case payment method to that applied in New
Jersey.
Approved rates were typically set to cover "allowable" costs at projected volumes.
If short-run marginal costs were substantially below average costs, however, hospi-
tals charging approved rates but experiencing higher-than-projected volumes could earn
substantial profits. The hospital economics literature gave some support to the notion
that short-run marginal costs were indeed below average costs, suggesting that the
profit incentive for higher-than-projected volume was real. Many experts also believed
that hospitals (acting in concert with physician on their medical staff) could respond to
this incentive by manipulating patient demand and encouraging greater use of services.
Since a primary stated objective of rate regulation was to reduce overall spending (and
not just unit costs), this incentive to increase volume was seen as problematic. To dimin-
ish (or even eliminate) this incentive, some states implemented "volume-adjustment"
provisions, whereby the allowable unit charge was reduced (increased) for hospitals
that experienced greater-than-projected (less-than-projected) volumes. At least one state
(Maryland) undertook a detailed analysis to determine, on a department by department
basis, the relationship of short-run marginal to average costs and factored the results of
this analysis into their volume adjustment formula.
The shift away from cost reimbursement by Medicare in the early 1980s, and the
emergence of managed care and capitation as viable cost-control mechanisms in the late
1980s undermined the principal political arguments for state rate setting. Both private
sector costs and state Medicaid payments could be managed under the new capitation
payment arrangements. Alternative strategies for funding uncompensated care, such as
insurance premiums or hospital taxes, were devised and implemented. These factors
contributed to a rapid and widespread demise of state rate-setting beginning in 1986 in
Wisconsin. In the ensuing decade every state program was terminated except for those
in Maryland and West Virginia.
2.3. Federal regulatory controls
Federal controls on hospital costs in the early 1970s were implemented as part of the
Nixon administration's economy-wide effort to fight inflation, the Economic Stabiliza-
tion Program (ESP). Controls began with the three-month freeze on all wages and prices
throughout the economy implemented in August of 1971. This broad policy initiative
was not designed with the complexity of the hospital sector in mind, with the result that
cost reimbursements to hospitals (which were the majority of their inpatient revenues)
were not covered under the freeze. New Phase II regulations specific to hospitals went
into effect in December of 1971; there were, however, substantial ambiguities in the reg-
ulations, particularly with respect to volume adjustment procedures, and it was not until
September of 1972 that these ambiguities were resolved and the process of computing
allowable revenue increases due to volume adjustments was clarified [Ginsburg (1978)].
1498 D.S. Salkever
The ESP Phase II regulations for hospitals remained in effect with only relatively mi-
nor changes until the cessation of the ESP in April of 1974 [Ginsburg (1976)]. These
regulations limited annual revenue growth due to cost-justified increases in prices and
wages to 6 per cent. Allowable increases in revenue due to volume growth were com-
puted on the basis of admissions, length of stay, and recent trends in intensity growth
(i.e., growth in use of specific tests and other services per day or per case) in the hospi-
tal sector. The Phase II regulations treated virtually all hospital costs as variable so that
growth in patient days or admissions translated into nearly proportionate increases in
allowable revenue increases [Altman and Eichenholz (1976)1.
With the lapse of Federal price controls in April 1974, Federal efforts to strengthen
direct price and investment regulation of hospitals in the ensuing years mainly took the
form of efforts to encourage State-level controls and to strengthen the state and local
health planning agencies. The major step in this direction was the enactment of P.L.
93-641 to promote state CON programs. A renewed and expanded Federal role was
proposed, following the change in administration in Washington, in the Carter adminis-
tration's Hospital Cost Containment Act of 1977. This bill included provisions to limit
the growth of total revenues of hospitals (Title I) and to limit hospital capital expen-
ditures and expansion of hospital bed supplies (Title II).6 After the introduction of the
bill and several amended alternatives in the Congress, an extended process of political
debate ensued.
During this period, a coalition of major national health care provider associations,
led by national hospital industry groups, reacted to the prospect of greatly increased
Federal controls by proposing an alternative in December of 1977. This alternative was
a voluntary cost containment effort whereby hospitals in each state would seek to limit
the annual rate of increase in costs to 2 per cent below the rate experienced in 1977.
Formal targets were set for each state for 1978 and 1979 and the so-called Voluntary
Effort program was initiated. The success of the Voluntary Effort in reaching its goals
for 1978, despite increased inflation in the economy generally, bolstered the opposition
to the Carter cost containment bill and contributed to its final defeat in November of
1979. With the threat of Federal regulation removed, the Voluntary Effort terminated
shortly thereafter [Davis et al. (1990, Chapter 2)].
2.4. Medicareprospective payment and the changingpolicy environment
Hospital rate regulation and investment controls were put in place during an era when
hospitals were paid primarily on the basis of reimbursement of incurred costs. Thus, the
private insurers with the largest market shares (Blue Cross plans) and the major public
purchasers (Medicare and Medicaid) paid hospitals in a manner that provided virtually
no efficiency incentives nor scope for price competition. The predictable result was a
massive growth in state spending that spurred the demand for state regulation.
6 For detailed analyses of the administration proposal and several alternative bills, see American Enterprise
Institute (1978) and Dunn and Lefkowitz (1978).
The implementation of the Medicare Prospective Payment System (PPS) in 1983

began a period of change in this dynamic. For approximately one-third of hospital in-
patients, incurred costs now had essentially no effect on per unit payment levels and
payment rates were set prospectively on a case-mix adjusted per case basis. 7 After a
phase-in period, a uniform national schedule of per case rates was applied to all gen-
eral hospitals, with adjustments for local labor market conditions, capital costs, medical
education activities, and provision of care to the poor.
The implementation of PPS dramatically altered hospitals' incentives. Reductions
in inpatient costs through shorter stays would now improve a hospital's bottom line;
reducing admissions for types of cases where costs exceeded Medicare payment rates
would have the same effect. Trend data for "community hospitals" in the U.S. in Table 1
suggest that hospitals did indeed react to this change in incentives. 8 In the first year of
PPS implementation, 1984, a relatively sharp drop was observed in inpatient length
of stay both for all patients and for Medicare patients. In subsequent years, the trend
leveled off until the substantial declines of the early and mid-1990s.
The initial years of PPS also saw a reduction in the rate of growth of inpatient ex-
penses in total and on a per case basis. Here again, it appears that the initial PPS impact
diminished in the latter 1980s while the most dramatic curbs on growth occurred dur-
ing the early and mid-1990s, with real inpatient expense per case actually declining in
1994 and in 1996. Trends in numbers of inpatient cases (admissions or discharges) also
changed during the early PPS years, with declines for Medicare and for all patients. One
might question, however, whether this trend shift was directly related to PPS since most
observers viewed PPS as providing incentives for increasing numbers of admissions.
An upward trend of this sort did not begin until 1987 for Medicare patients while total
admissions have been quite stable over the 1989-1996 period.
In summary, while PPS appears to have ushered in an era of diminished growth,
the economic landscape changed most dramatically with the rapid growth of managed
care and capitation, and the emergence of stronger price competition in the 1990s. As a
consequence, rate regulation largely disappeared from the health policy radar screen.
2.5. Hypotheses about regulatory structure, incentives and effectiveness
Students of health sector regulation have often suggested that variations in the struc-
ture and scope of regulatory programs imply variations in the effectiveness of these
programs. Several of these suggested hypotheses draw upon familiar ideas from the
7 Note, however, that for those DRG's that Medicare defined on the basis of surgical procedures, the connec-
tion between payment rates and resources used to treat each case was not completely severed, since surgical
treatment resulted in patients being classified in a DRG with a higher per case payment. For further discussion
of this point, see McClellan (1996).
8 Community hospitals are defined as non-Federal short-term general and special hospitals excluding hos-
pital units of institutions (e.g., colleges, prisons). As of 1995, they accounted for 82 per cent of the beds and
89 per cent of the expenditures for all U.S. Hospitals.
1500 D.S. Salkever
o^.
ON -
3 m
I I I I I I I i I
._ S oe
a -
SN
O N Cti
m O - O O - - ' It
O
d
Eg .E:
C CO
N N C 'C 'C C/ 'D T C/ V) C C a x 1 I

'C X r0 C C 1 N Sq oC n Ot t ON
'CI 3 ION
C/a ' N
10 X 10 N V x r 't t mi iCC
't O----
'C C)a
- C) aC
C) ur * C N X oN 'C ON ON ON oN 'C 'C

C)"
- ft Uo - < So o N~~~r-; 17ro

C| X
'o
O C X~ 'C oC oN _'' N ON ON C rC cN o/ o-N-
'0~~-
C
m
C C C
NC N C ICC
CC ' N 69
So
~N ~
CC N
0C C
C
~c C
~
8
CC
i
P a^
-
rncN
M '-
C
CC
C
- C' '~t *tONCC
/Nn - N
C/N ON
N
E
C
.e
Ca
- '11 CCCNO
q6~V 1
O-~'NO
lbl ~ ~ " l
O ON
~b,
…N cINNGl 'S'I 'C
e
1 C8
ON
-C
z
ON m MCO x 7 C t o - CO
,t CC-ON ON
N C,CN
-
ClNr r- ON 0)0V'-
CC- Iet ON
o r
1o 2 'C - r vN r C xlo x N
a, o O
a) C 'C C) NOONNNN NON 'C' m~cr-
'C N ON
C, ON N - - .s C
E-
-
CCCC , i rs .N N N N N MC _ C) C)'.
'f5 6S * t f C CC CC CC CCfC CC C CC C
u 0
wo
oo ox 0x' C r0
w4 csn 0o 00 woot- o
v ro I
'C N x O N O NC C'
.
0 -OP
C)
·a
a
'a -C
u
'C 'C C 'C C
~~~3NNNNNC'NCCC
_ _ Ns N N N9 Ns _
r r rr r r r r
ON xC C
r r IoC
N
9
'0
a 3 0
u
3-- N CC C o N NcN ' O C N
A rO
N o O
o N
00N CC 'D
C ONNw'C O C,CCO a
.21
E
_- CL oC)
C
ONON--CONONONONONONO
2 .b c Ct
ONC)
C)
° _:zS
* *
x
p N .,
ON1
.0
1
.. om im c r _a
*t N 04 -
.F P,
ON - N CC C/ 'C N oo ON oN - Ns CC < CN 'C ,: m
x C x C x x C x O CO N ON ON ON O O ON
C) N N N N N N N N O O O
general literature on the economics of regulation while others derive, at least in part,
from the special institutional characteristics of the health sector.
In the case of CON regulation, it has been suggested that more stringent sanctions
and a broader scope of regulation (e.g., lower dollar thresholds) would be predictive
of stricter regulatory controls. This implies, for example, that CON programs whose
sanctions which may include license suspensions or injunction-related fines should deter
investment more than Section 1122 programs whose sanctions only entailed loss of
reimbursement for the capital costs under Medicare and Medicaid for the disapproved
investment. Lower thresholds were presumed more effective in deterring investment
both because of a broader scope of control and also because lower thresholds made it
more difficult to segment large projects into smaller components each of which was
below the threshold and therefore exempt from review.
The role of local versus state agencies in CON reviews was also cited as a possible
explanatory factor for variations in CON impact. It was suggested that first-level re-
view by local agencies would result in higher approval rates for investment projects for
"moral hazard" reasons, that is, because local citizens would support expansions of hos-
pital service in their own communities that were financed in part by insurance premiums
and taxes paid by citizens of other localities within and outside the state.
Several hypotheses were also advanced about differential application of CON con-
trols. Regulatory "capture" theory was cited to support the expectation that investment
constraints would be tightest for new market entrants (i.e., proposals to build new hos-
pitals) but much more lenient for proposed service expansions by existing entrants. The
dominant position of nonprofit providers in the hospital market was invoked to support
the expectation that entry or expansion controls would be especially strict for investor-
owned (i.e., "for-profit") providers. The Friedman and Peltzman thesis about informa-
tional asymmetries in regulation of new drugs was invoked to argue that certain types of
investment (e.g., investment in bed capacity) would be more strictly controlled because
the consequences of regulatory leniency (i.e., low hospital occupancy rates) would be
more easily visible for this type of investment versus investment in new equipment.
A number of possibilities for perverse incentive effects were also noted. The premise
that providers faced borrowing constraints or an increasing marginal cost of capital was
cited to support the expectation that if CON controls restricted one type of investment
by incumbent firms, it would free up capital funds and thereby encourage other types of
investment. Some observers also noted the possible "franchise" nature of CON controls
which advantaged providers who applied first (or "early") to establish new services
(e.g., the first open-heart surgery service in a metropolitan area) and disadvantaged later
applicants since the new service was no longer "needed." It was suggested that this
"franchise" aspect of CON could encourage hospitals to speed up plans to invest in new
services to pre-empt competitors from successful CON applications and that the net
result might in fact be an increase in the rate of investment under CON controls.
The malleability of demand by well-insured consumers in response to provider in-
fluence was also cited as a possible basis for perverse incentive effects. For example,
if inpatient occupancy rates (i.e., days of care per year divided by "bed-days" of care
1502 D.S. Salkever
available) were used as indicators of "need" by the CON regulators, hospitals seeking
CON approval to expand might encourage their staff physicians to discharge inpatients
more slowly and thereby boost the occupancy rate and increase the apparent "need" for
the proposed new investment.
In the case of rate-regulation programs, observers noted variations in the process of
rate setting (i.e., budget review vs. formula), in the scope of coverage (e.g., all-payer
vs. only private payers), and in the incentive structures of the programs as factors that
could explain variability in program impacts. Budget review programs, which involved
more possibilities for negotiation and interactions between regulators and hospitals,
were viewed as more subject to "capture" effects and thus presumably less stringent
[Sloan and Steinwald (1980)]. Programs that applied to all payers were expected to
produce stronger constraints on costs because the consequences for hospitals of any dif-
ferentials between regulated rates and actual costs would be larger when rates applied
to all patients.
Two aspects of incentive structures received particular attention. One related to the
profit opportunities provided by the rate-setting program. In most programs, net rev-
enues earned by reducing costs below the required rates would be returned in full to
the hospital; however in some programs (e.g., New Jersey), rates were set as the lesser
of approved charges or actual costs so the opportunity to earn positive profits did not
exist. The length of the regulatory lag was also seen as having important incentive ef-
fects. In programs with annual budget reviews, cost savings in the previous year might
be incorporated into an even tighter budget approved for the following year, so positive
profits generated from these cost savings might accrue only over one or two years. The
formula-based system in New York was similar in that actual costs in year t - 2 (and
not rates in year t - 2) were the basis for setting rates in year t. Some programs, as in
Maryland, were specifically designed to include strong profit incentives. In Maryland,
full budget reviews were done infrequently and hospitals were given a long regulatory
lag (approximately five years) over which they could accumulate profits earned by cut-
ting costs.
The second key aspect of incentive structures concerned the extent to which the rate-
setting program yielded profits for providers who increased volume. As noted above, in
state programs rates were typically set with reference to average costs and when short-
rmn costs are below average costs, the hospital can increase profits by increasing volume
unless a volume-adjustment provision is incorporated in the rate-setting program. In the
Federal ESP program, increases in allowable revenues due to volume growth were also
based essentially on average rather than marginal revenues and costs. Given assump-
tions about the malleability of demand to provider influence, it was expected that hos-
pitals would respond to the profit incentive and endeavor to increase the volumes of
services provided. Another aspect of the New York and Massachusetts rate-setting pro-
grams that presumably reinforced this incentive to increase volumes was the use of an
occupancy rate penalty, which reduced rates for hospitals that failed to provide enough
patient days to exceed a minimum occupancy rate target set by the regulators [Bauer
(1978)].
Several strategies were used in rate-setting programs to counteract these incentives.

One was the application of volume-adjustment formulas discussed above. A second was
to set rates on the basis of larger units of service, such as cases treated, rather than on
the basis of individual tests, procedures, and days of inpatient care. As we previously
noted, a trend toward these per case or "DRG" systems began in the late 1970s. The
expectation that this would counteract incentives to increase volumes was based on the
assumption that providers would have less ability to influence the volume of admissions
demanded by their well-insured patients than the volume of individual tests or days of
care once these patients were already admitted. Extending this reasoning even further,
there were even several instances of experimental or demonstration programs where
selected hospitals were placed on a fixed overall budget regardless of the number of
admissions, thereby eliminating any incentive to increase volume by admitting more
patients and giving the providers the maximum incentive to substitute outpatient for
inpatient care.
All of the various discussions about profit incentives in rate-setting programs were,
to be sure, based on the premise that even nonprofit hospitals would respond to oppor-
tunities to earn positive profits. In fact, this premise may have been less accurate in the
financially secure climate of the early 1970s than it was in the early 1980s and beyond
when Medicare substantially abandoned its original cost-based approach to hospital
payment.
3. The demand for regulation: empirical models
Empirical models of the demand for regulation are relevant to two important concerns.
First, they allow us to test the various theories of regulation described above as expla-
nations of the political decisions to adopt regulatory programs. Second, they inform us
about potential biases, in studies comparing regulated versus nonregulated providers,
which may arise from the (hopefully) nonrandom nature of the political decision-
making process. Unfortunately, the literature on the demand for rate regulation and
CON regulation is sparse so the number of studies reviewed here is small.
Cone and Dranove (1986) developed an empirical model of the probability that a state
has a hospital rate-setting law, and estimated this model with a cross-section of data on
all states except Alaska and Hawaii. They focused on testing two specific hypotheses.
One is the Medicaid hypothesis, that enactment of Medicaid resulted in more gener-
ous benefits (due to Federal requirements) than the "optimal" or political equilibrium
level for the state, thereby encouraging states to use rate setting to limit these benefits.
The second is the "total cost" hypothesis, that voters in the state reacted to exogenous
shocks that transferred wealth to hospitals by imposing rate-setting limits to constrain
these wealth transfers. The results provided strong support for the Medicaid hypothesis,
and also indicated that the general political climate (as measured by the average Ameri-
cans for Democratic Action (ADA) rating for congressmen from the state) was strongly
predictive of adoption of a rate-setting law.
1504 D.S. Salkever
A similar study by Fanara and Greenberg (1985) reinforced Cone and Dranove's
major result. The authors found that the percentage of a state's budget spent on Medicaid
was a significant predictor of rate-setting adoption. They also found that the change in
hospital costs was positively related to adoption while the for-profit share of the hospital
market was negatively related to adoption,
Lanning, Morrisey and Ohsfeldt (1991) estimated demand for regulation models as
the first-stage in a two-stage model of regulatory effects on hospital and nonhospital
health care spending. Their models were estimated with time-series cross-section data
for states for the years 1969, 1972 and 1976-1982. They too reported that state-funded
Medicaid expenditures were a strong positive predictor of rate-setting adoption, while
hospital-beds per capita was negatively related to adoption. Political choice variables
(the ADA rating for senators and having the governor and the majority of both legislative
houses under the same party) were significant adoption predictors (positive and negative
respectively). They also applied the same model to predict the presence of a CON law
and once again reported strong evidence of a Medicaid effect.
Wendling and Werner (1980) estimated a model of CON adoption for a cross-section
of states and the single time interval 1968-1973. Adoption of CON was significantly
related to (1) a decline in the occupancy rate of hospitals and (2) a higher level of
hospital industry concentration. Declining occupancy rates are viewed by the authors as
a proxy for the benefit to the incumbent hospitals of entry controls while concentration is
interpreted as an inverse measure of the cost of organizing a coalition to lobby for CON
controls. The two results taken together are seen as confirmation that hospital industry
interests drove the CON adoption process. A null result for a measure of increases in
hospital costs led the authors to reject the "public interest" view that CON is adopted to
reduce excessive spending on hospital services.
In summary, it appears that the desire to constrain Medicaid spending was a key fac-
tor in promoting state legislative adoption of regulatory controls, but that indicators of
general political preferences also had significant explanatory power. While the evidence
reviewed here is very limited, it gives relatively little support to the view that providers
or other narrow interest groups were the principal demanders of hospital regulation.
4. Effects of regulation
4.1. Dependentvariables
Since the most common arguments for regulatory controls, and the stated purposes of
the laws enacting these controls, focused on the need to reduce the costs of services, it
is not surprising that the great bulk of the studies assessing regulatory effects have used
measures of hospital service costs as dependent variables. The three most common cost
measures are cost per day of inpatient service, cost per inpatient admission, and cost
per capita. The latter measure is available only when the unit of analysis is a geographic
area with a defined population such as a state or Metropolitan Statistical Area (MSA).
The cost per day and cost per case measures can be applied with data on individual
providers, geographic areas, or even individual patients as the unit of analysis. Revenue
measures have occasionally been used instead of or in addition to cost or expenditure
measures. In the hospital sector, in particular, revenue and cost measures will yield very
similar results since the large majority of provider revenue is generated by nonprofit
organizations and their reported profit margins are low. The interpretation of revenue
measures is also clouded by the differences between revenues based on list prices and
actual receipts ("gross" versus "net" patient revenues in the parlance of hospital finance)
which may be substantial because of discounting practices and the provision of uncom-
pensated care.
Empirical studies have also examined regulatory effects on various measures of in-
vestment. This reflects both the fact that some regulatory programs (e.g., CON) are
specifically directed at controlling investment, and the expectation that providers will
respond to rate regulation in the long run by altering their investment behavior. One
particular concern is that rate regulation will tend to stifle innovation and the adoption
of new technology because it is extremely difficult to provide for technological change
in determining rates or allowable year-to-year increases in rates.9 Dependent variables
in studies that examine regulation effects on investment typically include numbers of
beds in providers' facilities, dollar value of plant and/or equipment, and measures of
diffusion and adoption of specific types of new equipment (e.g., CAT scanners).l°0
A third major category of dependent variable included in empirical research is mea-
sures of the quantity of services used. Typical examples are numbers of inpatient ad-
missions, number of days of care, and average days per admission (i.e., average length
of stay). 1 From the perspective of the standard theory of a firm with any market power,
one would expect regulation to impact the quantity of services demanded (and sold) if it
9 Some, however, might view reductions in the rate of technology adoption or diffusion as a positive out-
come of regulation. This could be consistent with Feldstein's (1971) argument about insurance-induced bias
in technological progress away from cost-saving innovations, or with "medical arms-race" arguments that
hospitals engage in inefficient non-price competition with one another by adding new equipment which is
expensive and "underutilized". A discussion of a broad range of possible regulatory impacts on technology
diffusion is presented in Warner (1978).
10 While the number of beds is a commonly used measure of capital or capacity, some ambiguities in its
interpretation should be noted. The measure of beds typically used is the number of beds "set up and staffed"
so they are in some sense ready for use. This measure averaged over, say, a year, is often referred to as
the number of "statistical" beds. This measure may be well below the licensed bed capacity of the provider
facility. Licensed bed capacity is an alternative measure of capital stock but it may be a misleading measure
of actual current capacity when (as often happens) major portions of a facility are renovated and diverted
from use for inpatient care to other uses. Alternative measures of capacity, such as facility square footage or
plant asset value, may also present problems in terms of data availability (in the former case) and accuracy of
valuation data (in the latter case).
11 Average length of stay could be viewed as measuring quantity of days of care per patient; alternatively,
one might view it as a product characteristic relevant for judging quality or value of the "product" (i.e., the
inpatient admission).
1506 D.S. Salkever
impacts the price charged, the characteristics of the service, and/or the price or charac-
teristics of substitute or complementary goods and services provided by other regulated
providers.
In the literature on hospital regulation, however, the concern with quantity effects
goes beyond the expectation of movements along the product demand curve facing the
provider, or shifts in this curve as the provider's product characteristics change or other
providers' prices and products change. Impacts of regulation on investment can result
in changes in "availability" that also shift the demand curve for the provider's services
through time-price effects. It has been further argued that impacts on investment in beds
or equipment may affect the extent to which the provider seeks to manipulate or "cre-
ate" demand for services. 2 Also, we have already noted the possibility that financial
incentives under rate-setting programs could encourage hospitals to "create" demand
and thus increase the volume of services provided.
Several studies have also looked at regulatory effects on hospital excess capacity or
"reserve margin". Since reserve margin is defined as a function of bed capacity and the
volume of days of care provided, one can view the reserve margin dependent variable
as essentially a combination of volume and investment variables (i.e., patient days, and
beds). Reserve margin is a potentially relevant measure of product quality in that a
higher reserve margin implies a higher probability of an open bed when a nonscheduled
admission is demanded. An alternative and less positive view is that the "turnaway"
probability is already so low in hospitals that increases in reserve margins primarily
represent higher excess capacity costs.
4.2. Modelingframeworks
Specifications used for estimating regulatory effects can usually be interpreted either
as "behavioral" reduced-form equilibrium relationships or as technological or "quasi-
technological" relationships. For example, analyses of data on individual hospitals with
cost per inpatient day as the dependent variable and exogenous regressors describing
hospital preferences, input supply conditions, and product demand conditions could be
viewed as estimates of a reduced form derived from maximization of the hospital's
objective function subject to technology, product demand, and input price or supply
constraints. This same general model could be used to motivate reduced-form equilib-
rium expressions for cost per inpatient case, average length of stay (or other product
"quality" measures), volume of patient days or cases, total costs, and capital and labor
input quantities. 13
12 An early formulation of this argument was Roemer's Law, sometimes paraphrased as "a bed built is a
bed filled", or in a more modern rendition, "build it and they will come". See Roemer (1961). This concern
was certainly one of the conceptual foundations for the development of proposals to regulate investment via
CON laws.
13 For an example of a formal description of this type of behavioral model, see Salkever and Bice (1979), Ap-
pendix C. A similar formulation with product price dependent on product quality but not quantity is presented
in Sloan (1981).
Alternatively, an estimate of a model with cost (either per day, per case, or to-
tal) as the dependent variable and regressors relating to the volume, quality and mix
of days of care provided, and to input prices could be viewed as a technological or
"quasi-technological" cost function.1 4 Of course, in either the "behavioral" or "quasi-
technological" relationships, regulatory impacts are typically estimated by including
additional regressors describing the presence and/or characteristics of regulatory pro-
grams. Coefficient estimates for these regulatory variables are most easily interpreted
as measuring average deviations between the expected dependent variable values in the
absence of regulation and the observed dependent variable values. 15
There are other studies in the literature where a clear interpretation of the model
as either behavioral, reduced-form or "quasi-technological" is not possible. In some
instances, I have omitted these studies from this review because of the difficulty in
interpreting the meaning of their results.
4.3. Regulatory impacts on hospital costs
Our review of research findings regarding hospital costs proceeds mainly in chrono-
logical order. We did not attempt to organize results by the specific types of regulation
programs under study (i.e., CON regulation, mandatory state rate setting, and the Fed-
eral ESP) since a number of studies report results for more than one type of program. We
begin by reviewing results from studies that include statistical comparisons of regula-
tory "treatments" and control (non-regulation) regimes. Then we briefly review several
studies that have examined variations within regulatory programs.
4.3.1. Empirical researchon cost impacts of regulation
Some of the earliest econometric assessments of rate setting were carried out under a
series of five contracts let by the Federal government in 1974. Three of these contracts
dealt with situations of mandatory rate-setting laws (in "upstate" New York, "down-
state" New York, and Rhode Island) while one dealt with a voluntary program (in New
Jersey) and one with a private sector Blue Cross program (in Indiana). In all cases, the
unit of analysis was the individual hospital and the researchers selected comparison
group hospitals from other states to include in their analyses.
14 This estimated cost function might be described as "quasi-technological" if it included additional re-
gressors intended to capture shifts in technology or differences in efficiency; a typical example would be
inclusion of a 0-1 dummy for nonprofit status to allow for efficiency differences between nonprofit and
for-profit providers. For further discussion of the distinction between behavioral and technological or quasi-
technological hospital cost models, see Salkever (1983) and Rosko and Broyles (1988, Chapter 8).
15 Note that this interpretation for behavioral models does not imply that regulated providers are in fact in
equilibrium (in the sense of being at an interior solution of their maximization problem). One could develop
more elaborate strategic models that make more explicit the behavior of the regulators and the providers who
are under regulation, but the literature to date has not followed this approach.
1508 D.S. Salkever
The three studies of mandatory rate-setting laws [Dowling et al. (1976), Abt Asso-
ciates and Policy Analysis (1976), Thornberry and Zimmerman, n.d.] all estimated rate
setting effects within the context of the "quasi-technological" cost functions. The unit
of analysis was the individual hospital. Dependent variables included average cost per
day and per case in all three studies, and total costs in the upstate New York study. The
estimated cost functions were short-run, since measures of capital inputs (e.g., beds,
dummies for the presence of various facilities and services) were used as explanatory
variables. All three studies estimated rate-setting impacts by including "control" hos-
pitals from other locations in their analysis.' 6 The two New York studies covered the
period 1968-1974 while the Rhode Island study used data for 1969-1972.
The three studies generated a large volume of results based on alternative model spec-
ifications. A detailed review and synthesis of these studies [Salkever (1979, Chapter 4)]
found that taken together they provided only weak evidence that mandatory rate-setting
had reduced unit costs. 17 Possible explanations for this null overall result are the brief
time period used in these studies, the brief post-regulation experience captured by their
data (two to five years), the fact that some of the "control" hospitals (in Massachusetts)
were subject to rate controls in the last year of the study period, and the possible effect
of the Federal ESP program on the behavior of costs in the control hospitals.
An econometric assessment of CON impacts on hospital costs in the early years of the
CON experience was presented in Salkever and Bice (1979). They analyzed state-level
data for the years 1968-1972, so their results only reflect CON results in the earli-
est adopter states: California (1969), Connecticut (1969), Maryland (1968), New York
(1964), and Rhode Island (1968). Cost estimates are based on a reduced-.form short-run
behavioral cost model, with equilibrium costs a function of exogenous demand and sup-
ply determinants as well as hospital capital stock (beds per capita and plant assets per
bed). Their estimates of CON impacts on the capital stock variables are discussed in
Section 4.5 below.
Their estimated CON cost impacts combined direct CON effects (holding capital
stock constant) and indirect effects (working through CON impacts on capital stocks).
A CON dummy variable is included to estimate direct effects in the short-run cost
model. Results for the direct effect coefficients varied with the inclusion or exclusion of
state fixed effects, being significantly negative for both total cost per capita and cost per
inpatient day with fixed effects included and positive (and significant for cost per day)
16 The downstate New York study used hospitals from the Chicago, Cleveland and Philadelphia metropolitan
areas as controls while the upstate New York study used hospitals from northern Ohio, southeastern Michigan
and the Milwaukee metropolitan area. The Rhode Island study used Massachusetts hospitals as controls. All
control hospitals were not subject to rate-setting except for the Massachusetts hospitals that came under rate-
setting in 1972.
17 Other reviewers reached similar conclusions. For example, Hellinger (1979) found evidence of "a moderate
lessening in the pace of hospital cost inflation in the neighborhood of two to four percentage points per
year" but observed that "firm conclusions regarding ... effectiveness ... should not be drawn from (these)
evaluations .".
with fixed effects excluded. Combined direct plus indirect effects also varied in both
sign and magnitude with the estimation method, thus pointing to the conclusion that no
substantial positive or negative effects on either total cost per capita or cost per patient
day could be attributed to CON regulation.
The first econometric assessment of the cost impact of the ESP program was car-
ried out by Ginsburg (1978). He analyzed a database of pooled quarterly data from
individual hospitals that were respondents in the American Hospital Association's Na-
tional Hospital Panel Survey. Individual hospital data were combined into nine census
divisions spanning 44 quarters over the time period 1963-1973. The estimated mod-
els were reduced-form behavioral cost equations that included exogenous demand and
supply determinants, hospital fixed capital as measured by bed supply, dummy vari-
ables for hospital ownership and teaching affiliations, and a dummy variable for the
ESP program period (from the third quarter of 1971 through 1973). Dependent vari-
ables pertaining to costs were cost per admission and cost per patient day. Models were
estimated with a transformation for first-order auto-regression and with and without a
lagged dependent variable. Estimation results showed some evidence of negative im-
pacts for the ESP program on cost per case but not on cost per day in his static model.
In the lagged adjustment model, no significant cost impacts were observed. Ginsburg
also estimated models with dependent variables that were closely related to unit costs,
such as revenue per day and per case and an overall index of input use per case and per
day. Results for the ESP dummy with these dependent variables paralleled the results
from the unit cost regressions. Finally, results with the average hospital employee wage
as the independent variable indicated that the ESP program had a significantly negative
impact in both static and dynamic specifications.
One of the first major national studies of unit cost impacts of regulation was con-
ducted by Sloan and Steinwald (1980). Their database consisted of observations on
more than 1,200 hospitals for the years 1970 through 1975. Dependent variables were
average cost per patient day and average cost per admission. Explanatory variables in-
cluded local area product-demand conditions, hospital wages and collective bargaining
status, other hospital characteristics (teaching status, ownership, and size class based on
number of beds). In specifying their CON regulation variables, Sloan and Steinwald dis-
tinguished between young programs (in the first or second year of operation) and more
mature programs, and between comprehensive programs (that include authority to reg-
ulate service expansions and that have new equipment review thresholds of $100,000 or
less) and noncomprehensive programs. Cross-classification by these two characteristics
yields four different CON dummy variables. An additional dummy variable for the year
preceding the introduction of CON is included to capture "anticipatory" effects. An
additional capital regulation variable is the proportion of the Medicare and Medicaid
population in the local area multiplied by a dummy for the presence of a Section 1122
review program. Rate-setting variables included (1) the proportion of the population in
the local area covered by insurance arrangements for which mandatory formula-based
prospective rate-setting (PR) is in effect, (2) the proportion of the population in the lo-
1510 D.S. Salkever
cal area covered by insurance arrangements for which mandatory budget-review-based

PR is in effect, and (3) the fraction of the year that the Federal ESP program was in
effect.
Unit cost regressions were estimated using a variance-components method and with
lagged dependent variables. Results for the regulation variables were decidedly mixed.
Young, noncomprehensive CON programs had significantly positive coefficients in both
the cost per day and cost per case regressions while formula PR programs had a positive
effect that was only significant for cost per case. Budget-review PR was strongly neg-
ative in the cost per day model and insignificantly negative in the cost per case model.
The only significant CON coefficients were all positive (for the pre-CON dummy in
the cost per day model, for both young and mature noncomprehensive CON programs
in both cost per day and cost per case regressions, and for young comprehensive CON
programs in the cost per case model). The only regulatory variable that was strongly
negative in both cost per day and cost per case regressions was the ESP program, but the
estimated coefficient magnitudes implied unit cost reductions of only about 1 per cent
on a long-term basis. Based on these results, the authors conclude, "PR has, at best, a
very small negative effect on costs and input use". 18 The authors also conjecture that
their more significant results for the ESP program relative to those reported by Ginsburg
(1978) (reviewed above) may have been due to the inclusion of later data, including the
post-ESP year of 1975, in their analysis.
Coelen and Sullivan (1981) conducted a large national study of cost impacts, focus-
ing primarily on rate regulation programs. They analyzed hospital-level data on cost
per day and cost per admission, and county-level data on hospital cost per capita. The
hospital-level data set consisted of all short-term general hospitals in 15 states with var-
ious kinds of rate-setting or prospective payment programs and a 25 per cent sample
of short-term general hospitals in all other states. 19 In all, 2,693 hospitals were in the
study. The county-level data set included the 1,317 counties containing at least one of
the hospitals included in the hospital-level data. The period covered by the data analy-
sis was 1969-1978.
Explanatory variables in the regression models were individual year dummies, in-
dividual state dummies, and a variety of county characteristics. Regulation variables
included a state-specific dummies for CON and state-specific dummies for the state
rate-setting programs. Regression models were estimated both with the three dependent
variables as annual levels, and with these variables expressed as year-to-year annual
percentage changes. Estimation was by stepwise ordinary least-squares.
18 In addition to studying regulatory effects on unit costs, Sloan and Steinwald also examined effects on
several capital input measures and on four other measures of non-capital input use (three measures of labor
inputs per bed and nonlabor expense per bed). We review their findings on regulation and capital input use
below (in Section 4.5.1).
19 The 15 states were Arizona, Colorado, Connecticut, Indiana, Kentucky, Maryland, Massachusetts,
Minnesota, Nebraska, New Jersey, New York, Pennsylvania (western part of the state only), Rhode Is-
land,Washington, and Wisconsin.
In spite of the large number of models and regulatory variables, some clear patterns
were identified in Coelen and Sullivan's results. First, the results for the CON variables
were extremely diverse, with significant coefficients of both signs but relatively few
significant coefficients in all. Second, significant coefficients for the mandatory rate-
setting programs were consistently negative. While a number of programs did not show
any significant effects, those that did show cost reductions tended to do so in the later
years of the period (1975 to 1978). Third, evidence for negative rate-setting impacts
on costs were strongest for cost per day, second strongest for cost per case, and much
weaker for cost per capita. This pattern of results is consistent with hospitals responding
to stringent rate controls by increasing their volume of services, and in particular their
length of patient stay.
In another major national study of unit cost impacts, Sloan (1981) examined state-
level panel data on the 48 contiguous states and the District of Columbia for the years
1963-1978. Dependent variables were again average hospital cost per admission and
average hospital cost per day, but were defined to include only private, nonprofit short-
term general hospitals. Explanatory variables included product demand, factor price,
and regulation variables, as well as lagged dependent variable values and individual
state dummies. Regulation variables included: the fraction of hospital costs covered by
mandatory rate-setting programs that were less than three years old ("young PR" pro-
grams), the fraction of hospital costs covered by mandatory rate-setting programs that
were at least three years old ("old PR" programs), an Economic Stabilization Program
(ESP) indicator equal to 1 for the years 1972 and 1973 and to 0.33 for 1974, a pre-CON
dummy for the years prior to and during initial CON implementation, a "young" CON
dummy for the first two years after implementation, and an "old" CON dummy for
subsequent years, a Section 1122 variable equal to the Medicare plus Medicaid share
of hospital costs for states and years in which a Section 1122 was in place (excluding
the year of implementation), and a 1978 dummy to capture the effect of the national
Voluntary Effort (VE) cost containment program.
Estimation results for the CON variables yield minimal evidence in some specifica-
tions of negative cost impacts but no evidence of differences between "young" and "old"
programs or of anticipatory effects. Sloan concludes that the "parameter estimates are
too unstable to permit a conclusion that CON programs control hospital costs". Surpris-
ingly, Sloan finds fairly large and significant negative coefficients for the Section 1122
variable; most observers have viewed the Section 1122 program as considerably less
constraining than CON regulation. Results for the rate-setting and price control vari-
ables do show significant cost-control effects. The ESP results suggest a long-run im-
pact of about -10 per cent while the results for "old" PR programs suggest long-run
impacts of -7 to -20 per cent. Results for "young" PR programs show no cost con-
tainment impacts. In comparing these findings to earlier studies that showed little if any
effect of rate setting, Sloan conjectures that his findings may reflect the inclusion of
data for the years 1976-1978 in his sample and the fact that a number of the manda-
tory PR programs switched from "young" to "old" (based on Sloan's two-year criterion)
1512 D.S. Salkever
during these years. Finally, Sloan's estimates also show very strong negative effects for
the VE program. 2 0
Sloan (1983) updated this analysis by adding data for the years 1979 and 1980.
He also expanded his model to include dummy variables for voluntary and manda-
tory/advisory rate setting programs. (The latter require review of hospitals rate but do
not require compliance with the review agency's recommendations.) Estimation results
for the ESP, CON and mandatory rate-setting variables were quite similar to Sloan's
earlier findings.
Further corroboration of some of these empirical results is provided by Morrisey,
Sloan and Mitchell (1983) in their analysis of metropolitan area data for 27 large
metropolitan areas for the years 1968-1981. They estimate a sparse reduced form model
with the following dependent variables: hospital cost per day, hospital cost per admis-
sion, and hospital cost per capita. Regulatory variables include dummies for CON and
for the Federal ESP program. Rate-setting is described by a single dummy variable for
"young" mandatory programs (in their first two years) and by separate dummy vari-
ables for the five different mandatory programs included in the data (Maryland, Mas-
sachusetts, New Jersey, New York, and Washington). The other explanatory variables
are population, per capita income, the area cost-of-living, fraction of the population
over 65, share of hospital beds in teaching hospitals, and a time trend.
The authors report considerable variation in effects by state, with New Jersey and
New York (post-1975) having the largest negative effects on cost per admission and
cost per capita. Results for the other states, and particularly Maryland and Washington,
show stronger effects on cost per day than on cost per case and weakest effects on
cost per capita. This pattern indicates that unit cost reductions due to rate setting are
being offset by volume responses in terms of longer patient stays or more admissions.
Estimates combining the experience of all five states show significant negative effects
on all three dependent variables of about the same magnitude, implying reductions in
the rate of cost increase on the order of 2 to 3 per cent per year. The authors argue,
however, that this overall result is misleading because it is driven almost entirely by the
New Jersey and New York results. Unfortunately, no results are reported for the CON
and ESP variables.
Dranove and Cone (1985) used a cross-sectional first-difference regression with data
from 1982 and 1970 to estimate rate-setting impacts on average hospital costs at the
state level. They included the residuals from a regression on 1970 data as an explana-
tory variable to control for regression-to-the-mean effects. Their regulatory measure was
the number of years of rate regulation over the 1970-1982 period minus two (to allow
for program maturation). Results showed significantly negative effects on cost per day,
20 Sloan also reports the results of one regression on "profits" (more precisely, revenue divided by cost).
Results differ substantially from the cost regressions. The ESP, CON and Section 1122 variables all have
significantly negative effects on profits, the VE coefficient is strongly positive, and the PR variables have no
effect.
cost per admission, and cost per capita. Only six states were treated as regulated (Con-
necticut, Massachusetts, New Jersey, New York, Maryland, and Washington). Results
for rate regulation were not sensitive to inclusion of the 1970 residual, suggesting that
states that adopted regulation did not systematically experience higher-than-expected
costs in 1970; this is further evidence that the demand for regulation was not strongly
related to the level of costs.
Hadley and Swartz (1989) studied the impact of regulation on hospital costs with
data for the years 1980-1984 for almost 1,300 hospitals located in 43 large metropoli-
tan areas across the U.S. They present an estimate of a long-run quasi-technological cost
function which includes output measures for the numbers of inpatient admissions and
numbers of post-admission inpatient days (broken down into Medicare, Medicaid and
"other" payer groups), the casemix index for Medicare patients, the numbers of surgical
and nonsurgical outpatient visits, and the number of residents (as a measure of teaching
activity). Other explanatory variables in the cost function are a vector of input price
variables and exogenous demand and market structure variables (pertaining to insur-
ance coverage, ownership mix of hospitals in the market, and the physician/population
ratio), separate intercepts for each year, and several hospital characteristics (ownership
type, central city vs. suburban location). The variables of most direct interest as mea-
sures of regulation were: a dummy for state Medicaid-only rate regulation, a dummy for
other state rate regulation programs that do not cover all payers, year-specific dummies
for Maryland and New Jersey combined for the five years of their all-payer regulation
regimes (1980-1984), and year specific dummies for Massachusetts and New York for
the two years of their all-payer regulation regimes (1983 and 1984). No variables for
capital expenditure regulation were included in the model.
Coefficients were estimated by OLS without fixed or random effects. Results showed
significantly negative coefficients for the Massachusetts/New York dummies for the
years 1983 and 1984, for the Maryland/New Jersey dummy for 1984, and for the other
state regulatory dummy. Dummies for state Medicaid-only programs and for the years
1980-1982 for Maryland/New Jersey were positive, while the 1983 Maryland/New Jer-
sey dummy was negative and marginally significant. The authors conclude that state
rate-setting has in fact reduced costs of producing hospital services. They observe that
within the context of their cost function model, one cannot determine whether these
cost reductions represent increases in efficiency or reductions in service quality or both.
They also note that the magnitudes of the significantly negative coefficients for their
"all-payer" dummies are not substantially different than for the "other state regulatory"
dummy. Thus, they conclude that while state Medicaid-only regulation does not impact
on costs, other state rate regulation programs are as effective as all-payer rate regulation.
This latter finding is contrary to expectations and the authors suggest that it may be due
to the more stringent Medicare payment controls enacted in 1982 (under the 1982 Tax
Equity and Fiscal Responsibility Act) and the Medicare Prospective Payment System
(implemented in 1983).
Eakin (1991) analyzes the relationship between empirical values for an inefficiency
index for individual hospitals and indicators of regulation, using cross-section data on
1514 D.S. Salkever
331 hospitals for the period 1975-1976. The inefficiency indexes are developed by esti-
mating a non-minimum cost function, using the method described in Eakin and Kneis-
ner (1988) and treating numbers of inpatient cases and outpatient visits as the relevant
measures of output. Broad measures of case-mix categories are also included the cost
function, along with input price variables, a dummy variable for medical school affili-
ation, the average length of inpatient stays (in days), and the hospital's occupancy rate.
Since the latter variable is a function of the number of beds or capacity in the hospital,
it seems appropriate to regard the estimated cost functions as short-run in nature.
Eakin regresses the inefficiency index values on variables indicating ownership sta-
tus (for-profit vs. nonprofit and religious vs. secular), beds in the hospital, the hospi-
tal's share of total admissions in the county, HMO market share in the state, the mix
of patients in the hospital by five different payer categories and regional dummies. In
addition, regulatory dummies are included for a CON program in effect by 1974 in
the state, a state Section 1122 agreement, and mandatory state rate regulation in 1976.
His results indicate a significant reduction of approximately 1 per cent in allocative
inefficiency in response to rate regulation, an increase in allocative efficiency of ap-
proximately 1 per cent due to CON controls, and no significant effect of Section 1122
programs.
Second-stage results from Lanning, Morrisey and Ohsfeldt (1991) provide additional
evidence on rate-setting and CON effects on hospital and nonhospital health care spend-
ing per capita during the 1969-1982 period. Their regulatory variables are constructed
to include only "mature" programs that have been in effect for more than two years for
rate-setting and for more than three years for CON. They report strongly negative rate-
setting coefficients in both the per capita hospital and per capita nonhospital regressions;
corresponding CON coefficients are strongly positive. These findings are contrasted to
OLS results, which show significantly negative but smaller rate-setting effects and es-
sentially no CON effects. They conclude that "mature" rate-setting programs do indeed
restrain growth in health care costs (both hospital and nonhospital), that CON programs
tend to increase costs by protecting incumbent firms from competition, and that single-
equation estimates of regulatory effects are misleading because of the endogeneity of
regulation. They also report strongly positive cost impacts for the ESP program, a very
different finding from previous research.
Friedman and Coffey (1993) estimate a very simple regression model with state data
on the annual logarithmic change in hospital cost per case for the years 1981-1990.
Explanatory variables are limited to the contemporaneous logarithmic change in admis-
sions per capita, the year-to-year change in the number of years of experience with rate
setting in the state, a Medicare PPS dummy for the years after 1984 for states without
rate-regulation, the fraction of the state population in HMOs, and the residual from a
21
1980 cross-sectional regression on hospital cost per case. In one of their regressions
21 The states and years of rate-setting used by Friedman and Coffey to construct their rate-setting variables
were: New York (1970-1990), Rhode Island (1971-1990), Massachusetts (1971-1988), Wisconsin (1972-
1984), Connecticut (1973-1990), Maryland (1974-1990). New Jersey (1975-1990) and Washington (1975-
1988). They show the New Jersey program as terminating in 1992, after the end of their study period.
for the full 10-year period, they also include dummy variables for Maryland and New
York. Significantly negative coefficients are obtained for the years of rate-setting vari-
able and for the Maryland dummy in one model that excludes the HMO market share.
When the HMO share is added and the Maryland dummy is deleted, the years of rate-
setting coefficient is still significant and indicates a 0.7 per cent reduction the annual
growth rate of cost per case. The HMO market share is also strongly negative, as is the
PPS dummy. When the data are split into two periods, 1980-1984 and 1985-1990, ma-
jor differences between the periods are observed. Rate-setting is significantly negative
in the first period but has no effect in the second period. The opposite is true for the
HMO market share variable.
Antel, Ohsfeldt and Becker (1995) report results with pooled state data for the years
1968-1990. They measure rate regulation by the percentage of hospital revenues subject
to mandatory state regulation, and they also include dummy variables for CON, for the
years 1971-1974 (the years of the Nixon price controls), and for a state Section 1122
investment review program. In addition, they include interaction terms for these various
regulatory programs. They find that the coefficients for the rate-setting and CON vari-
ables in cost per day, cost per admission, and cost per capita regressions are uniformly
positive and occasionally significant or marginally significant. A number of the interac-
tion terms are significantly negative, but their magnitudes do not appear large enough
to clearly offset the positive coefficients on the main coefficients for rate-setting and
for CON. The authors conclude that their results imply "limited effectiveness of hospi-
tal cost regulation" though they also report strongly negative coefficients for the Nixon
price-control years.
4.3.2. Analyses of intra-programvariations
Several studies have examined hospital responses to rate regulation by analyzing cost
variations, within regulated states, in relation to the rules under which different hos-
pitals were functioning. Salkever, Steinwachs and Rupp (1986) compared hospitals in
Maryland that were under three different regimes: (1) standard regulation of rates for
specific services, (2) a prospective per case payment system that included profit incen-
tives, and (3) a stringent per case payment cap that exposed hospitals to the risk of loss
but included no profit incentives. Their data base was a panel of all short-term acute
care general hospitals in Maryland for the five fiscal years ending between July 1, 1976,
and June 30, 1981. Using both total inpatient cost and inpatient cost per case as de-
pendent variables, and estimating both behavioral and quasi-technological cost models,
the authors concluded that efficiency and cost behavior in regimes 1 and 2 were similar
while hospitals under regime 3 responded to the more stringent payment levels with
lower costs. This finding was interpreted as evidence that profit incentives, particularly
for nonprofit hospitals in a regulated state, are unlikely to induce cost-saving changes in
utilization patterns (such as shorter average length of stay and reduced numbers of tests
and procedures per case) but that such hospitals will respond with such changes when
they face the risk of substantial losses.
1516 D.S. Salkever
Thorpe and Phelps (1990) examined the responses of New York hospitals to differ-
ing levels of stringency in regulated rates and to a change in the regulatory "lag" in the
New York State program after 1982. The New York Prospective Hospital Reimburse-
ment Methodology (NYPHRM) program adopted in 1983 incorporated several impor-
tant changes relative to the previous form of rate setting in New York. First, a base rate
for each hospital computed from 1981 cost experience was adopted and rates in subse-
quent years were set by trending the 1981 base forward to recognize inflation. In the pre-
vious system, new base rates were calculated for each hospital every two years derived
from the hospital's own cost experience. Thus, the NYPHRM system involved a switch
from a system with a short regulatory lag to one with a long regulatory lag. Presum-
ably this shift strengthened hospitals' incentive to control costs. Second, when the new
NYPHRM base rates were computed, some hospitals had their rates reduced because of
disallowances of costs in excess of those experienced by their peer group. Thorpe and
Phelps compared the subsequent experience under NYPHRM of these hospitals with
disallowances versus those whose 1981 base rates did not include any disallowances.
In their empirical analysis, Thorpe and Phelps separated their study hospitals into two
groups (with and without disallowances) and examined the increases in cost for each
group over two periods 1980-1982 (pre-NYPHRM) and 1982-1985 (post-NYPHRM).
For each group and period, they regressed changes in cost per case on changes in vari-
ables describing the payer mix of the hospital's patients, casemix, county income and
population, and residents per bed in the hospital. Based on these regressions, the au-
thors computed predicted cost growth rate (above the allowable inflation trend factor)
for the two periods and found that the rate dropped by 3 per cent from pre- to post-
NYPHRM for all hospitals. This is consistent with the expected result of an increased
regulatory lag. The authors also found that the predicted rate of cost growth for hos-
pitals under stringent rate controls changed from about 1.5 per cent lower in the pre-
NYPHRM period to about 3.5 per cent lower in the post-NYPHRM period. Thus, the
hospitals subject to disallowances in fact responded with reduced rates of cost increase.
These hospitals also responded more strongly to the increased regulatory lag; Thorpe
and Phelps estimate a 4 per cent per year decline in cost growth for these hospitals while
the corresponding estimate for hospitals without disallowances was only 2 per cent.
4.4. Regulatory impacts on the quantity of hospital services used (utilization)
Regulatory programs can influence the quantity of hospital services used by affecting
the price of services, the availability (time-price) of services, and other characteris-
tics (e.g., quality) of services that influence the quantity of services demanded by con-
sumers. Regulatory programs could also directly influence quantity used by altering the
extent to which providers seek to "manipulate" consumer demand.
Based on his detailed review of the provisions of the Federal ESP program, Ginsburg
(1978) hypothesized that hospitals would respond to this program by increasing both
admissions and average days of stay per admission. He used the data and modeling
approach we reviewed above (Section 4.3) in connection with this unit cost analysis to
test this hypothesis. Contrary to expectations, he found no significant positive effect of

the ESP program on admissions and a significantly negative effect on length of stay.
The Salkever and Bice (1979) analysis of CON impacts on costs reviewed above also
estimated CON impacts on hospital inpatient days per capita using the same 1968-1972
state-level data base. The same short-run behavioral model of hospital decision-making
was used to specify a reduced-form regression equation in which inpatient days per
capita depended on capital stock (beds per capita and plant assets per bed), exogenous
demand and supply determinants, and a CON dummy variable. As in the case of costs,
direct impact estimates (holding capital stock constant) were measured by the CON
dummy coefficients while total impact included both this direct impact and indirect ef-
fects working through the effects of CON on capital stock and the effects of capital stock
on inpatient days used. Estimated direct effects were small, negative, and insignificant.
Total effects were negative and much larger (in the -2.5 to -9.0 per cent range), due
primarily to a negative CON impact on beds per capita and a strong estimated positive
relationship between beds per capita and inpatient days per capita.
Worthington and Piro (1982) used the large national hospital-level and county-level
data bases from the Coelen and Sullivan (1981) study to analyze rate-setting program
effects on the volume of services provided. Dependent variables were admissions per
bed, average length of stay, occupancy rate, and admissions per capita. Regressions
were estimated in level form and used the same explanatory variables as in Coelen and
Sullivan (1981). Individual state dummies were included in all regressions. Rate-setting
variables included one or more individual state dummies for the 15 different mandatory
and voluntary rate-setting programs under study. In states where significant program
changes occurred during the study period, or different sub-programs operated within
the same state, different dummies were used for each rate-setting regime. CON was
measured by two variables: an index of program "activism" and an index of program
"limitations".
In spite of the large number of rate-setting variables in the analysis, a fairly clear pat-
tern of results emerged. Indicators for several of the mandatory rate-setting programs
(in New York, Maryland, and Rhode Island) showed significantly positive effects on
hospital occupancy rates while none of the rate-setting program indicators showed sig-
nificant negative effects on occupancy rate. The programs with positive occupancy rate
effects also tended to show positive effects on length of stay. No significant effects were
observed for the CON variables.
In his analysis of costs with aggregate state data for nonprofit hospitals for the years
1963-1980, Sloan (1983) also estimates models for average length of patient stay. In a
lagged adjustment model, he finds that prospective payment programs have no effect on
length of stay while capital regulation variables (Section 122, young CON and old CON)
tend to have negative effects. Since the number of beds is not included in the model, it is
possible that the negative effect for capital controls could be working through negative
impacts on numbers of beds; this would be more plausible than any direct negative effect
(holding capital constant) of CON. He also reports a significantly negative coefficient
for his pre-CON variable. He then re-estimates the model in annual change form and
1518 D.S. Salkever
obtains similar results but the negative capital control (CON and Section 1122) impacts
are on the whole smaller and less significant.
In a companion analysis to their cost-function study described above, Hadley and
Swartz (1987, 1989) also estimated rate regulation impacts on volume. Their estimated
output equation included measures of local product demand conditions, input prices,
hospital size (number of beds), and other hospital characteristics (case-mix index, own-
ership, teaching status, and Census region). Assuming beds represent fixed capital, the
estimated model can be viewed as a short-run equilibrium output equation for the hos-
pital.
Separate volume regressions were estimated for three different payer categories
(Medicare, Medicaid, and other) of admissions and post-admission days, for surgical
outpatient visits, and for other outpatient visits. Regulatory variables were identical to
those used in their cost function (discussed in Section 4.3.1 above). Results for the state
Medicaid-only and state other rate regulation dummies did tend to suggest positive
volume responses in Medicare admissions and post-admission days and in Medicaid
post-admission days but negative volume effects on other admissions and other post-
admission days. Results for the year-specific dummies for the two state pairs (Mary-
land/New Jersey and Massachusetts/New York) were widely varied. On the whole, the
results do not support a conclusion of strong volume responses.
Salkever and Steinwachs (1988) exploited the variation among the three regulatory
regimes in the Maryland rate-setting program described above (Section 4.3) to test for
differential utilization responses to per case versus per service prospective payment.
Dependent variables were average length of stay and total admissions and the unit of
analysis was the hospital. The data base and the short-run behavioral models developed
for their length-of-stay and total admissions regressions paralleled the methods used in
their earlier cost study [Salkever, Steinwachs, and Rupp (1986)]. Their results indicated
that in comparison to per service payment, per case payment only had the expected
negative effect on length of patient stay for those hospitals where the per case payment
level was set well below average cost; for other hospitals paid on a per case basis a
negative length of stay effect was not observed. Their admissions results showed only
weak evidence that admissions increase when hospitals are paid on a per case basis.
This is consistent, of course, with the arguments advanced for per case payment that
it is more difficult for hospitals to manipulate demand for admissions in response to
financial incentives under per case payment than to encourage longer patient stays or
more tests in response to financial incentives under per diem or per service payment.2 2
In their comparison of hospitals with and without disallowances and the implemen-
tation of the New York NYPHRM system, Thorpe and Phelps (1990) also examined
changes in average length of inpatient stay. It is worth noting that the NYPHRM system
22 The authors also report some evidence that the stringent per case payment limit tended to reduce the
costliness of the case-mix in hospitals subject to these limits. This is consistent with the concern expressed by
some observers that strong financial incentives under per case payment systems could encourage hospitals to
select less costly cases.
also introduced a volume adjustment formula to account for the difference between av-
erage and marginal costs; in itself, this change should reduce the incentive to increase
net revenues by increasing length of stay. In fact, Thorpe and Phelps found that while
length of stay increased slightly in the 1980-1982 period, it actually decreased during
the post-NYPRHM period (1982-1985).
4.5. Regulation impacts on hospital investment
A number of studies of regulatory impacts on hospital investment have used dependent

variables based on summary measures of capital stock. The dollar value of total plant
assets has been used as an overall capital measure while the number of inpatient beds
has been used to describe the treatment volume capacity of the capital stock. The ra-
tio of these two measures, the dollar value of plant assets per bed, has been used as a
crude indicator of the sophistication or complexity of capital equipment used by hospi-
tals. Other studies of regulatory impact have looked at measures of the availability of
specific kinds of capital or equipment, such as the presence of cardiac catheterization
laboratories or the number of intensive care unit beds. These are reviewed later in this
section.
4.5.1. Studies based on summary capitalstock measures
One of the earliest studies of regulatory impacts on capital stock or investment was
the analysis by Salkever and Bice (1976, 1979) of the early CON experience using a
single cross-section of state-level data measuring changes in capital stock from 1968
to 1972, where total dollars of plant assets, numbers of beds, and dollars of plant as-
sets per bed were used as the measures of capital stock. The regression model included
measures for changes in product demand, the level of product demand relative to capac-
ity in 1968, and availability of funds. The CON variable measured the fraction of the
four-year period during which CON was in effect; this variable was only non-zero for
California, Connecticut, Maryland, New York and Rhode Island. Econometric results
showed insignificantly positive CON effects on change in total plant assets, strongly
negative effects on changes in beds, and significant positive effects on changes in plant
assets per bed. The authors interpreted this as evidence that CON controls were being
applied selectively, with proposals to expand bed supplies or to build new hospitals sub-
ject to higher disapproval rates and proposals for new equipment subject to much lower
disapproval rates.
Hellinger (1976) used a cross-section of aggregate state data for 1973 to estimate the
impact of CON on total plant assets in short-term hospitals. In addition to a dummy
indicator for CON, explanatory variables included a measure of hospital wages, man-
ufacturing wages (a proxy for state wealth), population density, and patient days. The
conceptual model could be viewed as an input demand function. In some regressions,
the prior year's plant assets figure is included to allow for lagged adjustment and in
some cases patient days is excluded from the model because of endogeneity concerns.
1520 D.S. Salkever
The CON dummy is equal to 1 for the states that had enacted a law by January 1, 1973.
Estimation results showed no significant effect of CON on total plant assets. Hellinger
also estimates lagged adjustment models for 1971 and for 1972 (with the hospital wage
variable excluded) and includes dummies for enactment of CON in the same year. He
finds positive CON effects and attributes this to anticipatory behavior on the part of
hospitals.
In their analysis of regulation impacts on hospital cost and input use, Sloan and Stein-
wald (1980) modeled impacts on two summary measures of capital input use: assets per
bed and total beds in the hospital. (Data and variable definitions for this study were
described in Section 4.3 above.) In their lagged adjustment model for assets per bed,
the only significant regulatory impact reported was a negative effect of the ESP pro-
gram. The authors also note that coefficient estimates for all five of their CON variables
are also negative in the assets per bed model but none is significant. In the total beds
regression, several of the CON variables are significantly positive (pre-CON, young
comprehensive CON) while no other regulatory variables are significant.
In his analysis of unit cost impacts based on state data for the 1963-1978 period de-
scribed above (Section 4.3), Sloan (1981) also estimates one regression model of the
annual percentage change in hospital beds for the years 1969-1978. The only explana-
tory variables are the regulatory variables described earlier and individual year dum-
mies. Neither the PR nor the CON variables have any perceptible impact on the rate of
growth in beds while the coefficient of the Section 1122 variable is, surprisingly, signif-
icantly positive. Given the high variability in the dependent variable and the poor fit of
the model (R 2 = 0.05), it is unclear how much weight can be given to these findings.
Cromwell (1987) reports the results of a study of rate-setting impacts on hospital in-
vestment using the data base from the national rate-setting study described above [Coe-
len and Sullivan (1981)] and studying the years 1970-1979.23 Hospital-level regressions
were estimated with the following dependent variables: annual change in net total fixed
assets, annual change in gross building and fixed equipment, annual change in gross ma-
jor movable equipment, and annual change in bed stock. A county-level regression is
also estimated with the dependent variable short-term beds per capita. Explanatory vari-
ables included county-level demand determinants, hospital characteristics, and regional
and year dummies. OLS estimation was employed. State-specific rate-setting dummies
were included, along with a single dummy for the existence of a CON program.
In general, the state-specific rate-setting coefficients were not significant. Among the
states with programs presumed to be most restrictive, only Massachusetts showed sig-
nificantly negative effects for two of the dependent variables (annual change in total
fixed assets and county beds per capita). Others showed at most one or no significant
negative coefficients. Since none of the rate-setting coefficients were significantly pos-
itive, Cromwell concluded that "few programs (had) clear retarding effects", though he
23 Cromwell notes that data for Indiana were excluded because the private (Blue Cross) rate-setting program
there was in place prior to the beginning of his study period.
did regard the negative effects on bed supply in New York and Massachusetts as par-
ticularly significant in view of a low level of capacity utilization elsewhere in the U.S.
He also reports that CON was estimated to have a significantly negative effect on bed
supply, but no significant effect on the other measures of investment.
4.5.2. Regulation impacts on diffusion/availabilityof specialized services
The landmark study of technology diffusion in hospitals was the analysis by Russell
(1979) of data for the years 1961-1975 on more than 2,700 hospitals in metropolitan
areas. The seven services studied in her analysis were intensive care units, respiratory
therapy departments, diagnostic radioisotope facilities, electroencephalography, cobalt
(radiation) therapy, open-heart surgery, and renal dialysis. Her estimated models could
be viewed as reduced-form input-demand equations with explanatory variables describ-
ing hospital characteristics (dummies for 3 different bed-size categories, type of own-
ership, special versus general service categories, and measures of teaching activity),
market demand characteristics for the metropolitan area, and hospital market structure
(four-firm concentration ratio) in the metropolitan area. The regulatory variables used
in her analysis were a dummy for the three states in which CON become effective be-
fore 1970 (Maryland, New York and Rhode Island) and a dummy for the 20 states in
which CON laws became effective during the years 1970 through 1973.
One set of regressions used the year adopted as the dependent variable and excluded
hospitals that had not adopted by 1975 from the analysis. These regressions pertained to
intensive care, respiratory therapy departments, diagnostic radioisotope facilities, and
electroencephalography. These regressions showed no significant effects of CON. An
additional regression was estimated, using data for all study hospitals, for the percent-
age of all beds in the hospital that were in the intensive care unit. In this case, a strong
negative effect was found for the early CON laws but no effects for the laws imple-
mented after 1969. Finally, three regressions were estimated with data for all study
hospitals to examine the probabilities that a hospital in 1975 offered cobalt (radiation)
therapy, open-heart surgery, and renal dialysis. In these regressions, early CON pro-
grams had a significantly negative effect on the availability of open-heart surgery, CON
programs adopted in 1970-1973 had a significantly negative effect on availability of
radiation therapy, and neither CON variable affected the availability of renal dialysis.
Since the four year-of-adoption regressions, with no significant CON effects, were lim-
ited only to hospitals that had adopted the service in question, one might view Russell's
overall results as offering some support for a constraining effect of CON on investment
in specialized services. 24
Cromwell and Kanak (1982) report on an analysis of hospital service adoption based
on the hospital-level data set described above in our summary of Coelen and Sullivan
24 Russell does note, however, that because her data are limited to metropolitan areas, the results for the early
CON adopter states are dominated by the experience in New York.
1522 D.S. Salkever
(1981). A set of 24 different services reported consistently in the data base over the
1969-1978 period were selected for analysis and were grouped into three broad cate-
gories: quality-enhancing, complexity-expanding and community. Regression equations
were estimated based on the total number of services in the hospital in each year and
the total within each of these three categories. The dependent variables were defined
as the annual percentage change in the number of services offered. Explanatory vari-
ables included the number of services offered lagged one year, hospital characteristics,
county-level hospital market structure variables, exogenous county-level product de-
mand determinants, and state and year dummy variables. Rate-setting variables were
again state-specific while an overall CON variable was included. The authors also esti-
mated linear-probability adoption models, using the same explanatory variables (except
for the lagged dependent variable) for 13 of the 24 specific services under study.
Evidence that mandatory state rate setting slowed the adoption rate for specific ser-
vices was inconsistent. In the regressions using the percentage change in numbers of
services offered as the dependent variable, only the New York program showed con-
sistent and significantly negative effects. All other programs had no significant effects
(except for Washington in the year 1978, which showed both positive and negative ef-
fects). In the linear-probability regressions on specific services, at least three mandatory
state programs showed significant negative effects for only three services: intensive care
units, EEG, and social work departments. Looking at the results across states, Maryland,
Massachusetts, New Jersey and New York showed the strongest tendencies for negative
impacts. Finally, while they do not present detailed results, the authors report that CON
had "no overall impact on diffusion ...
Romeo, Wagner and Lee (1984) analyzed a 1980 survey of hospitals to study dif-
fusion of three "cost increasing" technologies (fetal monitoring, volumetric infusion
pumps, and endoscopies) and two "cost reducing" technologies (automated bacterial
susceptibility testing and centralized energy management systems). For each technol-
ogy, they estimated models of adoption probability and year of adoption; they also mod-
eled the number of units of equipment adopted for the three "cost increasing" technolo-
gies. Their analysis was restricted to hospitals from two mandatory rate-setting states
(New York and Maryland), one state with a Blue Cross prospective payment program
(Indiana) and three "control" states (Pennsylvania, Missouri and Ohio). Explanatory
variables included measures of hospital output volume, size and ownership as well as
local market area characteristics and competitiveness. Separate dummy variables were
included in each regression for New York hospitals, for Maryland hospitals and for In-
diana hospitals. No significant effects were found for the Maryland program, while the
New York results suggested negative effects on numbers of equipment units for the "cost
increasing" technologies.
Robinson et al. (1987) used data from the 1983 Survey of Specialized Clinical Ser-
vices, conducted by the American Hospital Association, to study the influence of com-
petition and regulation on the probability that a hospital would offer specialized cardiac
services. The services studied were bypass surgery, any open-heart surgery, a cardiac
catheterization laboratory, and angioplasty. The regression model controlled for mea-
sures of teaching activity in the hospital, market-area population and numbers of cardiac
care physicians, hospital size (measured in beds), hospital ownership, numbers of births
in the hospital, and several rough measures of case complexity and case-mix. Market
areas were defined by a clustering process based on a minimum distance of 15 miles
between market areas. Competition was measured by the number of hospitals within
each market area. Regulatory variables included the years that CON programs were in
effect and state dummy variables for the six states judged to have the most stringent and
comprehensive rate-setting programs in 1983 (Connecticut, Maryland, Massachusetts,
New Jersey, New York, and Washington). Results for all hospitals indicated signifi-
cantly negative coefficients for all the New Jersey and New York dummies, for most of
the Connecticut and Massachusetts dummies, and for none of the Maryland and Wash-
ington dummies. Confining the analysis to hospitals with 21 or more competitors in
their market areas resulted in larger and significant negative state dummy coefficients
except for two Maryland and all Washington dummy variables. The authors viewed the
results as indicating that stringent rate-setting controls can slow the diffusion of spe-
cialized services, particularly in the most competitive hospital markets. No results were
reported for the CON variable.
4.6. Regulation impacts on reserve margins or excess capacity
The first study to estimate the impact of various types of regulation on hospital reserve
margins was Joskow (1980), who defined reserve margin as the average number of
statistical beds (i.e., beds set up and staffed) minus the average daily inpatient census for
the year. The study viewed the reserve margin, in an unregulated world, as the result of
a hospital's decision about the optimal number of beds. The model assumes that (1) the
hospital must supply an exogenously determined volume of days of care for which it is
reimbursed full costs, and (2) its objective is simply to maximize the number of patients
treated. 25
Joskow estimates the model with cross-sectional data on a random sample of 346
nonprofit hospitals for the year 1976 drawn from 46 states and the District of Columbia.
In addition to patient days, other exogenous predictors in the reserve margin model
are measures of competition (doctors per hospital, HMO penetration rate, and the
metropolitan-area Herfindahl index for hospitals based on numbers of beds). Regula-
tory measures include a single rate regulation/prospective payment dummy (for the nine
states he identified as having state rate-setting commissions in 1975 and the 15 states
in which Blue Cross plans had implemented some form of prospective payment), and
three alternative CON variables: (1) a dummy variable for each of the states that had
CON in effect by the beginning of 1976, (2) the number of years the CON program was
25 The assumption of an exogenously determined volume of days presents a challenge in interpretation, since
offering a higher reserve margin is viewed by the author as a form of nonprice competition intended to attract
more patients.
1524 D.S. Salkever
in effect, and (3) the square of that number of years. Regression results indicate negative
and often significant coefficients for the prospective payment dummy variable. Among
the three alternative CON measures the simple dummy is never significant but either the
number of years or its square is significantly negative.
A similar econometric specification is used in a hospital bed-supply equation esti-
mated by Mayo and McFarland (1989) with cross-section time-series data for 120 hos-
pitals in Tennessee over the period 1980-1984. In this model, the number of licensed
beds is the dependent variable, and the CON variable is a continuous measure of strin-
gency, defined as function of the cumulative rejection rate for all CON applications up
to the current year for the health services planning area (HSA) in which the hospital is
located. (Since all data are for Tennessee and its CON law was implemented in 1975,
there is no "control" data from hospitals not subject to CON in the study.) Other exoge-
nous explanatory variables are the hospital's average daily census, physician supply per
capita in the HSA, and the Herfindahl index (based on admissions) for the HAS. The
estimates confirm Joskow's earlier result of a strongly negative impact of CON controls
on bed supply.
Most recently, Graham and Cowing (1997) have estimated a reserve margin model
using a 1987 national cross-section of data on more than 3,000 hospitals. Their depen-
dent variable is the reserve margin defined as in Joskow (1980) divided by the average
number of statistical beds. Regulatory variables include a CON dummy, the number of
years CON has been in effect in a state, and a state hospital rate regulation dummy. Only
the measure of years of CON was significant, with a negative coefficient. A large num-
ber of other explanatory variables relating to hospital prices, licensed beds, and other
hospital and market characteristics are also included.
In summary, the main message from these three studies is that CON, and particularly
"mature" CON programs, do seem to constrain the number of hospital beds for given
levels of inpatient volume. This implies a negative CON effect on excess capacity or
reserve margins. The underlying models used in these studies do not fit neatly into
either a reduced-form behavioral or a structural model interpretation and thus the precise
interpretation of the measured CON effects in these studies is unclear. 26 Nevertheless,
they do provide useful corroboration of the evidence about CON impacts on investment
described above.
4.7. Overview of results
(A) Mandatory state rate-setting programs. One of the most interesting aspects of
the evidence on cost impacts of rate-setting is the variation in results over time. Early
studies based on data through 1974 or 1975 typically reported very little in the way of
cost reductions, while studies that incorporated data for the late 1970s and early 1980s
26 The questions about interpretation arise because these studies include explanatory variables, such as inpa-
tient volume, that could also be regarded as endogenous choice variables for the hospital.
reported more significant and substantial negative impacts on costs. At least two com-
peting hypotheses could explain this pattern. One is the learning curve view that young
programs will be ineffective because the program staff have not yet learned their craft
or because they have not had time to plug the loopholes that appeared in their new
programs. The other is the hypothesis that the control group actually changed more
over time than the regulatory programs. For example, in the early 1970s, the Federal
ESP may have held down costs in non-regulated hospitals thereby compressing the reg-
ulation vs. control differential. Similarly, the critical changes in the Medicare program
beginning in the early 1980s, and the emergence of managed care in the later 1980s may
also have dampened this differential. This control-group shift hypothesis suggests that
negative rate-setting impacts only occurred within a fairly narrow time window from the
mid-1970s to the mid-1980s. This view is consistent with the results from the somewhat
simplified model reported by Friedman and Coffey (1993). It is also consistent with the
absence of negative rate-setting impacts in the findings by Antel, Ohsfeldt and Becker
(1995) with more recent data. More studies using data from the latter part of the 1980s
and the early 1990s in models that are more complete that those used by Friedman and
Coffey, but following their approach of allowing for structural shifts as the "control"
conditions change, would be helpful for confirming this view.
The literature also suggests a rather narrow geographic window for negative rate-
setting impacts on costs. Empirical studies have applied increasingly detailed specifi-
cations of rate-setting programs to test hypotheses that distinguish among the various
types of programs: young vs. old, formula vs. budget review, all-payer vs. partial payer,
and so on. While evidence in support of these hypotheses have often been obtained,
other studies that examined state-by-state variations in impact have suggested that the
evidence supporting a number of these hypotheses is mainly driven by the experience of
two states (New York and New Jersey) [Morrisey et al. (1983)]. This observation raises
several important questions regarding the interpretation of empirical results on cost re-
ductions. First, to what extent are the observed reductions due to rate setting per se
rather than to the presence of other vigorous regulatory controls (e.g., CON, restrictive
Medicaid payment policies), or the interaction of rate-setting with these other controls,
or the breadth and depth of political support for cost containment (independent of the
particular instruments chosen for cost containment policy)? Second, while the results
for these two states demonstrate that rate-setting can reduce costs (at least unit costs),
can this result be achieved under other political conditions in other states?
The results relating to volume impacts of rate-setting generally contain good news for
proponents of regulation. There is some evidence from the Worthington and Piro (1982)
study that occupancy rates and average length of stay increased in response to rate set-
ting programs that used the patient day or individual services (including days of stay) as
a unit of payment. This evidence is consistent with the results for rate-setting reported
in Joskow's (1980) reserve margin study. Other studies suggest, however, that stringent
case-based rate regulation or appropriate volume adjustment provisions can offset in-
centives to increase volume under rate setting [Salkever and Steinwachs (1988), Thorpe
and Phelps (1990)]. This is also consistent with the fact that research using data from
1526 D.S. Salkever
the 1980s, when many rate-setting programs had moved to a case-based approach, do
not report evidence of positive occupancy or length of stay effects [Hadley and Swartz
(1989), Graham and Cowing (1997)].
Finally, evidence from the studies of investment and service diffusion indicates that
aggregate measures of investment (beds, plant assets per bed, plant assets) were gen-
erally not affected by rate-setting programs but that diffusion of some specialized pro-
grams (e.g., open-heart surgery, cardiac catheterization) was retarded by rate-setting. As
in the unit cost literature, the evidence suggests that the experience of New York and
one or two other states drove this result. This again raises the questions of interpretation
noted above.
(B) CONprograms. The evidence regarding CON impacts on hospital investment and
service expansion is decidedly mixed. The early study by Salkever and Bice (1976,
1979) indicated a constraining effects on investment in new beds. While this result is
consistent with the findings reported by Cromwell (1987), other later studies [Sloan and
Steinwald (1980), Sloan (1981)] using both individual hospital and aggregate data, and
incorporating distinctions between young and old CON programs failed to confirm this
result. 27 Results from the reserve margin/excess capacity literature, however, consis-
tently report strong negative effects of CON on excess capacity. Unless one attributes
this result to the direct effect of CON on service volume, conditional on beds, which
was insignificant in the studies reviewed above [Salkever and Bice (1979), Hadley and
Swartz (1989)], the reserve margin findings seem to corroborate the hypothesis that
CON does constrain bed supplies. 28
The few studies that have looked at CON impacts on plant assets per bed have not
produced a consensus; early evidence [Salkever and Bice (1976)] suggested a positive
effect but later studies did not support this finding. In contrast, early research on service
diffusion [Russell (1979)] produced some evidence that CON slowed diffusion of open-
heart surgery and cobalt therapy but later studies did not confirm these negative impacts.
Thus, it is fair to say that the literature to date does not support a finding of negative
CON impacts on diffusion of capital-intensive services. Note, however, that CON con-
trols became much more widespread than rate regulation so that the CON results from
later studies are not driven by the experiences of a small number of states. To the extent
that CON programs in rate-setting states such as New York or New Jersey were in fact
more restrictive than CON programs in other states, evidence of negative rate-setting
impacts on service diffusion may in fact be due at least in part to CON controls.
Finally, there is some evidence of direct CON impacts on unit costs in quasi-
technological cost functions. Eakin (1991) and Mayo and McFarland (1989) attribute
this to CON programs causing hospitals to operate with non-optimal levels of capital
27 Sloan and Steinwald note, of course, that CON effects on bed supply estimated from individual hospital
data can not capture effects arising from the CON impacts on entry or exit of hospitals from the market.
28 Note, however, that Worthington and Piro (1982) did not find positive CON effects on hospital occupancy
rates.
inputs. Lanning, Morrisey and Ohsfeldt (1991) suggest that positive CON impacts on
cost result from management responses to the protection which CON affords to incum-
bent firms against competition by new entrants. At a minimum, it seems fair to conclude
that direct CON effects on costs are not negative.
(C) The ESP program. Findings regarding the impact of the ESP program on costs
display more diversity than any other aspect of the literature we have reviewed. Gins-
burg (1978) reported no impact on units costs, strongly negative impacts are reported
by Sloan and Steinwald (1980), Sloan (1981, 1983) and Antel, Ohsfeldt and Becker
(1995) while strongly positive coefficients are reported by Lanning, Morrisey and Ohs-
feldt (1991). The only significant quantity and investment findings are a negative effect
on length of stay reported by Ginsburg (1978) and a negative effect on assets per bed
reported by Sloan and Steinwald (1980).
4.8. A review of selected research on PPS impacts on hospital costs
While the Medicare PPS is not, strictly speaking, a regulatory program since it does
not directly affect non-Medicare patients, it had a profound influence on the economic
environment in which regulatory programs operated. Thus, it is useful to briefly review
the studies of PPS impacts which parallel the literature we have already reviewed on
hospital regulation as a cost control strategy.
The national scope of the Medicare program presents an obvious difficulty in using
a control group vs. treatment group approach to empirical estimation of PPS cost im-
pacts. An alternative strategy used in the literature is to include explanatory variables
presumed to be correlated with the economic "bite" or "pressure" exerted by PPS on
each hospital. Indicators of this pressure, which do exhibit considerable cross-sectional
variation among hospitals include (1) the fraction of the hospital's patients who are
covered by Medicare and (2) the ratio of the hospital's pre-PPS (base-year) costs to the
fixed PPS rate. (Note that the PPS rate in the phase-in period of Fiscal Years 1984-1987
varied by region of the country.) For example, Zwanziger and Melnick (1988) construct
a PPS pressure index that is the product of these two indicators. They apply this index in
an empirical study of quasi-technological cost functions for general acute care hospitals
in California over the period 1980-1985. Interacting this PPS index with year dummies
for 1983-1985 yields year-specific estimates of PPS coefficients. Their dependent vari-
able is total hospital expenses and output variables include inpatient discharges, length
of stay, a case-mix index, and total outpatient revenue. Estimation results show signif-
icantly negative impacts for the index, with the magnitude increasing by a factor of
four from 1983 to 1985. The result implies that expenses grew significantly less rapidly
in hospitals that were most impacted by PPS (by virtue of high base-year costs and/or
heavier Medicare patient loads).
In the studies by Hadley and Swartz (1987, 1989) of 1980-1984 data which we re-
viewed above, early PPS effects in a quasi-technological total cost function are based on
coefficient estimates of a PPS dummy. While this formulation does not allow for vari-
ation in PPS effects by hospital, their results confirm the Zwanziger-Melnick finding
1528 D.S. Salkever
of significant cost reduction. Analogous results from their output-determination models

(also reviewed above) implied that PPS had significant negative effects on Medicare
and private payer lengths of stay, marginally significant positive effects on private payer
admissions and on outpatient surgical visits (for all payer categories combined), and
essentially no effects on Medicare admissions, Medicaid admissions, Medicaid length
of stay, and non-surgical outpatient visits.
Sloan, Morrisey and Valvona (1988) added another year of PPS experience to the data
and reported results that only partially confirmed Hadley and Swartz. Data aggregated
to the state level for the years 1977-1985 were employed and the models estimated
could be viewed as reduced-form equilibrium relationships. Explanatory variables de-
scribed demand conditions (population, income, insurance coverage) and supply condi-
tions (physician supply, wages) in each state. The state-level indicator of PPS was set at
0.25 times the fraction of state hospital revenues derived from the Medicare program for
1984, and at 0.5 times this fraction for 1985. The rationale for the 0.25 and 0.5 figures
was that the weights used in computing the PPS rates in 1984 (1985) was 0.25 (0.5) for
a national per case average and 0.75 (0.5) for the per case level based on each hospital's
own 1982 cost level (updated for inflation). No additional adjustment for fiscal pressure
was included.
Regression results indicated that PPS had significantly negative effects on total hos-
pital revenues per capita and that virtually all of this was the result of PPS having a
strongly negative effect on total adjusted admissions per capita. PPS also had signifi-
cantly negative effects on length of inpatient stays and labor costs per admission, but
significantly positive effects on cost per patient day and non-labor cost per admission.
No PPS effect was found on overall cost per adjusted admission. Sloan et al. comment
on the counter-intuitive finding that PPS reduced costs by reducing admission when one
might generally expect the hospital to have an incentive to increase admissions under
per case payment. They speculate that the implementation of reinvigorated utilization
controls through Medicare's Physician Review Organizations (PROs) may in fact ac-
count for this trend.29
Another reduced-form analysis with aggregate data, that includes substantially more
experience under PPS, is the Antel, Ohsfeldt and Becker (1995) analysis of state data
29 Hodgkin and McGuire (1994) point out that a decline in admissions could be explained by a PPS-induced
decline in the intensity" of care if one assumes that admission volume is determined by consumer demand
and that intensity is positively related to the demand for admissions. In their model, changes in the payment
system that imply reductions in intensity per case also imply reductions in the volume of admissions; such
changes include increased prospectivity of the reimbursement formula, and reduced generosity of the for-
mula. Unlike other writers, who typically allow (at least implicitly) for demand manipulation by the hospital,
Hodgkin and McGuire's model specifically rules this out. Their analysis demonstrates that both the level of
payment and the degree of prospectivity can influence hospital decisions. What is surprising, however, is that
the decline in admissions occurred during 1983-1985 while in the late 1980s, when the level of PPS pay-
ments became more stringent, admissions did not continue to decline. This would only be consistent with the
Hodgkin-McGuire argument if the initial change in prospectivity had a much stronger impact on intensity
than the subsequent decline in the level of generosity of PPS payments. They do provide some empirical
support for this view.
for the years 1968-1990 which we reviewed previously. The authors adopt the Sloan
et al. approach to specifying their PPS variable, but the fractional weight is increased to
0.75 in 1986 and to 1.0 in later years. (In both the Sloan et al. and Antel et al. studies,
the fraction is set at 0 for "waiver" states where all-payer state rate setting determines
Medicare payments.) Results when the PPS variable is included without interactions
indicates a significantly positive PPS effect on cost per day, a significantly positive PPS
effect on cost per admission, and no significant effect on cost per capita. Including
interactions of PPS with CON and rate-setting does not substantially change these main
effect estimates though there is some evidence of small, negative interactions between
PPS and CON in the cost per day and cost per admission regressions. When the model
with interactions is estimated with state-specific fixed effects, the significantly positive
main effects of PPS become more significant and a significant and positive effect is now
observed on cost per capita while the interaction effects remain very small. Antel et al.
do not report estimated PPS effects on length of stay or volumes of care.
In summary, the principal econometric evidence that PPS helped to control hospital
costs comes from estimates of quasi-technological cost functions and from estimates of
reduced-form length-of-stay regressions. A significant post-PPS decline in admissions
may also have helped to reduce expenditure growth though attribution of this effect to
PPS may not be accurate. In contrast to the econometric literature, a number of other
comparative studies that used trend comparisons, or very parsimonious regression mod-
els, do show more consistent evidence of overall cost savings after PPS implementation
[Gold et al. (1993), Coulam and Gaumer (1992), Russell (1989)]. On a priori grounds,
the conclusion that PPS did in fact slow the growth of hospital spending seems rea-
sonable; but it is troubling that the empirical evidence weakens when our econometric
models are more completely specified, and that an important element of this reduction
in cost growth may be attributable to a decline in admissions whose source we do not
yet understand.
One factor which could explain the difference in results between the quasi-
technological cost function estimates and the reduced-form model estimates of PPS
effects on cost per case is the influence of PPS on case mix. In other words, when we
control for case-mix [a la Zwanziger and Melnick (1988)] we may find that PPS reduced
the costs of producing a case, while a positive effect of PPS on case mix could cancel
this efficiency gain and yield insignificant reduced-form estimates. 3 0
There are several different reasons why PPS may in fact have had a positive impact
on case mix. First, the DRG case-mix classification system used in PPS is not purely a
diagnosis-based system. Instead, it includes a number of categories which are defined
in part by "intensity" of treatment; typical examples are situations where surgical puts
the patient in a different DRG treatment (e.g., cholecystectomy), with a higher per case
30 We should also exercise some caution in interpreting findings of reduced production costs per case as
"efficiency gains". Newhouse and Byrne (1988) point out that some of the post-PPS decline in inpatient
length of stay was the result of longer-stay patients being diverted to specialty hospitals that were not paid on
a per case basis under PPS.
1530 D.S. Salkever
payment, than nonsurgical treatment (e.g., biliary tract disorder). McClellan (1996) has
examined the implications of such a payment system by developing a formal model in
which hospital resource investment decisions, and physician choice of treatment "in-
tensity" are influenced by the differences in case payment levels between high "inten-
sity" (e.g., surgical) and low "intensity" (e.g., medical) DRGs. His empirical analysis
of costs and resource use by case type suggests that the Medicare per case payment are
roughly proportional to the costs for both high and low intensity treatments; correspond-
ing analysis of changes in admissions indicates that admission rates for high-intensity
DRGs increased (over the 1983-1988 period) relative to the rates for the corresponding
low-intensity alternative DRGs.
A second reason for an increase in Medicare case-mix costliness in response to PPS
is "upcoding" or DRG "creep", which occurs when hospitals make special efforts to
qualify patients for higher-paying DRGs. (This may occur, for example, when physi-
cians exert special efforts to ensure that all codable secondary diagnoses are in fact
recorded.) Analysis of empirical data by Ginsburg and Carter (1986) indicates that the
magnitude of the increase in the Medicare Case-Mix Index (CMI) from 1981 to 1984
was 9.2 per cent, and that 2.8 per cent of this increase was due to PPS-induced upcoding.
In a subsequent analysis, Carter, Newhouse and Relles (1990) studied the CMI change
from 1986 to 1987, which totaled roughly 2.7 per cent, and found that about one-fourth
of this change was due to upcoding.
5. Concluding observations
5.1. On the noncovergence of empirical results
In comparison to other empirical literatures in health economics, such as the literature

on the price elasticity of demand for medical care, one is struck by the wide variability
in the empirical results on regulatory impacts, and the regularity with which earlier find-
ings are reversed in subsequent studies. To a degree, this variability is a consequence
of the diversity of models and data sets employed in the research. A more fundamen-
tal explanation, however, is heterogeneity in the "treatment" and "control" conditions
examined by the researchers.
Aspects of the heterogeneity in "treatment" conditions have been recognized explic-
itly in many, and particularly the more recent, contributions to the literature. Regula-
tory variables have been defined to differentiate between young and mature programs,
between noncomprehensive and comprehensive programs, between formula-based rate
setting and budget review rate setting, and so on. As often as not, however, the hy-
potheses that motivated these differentiations have not been supported by the empirical
results. It is plausible to speculate that the heterogeneity in regulation programs which
has not been measured may in fact be more important than that which has been mea-
sured and included in the empirical models.
Heterogeneity in the "control" conditions raises some research design issues that may
be even more problematic. The economic and political "rules" of the U.S. health care
system have been far from stable in the post-Medicare era. The nationwide controls of
the ESP program in the early 1970s, the national Voluntary Effort in the late 1970s,
implementation of the Medicare PPS in 1983, the remarkable growth of managed care
and capitation beginning in the late 1980s, the virtual disappearance of cost-based re-
imbursement by the mid- 1990s - all of these major developments in the national health
care economy have surely affected the results of our empirical comparisons between
regulatory "treatments" and nonregulated "controls". What then are the possibilities for
generalizing our findings from a specific economic and political context into a turbulent
future in which the health care system may look very different than it does today?
At least the possibilities for addressing the issue of unmeasured "treatment" hetero-
geneity may be more apparent if the likely sources of this heterogeneity can be iden-
tified. Several reasonable conjectures about these sources can be offered. First, many
participants and expert observers of health care regulation have remarked on the coex-
istence of multiple regulatory programs within the same jurisdiction, and have argued
that the relationships among these programs are important determinants of their joint
effects. The clear implication of this line of argument for empirical model specification
is that tests of program interactions should be carried out. Only one of the studies in our
review in fact followed this approach.
Second, it seems intuitively obvious that the political climates within the states affect
the ways in which regulatory programs are applied. The very limited literature on the
demand for regulation certainly supports the general proposition that adoption of reg-
ulation is responsive to political factors. Is it not reasonable to expect that these same
factors would also influence the way in which regulations are applied? If so, it may be
that a structural modeling strategy which explicitly incorporates political influences on
regulatory "intensity" would be a promising initial step in capturing heretofore unmea-
sured "treatment" heterogeneity due to variations in states' political climates.
5.2. Future research directions
The rise and decline of regulation of hospital prices and investment has been matched,
with a short lag, by the even more dramatic rise and decline of research interest in
these subjects. This is unfortunate for several reasons. First, to paraphrase Mark Twain,
reports of the death of regulation have been greatly exaggerated. It is indeed true that
only two states continue to regulate hospital rates but, as noted above, 37 states still
have CON regulation in effect. Thus, research on regulation continues to be relevant
to understanding programs that are in place today in the U.S. health sector. Second,
I suspect that even the most atheistic health economists would be reluctant to totally
discount the possibility of resurrection, of regulation that is. This may take other forms.
What was hospital rate-setting in a previous life may reappear as rate-setting for large
HMOs that have captured market shares comparable to those of Blue Cross in bygone
days. We have already seen growing interest at the state and Federal level in regulating
the benefit packages offered, services provided, marketing practices, and other aspects
of HMO operations. It is not unreasonable to expect that lessons learned from studies of
1532 D.S. Salkever
hospital regulation will have some carryover to its transmigration into future regulatory
programs.
It is clear that we still have much to learn from our experience with hospital regula-
tion. As I noted above, convergence on a clear set of findings from econometric studies
of costs, output, and investment is still far off. I am optimistic that future research that
deals more adequately with the heterogeneity of treatments and controls will bring us
closer to convergence. In particular, we have barely begun to model the political forces
and "intensity" that is probably an important source of treatment heterogeneity. Such
analyses would have the potential to make more fundamental contributions to our under-
standing of health-sector regulation generally. Moreover, the econometric literature on
many other important regulatory impacts is virtually undeveloped. We know relatively
little about the impact of regulation on quality of services. The influence of regulation
on innovation and structural change in the health sector is of tremendous potential im-
portance. It has been argued, for example, that CON regulation retards the introduction
of hospital substitutes such as free-standing ambulatory surgery centers, and that regu-
lation has been used to block the growth of HMOs and other managed care plans. I am
not aware of any careful econometric studies of these important issues. Finally, we may
have much to learn from the experience of deregulation but we have barely begun to
exploit the data on this subject.
References
Abt Associates and Policy Analysis, Inc. (1976), "Analysis of prospective payment systems for Upstate New
York", Final Report on Contract HEW-OS-74-261 (Cambridge, MA).
Altman, S.H., and J. Eichenholz (1976) "Inflation in the health industry: Causes and cures", in: M. Zubkoff,
ed., Health: A Victim or Cause of Inflation (Prodist, New York) 7-30.
American Enterprise Institute (1978), Proposals for the Regulation of Hospital Costs (American Enterprise
Institute for Public Policy Research, Washington).
Antel, J.J., R.L. Ohsfeldt and E.R. Becker (1995), "State regulation and hospital costs", Review of Economics
and Statistics 77(3):416-422.
53:941-973.
Bauer, K.G. (1978), "Hospital rate setting - This way to salvation?", in: M. Zubkoff, I.E. Raskin and
R.S. Hanft, eds., Hospital Cost Containment: Selected Notes for Future Policy (Prodist, New York) 324-
369.
Carter, G.M., J.P. Newhouse and D.A. Relles (1990), "How much change in the case mix index is DRG
creep?", Journal of Health Economics 9(4):411-428.
Chayet, and P.C. Sonnenreich (1978), Certificate of Need: An Expanding Regulatory Concept (Medicine in
the Public Interest, Boston).
Coelen, C., and D. Sullivan (1981), "An analysis of the effects of prospective reimbursement programs on
hospital expenditures", Health Care Financing Review 2(3):1-40.
Cohen, H.A. (1978), "Experiences of a state cost control commission", in: M. Zubkoff, I.E. Raskin and
R.S. Hanft, eds., Hospital Cost Containment: Selected Notes for Future Policy (Prodist, New York) 401-
428.
Cone, K., and D. Dranove (1986), "Why did states enact hospital rate-setting laws?", Journal of Law and
Economics 29(2):287-302.
Cromwell, J. (1987), "Impact of state hospital rate setting on capital formation", Health Care Financing Re-
view 8(3):69-82.
Cromwell, J., and J. Kanak (1982), "The effects of prospective reimbursement programs on hospital adoption
and service sharing", Health Care Financing Review 4(2):67-88.
Coulam, R.F, and G.L. Gaumer (1992), "Medicare's prospective payment system: A critical appraisal",
Health Care Financing Review: 1991 Annual Supplement (March):45-77.
Davis, K., G.E Anderson, D. Rowland and E.P. Steinberg (1990), Health Care Cost Containment (The Johns
Hopkins University Press, Baltimore).
Dranove, D., and K. Cone (1985), "Do state rate setting regulations really lower hospital expenses?", Journal
Dowling, W., et al. (1976), "The impact of the Blue Cross and Medicaid Prospective Reimbursement Systems
in Downstate New York", Final Report on Contract HEW-OS-74-248 (Department of Health Services,
School of Public Health and Community Medicine, University of Washington, Seattle).
Dunn, W.L., and B. Lefkowitz (1978), "The hospital cost containment act of 1977: An analysis of the admin-
istration's proposal", in: M. Zubkoff, I.E. Raskin and R.S. Hanft, eds., Hospital Cost Containment: Selected
Notes for Future Policy (Prodist, New York) 166-214.
Eakin, B.K. (1991), "Allocative inefficiency in the production of hospital services", Southern Economic Jour-
nal 58(1):240-248.
Eakin, B.K., and T.J. Kneisner (1988), "Estimating a non-minimum cost function for hospitals", Southern
Economic Journal, January:583-597.
Fanara, P., and W. Greenberg (1985), " Factors affecting the adoption of prospective reimbursement programs
by state governments", in: J.A. Meyer, ed., Incentives vs. Controls in Health Policy (American Enterprise
Institute for Public Policy Research, Washington) 144-156.
Feldstein, M.S. (1971), The Rising Cost of Hospital Care (Information Resources Press, Washington).
Friedman, B., and R.M. Coffey (1993), "Effectiveness of state regulation of hospital revenues in the 1980s",
in: R. Helms, ed., Health Policy Reform: Competition and Controls (American Enterprise Institute, Wash-
ington, DC) 36-57.
Fuchs, V.R. (1972), "Health care and the United States economic system -An essay in abnormal psychology",
Milbank Memorial Fund Quarterly.
Ginsburg, P.B. (1976), "Inflation and the economic stabilization program", in: M. Zubkoff, ed., Health: A Vic-
tim or Cause of Inflation (Prodist, New York) 31-51.
Ginsburg, P.B. (1978), "Impact of the economic stabilization program on hospitals: An analysis with aggregate
data", in: M. Zubkoff, I.E. Raskin and R.S. Hanft, eds., Hospital Cost Containment: Selected Notes for
Future Policy (Prodist, New York) 293-323.
Ginsburg, P.B., and G.M. Carter (1986), "Medicare case-mix index increase", Health Care Financing Review
7(4):51-65.
Gold, M., K. Chu, S. Felt, M. Harrington and T. Lake (1993), "Effects of selected cost-containment efforts:
1971-93", Health Care Financing Review 14(3):183-225.
Graham, G.G., and T.G. Cowing (1997), "Hospital reserve margins: Structural determinants and policy im-
plications using cross-sectional data," Southern Economic Journal 63(3):692-709.
Hadley, J., and K. Swartz (1987), "The impacts of hospital rate regulation, competition, and changes in health
insurance coverage on hospital costs between 1980 and 1984", Working Paper 3470-03 (Urban Institute,
Washington).
Hadley, J., and K. Swartz (1989), "The impacts on hospital costs between 1980 and 1984 of hospital rate
regulation, competition, and changes in health insurance coverage", Inquiry 26(Spring):35-47.
Hellinger, F.J. (1976), "The effect of certificate-of-need legislation on hospital investment", Inquiry
13(June):187-193.
Hellinger, F.J. (1979), "Hospital rate-regulation programs and proposals: A survey and analysis", in: Topics
in Health Care Financing: Rate Regulation (Aspen Systems Corp.) 5-14.
Hodgkin, D., and T.G. McGuire (1994), "Payment levels and hospital response to prospective payment",
1534 D.S. Salkever
Joskow, P.L. (1980), "The effects of hospital competition and regulation on hospital bed supply and the reser-
vation quality of the hospital", The Bell Journal of Economics 11(2):421-447.
Kessel, R.A. (1958), "Price discrimination in medicine", Journal of Law and Economics 1:20-53.
Klarman, H.E. (1965), The Economics of Health (Columbia University Press, New York).
Lanning, J.A., M.A. Morrisey and R.L. Ohsfeldt (1991), "Endogenous hospital regulation and its effects on
hospital and non-hospital expenditures", Journal of Regulatory Economics 3:137-154.
Mayo, J.W., and D.A. McFarland (1989), "Regulation, market structure, and hospital costs", Southern Eco-
McClellan, M.B. (1996), "Medicare reimbursement and hospital cost growth", in: D.A. Wise ed., Advances
in the Economics of Aging (Un;iversity of Chicago Press, Chicago, IL).
Moore, J.D. (1997), "CON survival struggle: Certificate of need gone in many states but not dead yet", Modern
Health Care (August 11):32-36.
Morrisey, M.A., EA. Sloan and S.A. Mitchell (1983), "State rate-setting: An analysis of some unresolved
issues", Health Affairs 2(2):36-47.
Mushkin, S.J. (1958), "Toward a definition of health economics", Public Health Reports 73(9):785-793.
Newhouse, J.P., and D.J. Byrne (1988), "Did medicare's prospective payment system cause length of stay to
fall?", Journal of Health Economics 7(4):413-416.
Noll, R.G. (1975), "The consequences of public utility regulation of hospitals", in: Controls on Health Care
(Institute of Medicine, National Academy of Sciences, Washington) 25-48.
Pauly, M.V. (1968), "The economics of moral hazard: Comment", American Economic Review 58:531-537.
Pauly, M.V. (1971), "Indemnity insurance for health care efficiency", Economic and Business Bulletin
24(1):53-59.
Posner, R.A. (1971), "Taxation by regulation", The Bell Journal of Economics and Management Science
5(Spring):22-50.
Posner, R.A. (1974), "Theories of economic regulation", The Bell Journal of Economics and Management
Science 5(Autumn):335-358.
Robinson, J.C., D.W. Garnick and S.J. McPhee (1987), "Market and regulatory influences on the availability
of coronary angioplasty and bypass surgery in U.S. hospitals", The New England Journal of Medicine
317(2):85-90.
Roemer, M.I. (1961), "Bed supply and hospital utilization: A natural experiment", Hospitals
35(November):36-42.
Romeo, A.A., J.L. Wagner and R.H. Lee (1984), "Prospective reimbursement and the diffusion of new tech-
nology in hospitals", Journal of Health Economics 3(1):1-24.
Rosko, M.D., and R.W. Broyles (1988), The Economics of Health Care (Greenwood Press, Westport, CT).
Russell, L.B. (1979), Technology in Hospitals (The Brookings Institution, Washington).
Russell, L.B. (1989), Medicare's New Hospital Payment System: Is It Working? (The Brookings Institution,
Washington).
Salkever, D.S. (1979), Hospital-Sector Inflation (Lexington Books, Lexington, MA).
Salkever, D.S. (1983), "Cost implications of hospital unionization: New estimates and a review of recent
research", Advances in Health Economics and Health Services Research 4:225-255.
Salkever, D.S., and T.W. Bice (1976), "The impact of certificate of need controls on hospital investment",
Milbank Memorial Fund Quarterly: Health and Society (Spring):185-214.
Salkever, D.S., and T.W. Bice (1978), "Certificate-of-need legislation and hospital costs", in: M. Zubkoff,
I.E. Raskin and R.S. Hanft, eds., Hospital Cost Containment: Selected Notes for Future Policy (Prodist,
New York) 429-460.
Salkever, D.S., and T.W. Bice (1979), Hospital Certificate-of-Need Controls: Impact on Investment, Costs and
Use (American Enterprise Institute for Public Policy Research, Washington).
Salkever, D.S., D.M. Steinwachs and A. Rupp (1986), "Hospital cost and efficiency under per service and
per case payment in Maryland: A tale of the carrot and the stick", Inquiry 23(Spring):56-66.
Salkever, D.5., and D.M. Steinwachs (1988), "Utilization and case-mix impacts of per case payment in Mary-
land", Health Care Financing Review 9(3):23-32.
Scanlon, W.J. (1980), "A theory of the nursing home market", Inquiry 17(1):25-41.
Sloan, EA. (1981), "Regulation and the rising cost of hospital care", The Review of Economics and Statistics
63(4):479-487.
Sloan, FA. (1983), "Rate regulation as a strategy for hospital cost control: Evidence from the last decade",
Milbank Memorial Fund Quarterly/Health and Society 61(2):195-221.
Sloan, FA., M.A. Morrisey and J. Valvona (1988), "Effects of the medicare prospective payment system on
hospital cost containment: An early appraisal", The Milbank Quarterly 66(2):191-220.
Sloan, EA., and B. Steinwald (1980), "Effects of regulation on hospital costs and input use", Journal of Law
and Economics 23(April):81-109.
Thomberry, H., and H. Zimmerman (1972), "Hospital cost control: An assessment of the Rhode Island experi-
ence with prospective reimbursement, 1971 and 1972", Final Report on Contract HEW-OS-74-197 (Rhode
Island Health Services Research, n.d., Providence).
Thorpe, K.E., and C.E. Phelps (1990), "Regulatory intensity and hospital cost growth", Journal of Health
Economics 9(2):143-166.
Warner, K. (1978), "Effects of hospital cost containment on the development and use of medical technology",
The Milbank Memorial Fund Quarterly/Health and Society 56:187-211.
Wendling, W., and J. Werner (1980), "Nonprofit firms and the economic theory of regulation", Quarterly
Review of Economics and Business (Autumn):6-18.
Worthington, N.L., and P. Piro (1982), "The effects of rate setting programs on volumes of hospital services",
Health Care Financing Review 4(2):47-66.
Zwanziger, J., and G.A. Melnick (1988), "The effects of hospital competition and the medicare PPS program
on hospital cost behavior in California", Journal of Health Economics 7(4):301-320.
Chapter 29
THE ECONOMICS OF SMOKING*
FRANK J. CHALOUPKA
University of Illinois at Chicago and NBER
KENNETH E. WARNER
University of Michigan
Contents
Abstract 1541
Keywords 1541
1.1. Coverage 1543
1.2. Health consequences of tobacco consumption 1544
2. The impact of price on the demand for tobacco products 1546
2.1. Conventional studies of cigarette demand 1547
2.1.1. Analysis of aggregate data 1547
2.1.2. Analysis of individual level data 1550
2.2. Addiction models and cigarette demand 1556
2.2.1. Imperfectly rational addiction models 1556
2.2.2. Myopic addiction models 1557
2.2.3. Rational addiction models 1559
2.2.4. Critiques of the rational addiction model 1561
2.3. Behavioral economic analyses of cigarette demand 1563
2.4. Econometric studies of the demand for other tobacco products 1564
3. Cigarette and other tobacco taxation 1565
3.1. Comparative standards and the effects of tax on price 1566
3.1.1. Purposes and methods of taxation 1566
*This work has been supported by a grant from the Robert Wood Johnson Foundation of Princeton, NJ, to
the University of Michigan (Warner). The authors would like to thank the following individuals for helpful
comments on a draft of this chapter: Philip Cook, William Evans, Michael Grossman, Jeffrey Harris, Thomas
Hodgson, Chee-Ruey Hsieh, Teh-wei Hu, Prabhat Jha, Andrew Jones, Donald Kenkel, Willard Manning,
David Merriman, Joseph Newhouse, Rosalie Pacula, Tomas Philipson, Harold Pollack, Henry Saffer, and
Mark Showalter.
1540 EJ.Chaloupka and K.E. Warner
3.1.2. Effects of taxes on retail price 1567

3.1.3. Variations in cigarette tax across countries and states and the issue of smuggling 1569
3.2. Fairness standards 1571
3.2.1. Horizontal and vertical equity 1572
3.2.2. The benefit principle 1574
3.3. Public health standards 1574
3.3.1. The social cost of smoking 1575
3.3.2. The health benefits of increasing cigarette taxes 1577
3.4. Economic efficiency and the pursuit of an optimal cigarette tax 1579
3.4.1. Negative externalities associated with smoking 1579
3.4.2. Other efficiency considerations 1583
4. Advertising, promotion, and the demand for tobacco products 1584
4.1. Theoretical and conceptual issues 1585
4.2. Econometric evidence 1585
4.3. Findings from the noneconomic literature 1591
5. Other tobacco control policies and demand 1593
5.1. Health information and counter-advertising 1593
5.2. Restrictions on cigarette smoking 1596
5.3. Limits on youth access to tobacco products 1597
6. Agricultural policy and the macroeconomic implications of tobacco 1598
6.1. Size and nature of the tobacco industry 1599
6.1.1. The global industry 1599
6.1.2. The U.S. tobacco industry 1599
6.2. The impact of the U.S. tobacco agriculture regulatory system 1601
6.2.1. Nature of the system and its impact on tobacco farming 1601
6.2.2. Relevance of the tobacco program to smoking and health 1603
6.3. The contribution of the tobacco industry to the economy 1606
6.3.1. States and nations 1606
6.3.2. Tobacco farm communities 1609
7. Conclusion 1611
References 1612
Ch. 29: The Economics of Smoking 1541
Abstract
While the tobacco industry ranks among the most substantial and successful of eco-
nomic enterprises, tobacco consumption is associated with more deaths than any other
product. Economic analysis of the markets for tobacco products, particularly cigarettes,
has contributed considerable insight to debates about the importance of the industry and
the appropriate roles of public policy in grappling with the health consequences of to-
bacco. Certainly the most significant example of this phenomenon has been the rapidly
expanding and increasingly sophisticated body of research on the effects of price in-
creases on cigarette consumption. Because excise tax comprises an important compo-
nent of price, the resultant literature has played a prominent role in legislative debates
about using taxation as a principal tool to discourage smoking. In addition to informing
legislative debates, this literature has contributed both theory and empirical evidence to
the growing interest in modeling the demand for addictive products.
This chapter examines this body of research in detail, as well as a variety of equity
and efficiency concerns accompanying debates about cigarette taxation. Coverage also
includes economic analysis of the role of other tobacco control policies, such as restric-
tions on advertising, of special interest due to their prominence in debates about tobacco
control. The chapter concludes with consideration of research addressing the validity of
the tobacco industry's argument that its contributions to employment, tax revenues, and
trade balances are vital to the economic health of states and nations. This argument is
one of the industry's principal weapons in its battle against policy measures intended to
reduce tobacco product consumption.
Keywords
cigarettes, tobacco, smoking, price, taxation, addiction, public health, social costs,
externalities, public policy, advertising, counter-advertising, smoking restrictions,
tobacco agriculture, economic contribution
JEL classification:118
1542 FJ. Chaloupka and K.E. Warner
Historically one of the oldest and most important crops in the United States, tobacco has
become embroiled in the second half of the twentieth century in a struggle pitting Amer-
ican economic against public health interests. While the tobacco industry ranks among
the most substantial and successful economic enterprises in the U.S., tobacco products
are associated with more deaths than any other product [U.S. Department of Health and
Human Services (1989, 1998)]. Tobacco products, and particularly cigarettes, which ac-
count for 95% of U.S. tobacco product sales, are credited with approximately one-fifth
of the nation's annual death toll. Cigarettes cause fully a third of deaths during mid-
dle age. The leading cause of lung cancer and chronic obstructive pulmonary disease
mortality, as well as a major cause of cardiovascular death, cigarette smoking leads all
other causes of death in virtually all industrialized nations. According to an epidemio-
logical analysis sponsored by the World Health Organization, tobacco will become the
leading cause of death in developing countries during the first third of the 21st century.
By 2030, tobacco will be responsible for 10 million deaths annually worldwide (Peto et
al., forthcoming), a toll that will exceed by far that associated with any other cause of
disease [Murray and Lopez (1996)].
Formal economic analysis of tobacco dates back at least half a century [Tennant
(1950)]. At that time, most tobacco industry economic analysis was motivated by the
factors that prompted market analysis of any other product or service, such as a desire
on the part of an industry to understand the degree of price elasticity of demand for
its product, or the interest of government and academic economists in the causes and
implications of market concentration. Beginning in the late 1960s, however, following
publication of seminal British and American reports on smoking and health [Royal Col-
lege of Physicians (1962), U.S. Department of Health, Education, and Welfare (1964)],
the focus of economic research shifted from a general industrial organization orienta-
tion toward analysis self-consciously relevant to the public health damage wrought by
tobacco. The subsequent economic analysis has been motivated by a desire to determine
how economic forces influence tobacco consumption, with continuing emphasis on re-
fining the scientific rigor of the work; but the objective of much of the research is now
to determine how to harness economic forces and logic, how to use economic tools, to
decrease smoking, with the ultimate goal being to reduce the toll of tobacco.
Certainly the most important example of this phenomenon has been the rapidly ex-
panding and increasingly sophisticated body of research on the effects of price increases
on cigarette consumption. Because excise tax comprises an important component of
price, the resultant literature has played a prominent role in legislative debates about
Not all of the research is motivated by a desire to decrease smoking. Some authors express the opinion
that more respect should be accorded consumer sovereignty, despite the issues of addiction and youthful
initiation of smoking that have led many economists to perceive the market for cigarettes as suffering from
important market imperfections [Warner et al. (1995)]. See, for example, Viscusi (1992) and Tollison and
Wagner (1992).
using taxation as a principal tool to discourage smoking in individual states, in the U.S.
as a whole, and in numerous other countries as well. In the United States in the late
1990s, the findings of this literature have been showcased in the intense congressional
debate over whether to adopt comprehensive tobacco control legislation, with a major
price increase lying at the heart of all proposals [Chaloupka (1998)].
1.1. Coverage
This chapter examines in detail economic analysis of the relationships among taxation,
price, consumption, and disease outcomes, as well as considering how analysis has en-
lightened other debates about the economics of tobacco. The relationship between price
and cigarette consumption has been the focal point of economic research on smoking,
and the locus of increasingly sophisticated and interesting development of theory and
methodology. For these reasons, and because the resultant literature constitutes the most
important contribution of economics to understanding tobacco policy, this chapter's
principal emphasis is on this body of research. The chapter devotes special consider-
ation to recent attempts to model nicotine addiction in the context of rational economic
behavior. Borne of a generic interest in the role of addiction in economic behavior, new
theoretical models have received their best empirical testing through the use of data on
cigarette smoking. 2
The chapter also considers a variety of equity and efficiency concerns that invariably
accompany debates about cigarette taxation, including the validity of the externality or
social cost argument frequently invoked by the public health community in calls for
higher taxes; whether there are other legitimate grounds in economic theory to sup-
port increased excise taxation; whether cigarette tax increases are regressive, and if so
how much; and whether large tax hikes produce substantial cross-border smuggling of
cigarettes. Recent economic studies have explored subtle impacts of cigarette taxation
that receive attention here as well; for example, cross-price elasticity issues include how
cigarette taxation may shift demand toward other tobacco products, such as smokeless
tobacco, or, within the cigarette family, from lower to higher tar and nicotine cigarettes.
Coverage in this chapter also includes attention to economic analysis of the role of
advertising in the demand for cigarettes, as well as the role of restrictions or bans on ad-
vertising. The effects of advertising and of advertising and other marketing restrictions
are of special interest due to their prominence in debates about tobacco control at all
levels of government, from municipalities, which have restricted cigarette advertising
2 As is discussed below, numerous variables other than price influence the demand for cigarettes, including
consumers' knowledge of the hazards of smoking, parental and sibling smoking behavior, smoking by peers,
role modeling, income, and education [U.S. Department of Health and Human Services (1994)]. The near-
exclusive focus of this chapter on price and taxation reflects the facts that tax is the most policy-tractable
variable influencing the demand for cigarettes and that the economics literature has focused on price, taxation,
and addiction.
1544 EJ. Chaloupka and K.E. Warner
on public transit and on billboards, to international bodies, which have called for com-
plete bans [Roemer (1993)]. Although its value is constrained by obvious limitations,
econometric analysis offers insights into the role of advertising, and of advertising re-
strictions, on the demand for cigarettes.
The chapter also examines what is known about the influence of other tobacco control
policies on the demand for cigarettes, including the development and dissemination
of information on the health consequences of smoking; media advocacy by means of
"counter-advertising"; the adoption and implementation of laws or policies that limit
smoking in public places; and legal restrictions on youth access to tobacco products.
Given tobacco's role in employment, tax revenues, and, in selected countries, trade
balances, governments have a legitimate interest in the "health" of their tobacco in-
dustries. Tobacco companies tout the industry's economic contribution in attempts to
combat tobacco control policy measures. In recent years, independent economists have
countered the industry's economic argument by carrying out macroeconomic analyses
that examine the net contributions of tobacco to economies, rather than the gross contri-
butions featured by the industry. This literature, and its role in the debate over tobacco
control policy, are examined toward the end of the chapter. Also considered briefly is
the influence of tobacco agriculture support policies so prominent in the agricultural
policies of the United States and the European Union.
Despite its wide scope of coverage, this chapter does not examine all of the eco-
nomic contributions relating to smoking and health. To illustrate with two examples,
the chapter does not consider the growing literature on the cost-effectiveness of smok-
ing cessation interventions [Cromwell et al. (1997), Warner (1997)] and it omits the
newly emerging and potentially quite important analysis of the economics of the mar-
ket for nicotine replacement products [Oster et al. (1996), Hu et al. (1998)]. Another
limitation on coverage is that we consider almost exclusively English language pub-
lications, believed to comprise a very sizable majority of the peer-reviewed literature.
Further, reflecting the authors' knowledge of this field and familiarity with data, exam-
ples draw heavily, although not exclusively, on the U.S. experience. In particular, there
is little coverage of the economics of smoking in developing countries, the result pri-
marily of the dearth of studies on the subject. Although specific empirical conclusions
from a given country may not apply precisely to other nations, the general phenom-
ena described and findings presented should apply qualitatively to all countries, unless
otherwise indicated.
Before turning to the economics literature, the remainder of this introductory section
presents a brief "primer" on the health consequences of tobacco use. We deem this im-
portant background for understanding the nature and social significance of the economic
issues.
1.2. Health consequences of tobacco consumption

The health implications of tobacco have been contemplated for at least the past mil-
lennium. During the first half of that period, the predominant view held that tobacco
afforded users a wide variety of health benefits. The Amerindians employed tobacco as
an analgesic and as a treatment for such diverse ailments as intestinal problems, asthma,
rheumatism, headaches, toothaches, boils, worms, fevers, and the pains of childbirth
[Goodman (1993)].
Serious medical and scientific attention to the health consequences of smoking is a
phenomenon of the present century, primarily of its second half. 3 This is a reflection of
the development of the science of epidemiology during this period and of the relatively
modest number of victims claimed by tobacco prior to the 20th century. Before this
century, relatively few people reached the ages at which tobacco takes its greatest toll
(average life expectancy in the U.S. was 47 in 1900; currently it is 75). More impor-
tantly, widespread intensive use of the most dangerous form of tobacco consumption,
cigarette smoking, began only in the very late 1800s. Lung cancer, today the source of
30% of all cancer deaths in the U.S. [U.S. Department of Health and Human Services
(1989)], was a rarity until earlier cigarette smoking spawned the epidemic first widely
observed during the 1930s.
Although a few scientific studies associated smoking with disease prior to mid-
century [Broders (1920), Lombard and Doering (1928), Pearl (1938)], the first evidence
that strongly implicated smoking in disease (specifically, lung cancer) was published
in the 1950s [Wynder and Graham (1950), Doll and Hill (1954, 1956), Hammond and
Horn (1958a, 1958b)]. Since then, some 70,000 scientific articles have implicated smok-
ing in a wide variety of ailments, constituting the largest and best documented literature
linking any behavior to disease in humans [U.S. Department of Health and Human Ser-
vices (1994)].
Today, cigarette smoking is established as the leading cause of lung cancer (respon-
sible for approximately 90% of lung cancer deaths in the U.S.), the leading cause of
chronic bronchitis and emphysema (responsible for over 80% of chronic obstructive
pulmonary disease deaths), and a major cause of heart disease and stroke. Smoking also
causes aneurysms, atherosclerotic peripheral vascular disease, oral cavity and laryngeal
cancer, intrauterine growth retardation and neonatal death, including SIDS (Sudden In-
fant Death Syndrome). It is associated with additional cancers (bladder, pancreatic, re-
nal, gastric, and cervical) [U.S. Department of Health and Human Services (1989)], as
well as a host of other conditions affecting a wide variety of organ systems and disease
processes, including, for example, vision and hearing problems, slowed healing from
injuries, and increased susceptibility to certain infections [Napier (1996)]. Chronic in-
halation of environmental tobacco smoke (ETS) causes lung cancer in nonsmokers and
3 Concern about the health consequences of smoking predates the "modern era" by nearly four centuries. In
1604, for example, King James I of England lambasted smoking as "a custome lothsome to the eye, hatefull
to the Nose, harmefull to the braine, dangerous to the Lungs, and inthe black stinking fume thereof, neerest
resembling the horrible Stigian smoke of the pit that is bottomlesse" [as quoted inSullum (1998, p. 18)]. King
James subsequently raised the tax on tobacco by 1000%, deriving significant revenues for his coffers. This
illustrates the profound dilemma that has confronted policy decision makers ever since: whatever its health
consequences, tobacco has long been truly a "golden leaf" for farmers and politicians alike. Its role inthe very
earliest commerce between England and the American colonies is legendary, as is its role in contemporary
politics [Taylor (1984), Fritschler and Hoefier (1996)].
1546 F.J. Chaloupka and K.E. Warner
an assortment of diseases and functional limitations in the children of smokers [Envi-

ronmental Protection Agency (1994)]. ETS may be responsible for tens of thousands of
heart disease deaths annually [Glantz and Parmley (1995)].
All told, smoking is far and away the leading cause of premature death and of avoid-
able morbidity and disability in the United States and in most industrialized nations. As
indicated above, the intensification of smoking in the world's less affluent nations will
soon bring the same distinction to smoking in the developing countries. Barring sub-
stantial and unexpected decreases in tobacco use worldwide, a few decades hence the
global death toll from tobacco will dwarf all other causes, with the majority of deaths
occurring in the developing nations. The World Health Organization estimates that fully
500 million of the 5 billion people alive at the beginning of this decade will die as a
result of consumption of tobacco products [Peto et al. (1999)].
The mortality toll of tobacco reflects not only the lethality of tobacco products but
also the prevalence of their consumption. In the United States, approximately 45 million
adults, almost a quarter of the adult population, smoke cigarettes (down from a high of
42% in 1965 [U.S. Department of Health and Human Services (1989)]). Worldwide,
tobacco products are used by approximately one billion people. The large numbers of
tobacco consumers, combined with their frequent use of tobacco products, account not
only for the disease toll of tobacco, but also for the substantial size of the tobacco
industry. Important features of the structure and economic importance of the industry
are reviewed in Section 6 below.
2. The impact of price on the demand for tobacco products
Many researchers once viewed cigarette smoking and other addictive behaviors as ir-
rational and therefore not suitable for conventional economic analysis [Elster (1979),
Winston (1980), Schelling (1984b)]. They believed that the demand for cigarettes (and
other addictive substances) did not follow the basic laws of economics, including per-
haps the most fundamental law, that embodied in the downward-sloping demand curve.
As the now-substantial body of economic research demonstrates, however, the demand
for cigarettes clearly responds to changes in prices and other factors, as found in ap-
plications of both traditional models of demand and more recent studies that explicitly
account for the addictive nature of smoking.
Conceptually, economists use a relatively broad definition of price that includes not
only the monetary price of purchasing a product, but also the time and other costs as-
sociated with using the product. Restrictions on smoking in public places and private
work sites, for example, impose additional costs on smokers by forcing them outdoors
to smoke, raising the time and discomfort associated with smoking, or by imposing fines
for smoking in restricted areas. Similarly, limits on youth access to tobacco may raise
the time and potential legal costs associated with smoking by minors, while new in-
formation on the health consequences of tobacco use can raise the perceived long-term
costs of smoking. This section focuses on the effects of monetary price on demand,
while Section 5 below considers the effects of other aspects of full price.
In addition to price, a variety of other factors can affect the demands for cigarettes and
other tobacco products, including income, advertising and other promotional activities,
and tastes. In the industrialized nations, the relationship between income and cigarette
consumption has reversed. Early demand studies [for example, Ippolito et al. (1979),
Fujii (1980)] concluded that cigarette smoking was a normal good, with cigarette con-
sumption rising as income rose. More recent studies, however, have found that cigarettes
have become an inferior good, in that the likelihood of smoking declines as income rises
[Wasserman et al. (1991), Townsend et al. (1994)]. The effects of advertising and pro-
motion on the demand for cigarettes have been the subject of numerous studies; these
are reviewed in detail in Section 4 below. Finally, nearly all econometric studies of
cigarette demand use a variety of factors to control for tastes, including gender, race,
education, marital status, employment status, and religiosity. Given the focus of this
book on economics, the impact of these socio-demographic determinants of demand
will not be reviewed. 4
This section begins with a review of conventional studies of the impact of money
price on cigarette demand. This is followed by a discussion of economic models of
addiction and their applications to cigarette demand. Implications for the effects of price
on cigarette demand from the relatively new field of behavioral economics are then
reviewed. The section closes with a short consideration of the relatively limited research
on the effects of price on the demand for other tobacco products.
2.1. Conventionalstudies of cigarette demand
Numerous investigators have estimated the effects of price on cigarette demand using
conventional models of demand that do not account for the addictive nature of cigarette
smoking. Their studies have used diverse econometric and other statistical methods
on data from numerous countries. Many used aggregate time-series data for a single
geographical unit, while others employed pooled cross-sectional time series data; still
others used individual level data taken from surveys. The price elasticity estimates for
overall cigarette demand from recent studies fall within the relatively wide range from
-0.14 to -1.23, but most fall in the narrower range from -0.3 to -0.5.
2.1.1. Analysis of aggregatedata
Many recent studies use aggregate data and appropriate econometric methods to exam-
ine the effects of price on cigarette demand, controlling for income, tobacco control
4 The importance of these variables should not be downplayed, however. In many instances, these and other
variables, such as parental and peer smoking behavior and societal norms, are as important or more important
than the variables which economists have studied [U.S. Department of Health and Human Services (1994)].
Variations in these and other variables help to explain why large variations in prices across countries are often
not associated with comparably large variations in smoking prevalence. Economists' interests focus on the
marginal impact of price, advertising, and other economic variables on the demand for cigarettes.
1548 EJ. Chaloupka and KE. Warner
policies, and a variety of socioeconomic and demographic factors. The exceptions [Bal-
tagi and Goel (1987), Peterson et al. (1992)] compared changes in cigarette consump-
tion in states that had raised cigarette taxes to consumption in states where taxes had not
changed. The estimated price elasticities from these quasi-experimental studies, in the
range from -0.17 to -0.56, are consistent with those obtained from the econometric
studies.
Although there are numerous studies of the price-demand relationship in industri-
alized nations, until recently there were almost no estimates for developing countries.
Warner (1990) argued that price responsiveness in less developed countries is likely to
be greater than in more affluent countries, given the relatively low incomes and rela-
tively low levels of cigarette consumption by smokers in the poorer countries. Findings
from studies using data from Papua New Guinea [Chapman and Richardson (1990)],
China [Mao (1996), Xu, Hu and Keeler (1998)], South Africa [van der Merwe (1998a)],
Zimbabwe [Maranvanyika (1998)], and Taiwan [Hsieh and Hu (1997)] are consistent
with this argument.
Several difficulties are encountered in studies using time-series data. Particularly
troubling are the high correlations among many of the key independent variables and
price. Consequently, estimates of the impact of price and other factors on demand can be
sensitive to the inclusion and exclusion of other variables. Including highly correlated
variables can result in multicollinearity and unstable estimates for the parameters of in-
terest. Excluding potentially important variables, however, can produce biased estimates
of the impact of price on demand. Recent studies using state-of-the-art econometric
methods have addressed many of these difficulties [Seldon and Boyd (1991), Simonich
(1991), Flewelling et al. (1992), Sung et al. (1994), Barnett et al. (1995), Keeler et al.
(1996)]. Nearly all of the estimates from these studies have produced estimates for the
price elasticity of demand in a relatively narrow range, centered on -0.4.
Other problems are encountered when using pooled cross-sectional time-series data.
The measure of cigarette smoking employed in these studies is typically annual state-
level tax-paid cigarette sales. Interstate differences in cigarette prices, resulting from
wide variation in state cigarette taxes [Tobacco Institute (1998)], can lead to casual and
organized smuggling of cigarettes from low-tax to high-tax states, however [Advisory
Commission on Intergovernmental Relations (ACIR) (1977, 1985)]. As such, tax-paid
sales data are likely to overstate cigarette consumption in states with low cigarette taxes
and underestimate it in high tax states.5 Failing to account for this will produce upward-
biased estimates of the impact of price on cigarette demand. Many of the more recent
studies employing pooled time-series cross-sectional state data have controlled for the
potential for smuggling [ACIR (1977, 1985), Baltagi and Levin (1986), Chaloupka and
Saffer (1992), Keeler et al. (1996)]. These studies have also produced estimates of the
5 The same problem exists in time-series studies using aggregate country-level data for countries with rel-
atively high taxes and prices compared to neighboring countries. See Joossens (1998) for a discussion of
factors other than price that influence smuggling across country borders.
price elasticity of cigarette demand generally falling in a relatively narrow range cen-
tered on -0.4.
The fact that cigarette prices, sales, and consumption are simultaneously determined
creates an additional complication in the analysis of cigarette demand and supply. Fail-
ing to account for this simultaneity would lead to biased estimates of the price elasticity
of demand. Again, many of the recent studies employing aggregate time-series data for
a single country or other geographical unit, as well as many of those using pooled cross-
sectional time-series data, have avoided this problem by theoretically and empirically
modeling cigarette demand and supply [Bishop and Yoo (1985), Porter (1986), Showal-
ter (1991), Sung et al. (1994), Barnett et al. (1995), Tremblay and Tremblay (1995),
and Keeler et al. (1996)]. Other studies have taken advantage of natural experiments,
most notably 25-cent increases in the California and Massachusetts cigarette excise
taxes, to look at the impact of price on demand [Keeler et al. (1993), Hu et al. (1994,
1995b), Sung et al. (1994), Harris et al. (1996)]. After accounting for the potential si-
multaneity or taking advantage of natural experiments, most of these studies produce
estimates of the price elasticity of demand that fall into the same narrow range found in
other studies.
Finally, studies employing aggregate data are generally limited to examining the im-
pact of cigarette prices and other factors on aggregate or per capita measures of cigarette
consumption. Consequently, these studies are typically unable to evaluate the differen-
tial impact of prices on smoking by various population subgroups of particular interest,
especially youth and young adults. Nor can they differentiate between the impact of
price on smoking prevalence and quantity, or smoking initiation and cessation.
A few recent analyses have attempted to address these limitations. For example, Har-
ris (1994) used annual time-series data on U.S. smoking prevalence taken from the
National Health Interview Surveys, coupled with aggregate measures of cigarette con-
sumption, to estimate the effects of price on smoking prevalence and average cigarette
consumption by smokers for the period from 1964 through 1993. His estimate of the un-
conditional price elasticity of demand fell into the same narrow range generally found in
other studies. He estimated that approximately half of the impact of price was on smok-
ing prevalence, with the price elasticity of smoking participation being -0.238, while
the unconditional price elasticity of demand was -0.47. Townsend et al. (1994) looked
at the differential effects of price on cigarette smoking for various population subgroups
defined by age, gender, and socioeconomic status, using data aggregated from the 1972
through 1990 British General Household Surveys. They concluded that women were
more responsive to price than men, that both men and women in lower socioeconomic
groups were more sensitive to price than those that were better off, and that youth (16-
19 years) and young adults (20-24 years) were less responsive to price than adults. 6
6 As we discuss below, other studies have derived the opposite conclusions concerning the relative price
responsiveness by gender [e.g., Lewit and Coate (1982), Mullahy (1985), Chaloupka (1990)] and different
age groups [e.g. Lewit et al. (1981), Chaloupka and Grossman (1996)].
2.1.2. Analysis of individual level data
A relatively small but growing number of cigarette demand studies have used data on in-
dividuals taken from large-scale surveys. In general, their estimated price elasticities of
demand are comparable to those estimated using aggregate data. The use of individual-
level data helps avoid some of the problems inherent in using aggregate data. For ex-
ample, because an individual's smoking decisions are too small to affect the market
price of cigarettes, potential simultaneity biases are less likely. Similarly, individual-
level income data and measures of socio-demographic determinants of demand are less
correlated with price and policy variables than comparable aggregate measures.
Other problems persist but can be addressed somewhat more easily using individual-
level data. For example, failing to account for interstate differences in cigarette prices
will again produce a biased estimate of the price elasticity of demand (biased towards
0 in this case). Thus, given information on where an individual resides, studies using
individual-level data have employed a variety of approaches to control for potential
cross-border shopping in response to interstate price differentials. Some have limited
their samples to individuals who do not live near lower-price localities [Lewit and
Coate (1982), Wasserman et al. (1991), Chaloupka and Grossman (1996), Chaloupka
and Wechsler (1997)]. Others have included a measure of the price differential [Lewit
et al. (1981), Chaloupka and Pacula (1998, 1999, forthcoming)]. Still others have used a
weighted average price based on the price in the own-locality and other nearby localities
[Chaloupka (1991)].
As with the state tax-paid sales data, self-reported data on cigarette smoking yield
inaccurate measures of true consumption, given potential reporting biases. Based on
a comparison of self-reported consumption with aggregate sales data, Warner (1978)
demonstrated that survey-based self-reported consumption significantly and substan-
tially understated actual sales. Studies using individual-level survey data have implic-
itly treated underreporting as proportional to true consumption across groups of interest
(e.g., age, gender, or socioeconomic groups). If the assumption is true, estimates of the
price elasticity of demand will not be systematically biased. The assumption has yet to
be demonstrated, however.
Finally, as Wasserman et al. (1991) observed, studies using individual-level data may
be subject to a substantial ecological bias in that omitted variables affecting tobacco
use may be correlated with the included determinants of demand. Failing to account for
this can produce biased estimates for the included variables. For example, unobserved
sentiment against smoking may affect both cigarette sales and the strength of tobacco
control policies (including taxes and, consequently, prices). Ohsfeldt et al. (1999) con-
sidered this possibility in their analysis of cigarette smoking and other tobacco use that
employed data from the 1992/93 Current Population Survey Tobacco Use Supplements.
Surprisingly, after modeling cigarette taxes and other tobacco control policies as a func-
tion of cigarette smoking, various other indicators of sentiment against smoking, and
other factors, they found that taxes have a larger impact on demand.
Using individual-level data allows researchers to examine issues that generally can-
not be addressed with aggregate data. For example, most studies using individual-level
data separately consider the effects of price on the probability of smoking and on av-
erage cigarette consumption by smokers. In addition, several consider the differential
effects of price on demand for various population subgroups (defined by age or gender,
for example). Finally, some have taken advantage of retrospective or longitudinal data
to examine the effects of prices and other factors on smoking initiation and cessation
decisions.
The earliest of the cigarette demand studies employing individual-level data were
conducted by Lewit and his colleagues [Lewit et al. (1981), Lewit and Coate (1982)].
Lewit and Coate used data from the 1976 National Health Interview Survey to examine
the effects of price on cigarette smoking, estimating an overall price elasticity of demand
of -0.42 and an elasticity of smoking participation of -0.26. In addition, they found an
inverse relationship between (the absolute value of) price elasticity and age, estimating
a total price elasticity of demand for 20 through 25 year-olds more than double that of
persons 26 and older. The researchers found that most of the effect of price for young
adults was on the decision to smoke (participation elasticity of -0.74 and conditional
demand elasticity of -0.20), but was about evenly split for those over 35 years of age
(participation and conditional demand elasticities of -0.15). Finally, they also looked at
differences in price responsiveness by gender, concluding that men, particularly young
men, were very responsive to price, while women were generally insensitive to price.
Lewit et al. (1981) and Grossman et al. (1983) confirmed the Lewit and Coate (1982)
conclusion concerning the inverse relationship between price elasticity of cigarette de-
mand and age. Using data from Cycle III of the Health Examination Survey, Lewit
et al. estimated that the price elasticity of smoking participation for 12-17 year-olds
was -1.20, while the conditional demand elasticity was -0.25. Their estimated total
price elasticity of youth cigarette demand of -1.44 was more than three times Lewit
and Coate's (1982) estimate for adults. These conclusions were generally supported by
Grossman et al.'s (1983) analysis of data from the National Household Surveys on Drug
Use conducted during the 1970s.
Lewit et al. (1981) offered two reasons why youth should be more price sensitive than
adults, at least in the short run. First, given the addictive nature of smoking, long-term
adult smokers are likely to adjust less quickly to changes in price than youth who have
been smoking for a relatively short time, if at all. In addition, peer behavior is likely to
be much more influential for youth, multiplying the effects of price on youth smoking.
That is, an increase in cigarette price directly reduces youth smoking and then again
indirectly reduces it through its impact on peer smoking. Grossman and Chaloupka
(1997) offered two additional reasons. First, the faction of disposable income a young
smoker spends on cigarettes is likely to exceed that spent by an adult smoker. Sec-
ond, compared to adults, youth are more likely to be present-oriented. In the context
of an economic model of addictive behavior (discussed below), Becker et al. (1991)
predicted that changes in money price will have a greater impact on individuals with
1552 F.J. Chaloupka andK.E. Warner
higher discount rates since they give less weight to the future consequences of addictive
consumption.
The conclusion that youth cigarette demand is more price elastic than adult demand
was widely accepted until an influential 1991 Rand study by Wasserman and colleagues
(1991). These researchers evaluated adults' cigarette demand using data from several
of the National Health Interview Surveys from the 1970s and 1980s and youth demand
with data from the Second National Health and Nutrition Examination Survey of the
late- 1970s. Using a generalized linear model, the authors concluded that adult demand
in the earlier years of their data was relatively unresponsive to price, but that demand
had become more price elastic over time. Based on the trends in price elasticity, they
predicted an overall price elasticity of adult cigarette demand of -0.283 for 1988. Esti-
mates from a two-part model of adult cigarette demand implied that the effects of price
on the decision to smoke were almost double the impact of price on conditional de-
mand. However, the authors did not find a statistically significant impact of price on
youth smoking. They attributed their relatively low estimates of price elasticity, partic-
ularly those for youth, to the inclusion in their models of an index of restrictions on
smoking. These restrictions, which they note are positively correlated with price, had
not been included in most previous studies of cigarette demand. Indeed, they obtained
very similar estimates to Lewit and Coate (1982) when leaving the restriction index out
of models estimated using the 1976 survey data.
Several more recent studies of youth and young adult smoking have supported the
earlier conclusions reached by Lewit and his colleagues [Lewit et al. (1981), Lewit and
Coate (1982), Grossman et al. (1983)] that the price sensitivity of cigarette demand
is inversely related to age. Chaloupka and Grossman (1996) examined the impact of
price, numerous tobacco control policies (including smoking restrictions and limits on
youth access to tobacco), and a variety of other socioeconomic and demographic fac-
tors on youth smoking, using data from the 1992, 1993, and 1994 Monitoring the Future
Surveys of eighth, tenth, and twelfth grade students. They estimated a total price elas-
ticity of youth cigarette demand of -1.31, strikingly similar to the estimates obtained
by Lewit et al. (1981) 15 years earlier. In contrast to Lewit and his colleagues, how-
ever, Chaloupka and Grossman found that the effects of price on smoking participation
and conditional demand were similar (-0.68 for smoking participation and -0.64 for
conditional demand). Chaloupka and Pacula (1999, forthcoming) used the same data to
look at the differential response by gender and race, concluding that young men and
young blacks are more responsive to price than young women and young whites.
Chaloupka and Wechsler (1997) reached similar conclusions using data on young
adult smoking taken from the 1993 College Alcohol Survey. Also controlling for nu-
merous other determinants of cigarette demand, including a variety of restrictions on
smoking, they estimated a price elasticity of smoking participation of -0.53 and an un-
conditional price elasticity of demand of -1.11 for college students. Noting that their
sample was not a random sample of all young adults, Chaloupka and Wechsler sug-
gested that the price elasticity of cigarette demand by young adults may be even higher,
given the evidence that cigarette demand is relatively less elastic for more educated
or higher-income individuals [Townsend (1987), Chaloupka (1991), Townsend et al.

(1994), Farrelly et al. (1998)].
Farrelly and his colleagues (1998) found similar evidence for young adults and adults,
based on 13 waves of the National Health Interview Survey conducted between 1976
and 1992. They estimated that demand was more than twice as elastic for their sample
of young adults, ages 18 to 24 years (total elasticity of -0.58), as for their full sample
(total elasticity of -0.25). Similarly, they estimated that blacks were about twice as
responsive as whites to cigarette prices, while Hispanics were even more price sensitive.
In addition, they found that men were more price sensitive than women. Finally, they
estimated that individuals with family incomes below the sample median were about 70
percent more responsive to price than those with higher family incomes.
Additional support for the inverse relationship between price sensitivity and age is
provided by recent studies by Lewit and his colleagues (1997), Evans and Huang (1998),
and Tauras and Chaloupka (1999). Lewit and his colleagues used data for ninth grade
students in 1990 and 1992 collected in the 22 North American communities involved
in the National Cancer Institute's Community Intervention Trial for Smoking Cessation
(COMMIT). They found that both youth smoking participation and intentions to smoke
among young non-smokers were inversely related to price, with estimated price elastici-
ties of -0.87 and -0.95, respectively. Evans and Huang used state level aggregated data
on smoking prevalence constructed from the 1977 through 1992 Monitoring the Future
surveys to estimate a price elasticity of youth smoking participation of -0.20. Unlike
other studies on youth smoking that largely rely on the cross-sectional variation in state
cigarette taxes and prices, Evans and Huang took advantage of the long time period cov-
ered by their data and used the time series variation in state cigarette taxes to identify
the impact of cigarette taxes on smoking participation. While their estimated elasticity
for the 1977 through 1992 period is relatively low, Evans and Huang concluded that
youth smoking has become more price sensitive over time, estimating an elasticity of
-0.50 for youth smoking participation in the period from 1985 through 1992. Most re-
cently, Tauras and Chaloupka (1999) used the longitudinal data from the Monitoring the
Future surveys of high school seniors conducted from 1976 through 1993 to estimate
the price elasticity of smoking for young adults; respondents in their sample ranged in
age from 17 to 35 years. In models controlling for unobserved state and individual fac-
tors affecting demand, they estimated an overall price elasticity of demand centered on
-0.79.
In general, researchers examining the effects of price on smoking participation us-
ing individual-level data from cross-sectional surveys have assumed that much of the
price effect estimated for youth reflects the impact of price on smoking initiation, while
the estimate for adults is largely capturing the effects of price on smoking cessation.
A few recent studies have attempted to directly examine the impact of cigarette prices
on smoking initiation. With retrospective data from the smoking supplements to the
1978 and 1979 National Health Interview Surveys, Douglas and Hariharan (1994) stud-
ied the ages at which survey respondents reported that they began smoking. Based on
current state of residence, they matched data on cigarette prices to the survey data to esti-
1554 EFJ. Chaloupka and K.E. Warner
mate the impact of price on smoking initiation. They estimated a hazard model in which
"failure" was defined as a never smoker taking up smoking and used a relatively general
variation on standard duration methods: the split population duration model developed
by Schmidt and Witte (1989). This model allows for a large part of their sample to never
begin smoking. Finally, Douglas and Hariharan's theoretical and empirical framework
was based on the Becker and Murphy (1988) rational addiction model (described be-
low). As anticipated, Douglas and Hariharan found that a number of socioeconomic
and demographic factors had a significant effect on smoking initiation. However, their
estimates for cigarette prices were insignificant. Given the errors-in-variables problem
associated with both the retrospective data on smoking initiation and the cigarette price
data, they noted that price effects will be biased towards zero. Nevertheless, they found
no evidence that higher cigarette prices reduced smoking initiation.
Douglas (1998) extended this work by estimating a time-varying covariate model
that allows the hazard of smoking initiation to respond dynamically to changes in prices
and other factors. In addition to initiation, Douglas also estimated the hazard of smok-
ing cessation in a similar empirical framework, as well as estimating the impact of
smoking regulations and information on initiation and cessation (these findings are dis-
cussed later). Using data from the cancer risk factor supplement to the 1987 National
Health Interview Survey, Douglas again concluded that cigarette price has little im-
pact on smoking initiation. As with the earlier analysis, however, there are likely to be
errors-in-valiables problems that could account for this finding.
DeCicca et al. (1998a) employed data from the National Education Longitudinal Sur-
vey of 1988 to examine the impact of price on initiation of daily smoking. This data
set contains data on youth smoking at several points in time (eighth, tenth, and twelfth
grades). Treating the three waves as independent cross-sections, they obtained estimates
of the price elasticity of youth smoking participation comparable to other recent esti-
mates. In an effort to examine the impact of price on smoking initiation, they attempted
to exploit the longitudinal aspect of their data by looking at the probability of smoking
in twelfth grade for a sample that excluded those who were smokers in eighth grade.
Their estimates for the effect of cigarette taxes on the probability of starting to smoke
between the eighth and twelfth grade are not statistically significantly different from
zero, supporting the findings of Douglas and Hariharan (1994) and Douglas (1998) that
raised doubts about the hypothesis that higher cigarette prices lead to significant reduc-
tions in youth smoking. DeCicca et al. attributed the inconsistency in their two sets of
results to the possibility that cigarette tax rates are a proxy for unobserved sentiment
against cigarette smoking. If true, then estimates based on cross-sectional studies are
likely to significantly overstate the impact of price on smoking.
Dee and Evans (1998) reexamined the longitudinal data used by DeCicca et al., argu-
ing that their finding that price has no impact on smoking initiation was largely the result
of the way in which their sample was constructed. In particular, rather than following
DeCicca et al. in deleting the large number of observations with missing values for key
independent variables (including income, parental education, and number of siblings),
Dee and Evans included these along with dummy variables indicating observations for
which the data are missing. In addition, they included a variety of binary indicators
for categorically collected data, rather than constructing "continuous" measures from
these data as did DeCicca et al. (e.g., parental and family attributes). After making these
changes but otherwise following the same basic approach, Dee and Evans estimated a
negative and significant impact of cigarette taxes on smoking initiation. Their estimated
price elasticity of smoking onset is -0.63, consistent with several other recent studies
of youth smoking employing cross-sectional data.
In response to Dee and Evans (1998), DeCicca and his colleagues (1998b) conducted
a reanalysis of the NELS data that used an alternative approach for dealing with the
missing data problem. Where possible, they used information from the longitudinal
sample to fill in missing values; when this could not be done, they used a conditional
mean imputation approach. Their reanalysis produced somewhat more significant esti-
mates for the effect of cigarette taxes on the onset of daily smoking, with implied price
elasticities from alternative specifications ranging from -0.025 to -0.505; somewhat
smaller, less significant estimates were obtained from models using price rather than
tax. In addition, their estimates for samples based on race/ethnicity implied that higher
cigarette taxes significantly reduced smoking onset among Hispanics, but had little im-
pact on whites and blacks.
Clearly, the use of longitudinal data to examine the impact of cigarette tax and price
changes on smoking initiation and cessation is an important advance. The findings from
studies using relatively longer panels that control for unobserved state and/or individual
factors affecting demand [i.e., Evans and Huang (1998), Tauras and Chaloupka (1999)]
are consistent with the findings that price sensitivity is inversely related to age, as found
in several earlier studies based on cross-sectional data. The inconsistent findings from
a few recent studies [DeCicca et al. (1998a, 1998b), Dee and Evans (1998)] directly
addressing the effects of price on smoking initiation with a relatively short panel should
be viewed with caution.
Hu et al. (1995a) introduced an innovation in cigarette demand estimation, using data
from California's Behavioral Risk Factor Surveys for 1985 through 1991 to examine
the possible effects on adult smoking of the interdependence of cigarette smoking with
other risk factors, including alcohol use and obesity. Estimates of the smoking participa-
tion elasticity from models that included other behavioral risk factors were significantly
lower than when these factors were ignored, while conditional demand elasticities were
generally unaffected. Using two-part methods, Hu et al. estimated an overall price elas-
ticity of -0.46 from the models that included other risks, with the effects of price about
equally divided between smoking participation and conditional demand. The authors
noted, however, that their estimate of the price elasticity might be relatively high given
that they did not control for other tobacco control efforts.
Evans and Farrelly (1998) recently examined a phenomenon not previously studied
by economists. Using data from the 1979 Smoking Supplement and the 1987 Cancer
Control Supplement to the National Health Interview Surveys, the authors investigated
the compensating behavior by smokers in response to tax and price changes. The sup-
plements contain unique information on smokers' choices of types of cigarettes, which
Evans and Farrelly combined with data from the Federal Trade Commission on the tar
and nicotine content of cigarette brands to construct a variety of measures of daily smok-
ing intensity (including cigarette consumption, total length of cigarettes consumed, tar
intake, and nicotine intake). They also constructed comparable aggregate measures for
1964-1993 from the data used by Harris (1994) on aggregate smoking prevalence and
cigarette consumption. They found consistent evidence that, although smokers reduced
daily cigarette consumption in response to higher taxes, they also compensated in sev-
eral ways. In particular, smokers in high-tax states consumed longer cigarettes and those
that are higher in tar and nicotine, with young adults smokers also most likely to engage
in this compensating behavior. As a result, they argued that the perceived health bene-
fits associated with higher cigarette taxes are likely to be somewhat overstated. Given
this compensating behavior, Evans and Farrelly suggest that if cigarette taxes are to be
used to reduce the health consequences of smoking, then taxes based on tar and nicotine
content would be appropriate, an idea first suggested by Harris (1980).
2.2. Addiction models and cigarettedemand
The first discussion by an economist of the effects of addiction on demand can be found
in Marshall's (1920) Principlesof Economics, where he observed that
Whether a commodity conforms to the law of diminishing or increasing return,
the increase in consumption arising from a fall in price is gradual; and, further,
habits which have once grown up around the use of a commodity while its price
is low are not so quickly abandoned when its price rises again. (Appendix H,
Section 3, p. 807)
As Phlips (1983) noted, Marshall's statement clearly introduced the three basic dimen-
sions of addiction [U.S. Department of Health and Human Services (1988)] of gradual
adaptation (tolerance), irreversibility (withdrawal), and positive effects of habits (rein-
forcement) that are used in many of the more recent formal models of addictive be-
havior. Until recently, however, economists have either ignored the addictive nature of
goods such as cigarettes when estimating demand or have assumed that behaviors such
as smoking were irrational and could not be analyzed in the rational, constrained utility
maximizing framework of economics.
Many of the most recent studies of cigarette demand explicitly address the addictive
nature of cigarette smoking. Economic models of addiction can be divided into three
basic groups: imperfectly rational models of addictive behavior, models of myopic ad-
dictive behavior, and models of rational addictive behavior.
2.2.1. Imperfectly rationaladdiction models
Elster (1979), McKenzie (1979), Winston (1980), and Schelling (1978, 1980, 1984a,
1984b) best exemplify the economic models of imperfectly rational addictive behavior.
These models generally assume stable but inconsistent short-run and long-run prefer-
ences. This is seen, for example, in Schelling's (1978) description of a smoker trying to
"kick the habit":
Everybody behaves like two people, one who wants clean lungs and long life and
another who adores tobacco. ... The two are in a continual contest for control; the
"straight" one often in command most of the time, but the wayward one needing
only to get occasional control to spoil the other's best laid plan. (p. 290)
Thus, the farsighted personality may enroll in a smoking cessation program, only to be
undone by the shortsighted personality's relapse in a weak moment. Winston (1980)
formally modeled this behavior and described how this contest between personalities
leads to the evolution of what he called "anti-markets", which he defined as firms or
institutions that individuals will pay to help them stop consuming.
Strotz (1956) was the first to develop a formal model of such behavior, describing
the constrained utility maximization process as one in which an individual chooses a
future consumption path that maximizes current utility, but later in life changes this
plan "even though his original expectations of future desires and means of consumption
are verified" (p. 165). This inconsistency between current and future preferences only
arises when a non-exponential discount function is used.7 Strotz went on to suggest
that rational persons will recognize this inconsistency and plan accordingly, by pre-
committing their future behavior or by modifying consumption plans to be consistent
with future preferences when unable to pre-commit. Pollak (1968) went one step further,
arguing that an individual may behave naively even when using an exponential discount
function. Thaler and Shefrin (1981) described the problem similarly, referring to an
individual at any point in time as both a "farsighted planner and a myopic doer" (p. 392),
with the two in continual conflict. While these models present interesting discussions of
some aspects of addictive behavior, they have not been applied empirically to cigarette
smoking or other addictions.
2.2.2. Myopic addiction models
The naive behavior described in some of the imperfectly rational models of addiction
is the basis for many of the myopic models of addictive behavior. As Pollak (1975)
observed, behavior is naive in the sense that an individual recognizes the dependence
of current addictive consumption decisions on past consumption, but then ignores the
impact of current and past choices on future consumption decisions when making cur-
rent choices. Many of these models treat preferences as endogenous, allowing tastes
to change over time in response to past consumption [Gorman (1967), Pollak (1970,
1976, 1978), von Weizsacker (1971), Hammond (1976a, 1976b), El-Safty (1976a,
1976b)].
7 Vuchinich and Simpson (1999) provided an interesting application of this idea to the demand for alcoholic
beverages, comparing behavior under hyperbolic versus exponential discounting.
These models are similar in spirit to those in which tastes change in response to
factors other than past consumption, including advertising [Dixit and Norman (1978),
Galbraith (1958, 1972)] and prices [Pollak (1977)]. Others allow past consumption to
affect current consumption through an accumulated stock of past consumption [e.g.,
Houthakker and Taylor (1966, 1970)]. These models are comparable to those of the
demand for durable consumer goods that use a stock adjustment process [e.g., Chow's
(1960) model of the demand for automobiles, and Garcia dos Santos' (1972) analysis
of the demands for household durables]. As Phlips (1983) noted, however, the distinc-
tion between models with endogenous tastes and those with stable preferences within a
household production framework is purely semantic, since the underlying mathematics
of the two are the same.
The earliest theoretical models of demand in the context of myopic addiction can
be traced to the irreversible demand models [Haavelmo (1944), Duesenberry (1949),
Modigliani (1949), Farrell (1952)]. Farrell, for example, described an irreversible de-
mand function as one in which current demand depends on all past price and income
combinations. As a result, price and income elasticities are constant, but may differ for
increases and decreases in price and income. Farrell tested this model empirically, using
U.K. data on the demands for tobacco and beer from 1870 through 1938, in a model that
included not only current price and income, but also price, income, and consumption in
the prior year. In general, his estimates were inconclusive, although he did find limited
evidence of habit formation for tobacco use.
The notion of asymmetric responses to price and income reappeared in Scitovsky
(1976) and was applied to cigarette demand by Young (1983) and Pekurinen (1989),
using data from the U.S. and Finland, respectively. Both found that smoking was al-
most twice as responsive to price reductions as it was to price increases, which they
interpreted as evidence of addiction.
Most empirical applications of myopic models of addiction are based on the pioneer-
ing work by Houthakker and Taylor (1966, 1970) that formally introduced the depen-
dence of current consumption on past consumption by modeling current demand as a
function of a "stock of habits" representing the depreciated sum of all past consump-
tion. Houthakker and Taylor estimated demand functions for a variety of goods, in-
cluding cigarettes, using annual aggregates for the U.S. and several Western European
countries. Their estimates provided considerable support for their hypothesis of habit
formation in demand for almost all of the non-durable consumer goods they examined,
including cigarettes.
Mullahy (1985) took a similar approach in his empirical examination of cigarette de-
mand using individual level data from the 1979 National Health Interview Survey. In
his model, the stock of past cigarette consumption has a negative impact on the produc-
tion of commodities such as health and the satisfaction received from current smoking.
Mullahy used a two-part model to estimate cigarette demand, as well as instrumental
variables methods to account for the unobserved individual heterogeneity likely to be
correlated with the stock of past consumption. Mullahy found strong support for the
hypothesis that cigarette smoking is an addictive behavior, as shown by the positive and
significant estimates he obtained for the addictive stock in both the smoking participa-
tion and conditional demand equations. His estimates for price are quite similar to those
obtained by Lewit and Coate (1982), with the overall price elasticity of demand centered
on -0.47. In addition, Mullahy estimated that men were more price responsive than
women (total price elasticities of -0.56 and -0.39, respectively). Finally, using an in-
teraction between the addictive stock and price, Mullahy concluded that more-addicted
smokers (defined as those with a larger addictive stock) were less responsive to price
than their less-addicted counterparts. Other approaches to estimating myopic demand
models have similarly concluded that cigarette smoking is an addictive behavior and
that price has a significant impact on cigarette demand [e.g., Jones (1989), Baltagi and
Levin (1986)].
2.2.3. Rationaladdiction models
Several researchers have modeled addiction as a rational behavior. In this context, ra-
tionality simply implies that individuals incorporate the interdependence between past,
current, and future consumption into their utility maximization process. This is in con-
trast to the assumption, implicit in myopic models of addictive behavior, that future
implications are ignored when making current decisions. In other words, myopic be-
havior implies an infinite discounting of the future, while rational behavior implies that
future implications are considered, while not ruling out a relatively high discount rate.
Several of the rational addiction models, including those of Lluch (1974), Spinnewyn
(1981), and Boyer (1983), assume that tastes are endogenous. These models build on
the significant contributions of Ryder and Heal (1973), Boyer (1978), and others in the
optimal growth literature who have developed endogenous taste models with rational
behavior. Spinnewyn (1981) and Phlips and Spinnewyn (1982) argued that incorporat-
ing rational decision making into models of habit formation results in models that are
"formally equivalent to models without habit formation" [Spinnewyn (1981, p. 92)].
Thus, they argue, assuming rationality only leads to unnecessary complications.
This assertion was challenged by Pashardes (1986) who derived demand equations
for a rational consumer in which current consumption is determined by past consump-
tion and current preferences with full knowledge about the impact of current decisions
on the future costs of consumption. Pashardes found considerable empirical evidence to
support the hypothesis of rational behavior in general, as well as evidence that cigarette
smoking is an addictive behavior. Finally, he noted that expectations concerning the
future price and other costs of consumption played an important role in consumer be-
havior.
Becker and Murphy (1988) similarly rejected the notion that myopic behavior is em-
pirically indistinguishable from rational behavior in their theory of rational addiction.
They assumed that individuals consistently maximize utility over their life cycle, taking
into account the future consequences of their choices. In their model, utility at any point
in time depends on current addictive consumption, current non-addictive consumption,
and the stock of past addictive consumption. Tolerance is incorporated by assuming
that the marginal utility of the addictive stock is negative. Reinforcement is modeled
by assuming that an increase in the addictive stock raises the marginal utility of current
addictive consumption. Finally withdrawal is captured since total utility falls with the
cessation of addictive consumption.
Becker and Murphy (1988) and Becker et al. (1991) developed several hypotheses
from this basic model. First, addictive consumption displays "adjacent complementar-
ity"; that is, due to reinforcement, the quantities of the addictive good consumed in
different time periods are complements. As a result, current consumption of an addic-
tive good is inversely related to not only the current price of the good, but also to all past
and future prices. Consequently, the long-run effect of a permanent change in price will
exceed the short-run effect. 8 Moreover, in the Becker and Murphy model, the ratio of
the long-run to short-run price effect rises as the degree of addiction rises. In addition,
they predict that the effect of an anticipated price change will be greater than the impact
of a comparable unanticipated price change, while a permanent price change will have
a larger impact on demand than a temporary price change. Finally, price responsive-
ness varies with time preference: addicts with higher discount rates will be relatively
more responsive to changes in money price than those with lower discount rates. The
opposite will be true with respect to the effects of information concerning the future
consequences of addictive consumption. Thus, the model suggests that younger, less
educated, and lower income persons will be relatively more responsive to changes in
the money price of cigarettes, while older, more educated, and higher income persons
will be relatively more responsive to new information on the health consequences of
cigarette smoking.9
Strong adjacent complementarity, reflecting strong addiction, can lead to unstable
steady states in the Becker and Murphy model. This is a key feature of their rational
addiction theory, helping to explain the binge behavior and "cold turkey" quit behavior
observed among addicts. Furthermore, these unstable steady states imply that there will
be a bimodal distribution of consumption, again something that is observed for many
addictive goods. In addition, Becker and Murphy's model implies that temporary events,
including price reductions, peer pressure, or stressful events, can lead to permanent
addictions.
Chaloupka (1988, 1990, 1991, 1992) used data from the Second National Health and
Nutrition Examination Survey conducted in the late 1970s in the first empirical applica-
tion of the rational addiction model He found consistent evidence that cigarette smok-
ing was an addictive behavior and that smokers did not behave myopically. Chaloupka's
(1991) estimates of the long-run price elasticity of demand fell in the range from -0.27
to -0.48, larger than the elasticities obtained from conventional demand equations us-
ing the same data. In addition to estimating the rational addiction demand equations for
8 Myopic addiction models also predict that the long run price elasticity of demand will be larger than the
short run elasticity.
9 See Chaloupka (1988, 1990, 1992) or Becker et al. (1994) for a more formal discussion of these price
effects.
the full sample, Chaloupka also explored the implications of the Becker and Murphy
model with respect to the rate of time preference by estimating comparable demand
equations for subsamples based on age and educational attainment. Chaloupka's (1991)
estimates were generally consistent with the hypothesis that less educated or younger
persons behave more myopically than their more educated or older counterparts. In ad-
dition, less educated persons were more price responsive, with long-run price elasticities
ranging from -0.57 to -0.62, than were more educated persons, who were generally
unresponsive to price. Chaloupka (1990) also estimated separate demand equations for
subsamples based on gender, concluding that men behaved more myopically and were
relatively responsive to price (long-run price elasticity centered on -0.60) than women
(statistically insignificant effect of price on demand).
Similar findings were obtained by Becker et al. (1994) using aggregate, state-level
sales data for the U.S. over the period from 1955 through 1985. They found clear ev-
idence that smoking was addictive, as well as evidence of non-myopic, although not
fully rational, behavior.1 0 Estimates from other studies employing U.S. data [Keeler et
al. (1993), Sung et al. (1994)] and data from other countries, including Finland [Pekuri-
nen (1991)] and Australia [Bardsley and Olekalns (1998)], are generally consistent with
the hypothesis of rational addiction. In contrast, Duffy (1996a), Cameron (1997), and
Conniffe (1995), using annual time-series data for the U.K., Greece, and Ireland, re-
spectively, found little support for the rational addiction model. These latter studies,
however, are generally limited by the relatively small number of observations available
for their analyses, and by the use of several highly correlated regressors.
As noted above, Douglas (1998) used hazard models to examine the determinants
of smoking initiation and cessation in the context of the Becker and Murphy (1988)
rational addiction model. In contrast to his finding that price does not significantly affect
the hazard of smoking initiation, Douglas concluded that increases in price significantly
increase the likelihood (hazard) of smoking cessation. He estimated a price elasticity
for the duration of the smoking habit of -1.07 with respect to future price, consistent
with the hypothesis of rational addiction; paradoxically, past and cunrent prices were
not found to have a statistically significant effect on cessation. Similarly, his parametric
and non-parametric results imply that the hazard of smoking cessation has a positive
duration dependence, a finding Douglas suggested is consistent with rational addiction
in that the rational smoker will discount future health costs less as they become more
imminent.
2.2.4. Critiques of the rationaladdiction model
While the rational addiction model has gained acceptance among some economists,
many object to several assumptions of the model. Perhaps the most criticized aspect of
10 The authors concluded that there was insufficient information in the data to accurately estimate the discount
rate, but that their estimates were clearly inconsistent with myopic behavior.
1562 EJ.ChalouLpka and K.E. Warner
the model is the assumption of perfect foresight. As Winston (1980) explained, in the
context of the Stigler and Becker (1977) model:
[T]he addict looks strange because he sits down at period j = 0, surveys future
income, production technologies, investment/addiction functions, and consump-
tion preferences over his lifetime to period T, maximizes the discounted value of
his expected utility, and decides to be an alcoholic. That's the way he will get the
greatest satisfaction out of life. Alcoholics are alcoholics because they want to be
alcoholics, ex ante, with full knowledge of its consequences. (p. 302)
Similarly, Akerlof (1991) noted that individuals who become addicted in the rational
addiction model do not regret their past decisions, given that they are assumed to have
been fully aware of the consequences of their consumption of a potentially addictive
good when making those decisions.
A recent theoretical paper by Orphanides and Zervos (1995) addressed this and other
perceived inconsistencies of the rational addiction model that arise largely from the as-
sumption of perfect foresight. In particular, the authors introduced uncertainty into the
model by assuming that inexperienced users are not fully aware of the potential harm
associated with consuming an addictive substance. Instead, in their model, an individ-
ual's knowledge comes from the observed effects of the addictive good on others as
well as through his or her own experimentation with that good. More specifically, they
assume that the harmful effects (including addiction) of consuming a potentially addic-
tive good are not the same for all individuals, that each individual possesses a subjective
understanding of his or her potential to become addicted, and that this subjective belief
is updated via a Bayesian learning process as the individual consumes the addictive
good. Thus, an individual who underestimates his or her potential for addiction and ex-
periments with an addictive substance can end up becoming addicted. Rather than the
"happy addicts" implied by the rational addiction model [Winston (1980)], these ad-
dicts will regret becoming addicted. As Orphanides and Zervos noted, the incorporation
of subjective beliefs into the rational addiction model helps explain youthful experi-
mentation, the importance of peer influences, and other commonly observed facets of
addiction.
More recently, in a model focusing on cigarette smoking, Suranovic et al. (1999) also
reconsidered the Becker and Murphy (1988) model of rational addiction. As described
above, adjacent complementarity is a key feature of the rational addiction model. Sura-
novic et al. noted, however, that one implication of adjacent complementarity is that
efforts to reduce current consumption will lead to reductions in utility. These "quitting
costs" are an important feature of their model and help explain the seeming inconsis-
tency between smokers' stated wishes to quit smoking and their continued cigarette
consumption. In addition, they help explain why smokers engage in various behavior
modification treatments, such as the use of the nicotine patch, which help make quitting
easier.
A second point of departure from the Becker and Murphy model concerns the tim-
ing of the consequences of smoking, which Suranovic et al. assume are concentrated at
the end of a smoker's life. In addition, rather than assuming that individuals choose
a lifetime consumption path that maximizes the present value of their lifetime util-
ity, Suranovic et al. assume "boundedly rational" behavior, implying that individuals
choose current consumption only. As a result, their model suggests that aging is enough
to induce cessation among some smokers. As in the Becker and Murphy model, their
model implies that quitting "cold-turkey" is likely in the case of a strong addiction (one
where quitting costs rise rapidly for small reductions in consumption). However, in
contrast to Becker and Murphy, Suranovic et al. predicted gradual reductions in con-
sumption progressing to quitting in the case of relatively weak addictions. Interest-
ingly, some newly emerging epidemiologic evidence supports this prediction [Farkas
(1998)].
In addition, as Becker and Mulligan (1997) describe, addiction and time preference
may be related. As discussed above, the Becker and Murphy (1988) model of rational
addiction implied that people who discount the future more heavily were more likely
to become addicted. In their theoretical discussion on the determination of time pref-
erence, Becker and Mulligan suggest that addictive consumption, by raising current
utility at the expense of future utility, can make even rational persons behave more my-
opically.
Finally, Showalter (1998), in his analysis of the behavior of firms producing an ad-
dictive good, suggests an alternative interpretation for the finding in most empirical
applications of the rational addiction model that future consumption has a significant
impact on current consumption. Rather than resulting from rational behavior on the part
of consumers, Showalter shows that the same finding could result from myopic behav-
ior by consumers coupled with rational behavior by firms. In his empirical applications
of this model, Showalter finds that the rational and myopic demand models produce
similar predictions, but that neither does well in predicting actual behavior, a finding he
attributes to the difficulties of accurately forecasting prices.
2.3. Behavioraleconomic analyses of cigarette demand
Behavioral economics involves the application of the principles of consumer demand

theory to experimental psychology [Hursh and Bauman (1987)]. Over the past decade,
there have been numerous behavioral economic analyses of a variety of addictive be-
haviors, including cigarette smoking [Bickel and DeGrandpre (1996)]. These studies
examine the impact of price and other factors on the self-administration of a number of
addictive substances by humans as well as a variety of non-human species in a labora-
tory setting. Price, in this context, is defined as the response or effort required to receive
one dose of a drug [Bickel et al. (1993)]. As in standard economic analyses, an increase
in price is expected to lead to a reduction in the quantity of drug demanded. One advan-
tage of this experimental approach for the analysis of cigarette demand, both in general
and as it relates to policy debates specifically, is that it allows researchers to study the
effects on demand of changes in cigarette prices that are many times larger than the
price differences that are observed in the cross-sectional or time-series data that have
been used in the econometric studies of demand. One limitation of the approach, how-
ever, is that these methods are generally applicable only to dependent individuals. For
example, for ethical reasons (and others), they cannot be used to address issues related
to the effect of price on smoking initiation.
The behavioral economics of cigarette smoking is the most extensively researched
area in the behavioral economics of drug abuse [Bickel and Madden (1999)]. In a series
of papers, Bickel, DeGrandpre, and their colleagues have reported the results of research
on cigarette smoking conducted in their behavioral economics laboratory [Bickel et
al. (1991), DeGrandpre et al. (1992), DeGranpre et al. (1994), Bickel et al. (1995),
DeGrandpre and Bickel (1995), Bickel and DeGrandpre (1996), Bickel and Madden
(1999)]. These experiments typically involve individuals ages 18 and older who smoke
a pack or more of cigarettes per day who participate in between two and five three-hour
experimental sessions per week. l Price, in these experiments, is defined as the number
of complete pulls and resets of a plunger required to receive a preset number of puffs on
a cigarette. For example, 50 pulls on the plunger may be required to obtain two puffs on
a cigarette. Puffs are monitored by a puff-volume sensor so that each subject receives
essentially the same dose per puff [Bickel and Madden (1999)].
A wide range of prices is used in these experiments. In some of the experiments,
respondents were also presented with an opportunity to earn money for pulls on the
plunger that could then be spent on cigarettes. As in the econometric studies described
above, the behavioral economic analyses have consistently found an inverse relation-
ship between cigarette smoking and price. Estimates of the price elasticity of demand
obtained from these studies are surprisingly consistent with those obtained from econo-
metric studies. For example, Bickel et al. (1995) estimated a mean price elasticity of de-
mand of -0.56 for five subjects in an experiment in which price ranged from 12 to 1600
pulls per puff. A particularly interesting finding from the behavioral economics research
is that the price elasticity of demand rises as price rises. For example, DeGrandpre and
Bickel (1995) estimated a mean price elasticity of -1.58 for prices ranging from 400 to
4500 pulls per puff. These findings appear to be generalizable not only across drugs but
also across species [Bickel et al. (1990)].
2.4. Econometric studies of the demandfor other tobacco products
In contrast to the relatively large literature examining the impact of cigarette prices on
cigarette smoking, few studies look at the effects of price on the use of other tobacco
products, and fewer still consider cross-price effects for cigarettes and other tobacco
products. Much of this research has been conducted by Ohsfeldt and his colleagues
[Ohsfeldt and Boyle (1994), Ohsfeldt et al. (1997, 1999)]. Using state-level aggregates
constructed from the September 1985 tobacco use supplement to the Current Popula-
tion Survey, Ohsfeldt and Boyle (1994) examined the impact of state smokeless tobacco
1 For a discussion of a number of other requirements for the participants and more detail on the features of
these experiments, see Bickel and Madden (1999).
taxes and cigarette excise taxes on the prevalence of smokeless tobacco use by males
ages 16 years and older. The authors concluded that higher smokeless tobacco taxes
would significantly reduce the prevalence of smokeless tobacco use. In addition, Ohs-
feldt and Boyle found evidence of substitution among tobacco products, in that higher
cigarette excise taxes have a positive and significant effect on the prevalence of smoke-
less tobacco use. Given this finding, they suggested that the increase in smokeless to-
bacco use observed among young males in the 1980s was at least in part due to the
increases in state cigarette taxes which were rising more rapidly during this time than
state taxes on other tobacco products.
Similarly, Thompson and McLeod (1976) and Pekurinen (1989, 1991) concluded
that some Canadian and Finnish cigarette smokers, respectively, would switch from
manufactured cigarettes to less expensive hand-rolled cigarettes in response to increases
in the prices of manufactured cigarettes. Pekurinen also found a negative and significant
relationship between the demands for pipe tobacco and cigars and their own-prices. Leu
(1984), however, found little evidence of substitution among tobacco products by Swiss
tobacco users in response to changes in their relative prices.
The findings obtained by Ohsfeldt and Boyle based on aggregate data are confirmed
by their subsequent analyses using individual-level data from the September 1985 CPS
[Ohsfeldt et al. (1997)] and the September 1992, January 1993, and May 1993 CPS
Ohsfeldt et al. (1999)]. In the more recent analysis, the authors estimated an own-tax
elasticity for smokeless tobacco use of -0.10 for their sample of males ages 16 and
older and again concluded that smokeless tobacco products are substitutes for cigarettes.
In addition, as seen for cigarette demand, they estimated an inverse relationship between
the elasticity of demand for smokeless tobacco products and age.
This finding was confirmed by Chaloupka et al.'s (1997) recent analysis of smokeless
tobacco use among young males using data from the 1992, 1993, and 1994 Monitoring
the Future surveys of eighth, tenth, and twelfth grade students. The researchers esti-
mated an overall price elasticity of young males' smokeless tobacco use of -0.746, and
a participation elasticity of -0.523.
3. Cigarette and other tobacco taxation12
With retail price an important determinant of the demand for cigarettes, and excise tax
often a significant component of retail cigarette price, the issue of whether to increase
cigarette excise taxes has been highly visible in legislative debates on both governmen-
tal revenue raising and tobacco control for decades. Cigarette excise taxation offers an
unusual attraction for legislators: given the evidence on demand elasticities, increases
12 As most of the evidence and concern pertaining to tobacco taxation relates to cigarette excise taxation, we
refer specifically to cigarette taxation in most of the discussion in this section. However, we do present the
evidence pertaining to the taxation of other tobacco products where it applies.
in cigarette taxes of politically plausible magnitude will produce a public health bene-
fit, by discouraging smoking, particularly among children. At the same time, they will
generate additional revenues for the governmental unit in question, typically at a fairly
low administrative cost. Further, polls often find support for cigarette excise increases
among American voters, frequently even including smokers.
Still, the prospect of increased taxation raises a myriad of complicated philosophi-
cal and practical questions. Among the former are the following: What is the "right"
level of cigarette taxation, if any? What is the basis for determining that it is "right"?
Are cigarette taxes fair, given their distributional burdens in terms of both vertical
and horizontal equity? In particular, since more low-income than high-income peo-
ple smoke (in developed countries), will increased cigarette taxes impose an unfair
regressive burden on low-income taxpayers? What are the proper trade-offs between
the interests of individuals (liberty interests, tax burden) and the societal interest in
the public's health? Practical questions include the following: Given the oligopolistic
nature of the cigarette industry, as well as estimated supply and demand elasticities,
how will taxation affect cigarette price? With differences in tax rates by jurisdiction
defining much of the difference in prices across borders of states and nations, will a
given increase in excise tax in a relatively high-tax jurisdiction result in a significant
amount of smuggling from a neighboring low-tax jurisdiction? What are the revenue
implications of a tax increase of given magnitude? How will a given tax-induced price
increase influence smoking, and consequently, what impacts will it have on the public's
health?
Economists have made numerous important conceptual and empirical contributions
to the policy debate on cigarette taxation, primarily, although not exclusively, through
their evaluation of the relationship between cigarette price and consumption, reviewed
in the preceding section. In this section, we review economists' contributions to better
understanding the rationale for (or against), and additional effects of, cigarette excise
taxation.
3.1. Comparative standardsand the effects of tax on price
3.1.1. Purposesand methods of taxation
Cigarettes and other tobacco products have been taxed for centuries, primarily because
the relatively inelastic demand for these products makes them an easy source of rev-
enues. In the U.S., for example, tobacco has been taxed since colonial times, rising with
revenue needs and declining during more prosperous times. Since the Civil War, to-
bacco taxes have remained a part of the U.S. federal tax system, often increasing during
wartime and falling again in peacetime. Similar historical patterns are observed in many
other developed countries. However, the importance of tobacco taxes as a share of total
revenues has generally declined over time in most countries. In the U.S. for example,
tobacco taxes currently account for less than one-half of one percent of total federal
revenues, down from 3.36 percent of revenues in 1950.
In recent decades, the increased taxation of cigarettes and other tobacco products
has been motivated not only by the revenue generating potential of these taxes, but
also as a means to reduce cigarette consumption. Warner (1981b) concluded that the
information on the health consequences of cigarette smoking that began appearing in
the 1950s and early 1960s led a number of states to increase cigarette taxes as a way
of discouraging cigarette demand. More recently, a number of countries have adopted
or considered large tax increases on cigarettes and other tobacco products as a way
to reduce consumption. In Canada, for example, steady federal and provincial tobacco
tax increases throughout the 1980s and early- 1990s were motivated in part by concerns
over the health consequences of smoking. Similar concerns were behind the recent large
increases in cigarette taxes in California, Massachusetts, Arizona, Oregon, and other
U.S. states.
Large tax increases can generate both significant declines in cigarette smoking as
well as considerable increases in cigarette tax revenues. In Canada, for example, the
over 500 percent increase in taxes between 1982 and 1992 led to an increase in real
cigarette prices of 170 percent, reducing total cigarette smoking by 38 percent [Sweanor
and Martial (1994)]. Total federal and provincial cigarette tax revenues rose by 240
percent during this period, even with the development of a significant black market in
cigarettes.
Cigarettes and other tobacco products are taxed in a variety of ways. The most com-
monly used methods of taxation include specific excise taxes, value added and other
ad-valorem taxes, and import duties. Most cigarette excise taxes are specified as an
amount per x number of cigarettes (e.g., the U.S., Canada, and many others), while
others are based on the weight of tobacco contained in the cigarette (e.g., Australia and
Malaysia). Similarly, there are a variety of ad-valorem taxes, including the value added
taxes imposed by most European and many other countries, as well as the sales taxes
applied in most U.S. states and elsewhere. There are comparable differences in the types
of import duties applied by nearly all countries to tobacco products. Some of the distinc-
tive features of these taxes include: earmarking for tobacco-related education, counter-
advertising, and other health related activities (e.g., Finland, Denmark, Peru, Romania,
Nepal, and several U.S. states); the use of tax revenues to create the state-run Health
Promotion Foundations in several Australian states and the Health Sponsorship Council
in New Zealand, to fund sporting and artistic events previously backed by the tobacco
industry; and the differential taxes on cigarettes with high tar and nicotine content used
in previous years in the U.K. [WHO (1997), Roemer (1993)].
3.1.2. Effects of taxes on retailprice
Increases in cigarette and other tobacco taxes result in higher prices for these products.
When specific taxation is the primary form of tobacco taxation, however, the real value
of the tax will fall over time, unless regularly increased to account for inflation. Given
that taxes are an important component of price, one consequence of an excise tax sys-
tem for tobacco products with relatively infrequent tax increases is that the real price
of these products will fall over time as the prices of other goods and services increase
more rapidly. In the U.S., for example, due to the relative stability of federal and state
cigarette excise taxes throughout the 1970s, real cigarette prices fell by nearly 40 per-
cent. Between 1981 and 1996, however, real cigarette prices in the U.S. rose by over 65
percent, due in part to the tripling of the federal cigarette excise tax and numerous state
tax increases. 13 In contrast, the real value of an ad-valorem tax on tobacco products is
maintained when the prices of these products rise with the prices of other goods and
services.
The oligopolistic nature of the cigarette industry and the addictive nature of cigarette
demand have important implications for the effects of cigarette tax increases on cigarette
prices. In a perfectly competitive market with constant long-run costs of production, any
tax increase would be fully passed on to consumers. At the other extreme, a monopolist
would share the burden of the tax increase with consumers, with consumers bearing
relatively more of the burden when demand is relatively inelastic. In most developed
countries, the cigarette industry, however, is clearly at neither extreme, but is instead an
oligopoly. In the U.S., for example, the five leading cigarette producers accounted for
virtually the entire cigarette market, with the top three (Philip Morris, R.J. Reynolds,
and Brown & Williamson) controlling over 90 percent of the market [Federal Trade
Commission (1997)]. In some countries, however, particularly developing countries, a
domestic monopoly controls most of the market.
Most of the empirical analyses of the relationship between cigarette taxes and prices
are based on data from the U.S. Early studies produced generally inconsistent findings
[Barzel (1976), Johnson (1978), Sumner (1981), Sumner and Ward (1981), Bulow and
Pfleiderer (1983), Bishop and Yoo (1985), Sullivan (1985), Sumner and Wohlgenant
(1985), Ashenfelter and Sullivan (1987)]. One general weakness of these studies is that
they failed to account for the dynamic interaction of firms in an oligopolistic industry.
Instead, they generally assumed that the rules for firm behavior were established and
then worked backwards to estimate the degree of competition in the industry [Harris
(1987)].
More recent studies have attempted to model the dynamic nature of an oligopolistic
industry when estimating the impact of cigarette taxes on cigarette prices. Harris (1987)
used data on wholesale and retail cigarette prices as well as manufacturing costs to
estimate the change in cigarette prices that resulted from the doubling of the U.S. federal
cigarette tax in 1983. He concluded that the eight-cent tax increase led to a 17-cent price
increase that was not explained by increased manufacturing costs. Instead, Harris argued
that the scheduled tax increase served as a mechanism for a coordinated, oligopolistic
price increase.
Barnett and his colleagues (1995) noted that Harris' analysis did not fully account for
underlying trends in cigarette prices. Consequently, they argued that Harris attributed
13 Increases in the non-tax component of price, however, account for most of the rise in U.S. cigarette prices
between 1981 and 1996. During this period, cigarette taxes as a percentage of price fell from just over one-
third to under one-quarter.
too much of the increase in price to the increase in the tax since the upward trend in
cigarette prices predated the debate over the federal tax increase. Instead, they argued
that the introduction of generic cigarettes in 1981 allowed cigarette producers to engage
in coordinated increases in the prices of premium cigarettes since the generic cigarettes
would keep more price-sensitive smokers in the market.
Keeler and his colleagues [Keeler et al. (1996), Barnett et al. (1995), Sung et al.
(1994)] used national and state level data in empirical analyses of the effects of tax
increases on price. Their models account for the interaction of supply and demand,
the oligopolistic nature of the cigarette industry, and, in some, the addictive nature of
cigarette demand. Using annual state-level data for the period from 1960 through 1990,
Keeler et al. (1996) estimated that a one-cent increase in a state's cigarette tax would
raise retail prices in that state by 1.11 cents. In addition, they estimated that increases
in federal cigarette taxes would generate larger increases in cigarette prices than those
resulting from state tax hikes [Barnett et al. (1995)]. They attributed this finding to the
potential of cross-border shopping for cigarettes in response to a state tax increase. Fi-
nally, Keeler et al. (1996) concluded that cigarette producers price discriminate by state,
in that stronger state and local anti-smoking laws are offset by lower prices. However,
they noted that the effect of price discrimination is not large relative to retail cigarette
prices.
Based on the Becker and Murphy (1988) rational addiction model, Becker et al.
(1994) suggested an alternative explanation for the finding that cigarette prices increase
by more than cigarette taxes when taxes are raised. They argued that when taxes are
raised, cigarette companies will raise price by more to obtain maximum profit from
current, addicted smokers. These increased current profits help offset the future losses
from the reduced smoking initiation resulting from the price increase. Becker and his
colleagues explained this apparent paradox as follows: "If smokers are addicted and if
the industry is oligopolistic, an expected rise in future taxes and hence in future prices
induces a rise in current prices even though current demand falls when future prices are
expected to increase" (p. 413). They went on to explain that because of the addictive na-
ture of smoking, cigarette producers set prices below their short-run profit maximizing
level in order to "hook" consumers on their addictive product, thus raising the future
demand for this product. Showalter (1998) makes a similar argument with respect to
advertising, suggesting that cigarette producers might engage in apparently excessive
advertising in order to attract a few new customers.
3.1.3. Variations in cigarette tax across countriesand states and the issue of
smuggling
The share of cigarette taxes in cigarette prices varies widely among countries. In Den-
mark, Ireland, and the U.K., for example, over 80 percent of cigarette prices are ac-
counted for by cigarette taxes [Sweanor (1997)], while taxes in most others were be-
tween 65 and 80 percent.1 4 In contrast, among developed countries, cigarette taxes are
14 Based on prices and taxes as of December 31, 1996.

less than half of price only in the U.S., where they account for 35 percent of price, on
average (21 percent in the lowest taxing state and 47 percent in the highest). The large
difference in cigarette taxes leads to a five-fold difference in cigarette prices among
these countries. As the WHO (1997) has observed, inter-country differences in wages
and prices can understate the difference in price when expressed in a single currency.
When expressed in terms of minutes of labor required to earn the price of a pack of
cigarettes, the differences are even larger.
One consequence of the differences in cigarette taxes and prices, both across coun-
tries as well as among different taxing jurisdictions within countries, is the potential
for casual and organized cigarette smuggling and other forms of tax evasion. The
cigarette industry, for example, frequently argues that cigarette tax increases will ac-
tually lead to reductions in tax revenues due to smuggling and other tax evasion
[British American Tobacco (1994)]. The smuggling problem is exacerbated by the
relative ease with which tobacco products can be transported, the potential profits
from this illegal activity, the presence of an informal distribution network in many
countries, the availability of tax-free and duty-free cigarettes, and nonexistent or rel-
atively weak policies concerning cigarette smuggling and their lack of enforcement
[Joossens and van der Merwe (1997), Joossens and Raw (1995, 1998), Advisory Com-
mission on IntergovernmentalRelations (ACIR) (1977, 1985)]. Joosens and Raw (1995,
1998) argued that many of these other factors can be as important as price differ-
ences in spawning cigarette smuggling. For example, they noted that there is lit-
tle evidence of cigarette smuggling in some of the highest priced European coun-
tries, including France, Norway, Sweden and the U.K., while there is extensive ev-
idence of smuggling in those with relatively low prices, such as Spain and Italy.
Moreover, they concluded that much of the cigarette smuggling that does occur in
Europe and elsewhere is actually encouraged by the large, multinational tobacco
companies. Thursby and Thursby (1994) provided empirical support for this argu-
ment, based on their analysis of data from the U.S. from which they concluded
that increases in federal cigarette excise taxes lead to increased commercial cigarette
smuggling.
There have been relatively few econometric analyses of the impact of price differ-
entials on organized and casual cigarette smuggling. All of these studies are based on
annual state-level cigarette sales data from the U.S. and all have concluded that the
casual and organized smuggling of cigarettes from major tobacco producing states,
as well as other states with relatively low cigarette prices compared to neighboring
states, accounts for a significant share of sales in these states [Saba et al. (1995), Becker
et al. (1994), Chaloupka and Saffer (1992), Baltagi and Levin (1986), ACIR (1977,
1985), Manchester (1976)].
Perhaps the most widely cited example of the link between cigarette tax increases and
smuggling is the Canadian experience during the late-1980s and early-1990s. In 1980,
when Canada adopted an ad-valorem approach to cigarette taxation, Canadian cigarette
prices were somewhat higher than prices in the U.S. By 1984, the gap had widened
as Canadian cigarette taxes doubled and real cigarette prices rose by 25 percent. In
1984, in response to industry pressure, the ad-valorem tax was replaced by an excise
tax. Over the next few years, growth in Canadian cigarette taxes slowed, with most tax
increases taking place at the provincial level. In 1988, however, the federal government
mounted an aggressive anti-smoking campaign that included significant tax increases.
In 1989, the federal tax was raised by two cents per cigarette, followed by a three cents
per cigarette increase in 1991; provincial taxes continued to increase as well. By early
1994, the average Canadian tax per pack was $2.96 (in U.S. dollars), more than five
times the U.S. average [Sweanor and Martial (1994)].
The large tax and price disparities between the U.S. and Canada led to substantial
cigarette smuggling from the U.S. Smuggling was a relatively minor problem prior
to 1992; however, beginning in 1992, smuggling rapidly increased after the repeal of
a Canadian tax on cigarette exports. In addition, the smuggling problem was exacer-
bated by the long undefended border between the U.S. and Canada, relatively weak
border controls, and the high concentration of the Canadian population near U.S. bor-
ders [Sweanor and Martial (1994)]. Much of the black market trade was in cigarettes
originally produced in Canada, exported to the U.S. tax-free, and then smuggled back
into Canada; relatively little involved U.S. produced cigarettes given their use of a blend
of tobacco different from that preferred by Canadian smokers.
In response to an aggressive industry-sponsored campaign, the Canadian federal
cigarette tax was reduced by $5.00 per carton on February 9, 1994, with an agreement
to match provincial tax reductions of up to another $10.00 per carton. Quebec quickly
lowered its provincial tax by $11.00 per carton, for a total tax cut of $26.00 per carton,
cutting cigarette prices in half. Several other provinces followed and by the end of 1996,
the average tax per pack was less than $2.00. Canadian tax revenues fell and rates of
smoking increased, particularly among youth.
The variations in taxes across countries and within countries over time reflect a myr-
iad of practical and political considerations, with smuggling but one of them. In contrast
to the legislators who must set taxes based on such considerations, academic economists
approach the issue of the desirable level of cigarette taxation by contemplating the ap-
plication of a number of economic principles relating to both equity and efficiency. The
remainder of Section 3 examines these principles and their relevance to the determina-
tion of a theoretically optimal cigarette excise tax. The discussion also compares and
contrasts the perspective of the economist with that of the public health professional,
for whom different criteria define "optimality" in cigarette taxation.
3.2. Fairnessstandards
The search for an optimal tax encompasses considerations of efficiency and equity. Each
of these domains has featured prominently in policy debates on increasing cigarette
taxes as well. From an efficiency point of view, the principal economic theory argument
favoring imposition of a product-specific (excise) tax relates to the creation of negative
externalities through production or consumption of the product. The nature and extent of
such externalities with regard to smoking have been the subject of considerable debate
among economists, as is discussed later in this section. 15,16
First, however, we examine a central issue in cigarette taxation, namely whether it
violates widely accepted standards of fairness, with an emphasis on vertical equity.
In developed countries such as the U.S., proportionately more lower-income people
smoke than do those with high incomes. As a consequence, the burden of a tax on
cigarettes is experienced disproportionately by the poor. The tax is criticized as being
highly regressive.
3.2.1. Horizontaland vertical equity
In terms of tax policy, the principle of horizontal equity is that equals should be treated
equally. Clearly, cigarette taxation violates this principle, if one accepts that people
who are identical except for their smoking behavior should be deemed "equals". Ar-
guments in favor of cigarette taxation thus ignore this principle, while opponents of
taxation appeal directly to it (although rarely in the language of economists). Violation
of the horizontal equity principle has never been the focal point of critics' concern,
however. Rather, they have focused on questions of vertical equity, specifically the ap-
parent regressivity of cigarette taxes. Cigarette taxes would be regressive with respect
to income if poorer and more affluent consumers smoked at the same rate. The poten-
tial problem of regressivity is exacerbated, in many developed countries at least, by the
above-mentioned tendency for smoking prevalence to be inversely related to income.
Recent empirical analysis has muted this concern somewhat, concluding that the de-
gree of regressivity is substantially less than appears at face value. Using data from the
1984-1985 Consumer Expenditure Survey, analysts at the Congressional Budget Of-
fice (1990) found that expenditures on tobacco products increased with income, except
15Less frequently debated is a pragmatic consideration: whether cigarette taxation violates the Ramsey Rule,
namely that when dealing with consumption taxes, tax rates should vary inversely with the elasticity of de-
mand for products (holding supply elasticity constant). The purpose is to ensure that revenue-raising occurs
in a manner that will minimize distortions in consumers' choices among goods and services and in their de-
cisions of how much to spend and how much to save. Until recently, the empirical evidence has suggested
that cigarettes are an excellent target for taxation, consistent with the Ramsey Rule, given the consensus view
on their relatively low demand elasticities, discussed in the preceding section. Jones and Posnett (1988) esti-
mated that a 1% increase in the cigarette tax rate would generate about a 0.9% increase in revenue. As noted
in the preceding section, however, long-run demand elasticities may be as much as twice short-run elasticities,
as reflected in the findings associated with application of the rational addiction model to cigarette smoking
[Becker et al. (1994)]. Use of these greater long-run elasticities would suggest a lower level of tax efficiency.
[6 The issues raised in this discussion would apply also to many other behaviors and consumption goods,
such as consuming large quantities of fat (a risk factor for heart disease and cancer) or driving motorcycles (a
risk factor for serious injury). Despite the logical parallels, few economists or public health professionals have
advocated a fat tax or a special injury-related tax on motorcycles. Cigarettes (and alcohol; see the chapter by
Cook and Moore) have been identified as unique in terms of the magnitude of the health damage and negative
externalities they create. The "slippery slope" argument - once negative externalities are used to justify taxing
tobacco and alcohol, cars will be next, then fat, then salt, etc. - is not addressed in this chapter.
for those in the highest income quintile. In part this reflects an increasing intensity of
smoking (numbers of cigarettes) among smokers as income rises, and a propensity for
higher-income smokers to buy products that are more expensive per unit. As a percent-
age of post-tax income, however, tobacco spending varied inversely with income level,
with the lowest quintile spending 4% on tobacco.
When the CBO analysts examined tobacco spending as a percentage of expenditures
on all good and services, however, they found that the share of tobacco expenditures fell
gradually over the first four income quintiles (from 1.6 to 1.1%) and dropped sharply in
the top quintile (to 0.7%). Consequently, CBO concluded, if annual family expenditures
are more reflective of lifetime income than annual family income, then tobacco expen-
ditures are only slightly regressive over income classes. CBO also noted that younger
families spend a higher proportion of their income on tobacco and that their share of
tobacco spending as a percentage of total expenditures was higher as well.
To examine the distributional impact of a cigarette excise tax increase, CBO simu-
lated the effects of doubling the then (1990) 16-cent federal per pack excise tax. When
income tax brackets and transfer payments were indexed to account for the price in-
creases associated with the excise tax increase, the apparent regressivity was reversed;
the adjustments had the effect of lowering individual income taxes and raising transfer
payments. Relative to expenditures, CBO found the burden of the increased tax to be
closer to proportional than regressive. The largest share of the simulated tax increase
was paid for by families in the third and fourth income quintiles, with the smallest share
borne by families in the two lowest quintiles (first and second).
To control for the intertemporal nature of cigarette smoking, Lyon and Schwab (1995)
examined the distributional effects of cigarette (and other "sin") taxes across measures
of permanent or lifetime income. The authors did not find important differences in re-
gressivity patterns compared to studies based on current income.
Recent research on differences in the price elasticity of demand for cigarettes by var-
ious measures of socioeconomic status has produced findings that suggest that the de-
gree of regressivity normally attributed to cigarette taxation is considerably overstated.
Townsend and colleagues (1994) found that (the absolute value) of price elasticity was
inversely related to social class in Great Britain, with members of the highest social
class exhibiting little price responsiveness and those in the lowest social class having
an elasticity close to unity. In the U.S., Chaloupka (1991) concluded that less-educated
persons were more price responsive than the more-educated, while Farrelly and his col-
leagues (1998) found that cigarette demand by lower income persons was more elastic
than that by higher income persons. Given the high correlation between income and
both social class in Great Britain and education in the U.S., these studies indicate that
increased cigarette taxes would reduce observed differences in smoking among socioe-
conomic groups. This mitigates conclusions about regressivity that derive from analyses
that have failed to consider the inverse relationship between elasticity and income. The
latter has characterized all studies to date.
Regardless of whether regressivity proves to be a serious concern or not with regard
to cigarette taxation per se, analysts have pointed out that the goal of tax policy is for the
overall impact of a tax and expenditure system to be progressive or proportional, rather

than regressive [Congressional Budget Office (1990), Warner et al. (1995)]. Related to
this notion, various proposals to increase cigarette taxes have earmarked a portion of
revenues to expenditures designed to aid the poorer members of society. These have
ranged from smoking-specific expenditures (e.g., provision of smoking cessation ser-
vices for the poor) to proposals to fund health care for the poor in part with cigarette tax
revenues.
3.2.2. The benefitprinciple
Another perspective on fairness is that smokers should bear the costs of smoking that
they impose on other members of the society. This consists of two categories of costs:
those associated with diseases experienced by nonsmokers due to exposure to environ-
mental tobacco smoke; and smokers' own publicly-funded medical costs subsidized by
nonsmoking taxpayers. Related to the latter, excise taxation might be partially justified
on the basis of the benefit principle, the notion that people who derive benefits from
government activities should be taxed to cover their costs of production.
In the most direct application of this "user fee" concept, proponents argue that smok-
ers should pay, through cigarette taxes, for the publicly-funded health care that smoking
necessitates, primarily in the U.S. through the Medicaid program for the indigent and
the Medicare program for the elderly. Although appealing at first blush, the logic under-
lying this argument can be challenged. First, the tax is a blunt instrument: many smokers
who incur smoking-related health care expenditures will pay for them themselves, ei-
ther out of pocket or through private health insurance. Why should they subsidize the
health care costs of other smokers more than do nonsmokers? Similarly, many of today's
smokers, who pay the excise taxes, will cease smoking in time to avoid smoking-related
illnesses. Further, the costs that today's generation of smokers will experience in the fu-
ture are not easily predicted; perhaps some currently expensive smoking-related disease
conditions will be readily and inexpensively treatable in the future. Even if these costs
were knowable, they would need to be discounted to reflect the fact that they will not
occur, on average, for two or more decades. As it has been advanced to date, the "user
fee" argument has today's smokers paying for today's smoking-related health care costs
[Warner et al. (1995)].
In the case of smoking, the benefit principle is inextricably linked to the broader issue
of the negative externalities associated with smoking. We consider the evidence per-
taining to externalities below. First, however, we turn to fairness arguments in favor of
increasing cigarette taxes that emanate from the public health community. Economists'
analyses of the external costs of smoking are highly relevant to informing the public
health community's sense of the social costs of smoking.
3.3. Public health standards
The public health community has advocated large increases in tobacco taxes for two
reasons. One is the notion that smokers should cover the social costs of smoking, with
the public health conception of social costs including both private costs to smokers and
their families, as well as negative externalities [Cook (1991)]. The second is based on
a pragmatic realization: through its effects on prices, taxation will discourage many
people from smoking, particularly young people. As a consequence, literally hundreds
of thousands of premature deaths could be avoided by large increases in cigarette taxes.
In this context, cigarette taxation is viewed as a powerful policy tool with which to
foster improvements in the public's health.
3.3.1. The social cost of smoking
The public health community has long argued that smoking imposes large costs on so-
ciety and that smokers should bear the burden of these costs. Cost-of-smoking analyses
include three categories of costs: (i) the direct medical costs of preventing, diagnos-
ing, and treating smoking-related diseases; (ii) the indirect morbidity costs associated
with lost earnings from work attributable to smoking; 17 and (iii) the indirect mortality
costs related to the loss of future earnings due to premature smoking-produced deaths.
Combined, these total well over $100 billion in the U.S. 18 Although most of the cost-
of-smoking analysis has employed American data, estimates have also been prepared
for Canada [Forbes and Thompson (1983), Collishaw and Myers (1984)], Great Britain
[Atkinson and Townsend (1977)], China [Jin et al. (1995)], and other countries. In ad-
dition, numerous state-specific analyses have been performed in the U.S., most based
on the SAMMEC model (Smoking-Attributable Morbidity, Mortality, and Economic
Costs) [Shultz et al. (1991)].
The cost-of-smoking studies have employed a variety of methods of estimating
the different cost components, with attributable-risk methodology common in esti-
mating smoking-related disease incidence or prevalence and the human capital ap-
proach employed in placing a value on lost years of life [Hodgson and Meiners
(1982), Warner et al. (1999)]. Following a trend in cost-of-illness estimation in gen-
eral, more recent studies have adopted an incidence approach [Manning et al. (1989,
1991), Hay (1991), Hodgson (1988, 1992), Oster et al. (1984)], in contract with the
prevalence approach predominant in the earlier studies [e.g., Rice et al. (1986)]. The
prevalence approach values the present costs associated with all existing cases of
smoking-produced illness (including future lost earnings attributable to current deaths).
In contrast, the incidence approach values all of the future costs associated with new
cases of smoking-produced disease during the reference year. The former provides an
estimate of the current economic burden of smoking, while the latter is more useful
17 A significant amount of work loss is associated with smoking [U.S. Department of Health and Human
Services (1989)1. In addition, smoking may decrease productivity while smokers are on the job, due primarily
to the number and length of smoking breaks they take. The latter is rarely included in studies of the social
costs of smoking, although it has been raised during debates on indoor smoking restriction legislation.
18 Authors' calculations, updating estimates in Bartlett et al. (1994) and Miller et al. (1998) to contemporary
dollars.
1576 F:J.Chaloupka and K.E. Warner
for analyses of interventions that might interrupt the development of smoking-related

illness.
The cost-of-smoking studies can be faulted on numerous grounds. For one, they have
tended to omit certain categories of smoking-related health care, such as treatment of
bum victims from smoking-ignited fires19 and perinatal care for low-birth-weight ba-
bies of smoking mothers [U.S. Department of Health and Human Services (1998)]. Few
studies have considered the costs of treatment of diseases caused by environmental to-
bacco smoking. No study has yet attempted to value intangible costs, such as the pain
and suffering of smoking-related disease victims and their families. Ironically, these
intangible costs may well dominate all of those that are more readily quantified [Abt
(1975)].
Hodgson (1998) and his colleagues [Warner et al. (1999)] have argued that, as it
is being applied in this literature, the conventional attributable risk methodology con-
siderably underestimates smoking's burden on the health care system by virtue of its
failure to consider how smoking complicates the course of many illnesses not them-
selves directly associated with smoking. For example, diabetics who smoke often have
more complications of their diabetes than diabetics who do not smoke. Smokers re-
cover more slowly from surgeries of all types than do nonsmokers, thus extending post-
surgical hospital stays. HIV-infected smokers may be more likely to develop near-term
AIDS than are HIV-infected nonsmokers [Napier (1996)]. Inclusion of such costs in
cost-of-smoking analysis might lead to an increase in estimates of 50% or more.
The cost-of-smoking studies also fail to take account of a wide variety of direct costs
in addition to medical costs, including such items as the time and transportation costs
associated with getting patients to and through health care services, the direct costs of
home modifications to accommodate smoking-related disabilities, damage to buildings
due to smoking-produced fires, smoking-related maintenance costs in industrial settings
and homes, and the increased frequency of laundering necessitated by smoking. Omis-
sion of these nonmedical costs is standard practice in nearly all of the broader cost-of-
illness literature. Occasionally, such omissions are acknowledged, with the researchers
stating that they did not view such costs as large enough to warrant the additional re-
search.
The indirect morbidity and mortality costs have been criticized frequently as repre-
senting an inadequate means of valuing the avoidable premature loss of life. By relying
exclusively on the human capital approach, they place no value on life per se, clearly
contrary to the behavior of virtually all civilized societies which invest in preserving the
health of senior citizens after they retire. Many analysts concur that use of willingness-
to-pay measures would be preferable, but they bemoan the lack of suitable instruments
for assessing such values.
The cost-of-smoking analytical community includes individuals who have calculated
some of the economic "benefits" of smoking, including the reduction in Social Security
19 Cigarette smoking is the leading cause of home fires and the leading cause of bum deaths [Napier (1996)].
payments for smokers who die prematurely [Shoven et al. (1989)] and medical expendi-
tures avoided as a result of smokers' premature demise. The latter, in particular, has set
off a "battle of the studies" to ascertain whether the net medical expenditures associated
with smoking are positive or negative [Leu and Schaub (1983), Manning et al. (1989),
Hodgson (1992), Barendregt et al. (1997), Warner et al. (1999)]. The entire question
of whether such "negative costs", or cost offsets, should be included in the calculation
of smoking's social costs has become a major front in the academic battle over defi-
nition of the social costs of smoking. Viscusi (1995), for example, recently concluded
that consideration of medical and pension offsets makes the net social costs of smoking
small, if positive at all. Harris countered that in no other area of social policy analysis
is death treated as an economic benefit [Coalition on Smoking or Health (1994), Harris
(1993)].
The import of this intellectual debate is potentially substantial. At the center of the
public health community's advocacy of higher cigarette taxes is the social cost argument
that smokers (or the industry that feeds their addiction) are imposing a huge economic
burden on the society and ought to pay for it through higher excise taxes. Using the
public health construction of social cost, some analysts have concluded that in the U.S.,
the cigarette excise tax needs to be on the order of $3-4 or more to cover these costs
[e.g., Hay (1991)].
Economists of many political stripes have countered that, for purposes of estimat-
ing an optimal cigarette excise tax, the correct notion of social cost is the traditional
economist's measure of externalities, i.e., costs imposed by smokers on others, exclud-
ing their own family members. Economists' contributions to this debate are considered
below, following a brief discussion of the true heart of the public health case for higher
taxes: the health benefits that would result.
3.3.2. The health benefits of increasing cigarettetaxes
Through a variety of channels, the economics literature on tax, price, and demand has
reached the public health community [Scott and Dickert (1993), Coalition on Smoking
or Health (1994)]. Given the strength of the evidence linking price increases to demand
decreases, with the consensus that price elasticity is inversely related to age, the public
health community has become convinced that cigarette tax increases are one of the
most effective policy tools for decreasing smoking, especially among children. As a
consequence, increasing price, generally through a tax hike, is featured in nearly every
comprehensive tobacco control policy proposal.
The raison d'etre underlying the public health community's desire to see smoking
decline is to reduce the morbidity and disability and premature mortality associated
with smoking. Economists have taken the demand elasticity evidence and combined it
with data on the health consequences of quitting smoking (primarily adults) and not
starting (primarily children) to project the health gains that would be achieved with tax
increases of various magnitudes. For example, in the mid-1980s, the U.S. Congress had
to decide whether to permit a scheduled "sun-setting" of 8 cents of the then 16-cent
cigarette excise tax (increased from 8 cents in 1983). Warner (1986b) used the price
elasticity estimates of Lewit et al. (1981) for children and Lewit and Coate (1982) for
adults to estimate the consumption implications of permitting the sun-setting to occur
or, instead, increasing the tax by 8 or 16 cents per pack. He concluded that if the tax
were doubled to 32 cents, and the real value of the tax maintained thereafter, 800,000
youths would be deterred from starting to smoke and 2.7 million adults encouraged to
quit. Applying the conservative assumption that one of every four lifetime smokers dies
prematurely of a smoking-related illness (the proportion is now believed to be about
half), the analysis estimated that this tax increase would eventually reduce premature
deaths in persons then 12 years and older by 860,000.
Harris (1987) also evaluated the consumption and health implications of the 1983
doubling of the federal excise tax, considering the implications of various elasticity
estimates. He concluded that the tax increase had likely deterred 600,000 youths from
smoking and that, as a consequence of their avoidance of cigarettes and adults quitting,
54,000 of the youths and an additional 100,000 adults would survive to at least 65 years
of age.
The General Accounting Office (1989) employed the same elasticity estimates used
by Warner (1986b) and the same assumptions about premature mortality avoided to
evaluate the likely health benefits from a sustained real 21-cent federal tax increase in
1989, which they estimated would increase retail price by 15%. They predicted a fur-
ther reduction in youth smoking of 500,000, with a subsequent reduction in premature
mortality among these youths of 125,000.
Recently, Moore (1996) developed a more sophisticated econometric model that, in-
corporating state-level data on death rates from smoking-related diseases from 1954
through 1988, could be used to evaluate the impact of higher taxes on mortality. He de-
termined that a 10% increase in cigarette excise taxes would save approximately 5,200
lives each year. Similarly, Evans and Ringel (1999) examined whether or not higher state
cigarette taxes can be used to improve birth outcomes. Using data on approximately 10.5
million births in the U.S. over the period from 1989 through 1992, the authors estimated
a smoking prevalence elasticity of -0.5 for pregnant women and found that increased
cigarette taxes would significantly raise birth weight.
With the help of economic consultants, the Coalition on Smoking or Health (1994)
used relatively conservative estimates of price elasticity and of the mortality conse-
quences of smoking to estimate the health implications of alternative tax hikes advo-
cated by the Coalition's member organizations. The Coalition determined that a 75-cent
tax increase in 1992 would reduce premature deaths due to smoking by 900,000. A $2
per pack increase was estimated to save 1 million more lives than the 75-cent increase.
Chaloupka (1998) did the same for the price increases included in many of the recent
proposals for national tobacco legislation in the U.S. Based on Chaloupka and Gross-
man's (1996) estimates, Chaloupka estimated that a $1.50 increase in cigarette taxes and
prices, phased in quickly and maintained in real terms, would reduce overall cigarette
consumption by about 30 percent while cutting youth smoking prevalence almost in
half. Based on the CDC's (1996) estimates for the number of youth in the 1995 U.S.
cohort of 0 through 17 year olds who would eventually die prematurely from a smoking
related illness, Chaloupka estimated that this tax increase would prevent approximately
2.5 million deaths in this cohort.
To most members of the public health community, the health benefits of a tax in-
crease justify its imposition. However, public health professionals appeal to the social
cost argument to garner public and, especially, legislative support. Given the conceptual
as well as empirical problems with the public health community's construction of the
social cost of smoking, discussed above, the question remains as to whether economists
would find theoretical justification for increasing the cigarette tax in the analyses of ex-
ternal costs that have been performed to date, or for that matter in other considerations.
The next section addresses the economic efficiency issues and evidence.
3.4. Economic efficiency and the pursuit of an optimal cigarette tax
Most economists would concur that an economically optimal tax on cigarettes would
equate the revenues generated with the net external costs produced by smoking. 20 Here
we review the evidence pertinent to determining such a tax and evaluate additional con-
siderations that relate to the notion of optimality when considering taxation of an addic-
tive substance such as cigarettes.
3.4.1. Negative externalities associatedwith smoking
As discussed above, the public health community's definition of social costs incorpo-
rates both negative externalities and private costs. While economists agree that the latter
should not be considered as social costs in contemplating a corrective tax on cigarettes,
there is no complete consensus on precisely what consequences warrant inclusion, and
even for those for which there is consensus, estimates of the magnitude of the true
social externalities vary widely. Moreover, other tobacco control policies, particularly
restrictions on smoking in public places and private worksites, may be more efficient
approaches to dealing with some of these externalities.
One author found that including the costs of the long-term intellectual and physical
consequences of smoking-related low-birth-weight disabilities implied a tax of $4.80
per pack [Hay (1991)]. In contrast, other studies have found much smaller per-pack neg-
ative externalities, often less than existing excise tax rates. For example, evaluating data
from the RAND Health Insurance Experiment and the 1983 National Health Interview
Survey in an incidence-based cost analysis, Manning et al. (1989) concluded that, for
their mid-range estimates, the negative externalities of smoking totaled the equivalent of
20 Pigou (1962) suggested that, for goods with market prices less than their social costs, taxes could be used
to raise the marginal cost of consuming the good to the social marginal cost. For some goods, taxes could
generate revenues that exceed total external costs, reflecting the fact that the taxes are based on marginal
rather than average external costs [Cook and Moore (1993)].
1580 FEJ.Chaloupka and K.E. Warner
43 cents per pack (in 1986 dollars). Partially offsetting these negative externalities, how-
ever, were an estimated 27 cents in "external savings" resulting from smoking-related
premature deaths, meaning that the net negative externalities equaled 16 cents. The re-
searchers thus concluded that the empirical evidence did not justify raising the cigarette
tax on grounds of covering negative externalities. 21 In a later analysis that drew on the
Manning et al. study, Viscusi (1995) came to the same conclusion.
The Manning et al. study and Viscusi's reanalysis has been cited frequently by op-
ponents of a cigarette tax increase. Representing the Tobacco Institute at a Senate hear-
ing, Tollison (1994) identified the work by Manning and his colleagues, as well as that
of other prominent health economists, as rejecting the propriety of an economically-
motivated tax increase. The Congressional Research Service [Gravelle and Zimmerman
(1994)] cited the same study in the CRS's evaluation of the grounds for a tax increase
(which the authors found wanting). Updating the figures to 1995 dollars, the CRS esti-
mated the net negative externalities at 33 cents per pack of cigarettes, two-thirds of the
average 50 cents in federal and state taxes imposed on cigarettes in late 1993.
Although the Manning et al. study has dominated attention within the economics de-
bate about the marginal social costs of smoking, the study reveals several problems of
both omission and commission, many of which the authors have acknowledged [Man-
ning et al. (1991)]. Adjustments reflecting these issues often lead to qualitatively differ-
ent conclusions about the desirability of increasing the cigarette excise tax. In terms of
omission, most notably the study excluded a variety of costs associated with environ-
mental tobacco smoke (ETS) that, if included, would significantly increase the social
cost estimate. Many of the health consequences of ETS were not well appreciated when
Manning et al. undertook their research in the mid-1980s (e.g., the consequences of ETS
for heart disease). However, the authors made the decision to consider then-known ETS
costs as internal, based on the traditional economic assumption that the family is the
appropriate economic unit for consumption decisions, including the decision to smoke.
21 Early analysis of the offsetting savings associated with smokers' premature deaths is found in the work of
Leu and Schaub (1983). These authors estimated the lifetime medical expenditures of a cohort of Swiss males,
which included both smokers and nonsmokers, and compared them with the simulated expenditures of a hy-
pothetical cohort assumed to include only nonsmokers. The authors concluded that the lifetime expenditures
would be very similar for both cohorts, with the higher annual costs of smokers in the "real" cohort offset by
the additional years of medical expenditures in the longer-lived hypothetical no-smoker cohort. In estimating
the medical costs of nonsmokers in the hypothetical cohort, Leu and Schaub recognized that those who would
have been smokers in the "real" cohort would differ in ways other than just smoking fi-om those who would be
nonsmokers in both cohorts. They introduced the notion of the "non-smoker smoker type" as the conceptually
correct entity to evaluate in the hypothetical cohort, for those members who would have been smokers in
the "real" cohort. This useful distinction has been incorporated in the work of several economists since then,
including Manning et al. (1989). Implicitly, it is embodied in all attempts to evaluate smoking-related health
care costs in which smokers' other risk-taking behaviors are controlled [e.g., Bartlett et al. (1994), Miller et
al. (1998)].
Further, the authors reasoned, the adverse health consequences of smoking were largely
confined to the nonsmoking spouses of smokers. 2 2
Although few economists would challenge the sanctity of the family as the basic eco-
nomic unit, the assertion about the internal nature of ETS costs is less clear. Certainly,
some ETS costs are external to the family (e.g., airline attendants' ETS-induced lung
disease prior to the banning of smoking on flights). Others likely represent a mix of in-
ternal and external costs. For example, disease and developmental problems associated
with low birth-weight caused by mothers' smoking during pregnancy often have support
costs that spill over into the broader society, as social institutions are required to pick
up some of the medical, institutional, and other costs related to these conditions. Thus,
regardless of one's philosophical approach to the issue of intra-family health problems
caused by ETS, determining the appropriate distribution of costs between family and
the rest of society is a distinct challenge.
The potential role of ETS costs in reevaluating the net negative externalities associ-
ated with smoking is seen by considering the following figures. Manning et al. (1991)
noted that inclusion of the costs of 2,400 lung cancers from ETS (a fairly conservative
estimate of this toll [Environmental Protection Agency (1992)]) as external costs would
add approximately 19 cents per pack in external costs (updated to 1994 dollars). In addi-
tion, inclusion of the costs of neonatal care for smoking-related low-birth-weight babies
would add 3 cents to the total, while including fetal deaths attributable to smoking would
add yet another 19 cents. Deaths from smoking-related fires would add a further 9 cents.
The ETS costs would sky-rocket if one included the estimated 30,000-60,000 heart dis-
ease deaths recently associated with ETS [Glantz and Parmley (1995)], adding perhaps
70 cents to the total social costs per pack. Similarly, inclusion of the smoking-induced
respiratory tract infections and cases of aggravated asthma in children [Environmen-
tal Protection Agency (1992)] would boost the total further, as would inclusion of the
long-term developmental disabilities in smoking-related low-birth-weight babies [Hay
(1991)]. All told, the social costs per pack could easily mount toward several dollars if
all of the health hazards associated with ETS are real, many are treated as external to
the basic consuming unit, and if all or even a significant fraction of the associated costs
are included.
ETS cost estimation is also influenced substantially by whether one employs the hu-
man capital approach or willingness-to-pay to value the lives of persons who die pre-
maturely due to ETS exposure. Manning et al. used a conservative estimate of $1.66
million per premature death based on the range of estimates in the literature. 2 3 Using
22 At the time this analysis was undertaken, virtually all of the medical literature indicting ETS as a cause of
disease related to the experience of nonsmoking wives of smoking husbands [U.S. Department of Health and
Human Services (1986), Environmental Protection Agency (1992)]; there was no significant evidence of ETS
exposure or disease effects outside of the home.
23 As discussed above, the authors' estimates of external costs attributable to ETS were negligible, making
the choice of the cost of a premature death of little consequence to their calculations [Manning et al. (1989)].
However, had they included the full range of health consequences now attributed to ETS, the choice of a
value-of-life measure would have been of much greater importance.
willingness-to-pay, the Environmental Protection Agency (1994) obtained a $4.8 mil-

lion figure. In its analysis (which included the adverse effects of ETS on heart disease
and children's health, but not on fetal and perinatal health), the EPA estimated that
the benefits that would result from a ban on smoking in all worksites would total be-
tween $39 billion and $71 billion annually, the equivalent of from $2.45 to $4.45 per
pack of cigarettes. The EPA used fairly conservative assumptions concerning the ef-
fects of a worksite smoking ban on smoking and exposure to ETS. They assumed a ban
would reduce the number of current smokers by 3-6%, the number of future smokers
by 5-10%, and daily consumption among continuing smokers by 10-15%, for a total
long-run reduction in cigarette consumption of 14-22%. The EPA estimated that these
consumption reductions would reduce out-of-home ETS exposure by 90% and in-home
exposure by 6.4%. An earlier EPA report (1992) had concluded that an estimated 73%
of ETS exposure occurs outside the home. The EPA thus predicted that a worksite ban
would reduce total ETS exposure by 66.4%.
Quite independent of the treatment of the effects of ETS, the Manning et al. (1989)
study and "spin-offs", such as Gravelle and Zimmerman's (1994) review of the evidence
and Viscusi's (1995) reanalysis, raise the issue of which types of costs ought to be
included in calculations of external costs. Manning and colleagues' finding that current
taxes covered external costs would have been reversed had the authors not included
the value of pensions and Social Security benefits not realized by smokers by virtue of
their premature demise. As demonstrated by Shoven et al. (1989), smokers subsidize
nonsmokers' Social Security benefits by virtue of the smokers' early average ages of
death; the same relationship should hold for defined benefit pensions as well. Some
analysts have considered this a transfer, not subject to consideration as an externality,
rather than a compensating external benefit of smoking, as Manning et al. treated it.
However, Manning et al. did not consider it a transfer because they viewed the length of
life of smokers as endogenous. As the literature on the costs of smoking demonstrates
repeatedly, inclusion or exclusion of such costs can play a significant role in calculating
net external costs.2 4
The importance and complexity of the handling of such matters is illustrated by
their differential effects in different societies and at different times. For example, as
just noted, in the U.S. analysts have concluded that, by dying early, smokers subsidize
nonsmokers' Social Security payments [Shoven et al. (1989), Viscusi (1995)]. In the
United Kingdom, by contrast, research indicated that sickness benefits paid to smokers
and pensions paid to their dependents compensated for the lower direct pension ben-
efits paid to smokers as a consequence of their earlier average age of death [Atkinson
24 Other candidates for inclusion are the reduction in income taxes and insurance premiums paid by smokers
due to reduced earnings associated with smoking-related illnesses; smoking-related health care costs paid by
public insurance plans (and conceivably private, depending on how these are treated); and increased sick pay
and disability benefits paid during smoking-related illnesses. If such items are included, care must be taken
to consider both reduced payments by smokers into public revenues and altered patterns of consumption of
government-financed goods and services.
and Townsend (1977)]. In developing countries in which old-age expenses are largely a
private matter, the social "benefit" of smokers' dying early would not exist, and hence
would not offset any negative externalities of smoking. Of course, this could change
over time if and as such societies developed social security plans. Similarly, the magni-
tude of the offsets in the developed countries could change if and as benefit programs
in those countries were altered [Warner et al. (1995)].
As this discussion has demonstrated, calculation of the "true" net negative external-
ities associated with smoking is an exceedingly difficult challenge, one that involves
conceptual questions, epidemiologic and other data considerations, and "moving tar-
gets" in terms of both knowledge and institutional structures. The relevance of the task
to understanding optimal cigarette taxation recommends further research, despite its
difficulty.2 5
3.4.2. Other efficiency considerations
Several factors related to smoking complicate the task of defining an optimal cigarette
tax. Two essential realities about smoking - namely, that it is a behavior initiated al-
most exclusively during childhood [U.S. Department of Health and Human Services
(1994)] and that it is addictive [U.S. Department of Health and Human Services (1988)]
- give pause in treating cigarette consumption just like any other rational economic
behavior. A third reality - that many smokers are not truly well informed about the haz-
ards of smoking [U.S. Department of Health and Human Services (1989), Schoenbaum
(1997)]26 - also challenges a basic proposition of rational behavior. For smoking as for
other consumption, rational economic behavior presumes both the existence of adequate
knowledge on which to base consumption decisions and rational use of the knowledge.
In the absence of adequate knowledge, higher taxes might be justified [Cordes et al.
(1990)]. One may be particularly interested in applying this concept to teenagers, al-
though increasing taxes is a decidedly blunt instrument if its purpose is solely to better
"inform" youths about the risks of smoking [Warner et al. (1995)].
As discussed in an earlier section, the empirical applications of the rational addic-
tion model suggest that addicted adult smokers do not behave myopically in contem-
plating the relationship between cigarette consumption and past, present, and expected
future prices. However, as the evidence reviewed above indicated, youth exhibit much
more myopic cigarette consumption behavior than do adults, consistent with studies that
have found young smokers greatly underestimating the probability that they would still
be smoking five years later [U.S. Department of Health and Human Services (1994)].
25 An interesting example of the problems created by institutional structures in trying to assess the relevance
of ETS to determining an optimal tax lies in the effects of workplace smoking bans on the development
of ETS-related diseases: the more pervasive are workplace bans, the less ETS exposure nonsmokers will
experience, and hence the amount of ETS-related disease will decline. This, in turn, would decrease the
conceptually optimal tax.
26 For a contrary view, see Viscusi (1992).
1584 E. Chaloupka and K.E. Warner
Moreover, as illustrated by the Orphanides and Zervos (1995) model, the role of in-
formation (more specifically the lack of information on the potential for addiction) is
particularly important in the initiation process, and results in later regret.
A group of health economists who have studied the economics of smoking recently
concluded that protecting children from a future of nicotine addiction, with its as-
sociated health risks, was the most compelling reason favoring increased taxation of
cigarettes [Warner et al. (1995)]. They perceived high taxes as appropriate to balance an
environment in which children face numerous inducements to smoke, including multi-
billion dollar advertising and marketing campaigns by the cigarette companies, many
designed to attract children to smoking [U.S. Department of Health and Human Ser-
vices (1994)]. To address this imbalance, these economists supported such measures as
increased public education and increased enforcement of restrictions on youth access
to tobacco products, although the limited effectiveness of such measures is well docu-
mented (see Section 5 below). The economists observed, however, that these measures
do not address children's tendency to discount the future heavily, in a manner that, as
rational adults, they might come to regret. Taxation, they felt, is the best available pol-
icy instrument to address this problem, both conceptually and empirically. As observed
above, two recent analyses [Orphanides and Zervos (1995), Suranovic et al. (1999)] ex-
amining the initiation of smoking, continuation over time, and eventual consideration
of quitting (and difficulty in doing so) lend new insight into this issue of regret.
4. Advertising, promotion, and the demand for tobacco products
Cigarettes are one of the most heavily advertised and promoted products in the world.
In the United States, for example, the cigarette industry spent $5.1 billion on advertising
and promotion activities in 1996 [Federal Trade Commission (1998)] as a percentage of
sales, these expenditures have increased dramatically since 1980. Cigarette advertising
includes the more traditional advertising on television, radio, and billboards, in news-
papers, magazines, and transit facilities, and, most recently, on the internet. Spending
on promotion includes a wide variety of activities, including promotional allowances
to retailers, point-of-purchase promotional materials, direct mail advertising, the dis-
tribution of free samples, coupons, and specialty items, multiple pack promotions, and
retail value-added offers, as well as endorsements, sponsorship of cultural, sporting,
and other entertainment events, and sponsorship of community and other organizations.
Nearly 87% of all cigarette advertising and promotional expenditures in the U.S. in 1974
were devoted to traditional advertising; by 1996, in striking contrast, this had fallen to
just over 10%, with the balance going to the less-traditional promotional activities. Pro-
motional allowances ($2.15 billion in 1996) and coupons and retail value added ($1.31
billion in 1996) have been the largest spending categories in recent years.
This section begins with a brief discussion of the arguments related to cigarette ad-
vertising and demand, as well as a review of some of the economic issues related
to cigarette advertising. We then examine the econometric literature on the impact of
cigarette advertising on smoking, consider the limitations of the econometric approach,

and review the growing literature on the impact of restrictions and bans on cigarette
advertising and promotion. The section closes with a short description of the findings
on cigarette advertising and demand from the non-economics literature.
4.1. Theoretical and conceptual issues
The impact of cigarette advertising on cigarette smoking, particularly youth smoking,

has been the subject of extensive debate over the past several decades. The public health
community takes it as given that advertising encourages smoking and is a particularly
significant influence on smoking initiation among youth. The industry, on the other
hand, contends that cigarette advertising is a form of competition that has no impact on
overall cigarette smoking, but instead simply affects market share. In addition, the in-
dustry argues that advertising provides useful information to smokers about their prod-
ucts, including information on tar and nicotine content.
Warner (1986a) suggested several mechanisms through which cigarette advertising
and promotion could affect cigarette consumption. He identified four direct mecha-
nisms: (1) advertising can entice children and young adults to experiment with smoking
and to initiate regular smoking; (2) it can reduce current smokers' willingness to quit
smoking; (3) it can serve as a cue or stimulus that leads to increased daily cigarette
consumption by smokers; and (4) it can induce former smokers to resume their habit by
reinforcing the attractions of smoking. Two indirect mechanisms were: (1) discouraging
a full discussion of the health consequences of cigarette smoking in media dependent
on tobacco advertising; and (2) contributing to a social environment in which smok-
ing is perceived to be socially acceptable. The U.S. Surgeon General [U.S. Department
of Health and Human Services (1989)] added a third indirect mechanism, namely that
institutions dependent on tobacco industry promotional and other support may create
political opposition to strong tobacco control policies.
Warner and his colleagues [Warner (1985), Warner and Goldenhar (1989), Warner
et al. (1992)] have examined the first indirect mechanism empirically, concluding that
there is strong evidence that magazines' coverage of the hazards of smoking was signif-
icantly diminished as the magazines' share of advertising revenues from cigarette ad-
vertising rises. Warner et al. (1992) found that magazines that did not carry cigarette ad-
vertising were more than 40% more likely to cover the health consequences of smoking
than those with cigarette advertising. The difference was more pronounced for women's
magazines, with those that did not advertise cigarettes more than 230 percent more
likely to cover the hazards of smoking.
4.2. Econometricevidence
Other than this work by Warner and colleagues, research by economists has not ad-
dressed the hypothesized individual mechanisms by which advertising can influence
cigarette consumption. However, beginning with Schmalensee's 1972 study, there have
been numerous econometric studies of the impact of cigarette advertising on cigarette

demand, mostly for the U.S. and the U.K., but also for several other countries as well.
No consensus concerning the effects of advertising on smoking has emerged from this
research, however. Several recent reviews of this literature, drawing on many of the
same studies, reached opposite conclusions. Based on their review of 24 studies with
estimates of the advertising elasticity of cigarette demand, Andrews and Franke (1991)
concluded "that there is a significant relationship between advertising and cigarette con-
sumption across studies, independent of study design factors. However, the estimated
magnitude of this relationship varies depending on several study design factors" (p. 96).
Moreover, they concluded that the positive impact of advertising on cigarette demand
has declined over time, a factor they attributed to the maturation of cigarette markets in
most of the countries studied. Duffy (1996b), on the other hand, concluded that econo-
metric studies of the relationship between cigarette advertising and demand generally
have found that advertising has little or no impact on aggregate cigarette consumption.
In his now classic volume, The Economics of Advertising, Schmalensee (1972) in-
troduced several concepts that reemerge repeatedly in subsequent studies of cigarette
advertising and demand. In particular, he examined cigarette demand at both the in-
dustry and firm levels, allowed for the possibility that a firm's cigarette advertising
expenditures might be based on current sales, and modeled not only the effects of cur-
rent advertising expenditures on demand but also the impact of lagged expenditures.
As Schmalensee observed, failing to account for the potential endogeneity of cigarette
advertising expenditures would lead to biased estimates of the impact of advertising on
demand. Similarly, failing to account for the cumulative or "stock" effects of advertising
could lead to an omitted variables problem, although the evidence is mixed concerning
the durability of cigarette advertising [Boyd and Seldon (1990)]. Schneider et al. (1981)
extended the use of the "stock" of advertising to allow the marginal productivity of
advertising expenditures in the U.S. to fall after the 1971 ban on broadcast cigarette
advertising.
An additional issue raised by Schmalensee relates to the measurement of advertising
expenditures. In half of his industry-level models, he used the absolute level of cigarette
advertising, while in the other half, he used the ratio of cigarette advertising to total
advertising, providing an intuitive argument favoring the latter analogous to the use of
relative rather than absolute price. Subsequent theoretical models of consumer behavior
have made this argument more formally [Theil (1980), Duffy (1987)].
The numerous econometric studies on the impact of aggregate cigarette advertising
expenditures on aggregate cigarette consumption differ with respect to many of the is-
sues raised by Schmalensee, as well as in several other respects. Almost none follows
Schmalensee and treats advertising expenditures as endogenous. A few estimate firm-
specific demand [e.g., Roberts and Samuelson (1988)], while most examine industry
demand. Several introduce measures of past advertising expenditures; some do so di-
rectly [e.g., Goel and Morey (1995)], while others construct a cumulative "stock" of
advertising [e.g., Schneider et al. (1981)]. Most look at absolute or per capita measures
of cigarette advertising expenditures, while a few employ measures of expenditures on
cigarette advertising relative to overall advertising expenditures [e.g., Duffy (1996a)].

Many estimate single equation models of cigarette demand; several, however, estimate
simultaneous equations models of supply and demand [e.g., Porter (1986)]. Most esti-
mate conventional demand models, while some estimate myopic demand models that
include a measure of past cigarette consumption [e.g., Baltagi and Levin (1986)] or an
alternative approach [Fujii (1980)]; one estimates demand in the context of the rational
addiction model [Duffy (1996a)].
Schmalensee (1972) and many subsequent econometric studies based on aggregate
data from the U.S., U.K., and a few other countries found no statistically significant
effect of aggregate cigarette advertising expenditures on cigarette consumption [Hamil-
ton (1972), Lambin (1976), Grabowski (1976, 1978), Metra Consulting Group (1979),
Schneider et al. (1981), Johnson (1986), Baltagi and Levin (1986), Stavrinos (1987),
Tegene (1991), Duffy (1991, 1996a), Wilcox and Vacker (1992), U.K. Department of
Health (1992), Wilcox et al. (1994), Franke (1994), Goel and Morey (1995)]. A few of
these studies, however, did find some evidence that lagged cigarette advertising had a
significant positive impact on current cigarette consumption, providing some support for
the durability of advertising [Schmalensee (1972), U.K. Department of Health (1992),
Goel and Morey (1995)]. Several other studies have found that cigarette advertising has
a positive and significant impact on aggregate cigarette demand [McGuiness and Cowl-
ing (1975, 1980), Fujii (1980), Witt and Pass (1981), Reuijl (1982), Radfar (1985),
Bishop and Yoo (1985), Leeflang and Reuijl (1985), Abernethy and Teel (1986), Porter
(1986), Roberts and Samuelson (1988), Chetwynd et al. (1988), McAuliffe (1988), Kao
and Tremblay (1988), Seldon and Doroodian (1989), Seldon and Boyd (1991), Valdes
(1993), Tremblay and Tremblay (1995)]. However, the magnitude of the estimated ef-
fect is generally small. Differences in the design, data, and empirical methods account
for the inconsistent findings.
Based on the mixed evidence from the studies using aggregate data, aggregate
cigarette advertising appears to have a small or negligible impact on aggregate cigarette
sales. However, as a number of researchers have observed, the nature of competition in
the cigarette industry and the limitation of econometric analysis to estimating marginal
effects imply that there should be at most a small impact of aggregate cigarette adver-
tising expenditures on aggregate cigarette consumption [Cox (1984), Warner (1986a),
Warner et al. (1986), U.S. Department of Health and Human Services (1989), Saffer
(1995, 1998), Pollay et al. (1996)].
As Saffer (1995, 1998) has noted, advertising expenditures will have diminishing
marginal productivity. Drawing on the "advertising response functions" used to charac-
terize brand-level advertising in the empirical marketing literature, which clearly imply
diminishing returns to advertising at the brand level [Rao and Miller (1975), Ackoff and
Ernshoff (1975)], Saffer concluded that an "industry advertising response function" will
also show diminishing returns to increasing expenditures and marginal effects could be
well below average effects. Moreover, in a highly concentrated market with a virtual
absence of price competition, where the "personality" of the product is very important
to consumers, total advertising expenditures will almost certainly exceed the "rational"
1588 EJ5.Chaloupka and K.E. Warner
level associated with joint profit maximization that would be expected to show a signif-
icant positive impact on overall demand. Instead, in an effort to increase or protect mar-
ket share, firms will advertise beyond the level where one would expect to find a sizable
positive marginal effect of total advertising on total demand, assuming that the firms
do not collude in deciding upon the amount of advertising. A number of econometric
studies have looked at the impact of advertising on demand at the firm or brand level.
These generally have found that increases in advertising expenditures have a positive
and significant effect on market share [Telser (1962), Peles (1971), Schnabel (1972),
Grabowski (1978), Holak and Reddy (1986), Pollay et al. (1996)]. Pollay et al. (1996),
for example, estimated that "share of voice" (brand share of advertising expenditures)
has a significant impact on market shares, and that brand choice among teenagers is
about three times more sensitive to advertising than it is for adults (a result surprisingly
similar to the estimates for youth and adult price sensitivity).
In short, given that the econometric analyses of aggregate expenditures and consump-
tion are designed to assess the impact of a marginal change in advertising expenditures
on total cigarette sales, it is not surprising that most of these analyses estimate small
or insignificant effects of advertising on demand. In addition, critics of these analyses
suggest several methodological shortcomings, including: the lack of appropriate mea-
sures of advertising exposure and other problems with the measures of advertising em-
ployed; the failure to distinguish between the impact of advertising and promotional
activities; problems with the simultaneity between advertising expenditures and sales;
the omission of other key variables, such as concurrent counteradvertising; and more
[Cox (1984), Warner (1986a), Warner et al. (1986), U.S. Department of Health and Hu-
man Services (1989), Chapman (1989), U.K. Department of Health (1992), Luik (1994),
Saffer (1995, 1998), Duffy (1996b), Pollay et al. (1996)]. These and other critics have
suggested that more appropriate approaches include the examination of more disaggre-
gated data and the analysis of non-marginal changes in advertising expenditures, such as
those that result from significant restrictions or complete bans on cigarette advertising
and promotion.
To date, only one econometric study has examined the impact of cigarette advertising
employing individual level data and more appropriate measures of advertising exposure.
Lewit et al. (1981) used data on about 6,700 youth ages 12-17 years taken from Cycle
III of the U.S. Health Examination Survey conducted from 1966 through 1970. Based
on measures of televised cigarette advertising and counter-advertising, and self-reported
information on time spent watching television, Lewit and his colleagues estimated the
number of pro- and anti-smoking commercials each youth would have seen. Their es-
timates provide support for the hypothesis that televised pro-smoking advertisements
significantly increased youth smoking.
There are a number of studies examining the impact of restrictions and bans on
cigarette advertising on smoking. Many of the older studies look at the impact of the
U.S. ban on broadcast cigarette advertising that began January 2, 1971 [Ippolito et al.
(1979), Schneider et al. (1981), Bishop and Yoo (1985), Porter (1986), Baltagi and
Levin (1986), Kao and Tremblay (1988), McAuliffe (1988), Seldon and Dooroodian
(1989), Seldon and Boyd (1991), Simonich (1991), Franke (1994), Goel and Morey
(1995), Tremblay and Tremblay (1995)]. In general, these studies produced mixed evi-
dence on the impact of the ban on television and radio advertising. Most concluded that
the ban did not significantly reduce cigarette smoking in the U.S. A few suggested that
the marginal productivity of cigarette advertising fell after the ban [for example, Trem-
blay and Tremblay (1995)]. Several, including Hamilton (1972) and Warner (1979),
suggested that the net impact of the 1971 ban was to raise cigarette consumption be-
cause it also led to the elimination of effective anti-smoking commercials broadcast
under the Fairness Doctrine (discussed below). Schneider et al. (1981) supported this
argument empirically, concluding that the advertising ban led to a net increase of nearly
5% in per capita tobacco consumption, in part due to a price reduction resulting from the
reduced costs associated with less advertising. In addition, they argued, the advertising
ban limited the provision of information to smokers concerning the tar and nicotine con-
tent of different brands and, consequently, reduced the likelihood that smokers would
switch to lower tar and nicotine brands.
Others examined the impact of other country-specific restrictions, including: the
1965 U.K. ban on televised cigarette advertising [Atkinson and Skegg (1973), Witt and
Pass (1981)]; the ban on advertising in electronic media in Australia [Johnson (1986),
McLeod (1986)]; the Finnish extension of its television ad ban to other media [Pekuri-
nen (1989, 1991)]; and Spain's partial ban on broadcast advertising [Valdes (1993)].
Hamilton (1977) presented similar estimates from separate regressions for 11 countries
over the period 1948-1973. These studies also produced mixed evidence on the effec-
tiveness of these partial bans. In general, they suggested that the bans led to a temporary
reduction in cigarette smoking, but that they had little impact in the long run. However,
more extensive restrictions coupled with anti-smoking publicity, strong health warn-
ings, and other activities appear to have led to more permanent reductions in demand
[Pekurinen (1989, 1991)].
Still others have conducted cross-country analyses of the impact of restrictions and
bans on cigarette advertising and promotion [Hamilton (1977), Cox and Smith (1984),
Laugesen and Meads (1991), and Stewart (1993)]. These, too, have yielded mixed find-
ings. In addition to country-specific regressions, Hamilton (1977) included models pool-
ing some of the countries in his sample. As in the country-specific models, he found no
evidence that advertising restrictions reduced cigarette demand. Cox and Smith (1984)
took an indirect approach to estimating the impact of advertising bans on demand. Using
data from 15 OECD countries, they sorted countries by their use of legislative versus
voluntary strategies to reduce smoking, where limits on advertising reflected a more
legislative strategy. Based on a series of country-specific regression models, they con-
cluded that smoking declines more rapidly in countries that take a legislative approach
to tobacco control, suggesting that advertising restrictions are effective in reducing de-
mand.
Laugesen and Meads (1991) pooled annual aggregate data from 22 OECD countries
for the period 1960-1986 in their examination of the impact of advertising and promo-
tion restrictions. Rather than focusing on a specific type of restriction (e.g., a broadcast
1590 .EJ. Chaloupka and K.E. Warner
advertising ban), Laugesen and Meads constructed an advertising restriction index that
ranges from zero (no restrictions) to 10 (complete bans on advertising and sponsorships
coupled with multiple, strong warning labels on cigarette packaging). Estimates from
this model imply that cigarette consumption would be about 6% lower with the strongest
restrictions than it would be with no restrictions. In their preferred specification, to ac-
count for the lagged effects of advertising, the coefficient on the advertising restriction
index was interacted with time. In this specification, Laugesen and Meads found that
advertising restrictions actually had a positive effect on cigarette demand through the
early 1970s, but then reduced consumption after 1973. Estimates for the final year of
their data implied that each additional point in the restriction index reduced cigarette
consumption by about 1.5 percent, well above their estimate for the specification that
does not allow the effect to vary over time. Laugesen and Meads attributed the posi-
tive effects of the ban early in their sample to the industry's ability to substitute other
marketing activities for broadcasting advertising in response to early restrictions. How-
ever, its ability to substitute other media for banned media diminished over time as the
restrictions became more comprehensive.
Stewart (1992) raised a number of concerns about the approach taken by Laugesen
and Meads. Specifically, Stewart argued that errors in variables for the dependent and
several independent variables will bias the estimates on the advertising restriction coef-
ficients. In addition, he argued that the Laugesen and Meads approach failed to account
for unmeasured, country-specific factors (i.e., culture, tastes, and attitudes) that should
be important determinants of cigarette consumption, and that the omission of these fac-
tors leads to biased estimates of the advertising restriction coefficients. Laugesen and
Meads (1993) defended the estimates from their research, arguing that after correcting
for errors in the data, the estimates confirmed their earlier finding that bans on advertis-
ing significantly reduce cigarette consumption. The authors did not present these revised
estimates in their response, however.
Using data on 22 OECD countries for the period 1964-1990, Stewart (1993) pre-
sented his own empirical analysis of the impact of restrictions on cigarette advertis-
ing on demand. He estimated fixed effects models to control for unmeasured country-
specific influences on demand. Rather than using a comprehensive measure of restric-
tions on cigarette advertising, however, Stewart focused on bans on the televised ad-
vertising of cigarettes. Also, in contrast to Laugesen and Meads, Stewart did not allow
the impact of the advertising restriction to change over time. lie estimated that the ban
on cigarette advertising on television has had a positive but insignificant impact on
cigarette demand, consistent with the findings from several other studies on the effects
of broadcast advertising bans alone.
In late 1992 the United Kingdomn's Department of Health reviewed the evidence on
the impact of cigarette advertising and restrictions on advertising on cigarette demand
[U.K. Department of Health (1992)]. The "Smee Report", known by the name of the
project director, also contains two original econometric analyses on the impact of ad-
vertising restrictions, one for Norway and the other for Canada. Estimates from these
analyses suggest that the countries' relatively comprehensive advertising and promo-
tion bans did lead to significant reductions in smoking. Given this evidence as well as
that from the numerous qualitative and quantitative studies reviewed, the Smee Report
concluded that cigarette advertising has a positive impact on smoking and that bans on
advertising would reduce demand.
In a subsequent edited volume, critics of the Smee Report argued that it was flawed
in several ways, including the following: the literature review omitted several qualita-
tive and quantitative studies that found no impact of advertising or ad restrictions on
cigarette demand; the findings from some of the literature reviewed are misstated; the
empirical analyses contained methodological and other errors [Luik (1994)]. Stewart
(1994), for example, compared the estimates for the countries common to his 1993
econometric analysis with those presented in the Smee Report, concluding that adver-
tising bans in Norway, Finland and Canada have actually increased tobacco consump-
tion. 2 7
The Smee Report, Saffer (1998), Stewart (1993), and others have indicated several
factors that complicate the ability of econometric analysis to examine the impact on
cigarette demand of restrictions on cigarette advertising and promotion. The potential
endogeneity of advertising restrictions has not been carefully examined in any of the
econometric studies. Similarly, with the exception of Stewart (1993), social, cultural,
and other differences among countries have not been well controlled for in the econo-
metric research. However, efforts to control for these, using fixed effects modeling for
example, create severe multicollinearity problems that make it difficult to isolate the im-
pact of the advertising restrictions on demand from other key determinants. Similarly, as
Saffer noted, a majority of the studies to date have examined the impact of restrictions
on advertising in one or two media, leaving firms free to substitute towards other me-
dia and to develop new marketing approaches. The findings from several studies, which
suggest at best a temporary negative effect of a relatively limited set of advertising re-
strictions, are consistent with the argument that effective alternatives are developed in
response to the ban. So to are the findings from the few studies that have found that
relatively comprehensive restrictions significantly reduce demand.
4.3. Findingsfrom the noneconomic literature
While econometric methods are powerful tools for examining the demand for cigarettes
and other tobacco products generally, they are relatively ill-suited for evaluating the
effects of cigarette advertising and promotion and related restrictions, as described
above. Evidence from a number of other disciplines, however, supports the argument
that cigarette advertising and promotion directly and indirectly increase cigarette de-
mand [Warner (1986a), U.S. Department of Health and Human Services (1989, 1994,
1996, 1998), U.K. Department of Health (1992)].
27 The Smee Report's estimate for Finland is based on the work by Pekurinen (1989, 1991) rather than an
original econometric analysis.
1592 FJ.Chaloupka and K.E. Warner
A major source of noneconomic evidence is survey research and experiments that as-
sess reactions to and recall of cigarette advertising and smoking behavior, particularly
among children. These studies have concluded that cigarette advertising is effective in
getting children's attention and that the ads are recalled, with strength of interest cor-
related with current or anticipated smoking behavior or smoking initiation [U.S. De-
partment of Health and Human Services (1989, 1994), Food and Drug Administration
(1996), Goldstein et al. (1987), DiFranza et al. (1991), Evans et al. (1995), Pierce et
al. (1998)]. However, these studies generally cannot assess the potential endogeneity
between an interest in smoking and behavior [U.S. Department of Health and Human
Services (1989, 1996)].
Others have articulated logical arguments that conclude that cigarette advertising
and promotional activities are not consistent with the tobacco industry's claim that
the market for tobacco products is mature and that marketing activities are designed
to promote brand share rather than market expansion. For example, Tye et al. (1987)
calculated that cigarette firms' battling only for brand share did not make financial
sense in a U.S. market in which the top two firms now control 75% of cigarette sales
(and one company has 95% of smokeless tobacco sales) and in which brand loyalty
is notoriously strong. The authors argued that if the industry believed its own brand-
share argument, it would have welcomed the opportunity for a legislated ban on to-
bacco advertising, proposed in the U.S. Congress in the mid-1980s. Instead, the in-
dustry fought the ban vigorously. Similarly, Warner (1986a) noted that even if the in-
dustry is a mature or declining one, retaining existing consumers and recruiting new
ones would be particularly important in the cigarette market in which about 5% of
consumers are lost annually to cessation and death. Finally, while the overall market
may be mature, there are segments of the market that appear to be potential growth
markets, such as youth in the U.S., for whom smoking prevalence has risen through-
out the 1990s [University of Michigan News and Information Services (1997)], or spe-
cific minority groups, such as Hispanic females for whom smoking rates are well be-
low those of other groups of women [U.S. Department of Health and Human Services
(1998)]. Substantial evidence, including recently released internal industry documents
(www.house.gov/commerce/TobaccoDocs/documents.html), indicates that increasing
shares of advertising and promotion activities have been directed towards these growth
or potential growth markets [U.S. Department of Health and Human Services (1989,
1994, 1998), U.K. Department of Health (1992), King et al. (1998)],
Clearly, there is no "smoking gun" that proves that advertising and promotion play a
significant role in expanding or maintaining the market for tobacco products, or that they
do not. Examining all of the evidence collectively, Warner (1986a) concluded that it is
more likely than not that advertising and promotion do stimulate cigarette consumption.
However, he also characterized the extent of the influence of advertising as unknown and
possibly unknowable.
To date, economists' contributions to the relevant body of knowledge about cigarette
advertising have been less numerous, and likely less consequential, than in other areas
of smoking and health, such as the highly productive work on the relationship between
cigarette price and demand, reviewed in Section 2. Still, the econometric research in
this area has offered important insights into the challenge of evaluating the effects of
advertising. The door is open for creative new work to follow.
5. Other tobacco control policies and demand
This section focuses on the impact on cigarette demand of some of the more widely used
tobacco control policies in addition to taxation and advertising restrictions, including
the dissemination of information on the health consequences of smoking, restrictions
on smoking in public places and work places, and limits on youth access to tobacco
products. Other policies, such as the disclosure of tobacco product constituents and the
funding of school-based smoking prevention programs, have not been the subject of
economic analysis.
5.1. Health information and counter-advertising
In both the early 1950s and the mid-1960s, smoking-related "health scares" received
substantial public attention in the United States. The first was prompted by coverage in
the popular media of the then-new scientific evidence linking smoking to lung cancer
[Wynder and Graham (1950), Doll and Hill (1954)]. Illustrative was an article in the
December, 1952 Reader'sDigest entitled "Cancer by the Carton" [Norr (1952)]. The
second followed release of the first Surgeon General's report on smoking and health
[U.S. Department of Health, Education, and Welfare (1964)], the first official govern-
ment document to label smoking a cause of lung cancer and to call for "appropriate
remedial action". Media attention to the report ranked it as one of the year's most cov-
ered news stories.
The impact of these "health scares" has been the subject of extensive econometric
analysis [Sumner (1971), Hamilton (1972), Schmalensee (1972), Atkinson and Skegg
(1973), McGuiness and Cowling (1975), Thompson and McLeod (1976), Warner (1977,
1981a, 1989), Ippolito et al. (1979), Fujii (1980), Schneider et al. (1981), Leu (1984),
Porter (1986), Bishop and Yoo (1985), Kao and Tremblay (1988), Simonich (1991),
Pekurinen (1989, 1991), Meier and Licari (1997)]. In general, these and other stud-
ies concluded that cigarette smoking fell significantly in response to the new informa-
tion on its health consequences. Warner (1977, 1981a), for example, found that the
public scares in the early 1950s significantly reduced smoking in 1953 and 1954, but
that their negative impact diminished through the decade. He concluded that the 1964
Surgeon General's report led to an immediate 5% decline in cigarette consumption.
Schneider et al. (1981) estimated that U.S. per capita tobacco consumption was about
39 percent lower in 1978 than it would have been in the absence of the two health
scares.
1594 EJ. Chaloupka and KE. Warner
The evidence linking cigarette smoking to morbidity and premature mortality led to a
number of public policy efforts to disseminate information on the health consequences
of smoking. Numerous countries have adopted policies requiring health warning la-
bels on cigarette packaging and advertising; in general, these warnings have become
stronger and more prominent over time. Non-econometric evaluations of warning labels
have concluded that small, inconspicuous labels that provide little specific information
about the consequences of smoking are generally ineffective. However, multiple, strong,
and direct messages that are prominently displayed have been found to be effective
[World Health Organization (1997)]. The limited econometric evidence also suggests
that health warning labels have led to small but significant reductions in cigarette smok-
ing [Abernethy and Teel (1986), Tansel (1993), Meier and Licari (1997), Bardsley and
Olekans (1998)].
Mass media "counter-advertising" campaigns have been widely used to discour-
age cigarette smoking and other tobacco use. Econometric analyses of anti-smoking
publicity and paid counter-advertising generally, but not universally, have concluded
that these campaigns have significantly reduced cigarette smoking [Hamilton (1972),
Warner (1977, 1981a, 1989), Ippolito et al. (1979), Metra Consulting Group (1979), Fu-
jii (1980), Schneider et al. (1981), Lewit et al. (1981), Porter (1986), Abernethy and Teel
(1986), Baltagi and Levin (1986), Stavrinos (1987), Kao and Tremblay (1988), Pekuri-
nen (1989, 1991), Simonich (1991), Tansel (1993), Hu et al. (1994, 1995b, 1995c),
Tremblay and Tremblay (1995), Goel and Morey (1995), Hsieh et al. (1996)]. Much of
the econometric evidence from the U.S. is based on two major counter-advertising cam-
paigns: the anti-smoking messages broadcast in the late-1960s under the Federal Com-
munications Commission's Fairness Doctrine [Hamilton (1972), Warner (1977, 1981a,
1989), Ippolito et al. (1979), Fujii (1980), Schneider et al. (1981), Lewit et al. (1981),
Porter (1986), Baltagi and Levin (1986), Simonich (1991), Tremblay and Tremblay
(1995), Goel and Morey (1995)] and the anti-smoking media campaign in California
in the early 1990s, funded by an earmarked tax on cigarettes [Hu et al. (1994, 1995b,
1995c)].
From 1967 until January 2, 1971, the date television and radio advertising of
cigarettes was banned, anti-smoking messages were broadcast to "compensate" for pro-
smoking advertisements, initially at the rate of one anti-smoking message for every
eight cigarette ads and eventually at a 1:3 ratio. Television time for these counter-
advertisements was donated by broadcasters under the Fairness Doctrine which required
broadcasters to air both sides of a controversial issue if one side was being aired. Per
capita cigarette consumption dropped four years in a row, for the first time in history
[Warner (1977, 1979)]. Schneider et al. (1981) concluded that the counter-advertising
reduced per capita consumption by approximately 5%. Using individual-level data on
smoking among youth ages 12-17, taken from Cycle III of the Health Examination Sur-
vey, Lewit and his colleagues (1981) found that the anti-smoking messages significantly
reduced youth smoking prevalence.
In 1988, California voters passed Proposition 99, the California Tobacco Tax and
Health Promotion Act. The Act raised the state cigarette tax by 25 cents per pack
and earmarked 20 percent of new tax revenues for health education programs to re-
duce cigarette smoking, including a statewide media campaign. Similar tax increases
with funds earmarked for counter-advertising campaigns have been adopted in Mas-
sachusetts, Arizona, Oregon, and elsewhere. In addition, part of the funds received by
several other American states that have recently settled lawsuits with the tobacco in-
dustry are earmarked for counter-advertising campaigns. Hu and his colleagues (1994,
1995b) concluded that California's anti-smoking media campaign has significantly re-
duced smoking in California. They estimated an elasticity of cigarette sales with respect
to expenditures on the anti-smoking media campaign of -0.05 [Hu et al. (1995c)].
Comparing the impact of the tax increase with that of the media campaign, they esti-
mated that the tax increase reduced per capita cigarette sales by over 27 packs, while
sales declined by just under eight packs per person in response to the media cam-
paign. Early evidence from Massachusetts suggests a comparable decline in sales af-
ter that state's tax-funded anti-smoking campaign [Harris et al. (1996)]. Chaloupka and
Grossman (1996)] concluded that similar counter-advertising campaigns financed by
earmarked cigarette taxes lead to significant reductions in both the prevalence of youth
smoking and average cigarette consumption by young smokers.
Econometric evidence from Greece [Stavrinos (1987)], Finland [Pekurinen (1989,
1991)], Turkey [Tansel (1993)], Australia [Bardsley and Olekalns (1998)], and the U.K.
[Townsend (1998)] indicates that the U.S. experience is not unique. In each of these
studies, mass media campaigns aimed at reducing cigarette smoking by providing infor-
mation on the health consequences of smoking were estimated to have led to significant
reductions in smoking prevalence and in cigarette consumption.
The evidence described above clearly indicates that cigarette demand has declined
in response to dissemination of new information on the health effects of cigarette
smoking. Viscusi (1990, 1991, 1992, 1995) and others have concluded that individu-
als have heard and comprehended the health warnings and are making rational, well-
informed choices when it comes to smoking. Indeed, as noted above, Viscusi (1992)
believes that smokers overestimate the risk of dying from lung cancer as a result
of smoking. As such, he suggested [Viscusi (1992, 1995)] that the scope for fur-
ther government intervention to reduce cigarette smoking is relatively limited. Kenkel
(1991), however, concluded that while knowledge about the health effects of smok-
ing is relatively common and has significantly reduced smoking, it is incomplete.
Moreover, his estimates implied that improved health knowledge would lead to sig-
nificant changes in cigarette smoking, in contrast to his findings for health knowl-
edge concerning alcohol use and exercise. As was noted earlier, a wealth of addi-
tional evidence further supports the view that, while general knowledge concerning
the health consequences of smoking is relatively widespread, it is often superficial and
does not extend to risks other than those associated with lung cancer, heart disease,
and chronic lung disease [U.S. Department of Health and Human Services (1989)].
Further, many smokers, particularly including heavy smokers, do not personalize the
health risks that they acknowledge as applying to smokers "in general" [Schoen-
baum (1997)]. Warner et al. (1995), Brownson et al. (1992), Grossman et al. (1999),
1596 FEJ. Chaloupka and K.E. Warner
and others argue that some populations, particularly younger and less educated/low-
income groups, significantly understate the health consequences of smoking; for ex-
ample, children in particular may be prone to underestimate the risk of becoming ad-
dicted.
5.2. Restrictions on cigarette smoking
As information on the health consequences of exposure to environmental tobacco smoke

(ETS) has become more widespread [U.S. Department of Health and Human Services
(1986), U.S. Environmental Protection Agency (1992), Steenland (1992), Steenland et
al. (1996)], governments at all levels have adopted policies limiting smoking in public
places and private workplaces. Beginning with Arizona in 1973, states started adopting
"clean indoor air" laws with the explicit objective of limiting nonsmokers' exposure to
ETS [U.S. Department of Health and Human Services (1986)]. In general, these laws
prohibit smoking in elevators, health care facilities, public transportation, indoor cul-
tural and recreational facilities, government buildings, public meeting rooms, schools,
shopping malls, and retail stores. The most extensive laws also include restaurants and
private workplaces. A recent World Health Organization (1997) survey of tobacco con-
trol policies in 134 countries indicated that the vast majority of countries now have some
form of restriction on smoking in public places.
Although the restrictions are primarily intended to reduce nonsmokers' exposure to
ETS, they can also lead to significant reductions in cigarette smoking since they reduce
the smoker's opportunities to smoke or otherwise raise the "cost" of smoking. This is
particularly true for adult smokers restricted from smoking at the workplace. In addition,
restrictions on smoking may alter the perceived norms related to smoking by changing
attitudes concerning the social acceptability of smoking [U.S. Department of Health and
Human Services (1994)].
A number of recent econometric and other studies have examined the impact of smok-
ing restrictions on cigarette demand in the U.S. and elsewhere [Wasserman et al. (1991),
Chaloupka (1992), Chaloupka and Saffer (1992), Keeler et al. (1993), Chaloupka and
Grossman (1996), Evans et al. (1999), Chaloupka and Wechsler (1997), Chaloupka
and Pacula (1998, 1999, forthcoming), Ohsfeldt et al. (1999), Bardsley and Olekalns
(1998), Townsend (1998)]. In general, restrictions on smoking in public places and pri-
vate workplaces have been found to reduce both smoking prevalence and average daily
cigarette consumption among smokers. Wasserman et al. (1991) estimated that expand-
ing smoking restrictions from those limiting smoking in a small number of relatively
minor public places (elevators, waiting rooms, etc.) to more comprehensive restrictions,
including restaurants and private workplaces, would reduce overall per capita smoking
by almost 6%. Similarly, Chaloupka and Grossman (1996) and Chaloupka and Wechsler
(1997) concluded that strong restrictions on smoking significantly reduce both smok-
ing prevalence and average daily cigarette consumption for youth and young adults,
respectively.
Using annual state-level data for 1975-1985, Chaloupka and Saffer (1992) examined
the possibility that smoking restrictions are endogenous. They found that states with
the strongest restrictions, those with limits on smoking in private workplaces, were also
the states in which anti-smoking sentiment was relatively high and smoking was rela-
tively low. After accounting for this, Chaloupka and Saffer concluded that the strongest
restrictions had no impact on cigarette demand. However, they did find that relatively
comprehensive restrictions on smoking in public places (those including restaurants in
addition to a number of other public places) significantly reduced smoking even after
accounting for their potential endogeneity. In a more recent analysis of this issue, using
data from the September 1992, January 1993, and May 1993 tobacco use supplements
to the Current Population Survey, Ohsfeldt et al. (1999) concluded that the strongest
restrictions on smoking lead to significant reductions in smoking prevalence, after ac-
counting for their potential endogeneity.
Evans and colleagues (1999) examined whether workplace restrictions led to self-
selection, with nonsmokers attracted to worksites at which smoking was not permitted
and smokers seeking out worksites permitting smoking. Using data from the 1991 and
1993 National Health Interview Surveys, the authors examined self-reported informa-
tion on whether or not workers were in firms that had policies restricting smoking. If
the respondent answered affirmatively, more detailed information on the policies was
collected. Evans et al. estimated the impact of the restrictions on cigarette demand in a
simultaneous equations model that allows for individuals to self-select worksites based
on their smoking status and smoking policies. The authors found that, after accounting
for workers' potential self-selection, smoking bans diminished the probability of adult
smoking by 5%, while reducing average daily cigarette consumption among smokers
by 10%. As such, the authors concluded that recent declines in smoking among workers
relative to non-workers in the U.S. can be attributed to the growing number of workplace
bans on smoking.
5.3. Limits on youth access to tobacco products
According to the World Health Organization (1997), 43 countries ban the sale of
cigarettes to minors, typically by establishing a minimum legal purchase age for
cigarettes and restricting the distribution of free samples of to underage youth. The
non-economics literature provides mixed evidence on the effectiveness of these youth
access limits. A few studies have found that raising retailer compliance with the mini-
mum age laws reduces the prevalence of youth smoking [Jason et al. (1996), Forster et
al. (1998)]. Others, however, have found little impact on youth smoking, even with high
compliance by retailers [Rigotti et al. (1997)]. A few recent econometric analyses have
examined the impact of these limits on youth tobacco use in the U.S., generally finding
little or no impact on youth cigarette smoking and other tobacco use [Wasserman et
al. (1991), Chaloupka and Grossman (1996), Chaloupka et al. (1997), Chaloupka and
Pacula (1998)]. Chaloupka and Grossman (1996) attributed this to the relatively weak
enforcement of these laws.
Chaloupka and Pacula (1998) examined the impact of enforcement of and compli-
ance with the limits on youth access on youth smoking using data collected in a special
1994 survey of state activities related to the Synar amendment [Downey and Gardiner
(1996)]. This amendment requires states to establish minimum purchase ages for to-
bacco products and to demonstrate that these laws are being enforced by conducting
random, unannounced compliance checks of retailers selling tobacco products. Failure
to do so can lead to the loss of state block grant funds for substance abuse prevention
and treatment programs. Chaloupka and Pacula's estimates suggest that when the limits
on youth access are comprehensively and aggressively enforced and highly complied
with, they significantly reduce the prevalence of youth smoking.
6. Agricultural policy and the macroeconomic implications of tobacco
Most of the policy-relevant economic research on tobacco has focused on the argu-
ments in the cigarette demand function, discussed in the preceding sections. With a few
exceptions, the literature cited has addressed how policy variables directly influence
smoking by individuals. There is another domain in which economic issues arise and
economic analysis has produced important understanding, however: how policy effects
on the economic welfare of the industry indirectly influence smoking and health. In this
section, we examine the literature pertaining to two such issues, each of which has been
raised in the course of the social debate on the economic and health consequences of
tobacco.
The first involves economic policy intended to benefit the agricultural sector of the
U.S. tobacco industry: how the unorthodox regulation of domestic tobacco growing in
the U.S. affects the price and quantity of tobacco grown, and through this channel influ-
ences the price and consumption of cigarettes. The tobacco "subsidy" has been a source
of contention within the U.S. public health community for years, with most health pro-
fessionals believing that the "subsidy" encourages tobacco growing and thereby smok-
ing. As economists will appreciate immediately, the direct effect of a tobacco price
support system is the opposite: it discourages smoking by artificially inflating the price
of tobacco in cigarettes. Economic analysis has provided insight into the extent of this
effect, permitting policy analysts to consider it in the broader context of the overall
implications of the price support program.
The second issue addresses the broad question of how dependent nations' economies
are on preservation of a robust tobacco industry for employment, tax revenue, and a pos-
itive contribution to the trade balance. A central thrust of the tobacco industry's strategy
to combat tobacco control policies has long been to argue that, regardless of the health
consequences of its products, the economic vitality of America (and other countries) de-
pends on a strong tobacco industry. In recent years, macroeconomic research has been
undertaken in several countries to challenge the premise with empirical evidence.
Prior to addressing these issues, this section opens with background on both the
global and U.S. tobacco industries.
6.1. Size and nature of the tobacco industry
6.1.1. The global industry
In 1983,28 an estimated 47 million people directly owed their livelihoods, in whole or

in part, to tobacco cultivation, product manufacture, distribution, and retailing (repre-
senting 18.2 million full-time equivalent jobs), with 30 million of these in farming (11
million FTEs), nearly half in China alone.2 9 In addition, over 10 million people (8 mil-
lion FTEs) were employed in supplier industries, those providing materials and services
to the tobacco industry (e.g., harvesting tools and cigarette papers, insurance coverage
and transportation and shipping) [Agro-Economic Services (1987)].
In the 69 countries included in this evaluation of the global industry (covering 90%
of the world's population), an average of 0.3% of arable land was devoted to tobacco,
although tobacco accounted for 1.3% of full-time agricultural employment, reflecting
the labor intensity of tobacco growing. Tobacco constituted 1% of total agricultural
output, 0.8% of total manufacturing output, and 3 percent of total retail sales [Agro-
Economic Services (1987)].
International trade in tobacco and tobacco products represented 0.5% of total exports
and 0.4% of imports, the difference reflecting a robust international contraband trade
[Agro-Economic Services (1987)] which appears to have grown considerably in recent
years [Joosens (1998)]. For a handful of countries, tobacco exportation represents an
international economic lifeline. In Zimbabwe, for example, tobacco accounts for ap-
proximately one-third of the country's export revenues [Chapman and Wong (1990),
Zimbabwe... (1998)].
Tobacco's economic importance worldwide derives also from its use as a source of
governmental revenues. Excise (and other) taxation generates many tens of billions of
dollars annually. Several countries derive 10% or more of total government revenues
from tobacco taxation [Chapman and Wong (1990)].
6.1.2. The U.S. tobacco industry
In 1995, Americans spent $48.7 billion on tobacco products, most of it on just under
490 billion cigarettes. In addition, U.S. farms and cigarette companies shipped abroad,
respectively, $1.4 billion worth of unmanufactured tobacco leaf and $5 billion in man-
ufactured product [Gale (1997)].
The five core sectors of the tobacco industry - tobacco growing, auction warehous-
ing, product manufacturing, wholesale trade, and retail trade - collectively employ up
28 We are not aware of any contemporary figures on the size of the global industry. The data presented
here, covering the year 1983, suggest a rough order-of-magnitude estimate of the industry's importance,
albeit one that likely underrepresents the contemporary industry given that tobacco consumption has increased
worldwide annually since that year.
29 China is the world's largest producer and consumer of cigarettes.
to half a million Americans in tobacco-related activity [Gale (1997)]. Industry activity

generates an additional 650,000 to 2 million spin-off jobs, representing purchases from
suppliers and spending by the recipients of incomes from tobacco product sales. 30 In
1992, the emotional and political heart of the U.S. industry, tobacco farming, included
124,000 farms on which tobacco was grown in some 20 states. However, just three
states - North Carolina, Kentucky, and Tennessee - accounted for 82% of the 91,000
farms officially designated as tobacco farms (defined by the U.S. Department of Agri-
culture as farms on which tobacco accounts for at least 50% of sales). A further three
states - Georgia, South Carolina, and Virginia - accounted for an additional 10% of
tobacco farms. Collectively, these six southeastern tobacco states were responsible for
a comparable percentage of tobacco production as well [Grise (1995)].
Estimating tobacco farm employment is difficult because so few farmers rely exclu-
sively on tobacco for their full-time economic activity. Employing the arbitrary assump-
tion that each job is half time (1000 hours), the USDA estimated the number of tobacco
farming jobs in 1995 at 156,000 (i.e., fewer than 80,000 FTE jobs). The largest em-
ployment contribution associated with tobacco was in the retail sector, however, with
an estimated 257,000 jobs spread all over the country. The other politically "visible"
component of the industry - workers employed in cigarette manufacturing - numbered
only 25,600. They were concentrated primarily in only three cities in which the vast
majority of U.S. cigarettes are manufactured: Richmond, VA, Winston-Salem, NC, and
Louisville, KY [Gale (1997)].
Although the tobacco farmer is the "heart" of the industry in emotional and politi-
cal terms, economically the farmer appears to be little more than a minor appendage.
Domestically-grown tobacco represented only about 2% of the domestic retail tobacco
dollar in 1995, with imported tobaccos constituting another 1-2%.31 Once one accounts
for leases of tobacco-growing quotas, farm supplies and equipment, overhead, and mar-
keting costs, farm workers actually took home only about 30% of tobacco farm gross
receipts for their labor, under 1% of the nearly $50 billion Americans spent on cigarettes
30 Estimates of direct, indirect, and expenditure-induced employment are taken from several sources, includ-
ing Warner et al. (1996), Price Waterhouse (1992), Tobacco Merchants Association (1995), and Gale (1997).
The smallest total employment associated with tobacco industry activity is Gale's estimate of 1.2 million,
although Gale also reported a high estimate of direct employment (500,000). The largest total employment
estimate is that of the WEFA Group at 3 million jobs, although we believe that this and several other industry-
commissioned estimates rely on improbably large multipliers in estimating expenditure-induced employment.
Reviewing the Price Waterhouse analysis, Arthur Andersen Economic Consulting (1993) identified what they
believed to be serious methodological flaws and concluded that "employment and job loss figures are grossly
inflated".
Other analysts have suggested that the industry substantially underestimates employment associated with
tobacco, since they fail to include health care personnel who care for the victims of tobacco-produced diseases,
undertakers who bury them several years earlier than nonsmokers, professional launderers who clean and
repair smokers' clothing more frequently, and so on [Schelling (1986), Warner (1987)].
31 The tobacco crop value in 1995 was less than $2.6 billion, much of which was exported as raw leaf or in
manufactured cigarettes [Gale (1997)].
and other tobacco products. By comparison, the largest share of the tobacco dollar went
to manufacturing (38%), with additional major participants being wholesale and retail
trade (27%) and government, through excise taxation (26%) [Gale (1997)].
Although tobacco growing and cigarette manufacturing might seem like excellent
candidates for a highly competitive marketplace, they do not come close to conforming
to the economist's ideal of the smoothly functioning unregulated competitive market.
The manufacturing industry is characterized by a high degree of concentration, with
two companies, Philip Morris and R.J. Reynolds, selling three-quarters of all cigarettes
purchased in the U.S. and three others (Brown & Williamson, Lorillard, and Liggett)
accounting for the vast majority of the rest; Philip Morris alone captures half the mar-
ket (with one of its brand lines, Marlboro, accounting for more than half of the com-
pany's sales) [Kluger (1996)]. Four other companies round out this highly concentrated
oligopoly. Although one can readily imagine barriers to entry into the cigarette market
(e.g., brand-name marketing advantages, distribution channels, etc.), tobacco farming
seems a less obvious domain for noncompetitive forces. Nonetheless, this purely agri-
cultural endeavor is not subject to the conventional laws of supply and demand in the
U.S. Rather, a convoluted system of price supports and allotments regulates who can
grow tobacco, where and how much they can grow, and what minimum prices they can
expect at market. The existence of this system, and the concentration of the industry in
the six southeastern states, has been credited with responsibility for the unusual polit-
ical power wielded by the six tobacco bloc states in Congress [Taylor (1984), Warner
(1988), Gale (1997)]. We turn now to a consideration of the nature of this system and
its implications for both tobacco agriculture and smoking and health.
6.2. The impact of the U.S. tobacco agricultureregulatory system
6.2.1. Nature of the system and its impact on tobaccofarming3 2
Since the early 1930s, the U.S. federal government has implemented a variety of tobacco
farm programs designed to limit tobacco growing and prop up tobacco prices. Born of
Depression-era concerns about the traumatic effects on farmers of cyclical prices and the
vagaries of the weather, the programs have ensured stability but also limited innovation
in production techniques and farm size. Although the specifics of the programs have
varied over time, they have shared certain core elements in common: restriction of the
supply of tobacco, by restricting who can grow tobacco and how much they can grow,
and the assurance of minimum prices.
The tobacco farm programs began with the Agricultural Adjustment Act of 1933
through which cash payments were made to tobacco farmers who agreed to limit pro-
duction. The Agricultural Act of 1938 established the principle of marketing quotas,
32 This section's description of the tobacco agriculture program is based on material from Capehart (1997),
Grise (1995), and Zhang and Husten (1998).
1602 F1J. Chaloupka and K.E. Warner
with penalties for growers who exceeded them. Price supports were originally set at
75% of base-period prices and have varied up and down since then. The continued exis-
tence of the price support system rested on a vote of tobacco farmers every three years.
If two-thirds of the farmers supported the system, as they have during each vote (with
some minor exceptions), legally enforceable marketing quotas are put in place.
The growers of tobacco, and the acreage they can farm, are limited through a system
of allotments, in essence a license to grow tobacco, allocated to farms existing at the
time the system was established. Under the first of the allotment programs, only farm-
ers possessing the allotments, or renting or purchasing land with allotments, could grow
tobacco. Since 1962, however, farmers have been permitted to rent or purchase allot-
ments without having to use the allotment holder's land, although subject to a number
of restrictions as to type of tobacco and how far the quota could be transported (e.g.,
some quotas could be applied only within the county of the allotment holder). Supplies
of tobacco were thus limited through allotments and marketing quotas, as well as re-
strictions on imported tobacco. Quotas are established based on the intended purchases
by cigarette manufacturers, anticipated exports and imports, and the amount of tobacco
needed to achieve a specified level of reserves. The Secretary of Agriculture can further
adjust the quotas by +3% of the amount determined by formula.
The price support is based on a loan program through which farmers are guaranteed
a pre-specified minimum price. Farmers attempt to sell their tobacco at auction. If the
high bid does not at least match the loan price, a farmer-owned cooperative purchases
the tobacco at that guaranteed price, using money loaned by the USDA's Commodity
Credit Corporation (CCC). The cooperative stores the tobacco as collateral for the CCC
loan. When the cooperative later resells the tobacco, it forwards the proceeds to the
CCC to cover the loan principal and interest.
Until 1982, general tax revenues were used to cover CCC losses. Although the
amounts of money involved were modest by federal government program standards
[Warner (1988)], the image of taxpayers' subsidizing the growing of a product that
health officials characterized as deadly became politically untenable. As a result,
Congress passed the No-Net-Cost Tobacco Program Act of 1982 which eliminated the
taxpayer subsidy. Since then, farmers and buyers have paid an assessment per pound of
tobacco to cover any losses in the loan program. The federal government continues to
cover the cost of administering the program, as well as providing a variety of other ser-
vices to growers through the USDA. In 1993, the total federal outlay on these activities
was $26 million.
Over time, variations in the tobacco program have modified import policies and im-
posed and then removed domestic-content requirements for U.S. cigarette manufactur-
ers, as well as tinkering with quota determination and allocation, price support, and so
on. For a discussion of specific revisions of the relevant laws, see Capehart (1997) and
Grise (1995).
The net effect of the program on tobacco agriculture has been multi-faceted. It has
brought the stability to the tobacco farm economy that it was designed to ensure. It
has restricted domestic tobacco supplies. The program has clearly restricted tobacco
farm size and limited the development of more capital-intensive methods of tobacco
farming; indeed, the growing of tobacco entails greater labor intensity than for nearly
all other U.S. crops. The program has increased the price of domestic tobaccos and, as
a consequence, the price of cigarettes (discussed below). As a result of boosting prices,
it has likely restrained the amount of raw leaf exportation from the U.S. It has created
an economically entrenched political constituency, the allotment holders, the principal
beneficiaries of the program [Babcock and Foster (1992)].
6.2.2. Relevance of the tobacco program to smoking and health
The tobacco program has had two effects directly germane to the issue of smoking and
health, one ostensibly favorable, the other not. By restricting the supply of tobacco and
increasing its price, the program has likely boosted the price of the finished product,
cigarettes, and thereby decreased the quantity demanded. The extent of this effect -
arguably the less important of the two [Warner (1988), Zhang and Husten (1998)] -
has been studied by economists and is the focal point of this section. In contrast, by
allocating the right to grow tobacco, or to earn money by leasing allotments, to a select
group of citizens in the tobacco southeast, and by ensuring stable and relatively high
prices for farmers, the tobacco program has created a highly concentrated economic
and thus political interest that has long wielded substantial power within the halls of
Congress [Taylor (1984), Babcock and Foster (1992)]. Long-time observers of both
Congress and tobacco concur that the existence of the tobacco bloc has thwarted the
development of effective tobacco-and-health policies frequently over the past 35 years
[Taylor (1984), Fritschler and Hoefler (1996)].
Whether the consumption-discouraging effects of the increased price of tobacco or
the consumption-encouraging effects of a lack of aggressive federal tobacco control
policy have dominated has been the subject of some informed speculation but no for-
mal analysis. Having reviewed the evidence, both Warner (1988) and Zhang and Husten
(1998) concluded that the latter was more important than the former, but in each case
this conclusion rested as much on the finding that the direct effect of the tobacco pro-
gram on cigarette price was very small. It is here that economic analysis has provided
useful empirical evidence. 3 3
The economic effects of the tobacco price support program have been the subject of
formal economic analysis for at least three decades [Johnson (1965)]. Since the mid-
1980s, four analyses have estimated the impact of abandonment of the tobacco price
support program on tobacco supplies and prices. Using a simultaneous equations model
of the supply of and demand for tobacco and cigarettes, which included the possibility
of substitution of foreign for domestic tobaccos, Sumner and Alston (1985) estimated
33 This analysis has been particularly useful simply to help correct the misimpression of the lay public that
the tobacco price support program, or the "tobacco subsidy", as it is more commonly referred to, has directly
encouraged smoking by encouraging tobacco growing. The public has not generally appreciated that, to the
contrary, the program has limited the quantities of tobacco grown and brought to market.
that eliminating the program in 1983 would have reduced the price of U.S. tobacco by
20-30%. The authors estimated an increase in domestic tobacco output of 50-100% or
more with supply restrictions ended, with cigarette manufacturers likely to buy more
domestic tobacco and exports likely to double. Reflecting this expanded output, Sum-
ner and Alston estimated that tobacco growing revenues would have risen by 15-60%
despite the price decrease. Because domestic tobacco represented under 10% of the re-
tail price of cigarettes, the authors concluded that the price support program boosted the
retail price of cigarettes by no more than 3%.34 Employing a price elasticity of demand
of -0.3, they estimated that the direct effect of the price support program was to de-
crease the demand for cigarettes by about 1%. A decade later, economists at the USDA
produced a similar if less detailed analysis that supported Sumner and Alston's find-
ings. Grise (1995) concluded that the price support program raised domestic tobacco
prices by 30-40%. He estimated that this tobacco price effect raised cigarette prices
by 1-2%.
Still more recently, researchers at the federal Office on Smoking and Health (OSH)
in the Centers for Disease Control and Prevention analyzed contemporary data and de-
termined that the price support program increased tobacco prices by 18-23% [Zhang et
al. (1997)]. In contrast with the early 1980s, domestic tobacco accounted for only 3% of
cigarette retail price in 1991. This decline in the domestic tobacco farm value share of
retail cigarette value reflected several developments pertaining to the amount and price
of domestic tobaccos. First, the amount of tobacco employed in manufacturing a given
number of cigarettes has declined significantly since the 1980s, as it has since well be-
fore then (from the early 1950s to the present, the amount of tobacco per cigarette has
declined by over a third [Congressional Research Service (1994)]). This has resulted
from reduced wastage, in part as a consequence of new production technologies that
allow manufacturers to blend in parts of the tobacco plant previously discarded, such as
tobacco stems, and to expand the volume of tobacco per unit of weight (called "puff-
ing"). It also reflects a shift in demand from relatively large-barreled cigarettes, some
unfiltered, to filtered and small-diameter cigarettes.
A second reason for the decline in the domestic tobacco share of the cigarette dollar
is manufacturers' increasing reliance on less expensive imported tobaccos. At various
times, as much as a third or more of the tobacco in U.S. cigarettes has been imported.
From 1980 to 1991, for example, the imported tobacco share rose from 29% to 35%.
34 Sumner and Alston (1985) assumed that tobacco price increases attributable to the price support system
would be fully passed on to retail consumers. This is a reasonable assumption. Although the evidence is
mixed, most previous research has characterized the tobacco industry as a constant-cost industry. Research
has also demonstrated that the industry has exploited its oligopolistic character with a strong price-leadership
model, passing on more than 100% of federal excise tax increases [Harris (1987)]. A recent study concluded
that the industry engages in a minor amount of price discrimination by state, passing along slightly more
than
states' excise tax increases [Keeler et al. (1996)]. Obviously, the permeability of state borders limits the extent
of such price discrimination.
More recently, the share of imported tobaccos quickly decreased and then increased. 35
The 1993 Omnibus Budget Reconciliation Act (OBRA) included a provision requiring
that 75% of the tobacco in U.S.-manufactured cigarettes be domestically grown. Shortly
after the domestic content provision was implemented, it was determined to be incon-
sistent with the requirements of the General Agreement on Tariffs and Trade (GATT).
It was replaced in September 1995 by a complicated tariff-rate quota (TRQ) designed
to restrict imports but to conform to GATT requirements. 36
A third factor in the declining share of cigarette expenditures attributable to domestic
tobacco has been the stability of tobacco prices compared to more rapidly inflating
prices for the manufactured product. From 1980 to 1991, the farm price of tobacco rose
only 18%. During the same period, cigarette price increased 187%.
Accounting for the reduced role of domestic tobacco in cigarette price, Zhang and
colleagues concluded that the tobacco price support program likely increased the retail
price of cigarettes by no more than 1%. To assess the impact on smoking, the researchers
employed a more recent estimate of the price elasticity of demand for cigarettes and then
allocated half of the price response to decisions of whether or not to smoke and the other
half to number of cigarettes per day per continuing smoker. Given these assumptions,
they estimated that the direct effect of the tobacco price support program was to decrease
the number of smokers by 0.14%. As such, they concluded that the beneficial effect of
the price support program, from a public health point of view, was very modest at best. 3 7
In the most recent attempt to evaluate the implications of the tobacco price support
system, Brown (1998) examined a mix of likely provisions in comprehensive federal
tobacco control legislation that would directly affect domestic tobacco growing. Com-
bining the effects of eliminating the price support program with adoption of a $1.50 per
pack federal cigarette excise tax increase, he predicted a long-run decline in tobacco leaf
price of 20-30%, not inconsistent with the OSH estimates, which did not incorporate
an excise tax increase.
The consistency of the findings from these studies provides strong support for the
conclusion that the direct effect of the U.S. tobacco price support program on discour-
aging smoking, by virtue of raising cigarette prices, is very small.
35 Imported and domestic tobaccos are not perfect substitutes. Tobaccos come in numerous varieties, each
with its own characteristics, and soil and weather conditions combine to alter those characteristics from one
growing location to another. As a consequence, the world price of tobacco does not necessarily reflect the
marginal price. Some American-grown tobaccos, prized by cigarette manufacturers, can command a higher-
than-average price on world markets.
36 The TRQ imposes quotas on imported tobaccos by exporting country, with imports above quota levels
subject to a 350% ad valorem duty. However, most of the duty is refunded if the excess imported tobacco is
included in cigarettes made in the U.S. for export.
37 The authors' assumptions are subject to challenge. In particular, as indicated earlier in this chapter, most
of the research on cigarette price elasticities to date suggests that the dominant effect of increasing prices
on adult consumption is to lower daily cigarette consumption for continuing smokers, rather than decrease
smoking prevalence. The OSH authors' estimate of the impact on numbers of smokers is so small, however,
that alternative assumptions will not alter the qualitative conclusion that the impact on smoking is very small.
6.3. The contributionof the tobacco industry to the economy
6.3.1. States and nations
Since the late 1970s, the U.S. tobacco industry has commissioned numerous prominent
economics consulting firms to produce estimates of the industry's contributions to em-
ployment, incomes, and tax revenues for the country as a whole, the individual states,
and occasionally specific cities and counties [Wharton Applied Research Center (1979),
Chase Econometrics (1985), Price Waterhouse (1990, 1992), Tobacco Merchants As-
sociation (1995), American Economics Group (1996b)]. When tobacco control policy
measures have been under consideration by legislative bodies, industry representatives
have used the findings from these analyses to try to convince legislators that adoption of
the policy would inflict economic damage on the state's or nation's citizens, in particular
by causing widespread loss of jobs [Warner (1987)]. In a few instances, the industry's
consultants have observed in their formal written reports to their clients that alterna-
tive spending patterns would generate compensating employment [Chase Econometrics
(1985), American Economics Group (1996a)]. 38 When meeting with legislators, how-
ever, the industry's representatives have never mentioned this.
That the decline or demise of one economic activity would be replaced by alternative
economic activity, each of which would eventually produce comparable national levels
of employment, is obvious to economists. The compensating benefits of replacement
economic activities are not generally contemplated by the lay public, however. As such,
legislators, journalists, and other members of the public are susceptible to the industry's
argument that reduced purchase of tobacco products will lead to substantial economic
dislocation, and that such dislocation exacts a high price from communities. The indus-
try has used its estimates in two ways: to indicate the overall significance of tobacco in
the economies of the states and the nation as a whole; and to make projections of lost
jobs and tax revenues that would result from the adoption of specific tobacco control
policy measures.
To respond to this argument, economists have performed macroeconomic analyses
that essentially complete the analysis initiated by the industry's consultants. Where the
industry-sponsored studies estimate the gross economic contribution of tobacco - the
numbers of jobs, earnings, taxes paid - the independent studies estimate the net contri-
bution, i.e., the benefit of tobacco-related economic activity after one considers the im-
plications of redistribution of the same resources to alternative uses. Researchers at the
38 In the most telling example of this, buried in Chapter V of Volume 1 of the detailed technical report pre-
pared by analysts at Chase Econometrics (1985) is the acknowledgment that money not spent on tobacco
products would be reallocated to other spending, and that nationwide (combining tobacco and nontobacco
states), the economic results with and without tobacco "would be substantially the same". The report authors
explicitly observed that compensatory responses to the absence of tobacco spending "that would occur au-
tomatically within the Chase Econometrics Macroeconomic Model ... were constrained from taking place
within [the firm's] analysis".
University of Michigan employed the REMI Model (Regional Economic Models, Inc.)
[Treyz (1993)] to estimate how both declining tobacco consumption and the complete
elimination of tobacco consumption would affect employment in the state of Michi-
gan [Warner and Fulton (1994)] and in the principal regional economies of the U.S.,
as defined by the U.S. Bureau of Economic Analysis (BEA) [Warner et al. (1996)].
The study of the effects on Michigan was intended to demonstrate how declining to-
bacco consumption impacts the economy of a nontobacco state, since nontobacco states
comprise the large majority of U.S. states and they have often been the targets of the
industry's economic argument. The study of the regional economies of the U.S. was
intended to contrast effects within the southeast tobacco-state region (consisting of half
the 12 states in the BEA's southeast region) with implications for the 8 nontobacco re-
gions (one, "southeast nontobacco", consisting of the six nontobacco states in the BEA's
southeast region).
To illustrate the procedure (the basics of which were conceptually identical for the
two studies), in the regional analysis the researchers first generated a baseline forecast
of the economies of each of the 9 regions for the years 1993 to 2000, assuming no
changes in the expected pattern of spending on tobacco (which included an expected
annual decline in consumption, based on the trend in the decade preceding the period of
the simulation). The eight-year period selected for the simulation was intended to permit
analysis of dynamic short- and medium-run impacts on the regional economies. To eval-
uate the gross contribution of tobacco to employment, analogous to what the industry's
consultants have done, the researchers then generated an alternative forecast in which
all of the expected spending on tobacco was removed from the baseline forecast. Com-
parison of the two forecasts, with and without tobacco spending, permitted assessment
of the amount of employment associated with tobacco spending, by region, economic
sector, and year. To estimate the net employment implications resulting from consumers
devoting their former tobacco expenditures to other goods and services, the analysts re-
allocated this amount according to consumers' normal spending patterns, with tobacco
excluded. Net employment was estimated by comparing the employment projections in
the baseline simulation with those from the simulation in which tobacco spending is
reallocated to other goods and services.
To examine the implications of a more realistic scenario which might be expected if
effective tobacco control policy measures were adopted, namely an increasing rate of
decline in tobacco product consumption, the researchers assumed that the recent histor-
ical rate of decline would double. The simulations were repeated with the appropriate
amount of tobacco spending removed (gross model) and reinjected into the alternative
goods and services (net model).
In the first study, Warner and Fulton (1994) demonstrated that in a nontobacco state,
declining spending on tobacco products would increase the state's employment, and
that this effect would persist over several years. The finding reflected the fact that to-
bacco products represent imports for Michigan (and other nontobacco states). Since
some of the reallocated spending would be devoted to goods and services produced
within the state, more state spending would recycle within Michigan, thereby produc-
ing more Michigan-based jobs. Although this is obvious to economists, it represented a

revelation to many noneconomists in the policy community.
In the second study [Warner et al. (1996)], the researchers found that with either a
decline or the complete elimination of domestic tobacco spending, each of the eight
nontobacco regions would gain employment during the period studied, while the south-
east tobacco region would lose employment. The study further demonstrated that the
losses within the tobacco region would be considerably smaller than those suggested
by the industry's analyses, which again have never included the effects of alternative
spending, which would benefit the tobacco states as well as the nontobacco states. One
of the study's most important conclusions was that plausible declines in tobacco con-
sumption would have exceedingly small impacts on employment even in the southeast
tobacco region. Under the more realistic scenario, the estimated loss of 36,600 jobs in
the region by the year 2000 would amount to only 0.2% of regional employment.
Similar analyses have been performed in other countries. The idea was first intro-
duced by Allen (1993) in a qualitative consideration of the economic implications of the
tobacco industry in Canada. Subsequently, Buck and colleagues (1995) used an input-
output model to study the employment implications of tobacco in the United Kingdom.
Most recently, van der Merwe (1998b) evaluated the same issue in South Africa. De-
spite variations in basic methods (e.g., input-output vs. dynamic models) and more
specific assumptions (most notably, the nature of the alternative spending pattern), all
of these studies have arrived at the same conclusion: spending on tobacco does not gen-
erate greater employment for the country in question than would alternative spending
patterns.
The tobacco industry's consultants report on other industry-related impacts in addi-
tion to employment. The two most important are the tax revenues generated by spending
on tobacco and the positive contribution of tobacco to certain countries' trade balances.
No non-industry analysis has attempted to evaluate the net effects of reductions in to-
bacco spending on these two variables. The results are obvious, however. For most
countries, reductions in tobacco spending would produce reductions in government rev-
enues, reflecting the fact that cigarettes, unlike most other products, are subject to excise
taxation. Thus, governments that succeed in reducing tobacco consumption through to-
bacco control policies generally will need to seek alternative sources of revenue to re-
place those lost due to declining tobacco product sales. The one exception, of course,
is a sales reduction occasioned by an increase in an excise tax. In this instance, as is
discussed earlier in the chapter, government revenues will rise at the same time that
consumption falls.
Reductions in spending on tobacco could adversely affect the balance of payments
in those countries in which exports of tobacco and tobacco products exceed imports.
However, tobacco exports play a truly central role in the balance of payments in only
a handful of countries; most notably, Zimbabwe relies on tobacco for approximately
a third of its foreign exchange earnings [Chapman and Wong (1990), Zimbabwe...
(1998)]. In contrast to the net exporters, in countries in which tobacco product imports
exceed exports, decreasing consumption could improve the trade balance [Warner and
Fulton (1995)].
6.3.2. Tobaccofarm communities
As the regional analysis of the U.S. demonstrated, reductions in tobacco product sales
can harm the economies of specific areas of countries highly dependent on tobacco eco-
nomic activity. That analysis also demonstrated, however, that plausible policy-induced
decreases in tobacco consumption would have extremely modest effects on employment
within the United States' major tobacco region [Warner et al. (1996)].
Less clear, however, and likely more important from both a political and humanitar-
ian point of view, is the impact of declining tobacco sales on the local communities that
are most heavily dependent on tobacco farming or product manufacture. In the popular
mind in the U.S., large numbers of counties in North Carolina, Kentucky, and the four
other tobacco states are virtually wholly dependent on tobacco farming.3 9 Substantial
decreases in tobacco product sales would, it is widely believed, wreak havoc with these
communities' economies. Sympathy with this view has led to the inclusion of signifi-
cant benefits to tobacco farmers in all of the comprehensive tobacco control legislative
proposals under consideration by the U.S. Congress in 1998.
According to work by agricultural economists, however, the image distorts a more
benign reality. Relatively few tobacco counties in the U.S. are so dependent on tobacco
that plausible policy-induced decreases in tobacco consumption would inflict serious
economic hardship. Indeed, Gale (1998) stated recently that he expects merely a con-
tinuation of the kinds of economic adjustments that tobacco farmers have been making
for decades; and, he observed, tobacco farm communities today have more diversified
economies upon which to draw in making those adjustments than in years gone by. He
summarized the essence of the situation by noting, "Tobacco has an important historical
role in many Southern communities. Today, however, tobacco plays a minor economic
role in most local economies where it is grown" (p. 43).
That the importance of tobacco farming within the tobacco belt states has diminished
substantially is made clear by data supplied by the Economic Research Service (ERS)
of the USDA. From 1964 to 1993, the number of tobacco farms declined from 330,000
to 124,000.40 Domestic consumption of domestically-produced tobaccos has declined
from 1.6 billion pounds in 1952 to 900 million pounds in 1993. Adjusted for inflation,
the value of domestically grown tobacco has fallen.
39 Because, compared with tobacco growing, cigarette manufacturing involves many fewer, higher paid work-
ers whose employment is concentrated in three economically diversified cities, public sympathy resides more
with the farmers. The remainder of this section focuses exclusively on tobacco farming communities.
40 The decline in the number of farms is not matched by declining acreage devoted to tobacco. During the
most recent six years for which data are available, the number of farms fell from 179,000 to 124,000, but
acreage increased from 587,000 to 745,000 acres. This trend toward larger farms, permitted by relaxation of
some of the stricter limitations of the quota system, would greatly accelerate were the price support program
ended.
For most tobacco farmers, tobacco growing represents only part-time, seasonal work.
Further, most tobacco farms are small, with over 70% having annual gross sales of less
than $20,000. Nearly two-thirds of farm operators work off of their farms, as well as
on them, with 42% working off-farm at full-time jobs [Gale (1998)]. Also telling are
data indicating that the share of income from all farming, not just tobacco, in tobacco
counties fell from 5% in the early 1970s to well under 2% today. The ERS classifies
counties as "farm dependent" if earnings from all farming constitute at least a fifth of
the county's total earnings. By this definition, there were only 27 "farm dependent"
tobacco counties in the U.S. in the mid-1990s, out of 424 tobacco counties. 41
And among these farm dependent tobacco counties, only one derives a majority of
its farm receipts from tobacco. The next four most tobacco-dependent derive 25-35%
of their farm earnings from tobacco, while the remaining 22 counties each receives
less than 5% of its farm earnings from tobacco sales. Most tobacco counties are not
classified as "farm dependent". Across all tobacco counties, the USDA estimates that
tobacco sales account for approximately a fifth of total farm receipts. However, there are
a number of counties on the North Carolina-Virginia border and in eastern Kentucky in
which tobacco's share of farm sales exceeds 70%.
To put the role of tobacco into perspective, USDA calculates the ratio of tobacco
gross receipts to total proprietor and labor income within a county. By this measure, al-
most half of tobacco counties (199) have a tobacco-income ratio of less than 0.01. Only
33 counties have a ratio exceeding 0.1 [Gale (1998)]. USDA also calculates an index
of a tobacco county's ability to replace tobacco income through economic growth in
other sectors. The index measures the ratio of annual growth in inflation-adjusted local
personal income from all sources to tobacco gross receipts. USDA interprets an index
value exceeding 1.0 as meaning that the county is creating sufficient new economic op-
portunities to potentially completely replace tobacco income. Approximately half of all
tobacco counties have index values greater than 1.0 [Gale (1997)].
All told, the evidence indicates that America's tobacco farming communities are far
less dependent on tobacco than is widely believed. That abrupt declines in tobacco con-
sumption would inflict severe economic pain on selected individuals is almost certainly
true; that many others would experience temporary economic dislocation is certainly
possible. The notion, however, that realistic policy-induced decreases in tobacco con-
sumption would wreak havoc throughout much of the tobacco belt is simply not con-
sistent with the evidence. Appeals to the welfare of tobacco farmers may resonate po-
litically; but economically they appear to have little justification. 42 Indeed, the major
economic losers would be the allotment holders, a less politically-appealing group of
people.
41 A tobacco county is a county in which tobacco is grown for commercial purposes.

42 In the absence of data on the economic vitality of tobacco-growing regions in other countries, these con-
clusions cannot be extended to tobacco farm communities outside of the U.S.
7. Conclusion
In the complicated ethical, social, and political domain of tobacco policy, economic
analysis has introduced a base of objective and increasingly sophisticated knowledge
into debates in which rhetoric has often dominated. Particularly with regard to the cru-
cial issue of how price influences the demand for tobacco products, and how taxation
affects price, economists have contributed empirically-based insights that, in many in-
stances, have played essential roles in guiding the formulation of tobacco control policy.
Indeed, it is no exaggeration to credit the work of economists with the contemporary
global interest in using tobacco taxation as perhaps the primary tool of tobacco control
policy.
In the process of examining the empirical relationship between tobacco price and con-
sumption, economists have contributed to the evolving theoretical and methodological
literature on the effects of addiction on consumer demand. A "problem" in the tradi-
tional economic model of rational economic behavior, addiction is now receiving the
attention that promises important future contributions of both a conceptual and empiri-
cal nature. Public health policy making will be enriched in the process.
In addition to addressing issues of taxation, price, and demand, economic research
has also offered important understanding of the effects of other tobacco policy measures,
ranging from media counteradvertising to the introduction of restrictions on smoking
in public places. The work of economists has lent perspective to emotional issues in
debates on tobacco policy, such as the implications of tobacco control for employment
both inside and outside of tobacco-dependent regions of states and countries.
In other areas, economics research has been less successful in answering policy ques-
tions. A notable example involves the politically central issue of whether cigarette ad-
vertising increases consumption, and whether ad bans decrease it. Econometric research
has contributed empirical evidence to the debate, but without offering much by way of
resolution. In part this reflects limitations inherent in econometric methods; in part it
reflects the inadequacy of the data needed to quantify "advertising" (and exposure to it)
and evaluate its consequences. Recent work on the impacts of national advertising bans,
both partial and complete, shows promise but is decidedly in its infancy.
In addition to having enlightened debates on tobacco policy, economic analysis of
smoking serves a broader purpose as well, one not examined in the present chapter.
Constituting by far the largest body of economic research on the consumption of ad-
dictive substances, utilizing the best data available, economic research on smoking in-
forms both research and policy debates on other addictive substances [Warner et al.
(1990), Warner (1991)]. This is particularly important in the case of illicit drugs, such
as marijuana and cocaine, for which the availability of useful data has been severely
constrained. More generally, understanding the economics of tobacco lends insight into
a whole host of social, political, and economic issues, such as the political economy
of product regulation and the relationship between, and even meaning of, consumer
sovereignty and paternalism.
The use of tobacco, and particularly cigarette smoking, constitutes one of the great
public health plagues of the latter half of the 20th century, one sure to define much of
global health status far into the 21st century as well. As such, it is critical to understand
the determinants of tobacco use, perhaps especially those that can be addressed by pub-
lic policy. Using the conventional tools of their trade, often in novel and creative ways,
economists have been at the forefront of advancing knowledge in this central area of
public health. The impressive body of work described in this chapter augers a bright
future for the contribution of economics to grappling with what will soon become the
leading cause of disease and death worldwide.
References
Abernethy, A., and J.E. Teel (1986), "Advertising for cigarettes", Journal of Advertising 15(4):51-55.
Abt, C. (1975), "The social costs of cancer", Social Indicators Research 2:175-190.
Ackoff, R.L., and J.R. Ernshoff (1975), "Advertising research at Anheuser-Busch, Inc. (1963-1968)". Sloan
Management Review 16(3):1-15.
Advisory Commission on Intergovernmental Relations (1977), "Cigarette bootlegging: A state and federal
responsibility", A Commission Report (Advisory Commission on Intergovernmental Relations, Washing-
ton).
Advisory Commission on Intergovernmental Relations (1985), "Cigarette tax evasion: A second look" (Ad-
visory Commission on Intergovernmental Relations, Washington).
Agro-Economic Services Ltd., and Tabacosmos Ltd. (1987) The Employment, Tax Revenue and Wealth that
the Tobacco Industry Creates (Agro-Economic Services, London).
Akerlof, G.A. (1991), "Procrastination and obedience", American Economic Review 81:1-19.
Allen, R.C. (1993), The False Dilemma: The Impact of Tobacco Control Policy on Employment in Canada
(National Campaign for Action on Tobacco, Ottawa, ON).
American Economics Group (1996a), Economic Impact in the States of Proposed FDA Regulations Regarding
the Advertising, Labeling and Sale of Tobacco Products (American Economics Group, Washington).
American Economics Group (1996b), The U.S. Tobacco Industry in 1994: Its Economic Impact on the States
(American Economics Group, Washington).
Andrews, R.L., and G.R. Franke (1991), "The determinants of cigarette consumption", Journal of Public
Policy &Marketing 10(1):81-100.
Arthur Andersen Economic Consulting (1993), Tobacco Industry Employment: A Review of the Price Water-
house Economic Impact Report and Tobacco Institute Estimates of "Economic Losses from Increasing the
Federal Excise Tax" (Arthur Andersen Economic Consulting, Los Angeles, CA).
Ashenfelter,O., and D. Sullivan (1987), "Nonparametric tests of market structure: An application to the
cigarette industry", Journal of Industrial Economics 35(4):483-498.
Atkinson, A.B., and J.L. Skegg (1973), "Anti-smoking publicity and the demand for tobacco in the UK", The
Manchester School of Economic and Social Studies 41:265-282.
Atkinson, A.B., and J.L. Townsend (1977), "Economic aspects of reduced smoking", Lancet ii:492-495.
Babcock, B.A., and W.E. Foster (1992), "Economic rents under supply controls with marketable quota",
American Journal of Agricultural Economics 74(3):630-637.
Baltagi, B.H., and R.K. Goel (1987), "Quasi-experimental price elasticities of cigarette demand and the boot-
legging effect", American Journal of Agricultural Economics 69(4):750-754.
Baltagi, B.H., and D. Levin (1986), "Estimating dynamic demand for cigarettes using panel data: The effects
of bootlegging, taxation, and advertising reconsidered", Review of Economics and Statistics 68(1):148-
155.
Bardsley, P, and N. Olekalns (1998), "Cigarette and tobacco consumption: Have anti-smoking policies made
a difference?", Working Paper (Department of Economics, The University of Melbourne).
Barendregt, J.J., L. Bonneux and P.J. van der Maas (1997), "The health care costs of smoking", New England
Journal of Medicine 337(15): 1052-1057.
Barnett, P.G., T.E. Keeler and T.-W. Hu (1995), "Oligopoly structure and the incidence of cigarette excise
taxes", Journal of Public Economics 57(3):457-470.
Bartlett, J.C., L.S. Miller, D.P. Rice and W.B. Max (1994), "Medical care expenditures attributable to cigarette
smoking - United States 1993", MMWR 43:469-472.
Barzel, Y. (1976), "An alternative approach to the analysis of taxation", Journal of Political Economy
84(6):1177-1197.
Becker, G.S., M. Grossman and K.M. Murphy (1991), "Rational addiction and the effect of price on con-
sumption", American Economic Review 81:237-241.
Becker, G.S., M. Grossman and K.M. Murphy (1994), "An empirical analysis of cigarette addiction", Ameri-
can Economic Review 84(3):396-418.
Becker, G.S., and C.B. Mulligan (1997), "The endogenous determination of time preference", Quarterly Jour-
nal of Economics 112(3):729-758.
Becker, G.S., and K.M. Murphy (1988), "A theory of rational addiction", Journal of Political Economy
96(4):675-700.
Bickel, W.K., R.J. DeGrandpre and S.T. Higgins (1993), "Behavioral economics: A novel experimental ap-
proach to the study of drug dependence", Drug and Alcohol Dependence 33(2):173-192.
Bickel, W.K., and R.J. DeGrandpre (1996), "Modeling drug abuse policy in the behavioral economics labo-
ratory", in: L. Green and J.H. Kagel, eds., Advances in Behavioral Economics. Volume 3: Substance Use
and Abuse (Ablex Publishing Corporation, Norwood, NJ).
Bickel, W.K., R.J. DeGrandpre, S.T. Higgins and J.R. Hughes (1990), "Behavioral economics of drug self-
administration. I. Functional equivalence of response requirement and drug dose", Life Science 47:1501-
1510.
Bickel, W.K., R.J. DeGrandpre, S.T. Higgins, J.R. Hughes and G. Badger (1995), "Effects of simulated em-
ployment and recreation on cigarette smoking: A behavioral economic analysis", Experimental and Clinical
Psychopharmacology 3:467-476.
Bickel, W.K., R.J. DeGrandpre, J.R. Hughes and S.T. Higgins (1991), "Behavioral economics of drug self-
administration. II. A unit-price analysis of cigarette smoking", Journal of the Experimental Analysis of
Behavior 55(2):145-154.
Bickel, W.K., and G.J. Madden (1999), "The behavioral economics of smoking", in: EJ. Chaloupka,
M. Grossman, W.K. Bickel and H. Saffer, eds., The Economic Analysis of Substance Use and Abuse: An
Integration of Econometric and Behavioral Economic Research (University of Chicago Press, Chicago).
Bishop, J.A., and J.H. Yoo (1985), " 'Health scare', excise taxes and advertising ban in the cigarette demand
and supply", Southern Economic Journal 52(2):402-411.
Boyd, R., and B.J. Seldon (1990), "The fleeting effect of advertising: Empirical evidence from a case study",
Economic Letters 34:375-379.
Boyer, M. (1978), "A habit forming optimal growth model", International Economic Review 19:585-609.
Boyer, M. (1983), "Rational demand and expenditures patterns under habit formation", Journal of Economic
Theory 31:27-53.
British American Tobacco (1994), Tobacco Taxation Guide: A Guide to Alternative Methods of Taxing
Cigarettes and Other Tobacco Products (Optichrome The Printing Group, Woking, England).
Broders, A.C. (1920), "Squamous-cell epithelioma of the lip. A study of five hundred and thirty-seven cases",
Journal of the American Medical Association 74:656-664.
Brown, A.B. (1998), "Farm level effects of an increase in federal cigarette excise taxes under two scenarios:
Keep vs. eliminate the tobacco program", USDA Outlook Conference, March 9, Washington.
Brownson, R.C., J. Jackson-Thompson, J.C. Wilkerson, et al. (1992), "Demographic and socioeconomic dif-
ferences in beliefs about the health effects of smoking", American Journal of Public Health 82:99-103.
1614 FEJ. Chaloupka and K.E. Warner
Buck, D., C. Godfrey, M. Raw and M. Sutton (1995), Tobacco and Jobs (Society for the Study of Addiction
and the Centre for Health Economics, University of York, York, England).
Bulow, J.I., and P. Pfleiderer (1983), "A note on the effect of cost changes on prices," Journal of Political
Economy 91(1):182-185.
Cameron, S. (1997), "Are Greek smokers rational addicts?", Applied Economics Letters 4(7):401-402.
Capehart, T. (1997), "The tobacco program - a summary and update", Tobacco Situation & Outlook Report
TBS-238 (U.S. Department of Agriculture, Economic Research Service).
Centers for Disease Control (1996), "Projected smoking related deaths among youth - United States", Mor-
bidity and Mortality Weekly Report 45(44):966-970.
Chaloupka, EJ. (1988), "An economic analysis of addictive behavior: The case of cigarette smoking", Disser-
tation (City University of New York Graduate School, New York).
Chaloupka, F.J. (1990), "Men, women, and addiction: The case of cigarette smoking", Working Paper Number
3267 (National Bureau of Economic Research).
Chaloupka, F.J. (1991), "Rational addictive behavior and cigarette smoking", Journal of Political Economy
99(4):722-742.
Chaloupka, EJ. (1992), "Clean indoor air laws, addiction, and cigarette smoking", Applied Economics
24(2):193-205.
Chaloupka, EJ. (1998), "The impact of proposed cigarette price increases", Policy Analysis No. 9, Health
Sciences Analysis Project (Advocacy Institute, Washington).
Chaloupka, F.J., and M. Grossman (1996), "Price, tobacco control policies and youth smoking", Working
Chaloupka, F.J., and R.L. Pacula (1998), "Limiting youth access to tobacco: The early impact of the synar
amendment on youth smoking", Working Paper (Department of Economics, University of Illinois at
Chicago).
Chaloupka, F.J., and R.L. Pacula (1999), "An examination of gender and race differences in youth smoking
responsiveness to price and tobacco control policies", Tobacco Control, forthcoming.
Chaloupka, EJ., and H. Saffer (1992), "Clean indoor air laws and the demand for cigarettes", Contemporary
Policy Issues 10(2):72-83.
Chaloupka, EJ., J.A. Tauras and M. Grossman (1997), "Public policy and youth smokeless tobacco use",
Southern Economic Journal 64(2):503-516.
Chaloupka,F.J., and H. Wechsler (1997), "Price, tobacco control policies and smoking among young adults",
Chapman, S. (1989), "The limitations of econometric analysis in cigarette advertising studies", British Journal
of Addiction 84:1267-1274.
Chapman, S., and J. Richardson (1990), "Tobacco excise and declining consumption: The case of Papua New
Guinea", American Journal of Public Health 80(5):537-540.
Chapman, S., and W.L. Wong (1990), Tobacco Control in the Third World: A Resource Atlas (International
Organization of Consumers Unions, Penang, Malaysia).
Chase Econometrics (1985), The Economic Impact of the Tobacco Industry on the United States Economy in
1983 (Chase Econometrics, Bala Cynwyd, PA).
Chetwynd, J., P. Coope, R.J. Brodie and E. Wells (1988), "Impact of cigarette advertising on aggregate de-
mand for cigarettes in New Zealand", British Journal of Addiction 83:409-414.
Chow, G. (1960), "Statistical demand functions for automobiles and their use for forecasting", in: A.C. Har-
berger, ed., The Demand for Durable Goods (University of Chicago Press, Chicago) 149-178.
Coalition on Smoking Or Health (1994a), Saving Lives and Raising Revenue: The Case for a $2 Federal
Tobacco Tax Increase (Coalition on Smoking or Health, Washington).
Collishaw, N.E., and G. Myers (1984), "Dollar estimates of the consequences of tobacco use in Canada,
1979", Canadian Journal of Public Health 75(3):192-199.
Congressional Budget Office (1990), Federal Taxation of Tobacco, Alcoholic Beverages, and Motor Fuels
(US Government Printing Office, Washington).
Congressional Research Service (1994), "U.S. tobacco production: Prospects for the future", CRS Publication
No. 94-672 ENR (Congressional Research Service, The Library of Congress), August 24.
Conniffe, D. (1995), "Models of Irish tobacco consumption", Economic and Social Review 26(4):331-347.
Cook, P.J. (1991), '"The social costs of drinking", in: The Negative Social Consequences of Alcohol Use
(Norwegian Ministry of Health and Social Affairs, Oslo) 49-81.
Cook, P.J., and M.J. Moore (1993), "Taxation of alcoholic beverages", in: M.E. Hilton and G. Bloss, eds., Eco-
nomics and the Prevention of Alcohol-Related Problems. Research Monograph No. 25, NIH Publication
No. 93-3513 (US Department of Health and Human Services, Public Health Service, National Institutes of
Health, National Institute on Alcohol Abuse and Alcoholism, Rockville, MD) 33-58.
Cordes, J.J., E.M. Nicholson and F.J. Sammartino (1990), "Raising revenue by taxing activities with social
costs", National Tax Journal 43(3):343-356.
Cox, H. (1984), "Smoking, tobacco promotion, and the voluntary agreements", British Medical Journal
288:303-305.
Cox, H., and R. Smith (1984), "Political approaches to smoking control: A comparative analysis", Applied
Economics 16(4):569-582.
Cromwell, J., W.J. Bartosch, M.C. Fiore, et al. (1997), "Cost-effectiveness of the clinical practice recommen-
dations in the AHCPR guideline for smoking cessation", Journal of the American Medical Association
278(21): 1759-1766.
DeCicca, P., D. Kenkel and A. Mathios (1998a), "Putting out the fires: Will higher cigarette taxes reduce
youth smoking?", Presented at the Annual Meetings of the American Economic Association.
DeCicca, P., D. Kenkel and A. Mathios (1998b), "Putting out the fires: Will higher cigarette taxes reduce
youth smoking", Working Paper (Department of Policy Analysis and Management, Cornell University).
Dee, T.S., and W.N. Evans (1998), "A comment on DeCicca, Kenkel, and Mathios", Working Paper (School
of Economics, Georgia Institute of Technology).
DeGrandpre, R.J., and W.K. Bickel (1995), "Human drug self-administration in a medium of exchange",
Experimental and Clinical Psychopharmacology 3:349-357.
DeGrandpre, R.J., W.K. Bickel, S.T. Higgins and J.R. Hughes (1994), "A behavioral economic analysis of
concurrently available money and cigarettes", Journal of the Experimental Analysis of Behavior 61(2): 191-
201.
DeGrandpre, R.J., W.K. Bickel, J.R. Hughes and S.T. Higgins (1992), "Behavioral economics of drug self-
administration III. A reanalysis of the nicotine regulation hypothesis", Psychopharmacology 108(1-2):1-
10.
DiFranza, J.R., J.W. Rishards, P.M. Paulman, et al. (1991), "RJR Nabisco's cartoon camel promotes Camel
cigarettes to children", Journal of the American Medical Association 266(22):3149-3153.
Dixit, A., and V. Norman (1978), "Advertising and welfare", Bell Journal of Economics 9:1-17.
Doll, R., and A.B. Hill (1954), "The mortality of doctors in relation to their smoking habits; A preliminary
report", British Medical Journal 1:1451-1455.
Doll, R., and A.B. Hill (1956), "Lung cancer and other causes of death in relation to smoking. A second report
on the mortality of British doctors", British Medical Journal 2:1071-1081.
Douglas, S. (1998), "The duration of the smoking habit", Economic Inquiry 36(1):49-64.
Douglas, S., and G. Hariharan (1994), "The hazard of starting smoking: Estimates from a split population
duration model", Journal of Health Economics 13(2):213-230.
Downey, L.A., and J.A. Gardiner (1996), Reducing Youth Access to Tobacco: A Partial Inventory of State
Initiatives (Office of Social Science Research, University of Illinois at Chicago, Chicago).
Duesenberry, J.S. (1949), Income, Saving, and the Theory of Consumer Behavior (Harvard University Press,
Cambridge, MA).
Duffy, M. (1987), "Advertising and the inter-product distribution of demand: A Rotterdam Model approach",
European Economic Review 31(5):1051-1070.
Duffy, M. (1991), "Advertising and the consumption of tobacco and alcoholic drink: A system-wide analysis",
Scottish Journal of Political Economy 38(4):369-385.
Duffy, M. (1996a), "An econometric study of advertising and cigarette demand in the United Kingdom",
International Journal of Advertising 15:262-284.
Duffy, M. (1996b), "Econometric studies of advertising, advertising restrictions, and cigarette demand: A sur-
vey", International Journal of Advertising 15:1-23.
El-Safty, A.E. (1976a), "Adaptive behavior, demand, and preferences", Journal of Economic Theory 13:298-
318.
El-Safty, A.E. (1976b), "Adaptive behavior and the existence of Weizsacker's long-run indifference curves",
Journal of Economic Theory 13:319-328.
Elster, J. (1979), Ulysses and the Sirens: Studies in Rationality and Irrationality (Cambridge University Press,
Cambridge).
Environmental Protection Agency (1992), "Respiratory health effects of passive smoking: Lung cancer and
other disorders", EPA/600/6-90/006F (Environmental Protection Agency, Office of Research and Develop-
ment, Office of Air and Radiation).
Environmental Protection Agency (1994), "The costs and benefits of smoking restrictions: An assessment of
the smoke-free environment act of 1993", H.R. 3434 (Indoor Air Division, Office of Radiation and Indoor
Air, Environmental Protection Agency, Washington).
Evans, N., A. Farkas, E. Gilpin, et al. (1995), "Influence of tobacco marketing and exposure to smokers on
adolescent susceptibility to smoking", Journal of the National Cancer Institute 87:1538-1545.
Evans, W.N., and M.C. Farrelly (1998), "The compensating behavior of smokers: Taxes, tar and nicotine",
RAND Journal of Economics 29(3):578-595.
Evans, W.N., M.C. Farrelly and E. Montgomery (1999), "Do workplace smoking bans reduce smoking?",
American Economic Review 89(4):728-747.
Evans, W.N., and L.X. Huang (1998), "Cigarette taxes and teen smoking: New evidence from panels of
repeated cross-sections", Working Paper (Department of Economics, University of Maryland).
Evans, W.N., and J.S. Ringel (1999), "Can higher cigarette taxes improve birth outcomes?", Journal of Public
Farrell, M.J. (1952), "Irreversible demand functions", Econometrica 20:171-186.
Farrelly, M.C., and J.W. Bray (1998), "Office on Smoking and Health. Response to increases in cigarette
prices by race/ethnicity, income, and age groups - United States, 1976-1993", Morbidity and Mortality
Weekly Report 47(29):605-609.
Farkas, A.J. (1998), "When does cigarette fading increase the likelihood of future cessation?", Cancer Pre-
vention and Control Program (Cancer Center, University of California, San Diego, La Jolla, CA).
Federal Trade Commission (1997), "Competition and the financial impact of the proposed tobacco industry
settlement" (Federal Trade Commission, Washington, DC).
Federal Trade Commission (1998), "Federal Trade Commission Report to Congress for 1996: Pursuant to the
Federal Cigarette Labeling and Advertising Act" (Federal Trade Commission, Washington).
Flewelling, R.L., E. Kenney, J.P. Elder, J. Pierce, M. Johnson and D.G. Bal (1992), "First-year impact of
the 1989 California cigarette tax increase on cigarette consumption", American Journal of Public Health
82(6):867-869.
Food and Drug Administration (1996), "Regulations restricting the sale and distribution of cigarettes and
smokeless tobacco to protect children and adolescents", Final Rule (Federal Register, Washington).
Forbes, W.F., and M.E. Thompson (1983), "Estimating the health care costs of smokers", Canadian Journal
of Public Health 74(3):183-190.
Forster, J.L., D.M. Murray, M. Wolfson, et al. (1998), "The effects of community policies to reduce youth
access to tobacco", American Journal of Public Health 88(8): 1193-1198.
Franke, G.R. (1994), "U.S. cigarette demand, 1961-90: Econometric issues, evidence, and implications",
Journal of Business Research 30:33-41.
Fritschler, A.L., and J.M. Hoefler (1996), Smoking &Politics: Policy Making and the Federal Bureaucracy,
5th edn. (Prentice-Hall, Upper Saddle River, NJ).
Fujii, E.T. (1980), "The demand for cigarettes: Further empirical evidence and its implications for public
policy", Applied Economics 12:479-489.
Galbraith, J.K. (1958), The Affluent Society (Houghton Mifflin Company, Cambridge, MA).
Galbraith, J.K. (1972), The New Industrial State, 2nd edn. (Pelican Books, New York).
Gale, F. (1997), 'Tobacco dollars and jobs", in: Tobacco Situation &Outlook, TBS-239 (U.S. Department of
Agriculture, Economic Research Service) 37-43.
Gale, F. (1998), "Economic structure of tobacco-growing regions", in: Tobacco Situation & Outlook, TBS-
241 (U.S. Department of Agriculture, Economic Research Service) 40-47.
Garcia dos Santos, J. (1972), "Estimating the durability of consumers' durable goods", Review of Economics
and Statistics 54:475-479.
General Accounting Office (1989), Teenage Smoking: Higher Excise Tax Should Significantly Reduce the
Number of Smokers (General Accounting Office, Washington).
Glantz, S.A., and W.W. Parmley (1995), "Passive smoking and heart disease: Mechanisms and risk", Journal
of the American Medical Association 73(13): 1047-1053.
Goel, R.K., and M.J. Morey (1995), "The interdependence of cigarette and liquor demand", Southern Eco-
Goldstein, A.O., P.M. Fisher, J.W. Richards and D. Creten (1987), "Relationship between high school student
smoking and recognition of cigarette advertisements", Journal of Pediatrics 110(3):488-491.
Goodman, J. (1993), Tobacco in History: The Cultures of Dependence (Routledge, New York).
Gorman, W.M. (1967), "Tastes, habits, and choices", International Economic Review 8:218-222.
Grabowski, H.G. (1976), "The effect of advertising on the inter-industry distribution of demand", Explorations
in Economic Research 3:21-75.
Grabowski, H.G. (1978), "The effects of advertising on intra-industry shifts in demand", Explorations in
Economic Research 3:675-701.
Gravelle, J.G., and D. Zimmerman (1994), "Cigarette taxes to fund health care reform: An economic analy-
sis", CRS Publication No. 94-214E (Congressional Research Service, The Library of Congress, Washing-
ton).
Grise, V.N. (1995), "Tobacco: Background for 1995 farm legislation", Agricultural economic report No. 709
(US Department of Agriculture, Economic Research Service, Washington).
Grossman, M., and EJ. Chaloupka (1997), "Cigarette taxes: The straw to break the camel's back", Public
Health Reports 112(4):290-297.
Grossman, M., EJ. Chaloupka and H. Saffer (1999), "Addiction", in: The New Palgrave Dictionary of Eco-
nomics and the Law (Macmillan Reference Ltd., London), forthcoming.
Grossman, M., D. Coate, E.M. Lewit and R.A. Shakotko (1983), Economic and Other Factors in Youth Smok-
ing (National Science Foundation, Washington).
Haavelmo, T. (1944), "The probability approach in econometrics", Econometrica 12:96-124.
Hamilton, J.L. (1972), "Advertising, the health scare, and the cigarette advertising ban", Review of Economics
and Statistics 54:401-411.
Hamilton, J.L. (1977), "The effect of cigarette advertising bans on cigarette consumption", in: J. Steinfeld,
W. Griffiths, K. Ball and R.M. Taylor, eds., Proceedings of the Third World Conference on Smoking and
Health. Vol. II: Health Consequences, Education, Cessation Activities, and Social Action (US Department
of Health, Education, and Welfare, Washington) 829-840.
Hammond, E.D., and D. Horn (1958a), "Smoking and death rates - report on forty-four months of follow-up
on 187,783 men. I. Total mortality", Journal of the American Medical Association 166:1159-1172.
Hammond, E.D., and D. Horn (1958b), "Smoking and death rates - report on forty-four months of follow-up
on 187,783 men. II. Death rates by cause", Journal of the American Medical Association 166:1294-1308.
Hammond, P.J. (1976a), "Changing tastes and coherent dynamic choice", Review of Economic Studies
43:159-173.
Hammond, P.J. (1976b), "Endogenous tastes and stable long-run choice", Journal of Economic Theory
13:329-340.
Harris, J.E. (1980), "Taxing tar and nicotine", American Economic Review 70(3):300-311.
Harris, J.E. (1987), "The 1983 increase in the federal cigarette excise tax", in: L.H. Summers, ed., Tax Policy
and the Economy, Vol. 1 (MIT Press, Cambridge, MA).
Harris, J.E. (1993), Testimony before the Committee on Ways and Means, U.S. House of Representatives.
Public Hearings on the Financing Provisions of the Administration's Health Security Act, November 18
(Washington, DC).
Harris, J.E. (1994), A Working Model for Predicting the Consumption and Revenue Impacts of Large In-
creases in the U.S. Federal Cigarette Excise Tax. Working Paper No. 4803 (National Bureau of Economic
Research, Cambridge, MA).
Harris, J.E., G.N. Connolly and B. Davis (1996), "Cigarette smoking before and after an excise tax increase
and an antismoking campaign - Massachusetts, 1990-1996", Morbidity and Mortality Weekly Report
45(44):966-969.
Hay, J.W. (1991), "The harm they do to others: A primer on the external costs of drug abuse", in: M.B.
Krauss and E.P. Lazear, eds., Searching for Alternatives: Drug-Control Policy in the United States (Hoover
Institution Press, Stanford, CA) 200-225.
Hodgson, T.A. (1988), "Annual costs of illness versus lifetime costs of illness and implications of structural
change", Drug Information Journal 22:323-341.
Hodgson, T.A. (1992), "Cigarette smoking and lifetime medical expenditures", Milbank Quarterly 70(1):81-
125.
Hodgson, T.A. (1998), "The health care costs of smoking", New England Journal of Medicine 338(7):470.
Hodgson, T.A., and M.R. Meiners (1982), "Cost-of-illness methodology: a guide to current practices and
procedures", Milbank Memorial Fund Quarterly 60:429-462.
Holak, S.L., and S.K. Reddy (1986), "Effects of a television and radio advertising ban: a study of the cigarette
industry", Journal of Marketing 50:219-227.
Houthakker, H.S., and L.D. Taylor (1966), Consumer Demand in the United States, 1929-1970: Analyses and
Projections (Harvard University Press, Cambridge, MA).
Houthakker, H.S., and L.D. Taylor (1970), Consumer Demand in the United States, 1929-1970: Analyses and
Projections, 2nd edn. (Harvard University Press, Cambridge, MA).
Hsieh, C.-R., T.-W. Hu (1997), "The demand for cigarettes in Taiwan: Domestic versus imported cigarettes",
Discussion Paper No. 9701 (The Institute of Economics, Academia Sinica, Nankang, Taipei).
Hsieh, C.-R., L.-L. Yen, J.-T. Liu and C.J. Lin (1996), "Smoking, health knowledge and anti-smoking cam-
paigns: An empirical study in Taiwan", Journal of Health Economics 15(1)87-104.
Hu, T.-W., J. Bai, T.E. Keeler., P.G. Barnett and H.-Y. Sung (1994), "The impact of California Proposition 99,
a major anti-cigarette law, on cigarette consumption", Journal of Public Health Policy 15(1):26-36.
Hu, T.-W., Q.-F. Ren, T.E. Keeler and J. Bartlett (1995a), "The demand for cigarettes in California and be-
havioural risk factors", Health Economics 4(1):7-14.
Hu, T.-W., H.-Y. Sung and T.E. Keeler (1995b), "Reducing cigarette consumption in California: Tobacco taxes
vs an anti-smoking media campaign", American Journal of Public Health 85(9):1218-1222.
Hu, T.-W., H.-Y. Sung and T.E. Keeler (1995c), "The state antismoking campaign and the industry response:
The effects of advertising on cigarette consumption in California", American Economic Review 85(2):85-
90.
Hu, T.-W., H.-Y. Sung, T.E. Keeler, M. Marciniak, A. Keith and R. Manning (1998), Cigarette Consumption
and Sales of Nicotine Replacement Products (University of California, Berkeley, CA).
Hursh, S.R., and R.A. Bauman (1987), "The behavioral analysis of demand", in: L. Green and J.H. Kagel,
eds., Advances in Behavioral Economics, Vol. 1 (Ablex Publishing Corporation, Norwood, NJ) 117-165.
Ippolito, R.A., R.D. Murphy and D. Sant (1979), Staff Report on Consumer Responses to Cigarette Health
Information (Federal Trade Commission, Washington).
Jason, L.A, W.D. Billows, D.L. Schnopp-Wyatt and C. King (1996), "Long-term findings from Woodridge in
reducing illegal cigarette sales to older minors", Evaluation and the Health Professions 19:3-13.
Jin, S.G., B.Y. Lu, D.Y. Yan, et al. (1995), "An evaluation of smoking-induced health costs in China (1988-
1989)", Biomedical and Environmental Sciences 8:342-349.
Johnson, L.W. (1986), "Advertising expenditure and the aggregate demand for cigarettes in Australia", Inter-
national Journal of Advertising 1:45-58.
Johnson, P.R. (1965), "The social cost of the tobacco program", Journal of Farm Economics 47:242-255.
Johnson, T.R. (1978), "Additional evidence on the effects of alternative taxes on cigarette prices", Journal of
Political Economy 86(2, Pt. 1):325-328.
Jones, A.M. (1989), "A systems approach to the demand for alcohol and tobacco", Bulletin of Economic
Research 41:85-105.
Jones, A.M. (1994), "Health, addiction, social interaction and the decision to quit smoking", Journal of Health
Jones, A., and J. Posnett (1988), 'The revenue and welfare effects of cigarette taxes", Applied Economics
20:1223-1232.
Joossens, L. (1998), "Tobacco smuggling: An optimal policy approach", in: I. Abedian, R. van der Merwe,
N. Wilkins and P. Jha, eds., The Economics of Tobacco Control: Towards an Optimal Policy Mix (Applied
Fiscal Research Centre, University of Cape Town, Cape Town, South Africa).
Joossens, L., and M. Raw (1995), "Smuggling and cross border shopping of tobacco in Europe", British
Joossens, L., and M. Raw (1998), "Cigarette smuggling in Europe: Who really benefits?", Tobacco Control
7, in press.
Joossens, L., and R. van der Merwe (1997), "Cigarette trade and smuggling", Project Update #7, The Eco-
nomics of Tobacco Control Project (Cape Town, South Africa).
Kao, K., and V.J. Tremblay (1988), "Cigarette 'health scare', excise tax, and advertising ban: Comment",
Southern Economic Journal 54(3):770-776.
Keeler, T.E., T.-W. Hu, P.G. Barnett and W.G. Manning (1993), "Taxation, regulation and addiction: A demand
function for cigarettes based on time-series evidence", Journal of Health Economics 12(1):1-18.
Keeler, T.E., T.-W. Hu, P.G. Barnett, W.G. Manning and H.Y. Sung (1996), "Do cigarette producers price-
discriminate by state? An empirical analysis of local cigarette pricing and taxation", Journal of Health
Kenkel, D.S. (1991), "Health behavior, health knowledge, and schooling", Journal of Political Economy
99(2):287-305.
King, C., M. Siegel, C. Celebucki and G.N. Connolly (1998), "Adolescent exposure to cigarette advertising
in magazines: An evaluation of brand-specific advertising in relation to youth readership", Journal of the
American Medical Association 279(7):516-520.
Kluger, R. (1996), Ashes to Ashes: America's Hundred-Year Cigarette War, the Public Health, and the Un-
abashed Triumph of Philip Morris (Alfred A. Knopf, New York).
Lambin, J.J. (1976), Advertising, Competition, and Market Conduct in Oligopoly Over Time (North-Holland,
Amsterdam).
Laugesen, M., and C. Meads (1991), "Tobacco advertising restrictions, price, income, and tobacco consump-
tion in OECD countries 1960-1986", British Journal of Addiction 86:1343-1354.
Laugesen, M., and C. Meads (1993), "The authors reply to "Tobacco consumption and advertising restrictions:
a critique of Laugesen and Meads (1991)" ", International Journal of Advertising 12:75-80.
Leeflang, P.S.H., and J.C. Reuijl (1985), "Advertising and industry sales: An empirical study of the West
German cigarette market", Journal of Marketing 49:92-98.
Leu, R.E. (1984), "Anti-smoking publicity, taxation, and the demand for cigarettes", Journal of Health Eco-
nomics 3(2): 101-116.
Leu, R.E., and T. Schaub (1983), "Does smoking increase medical expenditures?", Social Science and
Medicine 17:1907-1914.
Lewit, E.M., and D. Coate (1982), "The potential for using excise taxes to reduce smoking", Journal of Health
Economics 1(2):121-145.
Lewit, E.M., D. Coate and M. Grossman (1981), "The effects of government regulation on teenage smoking",
Journal of Law and Economics 24(3):545-569.
Lewit, E.M., A. Hyland, N. Kerrebrock and K.M. Cummings (1997), "Price, public policy and smoking in
young people", Tobacco Control 6(S2):17-24.
Lluch, C. (1974), "Expenditure, savings, and habit formation", International Economic Review 15:786-797.
Lombard, H.L., and C.R. Doering (1928), "Classics in oncology. Cancer studies in Massachusetts. 2. Habits,
characteristics and environment of individuals with and without cancer", New England Journal of Medicine
198:481-487.
Luik, J.C., ed. (1994), Do Tobacco Advertising Bans Really Work? A Review of the Evidence (The Niagara
Institute, Niagara-on-the-Lake, ON).
Lyon, A.B., and R.M. Schwab (1995), "Consumption taxes in a life-cycle framework: Are sin taxes regres-
sive?", Review of Economics and Statistics 77(3):389-406.
Manchester, P.B. (1976), "Interstate cigarette smuggling". Public Finance Quarterly 4:225-238.
Manning, W.G., E.B. Keeler, J.P. Newhouse, E.M. Sloss and I. Wasserman (1989), "The taxes of sin: Do
smokers and drinkers pay their way?", Journal of the American Medical Association 261(11):1604-1609.
Manning, W.G., E.B. Keeler, J.P. Newhouse, E.M. Sloss and J. Wasserman (1991), The Costs of Poor Health
Habits (Harvard University Press, Cambridge, MA).
Mao, Z.Z. (1996), "Demand for cigarettes and price policy: A time-series analysis" (in Chinese), Working
Paper (School of Public Health, West China University of Medical Sciences).
Maranvanyika, E. (1998), "The search for an optimal tobacco control policy in Zimbabwe", in: I. Abedian,
R. van der Merwe, N. Wilkins and P. Jha, eds., The Economics of Tobacco Control: Towards an Optimal
Policy Mix (Applied Fiscal Research Centre, University of Cape Town, Cape Town, South Africa).
Marshall, A. (1920), Principles of Economics, 8th edn. (Macmillan, London).
McAuliffe, R. (1988), "The FTC and the effectiveness of cigarette advertising regulations", Journal of Public
Policy and Marketing 7:49-64.
McGuinness, T., and K. Cowling (1975), "Advertising and the aggregate demand for cigarettes", European
McGuinness, T., and K. Cowling (1980), "Advertising and the aggregate demand for cigarettes: A reply",
European Economic Review 14(1):127-130.
McKenzie, R.B. (1979), "The non-rational domain and the limits of economic analysis", Southern Economic
Journal 46(1):145-157.
McLeod, P.B. (1986), "Advertising bans, tobacco and cigarette consumption", Economic Letters 20(4):391-
396.
Meier, K.J., and M.J. Licari (1997), "The effect of cigarette taxes on cigarette consumption, 1955 through
1994", American Journal of Public Health 87(7):1126-1130.
Metra Consulting Group (1979), The Relationship between Total Cigarette Advertising and Total Cigarette
Consumption in the UK (Metra Consulting Group, London).
Miller, L.S., X. Zhang, D.P. Rice and W. Max (1998), "State estimates of total medical expenditures at-
tributable to cigarette smoking, 1993", Public Health Reports 113:447-458.
Modigliani, E (1949), "Fluctuations in the savings-income ratio: A problem in economic forecasting", in:
Studies in Income and Wealth, Vol. 11 (National Bureau of Economic Research, New York) 371-443.
Moore, M.J. (1996), "Death and tobacco taxes", RAND Journal of Economics 27(2):415-428.
Mullahy, J. (1985), "Cigarette smoking: habits, health concerns, and heterogeneous unobservables in a micro-
econometric analysis of consumer demand", Dissertation (University of Virginia, Charlottesville, VA).
Murray, C.J.L., and A.D. Lopez, eds. (1996), The Global Burden of Disease, Vol. I: A Comprehensive As-
sessment of Mortality and Disability from Diseases, Injuries, and Risk Factors in 1990 and Projected to
2020 (Harvard University Press, Cambridge, MA).
Napier, K. (1996), Cigarettes: What the Warning Label Doesn't Tell You. The First Comprehensive Guide to
the Health Consequences of Smoking (American Council on Science and Health, New York).
Norr, R. (1952), "Cancer by the carton", Reader's Digest 61(December):7-8.
Ohsfeldt, R.L., and R.G. Boyle (1994), "Tobacco excise taxes and rates of smokeless tobacco use in the US:
An exploratory ecological analysis", Tobacco Control 3(4):316-323.
Ohsfeldt, R.L., R.G Boyle and E.I. Capilouto (1997), "Effects of tobacco excise taxes on the use of smokeless
tobacco products", Health Economics 6(5):525-532.
Ohsfeldt, R.L., R.G. Boyle and E.I. Capilouto (1999), "Tobacco taxes, smoking restrictions, and tobacco use",
in: F.J. Chaloupka, M. Grossman, W.K. Bickel and H. Saffer, eds., The Economic Analysis of Substance
Use and Abuse: An Integration of Econometric and Behavioral Economic Research (University of Chicago
Press, Chicago).
Orphanides, A., and D. Zervos (1995), "Rational addiction with learning and regret", Journal of Political
Economy 103:739-758.
Oster, G., G.A. Colditz and N.L. Kelly (1984), The Economic Costs of Smoking and Benefits of Quitting
(D.C. Heath and Company, Lexington, MA).
Oster, G., T.E. Delea, D.M. Huse, et al. (1996), "The benefits and risks of over-the-counter availability of
nicotine polacrilex ("nicotine gum")", Medical Care 34:389-402.
Pashardes, P. (1986), "Myopic and forward looking behavior in a dynamic demand system", International
Pearl, R. (1938), "Tobacco smoking and longevity", Science 87:216-217.
Peles, Y. (1971), "Rates of amortization of advertising expenditures", Journal of Political Economy
79(5):1032-1058.
Pekurinen, M. (1989), "The demand for tobacco products in Finland", British Journal of Addiction 84:1183-
1192.
Pekurinen, M. (1991), Economic Aspects of Smoking: Is There a Case for Government Intervention in Fin-
land? (Vapk-Publishing, Helsinki).
Peterson, D.E., S.L. Zeger, P.L. Remington and H.A. Anderson (1992), "The effect of state cigarette tax
increases on cigarette sales 1955-1988", American Journal of Public Health 82(1):94-96.
Peto, R., A.D. Lopez and L. Boqi (1999), "Global tobacco mortality: Monitoring the growing epidemic", in:
R. Lu, J. Mackay, S. Niu and R. Peto, eds., The Growing Epidemic (Springer-Verlag, Singapore), in press.
Phlips, L. (1983), Applied Consumption Analysis. Advanced Textbooks in Economics (North-Holland, Am-
sterdam).
Phlips, L., and F. Spinnewyn (1982), "Rationality versus myopia in dynamic demand systems", in: R.L.
Basman and G.F. Rhodes, Jr., eds., Advances in Econometrics, Vol. 1 (JAI Press, Greenwich, CT) 3-33.
Pigou, A.C. (1962), A Study in Public Finance, 3rd rev. edn. (Macmillan, London).
Pierce, J.P., W.S. Choi, E.A. Gilpin, et al. (1998), "Tobacco industry promotion of cigarettes and adolescent
smoking", Journal of the American Medical Association 279(7):511-515.
Pollak, R.A. (1968), "Consistent planning", Review of Economic Studies 35:201-208.
Pollak, R.A. (1970), "Habit formation and dynamic demand functions", Journal of Political Economy 78:745-
763.
Pollak, R.A. (1975), "The intertemporal cost of living index", Annals of Economic and Social Measurement
4:179-195.
Pollak, R.A. (1976), "Habit formation and long-run utility functions", Journal of Economic Theory 13:272-
297.
Pollak, R.A. (1977), "Price dependent preferences", American Economic Review 67:64-75.
Pollak, R.A. (1978), "Endogenous tastes in demand and welfare analysis", American Economic Review
68:374-379.
Pollay, R.A., S. Siddarth, M. Siegel, et al. (1996), "The last straw? Cigarette advertising and realized market
shares among youths and adults, 1979-1993", Journal of Marketing 60:1-16.
Porter, R.H. (1986), "The impact of government policy on the U.S. cigarette industry", in: P.M. Ippolito and
D.T Scheffman, eds., Empirical Approaches to Consumer Protection Economics (US Government Printing
Office, Washington).
Price Waterhouse (1990), The Economic Impact of the Tobacco Industry on the United States Economy (Price
Waterhouse, Arlington, VA).
Price Waterhouse (1992), The Economic Impact of the Tobacco Industry on the United States Economy (Price
Waterhouse, Arlington, VA).
Radfar, M. (1985), "The effect of advertising on total consumption of cigarettes in the UK", European Eco-
Rao, R., and P. Miller (1975), "Advertising/sales response functions", Journal of Advertising Research 15:7-
15.
Reuijl, J.C. (1982), On the Determination of Advertising Effectiveness: An Empirical Study of the German
Cigarette Market (Kluwer-Nijhoff, Boston).
Rice, D.P., T.A. Hodgson, P. Sinsheimer, W. Browner and A.N. Kopstein (1986), "The economic costs of the
health effects of smoking, 1984", Milbank Quarterly 64(4):489-547.
Rigotti, N.A., J.R. DiFranza, YC. Chang, et al. (1997), "The effect of enforcing tobacco-sales laws on adoles-
cents' access to tobacco and smoking behavior", New England Journal of Medicine 337(15):1044-1051.
Roberts, M.J., and L. Samuelson (1988), "Am empirical analysis of dynamic, nonprice competition in an
oligopolistic industry", RAND Journal of Economics 19(2):200-220.
Roemer, R. (1993), Legislative Action to Combat the World Tobacco Epidemic, 2nd edn. (World Health
Organization, Geneva).
Royal College of Physicians (1962), "Smoking and health", Summary and Report of the Royal College of
Physicians of London on Smoking in Relation to Cancer of the Lung and Other Diseases (Pitman Publish-
ing, New York).
Ryder, H.E., and G.M. Heal (1973), "Optimal growth with intertemporally dependent preferences", Review
of Economic Studies 40:1-31.
Saba, R.P., T.R. Beard, R.B. Ekelund and R.W. Ressler (1995), "The demand for cigarette smuggling", Eco-
nomic Inquiry 33(2):189-202.
Saffer, H. (1995), "Alcohol advertising and alcohol consumption: Econometric studies", in: S.E. Martin, ed.,
The Effects of the Mass Media on the Use and Abuse of Alcohol (National Institute on Alcohol Abuse and
Alcoholism, Bethesda).
Saffer, H. (1998), "Economic issues in cigarette and alcohol advertising", Journal of Drug Issues 28(3):781-
793.
Schelling, T.C. (1978), "Egonomics, or the art of self-management", American Economic Review 68:290-
294.
Schelling, T.C. (1980), "The intimate contest for self-command", The Public Interest 60:94-113.
Schelling, T.C. (1984a), Choice and Consequence (Harvard University Press, Cambridge, MA).
Schelling, T.C. (1984b), "Self-command in practice, in policy, and in a theory of rational choice", American
Schelling, T.C. (1986), "Economics and cigarettes", Preventive Medicine 15(5):549-560.
Schmalensee, R.L. (1972), On the Economics of Advertising (North Holland, Amsterdam).
Schmidt, P., and A.D. Witte (1989), "Predicting criminal recidivism using 'split population' survival time
models", Journal of Econometrics 40(1):141-159.
Schnabel, M. (1972), "An oligopoly model of the cigarette industry", Southern Economic Journal 38(3):325-
335.
Schneider, L., B. Klein and K. Murphy (1981), "Government regulation of cigarette health information",
Schoenbaum, M. (1997), "Do smokers understand the mortality effects of smoking? Evidence from the Health
and Retirement Survey", American Journal of Public Health 87(5):755-759.
Scitovsky, T. (1976), The Joyless Economy: An Inquiry into Consumer Satisfaction and Human Dissatisfac-
tion (Oxford University Press, Oxford).
Scott, E., and J. Dickert (1993), "From research to policy: The cigarette excise tax", Case Studies in Public
Policy and Management, #C16-93-1233.0 (John E Kennedy School of Government, Cambridge, MA).
Seldon, B.J., and R. Boyd (1991), "The stability of cigarette demand", Applied Economics 23:319-326.
Seldon, B.J., and K. Doroodian (1989), "A simultaneous model of cigarette advertising: Effects on demand
and industry response to public policy", Review of Economics and Statistics 71:673-677.
Shoven, J.B., J.O. Sundberg and J.P, Bunker (1989), "The social security cost of smoking", in: D.A. Wise,
ed., The Economics of Aging (University of Chicago Press, Chicago) 231-254.
Showalter, M.H. (1991), "Essays in applied econometrics. Essay III: Monopoly behavior with intertemporal
demands", Dissertation (Massachusetts Institute of Technology, Cambridge, MA).
Showalter, M.H. (1998), "Firm behavior in a market with addiction: The case of cigarettes", Working Paper
(Department of Economics, Brigham Young University).
Shultz, J.M., T.E. Novotny and D.P. Rice (1991), "Quantifying the disease impact of cigarette smoking with
SAMMEC II software", Public Health Reports 106:326-333.
Simonich, W.L. (1991), Government Antismoking Policies (Peter Lang Publishing, New York).
Spinnewyn, F. (1981), "Rational habit formation", European Economic Review 15:91-109.
Stavrinos, V.G. (1987), "The effects of an anti-smoking campaign on cigarette consumption: Empirical evi-
dence from Greece", Applied Economics 19(3):323-329.
Steenland, K. (1992), "Passive smoking and the risk of heart disease", Journal of the American Medical
Association 267:94-99.
Steenland, K., M. Thun, C. Lally and C. Heath Jr. (1996), "Environmental tobacco smoke and coronary heart
disease in the American Cancer Society CPS-II cohort", Circulation 94(4):622-628.
Stewart, M.J. (1992), "Tobacco consumption and advertising restrictions: A critique of Laugesen and Meads
(1991)", International Journal of Advertising 11:97-118.
Stewart, M.J. (1993), "The effect on tobacco consumption of advertising bans in OECD countries", Interna-
tional Journal of Advertising 12:155-180.
Stewart, M.J. (1994), "Advertising bans in Norway, Finland and Canada: A comparison of C. Smee (1992)
and M.J. Stewart (1992)", in: J.C. Luik, ed., Do Tobacco Advertising Bans Really Work? (The Niagara
Institute, Niagara-on-the-Lake, ON).
Stigler, G., and G.S. Becker (1977), "De gustibus non est disputandum", American Economic Review 67:76-
90.
Strotz, R.H. (1956), "Myopia and inconsistency in dynamic utility maximization", Review of Economic Stud-
ies 23:165-180.
Sullivan, D. (1985), "Testing hypotheses about firm behavior in the cigarette industry", Journal of Political
Economy 93(3):586-598.
Sullum, J. (1998), For Your Own Good: The Anti-Smoking Crusade and the Tyranny of Public Health (The
Free Press, New York).
Sumner, D.A. (1981), "Measurement of monopoly behavior: an application to the cigarette industry", Journal
of Political Economy 89(5):1010-1019.
Sumner, D.A., and J.M. Alston (1985), Removal of Price Supports and Supply Controls for U.S. Tobacco: An
Economic Analysis of the Impact (National Planning Association, Washington).
Sumner, D.A., and M.K. Wohlgenant (1985), "Effects of an increase in the federal excise tax on cigarettes",
American Journal of Agricultural Economics 67(2):235-242.
Sumner, M.T. (1971), "Demand for tobacco in the U.K.", The Manchester School of Economic and Social
Studies 39:23-36.
Sumner, M.T., and R. Ward (1981), "Tax changes and cigarette prices", Journal of Political Economy
89(6): 1261-1265.
Sung, H.-Y., T.-W. Hu and T.E. Keeler (1994), "Cigarette taxation and demand: An empirical model", Con-
temporary Economic Policy 12(3):91-100.
Suranovic, S.M., R.S. Goldfarb and T.C. Leonard (1999), "An economic theory of cigarette addiction", Jour-
nal of Health Economics 18(1):1-29.
Sweanor, D.T. (1997), Global Cigarette Taxes and Prices (Smoking and Health Action Foundation, Ottawa,
Canada).
Sweanor, D.T., and L.R. Martial (1994), The Smuggling of Tobacco Products: Lessons from Canada (Non-
Smokers' Rights Association/Smoking and Health Action Foundation, Ottawa, Canada).
Tansel, A. (1993), "Cigarette demand, health scares and education in Turkey", Applied Economics 25(4):521-
529.
Tauras, J.A., and EJ. Chaloupka (1999), "Price, clean indoor air laws, and cigarette smoking: Evidence
from longitudinal data for young adults", Working Paper Number 6937 (National Bureau of Economic
Research).
Taylor, P. (1984), The Smoke Ring: Tobacco, Money, and Multinational Politics (Pantheon Books, New York).
Tegene, A. (1991), "Kalman filter and the demand for cigarettes", Applied Economics 23(7): 1175-1182.
Telser, L.G. (1962), "Advertising and cigarettes", Journal of Political Economy 70(5):471-499.
Tennant, R.B. (1950), The American Cigarette Industry: A Study in Economic Analysis and Public Policy
(Yale University Press, New Haven, CT).
Thaler, R., and H.M. Shefrin (1981), "An economic theory of self control", Journal of Political Economy
89:392-406.
Theil, H. (1980), System-wide Explorations in International Economics, Input-Output Analysis, and Market-
ing Research (North Holland, Amsterdam).
Thompson, M.E., and I. McLeod (1976), "The effects of economic variables upon the demand for cigarettes
in Canada", Mathematical Scientist 1:121-132.
Thursby, J.G., and M.C. Thursby (1994), "Interstate cigarette bootlegging: Extent, revenue losses, and effects
of federal intervention", Working Paper No. 4763 (National Bureau of Economic Research).
Tobacco Institute (1998), The Tax Burden on Tobacco. Historical Compilation 1995, Vol. 30 (Tobacco Insti-
tute, Washington).
Tobacco Merchants Association (1995), Tobacco's Contribution to the National Economy (Tobacco Mer-
chants Association, Princeton, NJ).
Tollison, R.D. (1994), "Testimony", Hearing, April 28, 1994 (Committee on Finance, United States Senate.
Tax Treatment of Organizations Providing Health Care Services, and Excise Taxes on Tobacco, Guns and
Ammunition; US Government Printing Office, Washington) 40.
Tollison, R.D., and R.E. Wagner (1992), The Economics of Smoking (Kluwer Academic Publishers, Boston).
Townsend, J.L. (1987), "Cigarette tax, economic welfare, and social class patterns of smoking", Applied
Townsend, J.L. (1998), "U.K. smoking targets: Policies to attain them and effects on premature mortality", in:
I. Abedian, R. van der Merwe, N. Wilkins and P. Jha, eds., The Economics of Tobacco Control: Towards an
Optimal Policy Mix (Applied Fiscal Research Centre, University of Cape Town, Cape Town, South Africa).
Townsend, J.L., P. Roderick and J. Cooper (1994), "Cigarette smoking by socioeconomic group, sex, and age:
Effects of price, income, and health publicity", British Medical Journal 309(6959):923-926.
Tremblay, C.H., and V.J. Tremblay (1995), "The impact of cigarette advertising on consumer surplus, profit,
and social welfare", Contemporary Economic Policy 13(1):113-124.
Treyz, G.I. (1993), Regional Economic Modeling: A Systematic Approach to Economic Forecasting and
Policy Analysis (Kluwer Academic Publishers, Boston).
Tye, J.B., K.E. Warner and S.A. Glantz (1987), "Tobacco advertising and consumption: Evidence of a causal
relationship", Journal of Public Health Policy 8:492-508.
UK Department of Health (1992), Effect of Tobacco Advertising on Tobacco Consumption: A Discussion
Document Reviewing the Evidence (U.K. Department of Health, Economics and Operational Research
Division, London).
US Department of Health and Human Services (1986), "The health consequences of involuntary smoking",
A Report of the Surgeon General (US Department of Health and Human Services, Public Health Service,
Centers for Disease Control, Center for Health Promotion and Education, Office on Smoking and Health,
Rockville, MD; US Government Printing Office, Washington).
US Department of Health and Human Services (1988), "The health consequences of smoking: Nicotine addic-
tion", A Report of the Surgeon General, DHHS Publication No. (CDC) 88-8406 (US Department of Health
and Human Services, Public Health Service, Centers for Disease Control, Center for Health Promotion and
Education, Office on Smoking and Health, Rockville, MD).
US Department of Health and Human Services (1989), "Reducing the health consequences of smoking: 25
years of progress", A Report of the Surgeon General, DHHS Publication No. (CDC) 89-8411 (US Depart-
ment of Health and Human Services, Public Health Service, Centers for Disease Control, National Center
for Chronic Disease Prevention and Health Promotion, Office of Smoking and Health, Atlanta).
US Department of Health and Human Services (1994), "Preventing tobacco use among young people", A Re-
port of the Surgeon General (US Department of Health and Human Services, Public Health Service, Cen-
ters for Disease Control, National Center for Chronic Disease Prevention and Health Promotion, Office on
Smoking and Health; US Government Printing Office, Washington).
US Department of Health and Human Services (1996), "The FTC cigarette test method for determining tar,
nicotine, and carbon monoxide yields of U.S. cigarettes", Report of the NCI Expert Committee, NIH Pub-
lication No. 96-4028, August 1996 (National Cancer Institute, Bethesda, MD).
US Department of Health and Human Services (1998), "Tobacco use among U.S. racial/ethnic minority
groups", A Report of the Surgeon General (US Department of Health and Human Services, Public Health
Service, Centers for Disease Control, National Center for Chronic Disease Prevention and Health Promo-
tion, Office on Smoking and Health; Government Printing Office, Washington, DC).
US Department of Health, Education, and Welfare (1964), "Smoking and health", Report of the Advisory
Committee to the Surgeon General of the Public Health Service, PHS Publication No. 1103 (U.S. Depart-
ment of Health, Education, and Welfare, Public Health Service, Center for Disease Control).
University of Michigan News and Information Services (1997), "Cigarette smoking rates may have peaked
among younger teens", Press Release, December 20 (University of Michigan News and Information Ser-
vices, Ann Arbor).
Valdes, B. (1993), "Cigarette consumption in Spain: Empirical evidence and implications for public health
policy", Applied Economics 25(20):149-156.
van der Merwe, R. (1998a), "The economics of tobacco control in South Africa", in: I. Abedian, R. van der
Merwe, N. Wilkins and P. Jha, eds., The Economics of Tobacco Control: Towards an Optimal Policy Mix
(Applied Fiscal Research Centre, University of Cape Town, Cape Town, South Africa).
van der Merwe, R. (1998b), "The implications of falling tobacco consumption for employment in South
Africa", in: The Economics of Tobacco Control in South Africa (School of Economics, University of Cape
Town, Cape Town) 107-130.
Viscusi, W.K. (1990), "Do smokers underestimate risks?", Journal of Political Economy 98(6):1253-1269.
Viscusi, W.K. (1991), "Age variations in risk perceptions and smoking decisions", Review of Economics and
Statistics 73(4):577-588.
Viscusi, W.K. (1992), Smoking: Making the Risky Decision (Oxford University Press, New York).
Viscusi, W.K. (1995), "Cigarette taxation and the social consequences of smoking", in: J.M. Poterba, ed., Tax
Policy and the Economy (Massachusetts Institute of Technology Press, Cambridge, MA) 51-101.
von Weizsacker, C.C. (1971), "Notes on endogenous change of tastes", Journal of Economic Theory 3:345-
372.
Vuchinich, R.E., and C.A. Simpson (1999), "Delayed reward discounting in alcohol abuse", in:
EJ. Chaloupka, M. Grossman, W.K. Bickel and H. Saffer, eds., The Economic Analysis of Substance
Use and Abuse: An Integration of Econometric and Behavioral Economic Research (University of Chicago
Press, Chicago).
Warner, K.E. (1977), "The effects of the anti-smoking campaign on cigarette consumption", American Journal
of Public Health 67:645-650.
Warner, K.E. (1978), "Possible increases in the underreporting of cigarette consumption", Journal of the
American Statistical Association 73(362):314-318.
Warner, K.E. (1979), "Clearing the airwaves: The cigarette ad ban revisited", Policy Analysis 5:435-450.
Warner, K.E. (1981a), "Cigarette smoking in the 1970s: The impact of the antismoking campaign on con-
sumption", Science 211:729-731.
Warner, K.E. (1981b), "State legislation on smoking and health: A comparison of two policies", Policy Sci-
ences 13(2):139-152.
Warner, K.E. (1985), "Cigarette advertising and media coverage of smoking and health", New England Jour-
nal of Medicine 312:384-388.
Warner, K.E. (1986a), Selling Smoke: Cigarette Advertising and Public Health (American Public Health
Association, Washington).
Warner, K.E. (1986b), "Smoking and health implications of a change in the federal cigarette excise tax",
Warner, K.E. (1987), "Health and economic implications of a tobacco-free society", Journal of the American
Medical Association 258:2080-2086.
Warner, K.E. (1988), "The tobacco subsidy: Does it matter?", Journal of the National Cancer Institute
80(2):81-83.
Warner, K.E. (1989), "Effects of the antismoking campaign: An update", American Journal of Public Health
79:144-151.
Warner, K.E. (1990), "Tobacco taxation as health policy in the Third World", American Journal of Public
Health 80:529-531.
Warner, K.E. (1991), "Legalizing drugs: Lessons from (and about) economics", Milbank Quarterly 69:641-
661.
Warner, K.E. (1997), "Cost-effectiveness of smoking cessation therapies: Interpretation of the evidence and
implications for coverage", PharmacoEconomics 11:538-549.
Warner, K.E., F.J. Chaloupka, P.J. Cook, et al. (1995), "Criteria for determining an optimal cigarette tax",
Tobacco Control 4:380-386.
Warner, K.E., T. Citrin, G. Pickett, et al. (1990), "Licit and illicit drug policies: A typology", British Journal
of Addiction 85:255-262.
Warner, K.E., V.L. Ernster, J.H. Holbrook, et al. (1986), "Promotion of tobacco products: Issues and policy
options", Journal of Health Politics, Policy and Law 11:367-392.
Warner, K.E., and G.A. Fulton (1994), "The economic implications of tobacco product sales in a nontobacco
state", Journal of the American Medical Association 271(10):771-776.
Warner, K.E., and G.A. Fulton (1995), "Importance of tobacco to a country's economy: An appraisal of the
tobacco industry's economic argument", Tobacco Control 4:180-183.
Warner, K.E., G.A. Fulton, P. Nicolas and D.R. Grimes (1996), "Employment implications of declining to-
bacco product sales for the regional economies of the United States", Journal of the American Medical
Association 275:1241-1246.
Warner, K.E., and L.M. Goldenhar (1989), "The cigarette advertising broadcast ad ban and magazine coverage
of smoking and health", Journal of Public Health Policy 10:32-41.
Warner, K.E., L.M. Goldenhar and C.G. McLaughlin (1992), "Cigarette advertising and magazine coverage
of the hazards of smoking: A statistical analysis", New England Journal of Medicine 326:305-309.
Warner, K.E., T.A. Hodgson and C.E. Carroll (1999), "The medical costs of smoking in the United States:
estimates, their validity, and their implications", Tobacco Control 8:290-300.
Wasserman, J., W.G. Manning, J.P. Newhouse and J.D. Winkler (1991), "The effects of excise taxes and
regulations on cigarette smoking", Journal of Health Economics 10(1):43-64.
Watkins III, B.G. (1990), '"The tobacco program: An econometric analysis of its benefits to farmers", Ameri-
can Economist 34(1):45-53.
Wharton Applied Research Center (1979), "A study of the tobacco industry's economic contribution to the
nation, its fifty states, and the District of Columbia" (Wharton Applied Research Center and Wharton
Econometric Forecasting Associates, Inc.; University of Pennsylvania, Philadelphia, PA).
Wilcox, G.B., M. Tharp and K.-T. Yang (1994), "Cigarette advertising and consumption in South Korea",
International Journal of Advertising 13:333-346.
Wilcox, G.B., and B. Vacker (1992), "Cigarette advertising and consumption in the United States", Interna-
tional Journal of Advertising 11:269-278.
Winston, G.C. (1980), "Addiction and backsliding: A theory of compulsive consumption", Journal of Eco-
nomic Behavior and Organization 1(4):295-324.
Witt, S.F., and C.L. Pass (1981), "The effects of health warnings and advertising on the demand for cigarettes",
Scottish Journal of Political Economy 28(1):86-91.
World Health Organization (1997), Tobacco or Health: A Global Status Report (World Health Organization,
Geneva).
Wynder, E.L., and E.A. Graham (1950), "Tobacco smoking as a possible etiologic factor in bronchiogenic
carcinoma: A study of 684 proved cases", Journal of the American Medical Association 143:329-396.
Xu, X., T.-W. Hu and T.E. Keeler (1998), "Optimal cigarette taxation: Theory and estimation", Working Paper
(Department of Economics, University of California, Berkeley).
Young, T. (1983), "The demand for cigarettes: Alternative specifications of Fujii's model", Applied Eco-
nomics 15:203-211.
Zhang, P., and C. Husten (1998), "Impact of the tobacco price support program on tobacco control in the
United States", Tobacco Control 7:176-182.
Zhang, P, C. Husten and G. Giovino (1997), The Impact of the Price Support Program on Cigarette Con-
sumption in the United States (Office on Smoking and Health, Centers for Disease Control and Prevention,
Atlanta).
Zimbabwe... (1998), "Zimbabwe tobacco climb on week, seen firming", Reuters World Report, June 12,
1998 (off the Internet).
Chapter 30
ALCOHOL*
PHILIP J. COOK and MICHAEL J. MOORE
Duke University andNBER
Contents
Abstract 1630
2. Trends and patterns in alcohol consumption 1632
2.1. Aggregate data 1632
2.2. Individual differences 1633
2.3. Expenditures 1634
3. Framework 1634
4. Demand for alcoholic beverages 1638
4.1. Measurement issues 1639
4.2. Drinking by youths 1641
4.3. Heavy drinking 1642
4.4. Social influence 1643
4.5. Advertising 1644
5. Consequences of alcohol consumption and taxation 1645
5.1. Motor-vehicle mortality 1646
5.2. Cirrhosis 1649
5.3. Medical care 1650
5.4. Heart disease 1651
5.5. Crime and suicide 1652
6. Productivity 1653
6.1. Direct effects of drinking on productivity 1654
6.2. Drinking and human capital 1657
7. Evaluation of alcohol taxation and other alcohol-control measures 1658
7.1. Background 1658
7.2. Equity criteria 1659
7.3. Economic efficiency 1662
7.4. Total social cost 1664
8. Concluding observations 1665
References 1666
*Thanks to Michael Grossman, Will Manning, and Harri Sintonen for their very helpful comments. All errors
remain the authors'.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.. Newhouse
1630 P.J. Cook and MJ. Moore
Abstract
Excess drinking is associated with lost productivity, accidents, disability, early death,
crime, neglect of family responsibilities, and personality deterioration. These and re-
lated concerns have justified special restrictions on alcoholic-beverage commerce and
consumption. The nature and extent of government involvement in this arena vary
widely over time and place, and are often controversial. Economists have contributed
to the evaluation of alcohol policy through empirical work on the effects of alcohol-
control measures on consumption and its consequences. Economics has also provided
an accounting framework for defining and comparing costs and benefits of interven-
tions, including excise taxes.
Outside of the policy arena, economists have analyzed alcohol consumption in the
context of stretching the standard model of consumer choice to include intertemporal
effects and social influence. Nonetheless, perhaps the most important contribution by
economists has been the repeated demonstration that there is nothing unusual about
alcohol in at least one essential respect: consumers drink less ethanol (and have fewer
alcohol-related problems) when alcohol-beverage prices are increased.
Important econometric challenges remain, including the search for a satisfactory res-
olution to the conflicting results on the effect of price changes on consumption by con-
sumers who tend to drink heavily. There are also unresolved puzzles about the rela-
tionship between drinking and productivity; even after controlling for a variety of other
characteristics, drinkers tend to have higher earnings than abstainers, and women's earn-
ings (but not men's) tend to increase with alcohol consumption.
JEL classification: II1,J2, H2, D1, D6

Ch. 30: Alcohol 1631
1. Introduction
While the production and sale of alcoholic beverages constitute a minor share of na-
tional product in advanced economies, the health-related and socioeconomic conse-
quences of alcohol consumption are quite severe. Excess drinking is associated with
lost productivity, disability, early death, crime, neglect of family responsibilities, per-
sonality deterioration, and other problems. These and related concerns have long en-
gendered public support for treating alcohol differently from other commodities. In the
private sector, religious teachings and cultural norms are reinforced by rules restricting
drinking in the workplace, schools, and other gathering places. There is also pervasive
government regulation of the production and marketing of alcoholic beverages, coupled
with high excise taxes. The nature and extent of government involvement in this arena
vary widely over time and place, and are often controversial.
Economists have contributed to the evaluation of alcohol policy through empirical
research on the effects of alcohol-control measures on alcohol consumption and its con-
sequences. They have also made progress in developing and implementing a social-
cost-accounting framework for evaluating such measures that gives due weight to the
benefits of drinking, and attempts to distinguish consequences (both good and ill) that
are borne by the drinker and those borne by third parties.
Outside of the policy arena, economists have been interested in studying consumer
behavior with respect to alcohol because of several intriguing characteristics that dis-
tinguish it from other commodities. For one thing, drinking is habit forming; current
consumption may have a profound influence on future tastes. There are other delayed
consequences of current drinking as well, ranging from hangover and accidental injury
in the short term, to loss of reputation, reduced earnings, and organ damage over the
longer run. Some people develop such a strong taste for alcohol that they are willing to
incur great personal costs for the sake of their habit. The economics of self-control and
addiction has become an active field, engaging both economists and behavioral scien-
tists, with alcohol as a primary referent.
A second characteristic of drinking behavior is the importance of social context.
A primary use of alcohol is as a social lubricant, and both the availability of drinks
and the pleasure of drinking may depend on the nature of the occasion. The claim that
drinking is a socially contagious activity, long asserted by sociologists, is just beginning
to emerge as a focus for economists.
In this chapter we review the economics literature on drinking and its consequences,
and suggest some promising future research directions. The next section describes the
trends and patterns of alcohol consumption within and across populations, and charac-
terizes some of the recent trends. Sections 3 and 4 then review the theory and estimation
of the demand for alcoholic beverages, noting the special challenges to empirical work
in this area and then reviewing findings with respect to the effect on various measures
of consumption of income, own price, price of other intoxicants, and social influence.
Section 5 reviews findings on how drinking affects health status and socially relevant
behaviors. The focus is on the economists' contributions to estimating the effects of
1632 P.J. Cook and M.J. Moore
alcohol-control measures on these outcomes; these contributions are of direct relevance

to the policy debate and have also been important to our understanding of the epidemi-
ology of alcohol-related problems. A particularly important consequence of drinking in
policy debates historically has been its effects on productivity. The literature in this area
is reviewed in Section 6.
Section 7 reviews the normative literature on alcohol-control measures, with par-
ticular focus on alcohol-excise taxes. We conclude with some suggestions for future
research directions.
2. Trends and patterns in alcohol consumption
Alcoholic beverages include wines, beers, and spirits. Wines are fermented from the
sugars in fruits or berries (most commonly grapes) or other sources [Keller (1978)].
Beers are fermented from grains after the starch in them is first converted to sugar.
Spirits are distilled from wines or beers. The alcohol in all these beverages is ethyl
alcohol, also known as ethanol. In beers the alcohol content ranges from 2 to 8 percent;
unfortified wines contain between 8 and 14 percent (with 14 being the upper limit from
the fermentation process). Fortified wines, including vermouth, sherry, and port, contain
about 20 percent, which is achieved by the addition of brandy or pure alcohol. Spirits
usually contain between 40 and 50 percent alcohol.
Consumption of alcoholic beverages varies widely across and within populations,
and over time. Here we report some of the more notable patterns, first for national-level
data, and then for data on the distribution among individuals.
2.1. Aggregate data
In the United States, apparent consumption of ethanol per capita (age 14 and over)
peaked in 1980 and 1981 at 2.76 gallons after 20 years of steady growth [Williams et
al. (1996)]. This peak volume, which is equivalent to about two drinks of beer, wine, or
spirits per day for every adult, is the highest recorded in the 20th century. Drinking has
declined since then, and was down 20 percent by 1994. The decline has mostly been in
the consumption of spirits; as a result, the share of alcohol from spirits declined from
37 to 30 percent of the total, while beer increased its share from 50 to 57 percent and
wine held a constant 13 percent share.
These statistics are estimates based on state excise-tax records and industry reports,
and are subject to error. They take no account of wastage, illicit production and imports,
and legal home production, all of which may affect observed trends as well as levels
of consumption. For example, we would expect the gap between measured and actual
consumption to be influenced by excise tax rates, the real value of which trended down-
ward during the post-War period. Another potential source of error is the conversion
from beverage volume to ethanol volume. In recent years the conversion factors, which
are estimates of the relevant averages for each beverage category, are 4.5 percent for
beer, 12.9 percent for wine, and 41.1 percent for spirits [Stinson et al. (1997)1.
The consumption trends observed in the United States are similar to those in most
other economically advanced countries. Between the Second World War and the 1970s,
consumption increased in almost all countries that were able to offer reasonably accu-
rate statistics, with the largest growth rates recorded by countries that started from a
relatively low level. Hence the trend was towards narrowing of international differences
in ethanol consumption [Sulkunen (1983)]. Consumption per capita fell in most OECD
countries during the 1980s [Edwards et al. (1994, p. 35)].
Based on 1995 data, apparent per capita alcohol consumption in the United States is
about the same as Canada and the United Kingdom and lower than in Western Europe
with the exception of several Nordic countries. At the very top of the world drinking list
are Luxembourg, France, Portugal, Hungary, and Spain where consumption is half again
as high as in the U.S.' The Japanese average about the same alcohol intake as North
Americans, while the Chinese and residents of other Asian countries are estimated to
drink a good deal less, as do Latin Americans.
2.2. Individual differences
Per capita consumption statistics conceal the wide variation among individuals in alco-
hol consumption. The National Household Survey on Drug Abuse for 1996 estimated
that in the United States, 65 percent of adults had at least one drink during the previous
12 months, while just half drank during the previous month. 2 The prevalence of self-
reported drinking decreases in middle age and is much lower for women than men. For
both men and women the prevalence of drinking increases with education and family
income, and is lower for blacks than whites. Survey data also provide information on
how much alcohol is typically consumed by those who do not abstain. The proportion
of adults who "binged" (5 or more drinks on a single occasion) in the previous month
follows the same patterns with respect to age and sex, but with respect to race and edu-
cation there is little difference across groups.
The distribution of alcohol consumption among drinkers is highly skewed to the right
in most every population that has been studied. The log-normal distribution provides a
reasonably good fit to the empirical histogram (Edwards et al. (1994)). Hence consump-
tion is highly concentrated among a relatively few people. Gerstein (1981) conveys this
result with the following image:
"If we were to reduce the overall U.S. consumption curve to a representative sam-
ple of 10 drinking-age adults, their annual consumption of absolute ethanol would
not be very different from the following rough approximation: 3 nondrinkers, 3
drinking a gallon among them, and the others drinking 1.5, 3, 6, and 15 gallons,
respectively." (p. 193)
See NTC Publications (1996).

2 See National Household Survey on Drug Abuse (1996).
The heaviest drinkers are of considerable importance to the alcoholic-beverage industry;

for example, if the top decile somehow could be induced to cut their consumption level
to that of the next lower group (the ninth decile), then total ethanol sales would fall by
over one third.
Actually it is unlikely that any available intervention (short of rigorously enforced ra-
tioning) would truncate the distribution in this fashion. International comparative stud-
ies of drinking distributions have found remarkable consistency in their shape. As per
capita consumption by drinkers increases (over time or across populations), the con-
sumption associated with each quantile tends to increase in proportion [Edwards et al.
(1994, pp. 83-90)]. Hence, we would expect that as per capita consumption declines, as
it has in many Western countries in recent years, then the prevalence of heavy drinking
(defined by any absolute standard) will decline also - indeed, given that the downward
shift is proportional at all levels, the prevalence of heavy drinking will decline more
rapidly than average consumption. This possibility is compatible with trends in the ad-
verse consequences of heavy drinking since the drinking downturn in the early 1980s
(see Section 5).
2.3. Expenditures
Retail sales of alcoholic beverages in the United States totaled $99 billion in 1996,
equally divided between on-premise and off-premise [Adams Business Media (1997)].
That represents just 2 percent of personal consumption expenditures in that year, al-
though in fact not all sales are paid for as household consumption items - by one es-
timate, employers pick up the tab for about 20 percent of sales [Sammartino (1990,
p. 76)]. While the division of expenditures between on-premise and off is 50-50, most
alcohol by volume is consumed off-premise: 79 percent of spirits and wine, and 76
percent of beer, in 1996 [Adams Business Media (1997)].
While these details differ from place to place, it is true throughout the developed
world that the alcoholic beverage industry is relatively minor in terms of direct eco-
nomic importance. But the health and social consequences of drinking amplify its im-
portance.
3. Framework
Economists who study alcoholic beverages have focused on those qualities that are dis-
tinctive and important about this class of commodities. First, alcohol is an intoxicant;
consumed in sufficient quantity in a single session it impairs mental and physical func-
tioning and is potentially toxic. Second, alcohol consumption has direct intertemporal
consequences; past consumption affects tastes for future consumption in a predictable
way (habit formation), and chronic alcohol use affects physical and mental health over
the course of years or decades. In the aggregate, drinking has important economic ef-
fects through its influence on the public health and safety, productivity, family function-
ing, criminal activity, and so forth. These special features define much of the research
agenda for economists and other social scientists.
In particular, the dependent variables used in demand studies are not only measures
of the overall quantity per unit of time, but also measures of the timing and circum-
stances of consumption and its consequences. The determinants of quantity demanded
include not only prices and income, but also measures of past and anticipated future
consumption, and the availability of other intoxicants.
The research agenda has also been shaped by the special features of alcoholic-
beverage markets. Commerce in alcoholic beverages is subject to extensive government
regulation, so quantity demanded is mediated not just by price but a variety of other
alcohol-control measures. Indeed, at different times and places the manufacture or trade
in all or at least some types of alcohol has been prohibited. Short of prohibition, com-
mon legal restrictions specify when, where, how, and to whom alcohol may be sold.
These restrictions include government monopolization of manufacture or trade, or if
private vendors are permitted, special licensing requirements. Also common are limits
on advertising and hours of operation, prohibition on sale to minors or drunks, restric-
tion on the content and labeling of beverages, and quantity rationing. 3 All such control
measures are potentially grist for the research mill.
A schematic diagram of the various relationships that comprise the economic liter-
ature on alcohol is presented in the figure below. In this diagram, "current drinking"
]
Figure 1. A causal model of drinking and its consequences.
3 For example, beginning during World War I (and ending in the mid 1950s) Sweden maintained a com-
plex rationing system in which citizens committees determined how much spirits each adult member of the
community could purchase based on such factors as his or her age, family and social responsibilities, and
reputation [Norstrbm (1987)].
patterns (both on and off the job) may affect productivity, which in turn determines in-
dividual earnings. The diagram also depicts an indirect effect via the influence of drink-
ing history on the accumulation of human and health capital. The final link represents
"reverse causation", in which current consumption is affected by earnings.
To fix ideas for our review, we specify a simple structural model of alcohol consump-
tion and its consequences that incorporates some of these considerations. The model
specifies intertemporal demand imbedded in a health-production process of the sort de-
veloped by Grossman (1972).
We specify the demand for alcohol in the rational-addiction form developed in semi-
nal papers by Becker and Murphy (1988) and by Becker, Grossman and Murphy (1991).
Prior empirical and theoretical research developed the notion of intertemporal depen-
dency via models of habit formation, in which current utility depends on past as well as
current consumption of the addictive good. These habit-formation models implicitly as-
sumed that the consumer was myopic, in the sense that he ignored the possibility that his
current choices would affect future tastes or health. A "rational" consumer, on the other
hand, would recognize the future consequences of current drinking decisions and take
them into account in planning a sequence of consumption levels. Current demand then
becomes a function of past consumption, current prices and income, and expectations
concerning future drinking and prices.
Actually, the future may loom more broadly in affecting current drinking than is im-
plied by the standard rational-addiction model. Given the habit-forming nature of drink-
ing, and its other consequences, it is reasonable that some consumers will moderate their
drinking in response to expectations concerning the effect of current consumption on fu-
ture schooling, employment, family, and health status. We will ignore these influences
in presenting the model, but return to them in the discussion.
To capture the potential for habituation, let current-period utility be a function of
health, other goods, and current and past consumption of alcohol. Thus, the period-
specific utility function is Ut = U(Ht, Xt, At, At- 1), where Ht denotes health capital
in period t, Xt other goods, and At the consumption of alcohol.
Let the current wage be denoted by Wt, labor hours by Lt, and non-labor income by
It. Income in any period, Yt, is then WtLt + It. For simplicity, assume full depreciation
of the addictive stock in each period, with an individual discount factor = 1/(1 + r),
where r is the discount rate. The individual chooses to maximize lifetime utility, V t ,
which is the discounted sum of the period-specific utilities. That is,
01
V = J>t-lut. (1)
t=l
One "outcome function" will be treated here: a health-production function. 4 This is

a technological relationship, in the sense that variation in the health output is produced
4 See Grossman (1972).

by variation in the underlying input of alcohol. Alcohol may impair the ability to per-
form various tasks, such as driving a car or operating a machine, or it may damage
health directly, as in the case of some heart diseases, cirrhosis, high blood pressure, and
hemorraghic strokes. On the other hand, moderate alcohol consumption appears to have
beneficial effects, helping prevent certain types of heart disease.
Health is produced using medical care, M, and A as inputs, conditional on the preex-
isting health-capital stock and on other determinants of health, VH
Ht = H(Mt, At, Ht-1, VH). (2)
The effect of M on H is assumed positive; the marginal effect of A may depend on the
level of A; at sufficiently high levels the marginal effect is surely negative.
In general, the assumed shape of the production relationship given by Equation (2) is
crucial in determining the validity of empirical estimates, and in inferences made about
the efficacy of policy interventions aimed at reducing the harmful effects of alcohol
consumption. We will expand on this issue in a subsequent section.
The individual has a unit of time available each period to spend working, L, or relax-
ing, R. The time constraint is then
L+R= 1. (3)
Under certain assumptions, and if the individual maximizes (1) subject to the con-
straints given by (2), (3) and the intertemporal budget constraint, we can derive linear
demand functions (or more precisely, equations of motion) for A and M, and a linear
labor supply function. In particular, if Equations (1) and (2) are quadratic, and holding
the marginal utility of wealth constant, we have:
At = ao + alAt+ o+2PAt + o3At-I

o + 4Wt +a(5P +o6Zt + At (4)
Mt = 8o +81 At,+1 + 2 PAt + 3At-I +S4Wt 85 PMt + 6Z6t + EMt (5)
and
Lt = yo + y 1A,+l + Y2PAt + y3 At_ 1 + y4Wt + Y5PMt + Y 6 Zt + Lt, (6)
where PA and PM represent the prices of alcohol and medical care respectively, and Z
represents one or more covariates.
The interior solution represented here is clearly more appropriate for aggregate data,
as zero values are a common feature of microeconomic data on alcohol consumption,
medical care, and labor supply. The Kuhn-Tucker approach to modeling commodity
demands for individual-level data is developed in Wales and Woodland (1983) and in
Pudney (1989). 5
In addition to these equations of motion, we could also estimate the production rela-
tionship given by the health-production function in Equation (2). In the model developed
above, this was assumed to be quadratic and, to be fully consistent with that model, a
quadratic version of (2) could be estimated. Most estimates of (2) that have appeared
in the literature have been linearized versions of these production relationships. Health-
production functions have been the focus of the epidemiological literatures on the health
effects of drinking, where health is measured variously by mortality, morbidity, and in-
dicators of physical or mental health. In some applications, health is proxied by the
use of medical inputs as well. The measures of alcohol consumption in these health-
production productions in some studies allow for non-linearities, either through the use
of dummy variables relating to intensity of use, or with quadratic terms in quantity of
alcohol consumed.
The typical health production relationship estimated in the literature, however, has
the form:
Ht = o a + Mt
2M + 3 Y, + 4Zt + Ht. (7)
4. Demand for alcoholic beverages
Econometric studies of demand for alcoholic beverages have been conducted with a
wide variety of data sets. Edwards et al. (1994) tabulate results from such studies for 18
countries. Estimated elasticities for beer, wine, and spirits differ widely over time, place,
data set, and estimation method, but one conclusion stands out: In almost every case the
own-price elasticities are negative. In that fundamental respect, at least, it appears that
alcoholic beverages are like other commodities.
Clements et al. (1997) report results for their estimates of systems of demand equa-
tions for Australia, Canada, Finland, New Zealand, Norway, Sweden, and the U.K., in
each case using aggregate data covering about 30 years. Their average own-price elas-
ticities are -0.35 for beer, -0.68 for wine, and -0.98 for spirits. Beer has the lowest
own-price elasticity of the three beverage types in all 7 countries. This pattern is well
established though still somewhat mysterious. 6
The focus of these studies is on average (per capita) or total quantity, standard fare
for empirical analysis of markets. However, since our interest here is on health effects,
the average is of less interest than other measures of a population's consumption of
alcohol, including the prevalence of abstention, of bingeing, and of chronic heavy drink-
5 Estimation using micro data must also reflect this feature. Two approaches have been suggested for data
such as these: The Tobit estimator, and the two-part estimator of Duan et al. (1984). The two part estimator
appears to be generally more preferable on a statistical basis.
6 In 31 out of 38 demand studies tabled in Edwards et al. (1994) that included estimated results for both beer
and spirits, the former was less price-elastic than the latter. A thorough review of econometric studies using
data for the United States [Leung and Phelps (1993)] provides additional confirmation.
ing. Distinctions by age, sex, and drinking history of the consumer are also relevant to
consequences. Analysis of this sort cannot be performed with aggregate data, but rather
requires data on individuals. What follows, then, is a review of studies that utilize survey
data on individual respondents, or, in a few cases, experimental data.
4.1. Measurement issues
There are a number of surveys based on U.S. national samples that include items on al-
cohol consumption [Sindelar (1993)]. Given the importance of intertemporal influences
in analyzing demand, panel data are particularly useful. In many respects the most com-
plete data are provided by the National Longitudinal Survey of Youth (NLSY), which
has interviewed its initial sample of over 12,000 youths every year since 1979, at which
time they were aged 14-21. The questionnaires included items on alcohol consumption
from 1982-1985 and again in more recent years, and also have included standard inven-
tories on alcohol dependence and abuse. Over the years a wealth of other information
has been collected on parents and siblings, schooling, work, earnings and other income,
use of other intoxicants and tobacco, health status, family formation, and other mat-
ters. Price information is not obtained from the respondents, but can be imputed from
knowledge of the respondent's place of residence. Of course the NLSY data are limited
to a particular set of cohorts. The literature includes results from the NLSY, but from a
variety of other surveys as well.
The drinking items in these surveys are of suspect quality. Self-reports of alcohol con-
sumption tend to understate actual consumption. In general population surveys, com-
parisons of self-reported drinking with sales data suggest that such surveys typically
capture only 40-60 percent of actual consumption [Midanik (1982)]. In addition to the
obvious explanation, that respondents tend to underreport their drinking, it is also likely
that national household surveys have a somewhat biased sample. Some heavy-drinking
groups are underrepresented either because they are not in the sampling frame (e.g.,
homeless people and those in institutions) or because it is difficult to contact them and
gain their cooperation (e.g., people with transitory life styles) [Polich and Orvis (1979,
p. 56)].
With surveys, the details matter, as illustrated by a comparison of two sets of survey
results for high-school seniors in 1982. We compared NLSY data with data from an
ongoing survey of cohorts of high-school students called Monitoring the Future (MTF),
finding that MTF generated much higher estimates of drinking prevalence and of binge
drinking than NLSY [Cook, Moore and Pacula (1993)]. One likely explanation is that
MTF questionnaires were administered in the classroom, while NLSY respondents were
interviewed at home. In subsequent years the drinking results for these two samples of
respondents converge. 7
7 NLSY's subsequent surveys continued to be at the respondent's home, but that home was increasingly
likely to be the respondent's than his or her parents'. The MTF's followup surveys, unlike the initial survey,
were mailed to the respondents' home. See also Hoyt and Chaloupka (1994).
The consequences of response error for econometric results depend on its structure
across respondents and over time. In fact there is little direct evidence on whether re-
sponse errors tend to be additive or proportional, random or systematic. It does seem
reasonable to view self-reported consumption by a respondent as a lower bound on true
consumption; those who report bingeing really do, and so do some of those who claim
to be abstainers or light drinkers. But there is little direct evidence on this matter. 8
The implications of measurement error for estimates of Equations (4)-(7) are far from
clear. If the drinking variable is used as an explanatory variable, bias will be introduced
into the estimated effect of this variable, and into the estimated coefficients of the re-
maining explanatory variables. If the drinking variable is the dependent variable, and
measurement errors are correlated with quantity consumed or with explanatory vari-
ables, then the coefficient estimates will be biased as well.
The measurement-error problems become particularly acute in the microeconometric
literature on alcohol demand. Since the dependent variables of interest in this literature
are typically discrete or bounded, nonlinear estimation techniques are usually employed.
Even random errors in the dependent variable impart bias to estimated coefficients. The
importance of individual heterogeneity in alcohol demand and human- or health-capital
models has led some researchers to resort to fixed-effect-type models using panels of in-
dividual data. Nonlinear fixed-effect models, which are notoriously difficult to estimate
[Chamberlain (1984)], become even more so in the presence of measurement errors.
Data problems also arise in the choice of an indicator of price. In any one market, the
price of a drink differs widely depending on the type of beverage, the brand, the type
of retailer, and whether the purchase is for consumption on-premise or off [Grunewald
et al. (1996)]. What is needed for cross-section or intertemporal comparisons is a price
index of some sort. In the absence of a local-area index, most investigators have used
either the state excise tax rate (usually just on beer), or the average price of a 6-pack of
a popular brand of beer (available for each of a number of urban areas from data com-
piled by the American Chamber of Commerce Researchers Association, or ACCRA),
sometimes adjusted for an index of the local cost of living. This ACCRA price estimate
is in principle a more sensitive indicator of local market conditions, but is of course
measured with greater error than the excise tax rate and is not computed for all areas of
residence.
8 Rare exceptions are Boland and Roizen (1973) and Popharn and Schmidt (1981). Both studies compare
self-reports to sales data. The former study reported that heavier purchasers were more likely to give accurate
information to an interviewer. The latter reached the opposite conclusion, but based on a somewhat faulty
analysis of their data. They compare the distribution across buyers of the number of bottles actually purchased
to the distribution of the number of bottles reported purchased in the survey. Both distributions are skewed to
the right, but the mean of the self-reports is less than the mean of the purchase records. The authors calculate
the percentage of the population in each of a series of intervals defined by the number of bottles. They find that
the ratio (sales to reported purchases) of the corresponding percentages from the two distributions increases
with the number of bottles, and offer that as evidence that underreporting is higher at higher quantities of
purchases. But in fact this pattern is compatible with a circumstance in which every drinker underreported by
the same percentage.
4.2. Drinking by youths
Much of the econometric research has focused on drinking by youths. Teenagers and
young adults are of special concern for several reasons. First, youths exhibit relatively
high rates (compared with their elders) of binge drinking and involvement in motor-
vehicle accidents and violent crime [Grossman et al. (1994)]. Second, to the extent
that drinking is habit forming, youthful drinking sets the pattern for later consumption.
And third, drinking behavior during the transition from adolescence to adulthood may
have important consequences for human capital and family formation [Cook and Moore
(1993c)].
Beginning with Grossman et al. (1987) and Coate and Grossman (1988), a series
of studies have documented the sensitivity of youthful drinking to both minimum-
drinking-age (MDA) laws and to beer prices. (The focus on beer is dictated by the fact
that most ethanol consumed by youths in the U.S. is in the form of beer.) A number of
studies using five different data sets suggest that how often youths drink and how often
they binge are both importantly related to price and MDA [Grossman et al. (1994)]. An
exception is Chaloupka and Wechsler (1996)], that analyzes drinking by college stu-
dents using a large one-time survey. They found that the price of beer has no discernible
effect on drinking practices of male students. A possible explanation, they suggest, is
that much of their alcohol consumption is in group settings where individuals do not pay
by the drink. Price also has a significant effect on the likelihood of abstention, though
ironically this is not true for the MDA [Moore and Cook (1995)].
Evidence that drinking is habit forming for youths comes from analyses of panel data.
Moore and Cook (1995), for example, analyze four waves of NLSY data. The likelihood
of drinking in 1985 is related to whether or not the respondent reported drinking in pre-
vious years according to the following equation, in which dt is an indicator of whether
the respondent consumed any alcohol during the 30 days prior to the interview in year t:
Pr[d85 = 1] = 0.16 + 0.36 d84 + 0.23 d83 + 0.16 d8 2.
Thus past drinking predicts future drinking, and the recent past predicts more closely
than the distant past. Of course, this result does not prove habit formation. It may in part
reflect persistent differences among youths with respect to their taste for alcohol (or
their circumstances). Distinguishing between state dependence (habit formation) and
persistent heterogeneity is a common problem in labor economics and other areas of
application [Heckman (1981)]. One approach for ruling out the heterogeneity explana-
tion is through use of instrumental variables for past drinking. We took the somewhat
more direct approach of estimating the effect of alcohol availability (as indicated by
excise tax and MDA) at age 14 on drinking a few years later. The results confirm the
habit-formation explanation for both drinking and bingeing.
The rational-addiction framework described in Section 3 suggests that current drink-
ing choices will be influenced by the future as well as the past and present. Youths who
anticipate higher alcohol prices in the future, or (perhaps more important) foresee cir-
cumstances such as childbearing in which drinking will have increased nonmonetary
costs, will curtail current drinking if they believe that their drinking is habit forming.
Demand estimation in the presence of rational addiction requires indicators of antici-
pated prices and nonmonetary costs. Empirical implementation of this framework has
proven difficult in practice [Chaloupka and Grossman (1994), Moore and Cook (1995),
Grossman et al. (1998)].
So far our review has been concerned with a single intoxicant, alcohol. Other in-
toxicants, especially marijuana but other illicit drugs as well, are widely available to
youths in the U.S. It is of no small concern whether the MDA and other alcohol-control
measures may lead youths to substitute illicit drugs for alcohol.
The 30-day-prevalence results from the NLSY data in 1984 suggest to the contrary
that drinking and marijuana use go together: 27 percent of drinkers, but only 5 percent of
abstainers, reported using marijuana [Pacula (1998)]. But this pattern of multi-drug use,
while suggestive, does not demonstrate that alcohol and marijuana are complements.
As before, it may simply reflect heterogeneity in the taste for intoxicants. More direct
tests have had mixed results. Pacula (1998) reports that higher beer prices significantly
reduce the demand for both alcohol and marijuana, suggesting contemporaneous com-
plementarity. Chaloupka and Laixuthai (1997) find, on the other hand, that an increase
in the full price of marijuana (indicated by the stringency of state laws and a measure of
retail price) increases drinking and bingeing, suggesting that alcohol and marijuana are
substitutes.
This issue is resolved to some extent by studies of the effect of alcohol-control mea-
sures on highway fatalities reviewed in Section 5. The indirect effect (via marijuana use)
of such measures on intoxicated driving is implicit in these results. The evidence that
raising the MDA reduces youth involvement in fatal crashes suggests that substitution
to marijuana (if any) is either relatively small or benign.
4.3. Heavy drinking
Some people acquire such a strong taste for alcohol that they are willing to sacrifice
their health and much else for the sake of continued heavy drinking. Symptoms of
strong commitment to alcohol are the basis for a diagnosis of alcohol dependence, a
form of mental illness more commonly known as alcoholism. It is a widespread pre-
sumption that alcoholics will do whatever is necessary to maintain their drinking at a
high level, including substituting cheaper sources of ethanol or cutting back on other
living expenses. In particular, an increase in the excise tax would likely make their lives
more difficult but not cause them to cut back.
Experimental evidence, however, suggests that this image is not entirely correct [Ba-
bor (1985)]. In an early experimental study, Mello et al. (1968) compared drinking pat-
terns of 14 male alcoholics as a function of the cost of a drink. Subjects who were
required to work twice as hard for their alcohol drank half as much as comparable sub-
jects in an identical situation. Other experiments with inpatient alcoholics found that
their drinking could be reduced by contingent loss of privileges and financial incentives
for abstinence [Mello (1972), Babor et al. (1978)]. Another study recruited 34 adult
males from the community for an experiment comparing the responsiveness of the 20
casual drinkers and the 14 heavy drinkers to changes in price. The response of both
groups to a "happy hour" in which prices were cut in half was to approximately double
the number of drinks they consumed [Babor et al. (1978)].
The evidence from outside the laboratory is primarily from studies that use the cir-
rhosis mortality rate as an indicator of the prevalence of chronic heavy drinking. These
studies generally confirm the experimental results; they are reviewed in detail in Sec-
tion 5.
A different conclusion emerges from the application of quantile regression analysis to
survey data. Manning et al. (1995) used data from the National Health Interview Survey
to analyze how price elasticity changes with drinking level. They find that the decision
to drink is responsive to price level, and that among those who report drinking, the esti-
mated price elasticity follows a U-shaped pattern with respect to relative consumption.
The elasticity peaks at - 1.2 at the median, and approaches zero at higher levels. At the
highest level of consumption, representing the 95th percentile, the estimated elasticity
is slightly positive. This result appears to confirm the conventional wisdom that heavy
drinkers will find a way to maintain their drinking in the face of modest changes in
price. The apparent contradiction between this result and other findings has not been
explained.
4.4. Social influence
While the decision of how much to drink depends to an extent on individual tastes and
financial circumstances, it is also true that drinking is a social activity. The utility of
taking a drink at a particular time and place depends not only on individual tastes but
also on the social setting; for many people, drinking with others is more enjoyable than
drinking alone, and associating with those who are not drinking, or who disapprove
of drinking, may lead one to substitute another type of beverage. Of course the social
setting may also influence drinking decisions directly, by determining the availability
of drinks.9 Given these mechanisms of social influence, it seems reasonable to presume
that individual drinking is influenced by the "wetness" of the social environment.
If individual drinking decisions are positively linked to the drinking practices of oth-
ers, then there will be a "social multiplier" in the response of aggregate alcohol con-
sumption with respect to prices, income, and other external influences; social influence
will amplify the direct effects of such variables. This mechanism may be particularly
important for initiation into drinking. Indeed, the assumption that peers are central to
adolescent alcohol and drug use is reflected in the social-influence paradigm underlying
many prevention programs [Bauman and Ennett (1996)].
9 At a more fundamental level, the consumption of alcohol is subject to a wide range of cultural influences,
including religious strictures, holiday traditions, popular entertainment, and other sources of alcohol-related
norms. These may provide another, more slow moving, feedback effect, responding and amplifying trends in
drinking.
One type of evidence in support of this view is that adolescents whose friends drink
are far more likely to drink themselves. For example, Norton et al. (1998) studied drink-
ing in 36 schools, finding that adolescents in schools with a high prevalence of drinking
were more likely to drink themselves. This result holds after controlling for various
individual, household, and neighborhood characteristics. Indeed, the estimated effect
is very large, suggesting that an increase of 10 percentage points in group drinking is
associated with an equal increase in the likelihood of individual drinking.l°
But this result is compatible with several other mechanisms besides social influence
[Manski (1995)]. First, it may be the result of an endogenous selection process where
some parents consider the behavior of the local adolescents in deciding where to live.
Second, it may reflect a "contextual" effect, where the individual's drinking behavior is
influenced by other characteristics of the group (commitment to getting a good educa-
tion) but not by their drinking per se. Third, it may be true that youths within the same
group share some important but unobserved aspect of the environment, such as whether
local merchants are willing to sell alcohol to youths.
In the absence of experimental data, the identification problem here is severe.
Instrumental-variables methods may provide estimates of the extent to which an indi-
vidual's drinking is influenced by the group without demonstrating that it is the group's
drinking (rather than some correlate of their drinking) that is the direct cause [Norton
et al. (1998)]. This distinction is important because if it is not the group's drinking per
se, but rather some other mechanism of social influence, then there is no social multi-
plier (Gaviria and Raphael, in press). For example, if higher-education plans influence
drinking, and youths are influenced by their school peers in evaluating the prospect of
higher education, then a student attending a school where most students have no plans
for higher education will, like her peers, tend to drink more than otherwise, even if
drinking is not contagious.
But assuming that drinking is subject to a positive-feedback effect through the group,
the result is to increase the elasticity of market demand with respect to own price and
other determinants. Further, for a given price regime, aggregate quantity consumed may
depend on the extent of social mixing among individuals with different drinking propen-
sities, a factor which may be deemed relevant, for example, in setting housing rules on
college campuses.
4.5. Advertising
Individuals learn about alcoholic beverages and drinking not only from friends, but
from a variety of other sources as well, including the popular media, church, the class-
room, consultations with health-care providers, labels on beverage containers, and so
forth. From this array of potential influences on the demand for alcohol, the economists'
contribution has been largely confined to the analysis of commercial advertising. The
0 Of course this result cannot hold over the entire range of possible prevalence levels, since it would rule out
the existence of an equilibrium value for group prevalence of drinking.
qualitative issues here are much the same as in the case of tobacco, and we refer readers
to the excellent discussion of advertising in that chapter of this volume.
Producers sponsor ads in order to increase the demand for their particular brands, and
it is possible that the cumulative effect has more to do with the distribution of market
share than with overall quantity of drinking. Econometric studies of commercial adver-
tising have reported differing results with respect to the estimated effect on consump-
tion, as reviewed in Saffer (1995). Saffer (1997) argues that we would expect the effects
of advertising to be subject to diminishing returns, and that rivalry among producers
may yield investments in advertising that at the margin have little effect on consump-
tion. Thus data from a regime where advertising is unconstrained provide information
on the (possibly negligible) marginal effects, but not on the overall effects, of advertis-
ing. That may help explain the null results reported by Nelson and Moran (1995), in
their study of U.S. national time-series data, and of Gius (1996), in his study of brand
advertising for 15 brands of distilled spirits; these and a number of earlier studies report
a negligible effect of overall advertising on aggregate consumption.
Saffer (1991) sought to measure infra-marginal effects of advertising by analyzing
national ad bans. His study was based on a time series of cross-section data on 17 coun-
tries over 14 years. He found that after controlling for price and other factors, a ban on
spirits advertising was associated with a 16 percent reduction in ethanol consumption,
and that a ban on advertising of all types of alcohol lowered consumption by an addi-
tional 11 percent. These results can be challenged on the basis that the "assignment" of
ad bans to countries is not necessarily exogenous. It may also be true that advertising
influences prices, as would be true if advertising strengthens brand loyalty and thus re-
duces own-price elasticity. In this case the "ban" coefficient provides a biased estimate
of the full effect.
Public concern about alcohol advertising is more focused on alcohol abuse than on
average consumption levels, and so it is of particular interest to measure the effect of
commercial advertising on such outcome measures as highway fatalities and cirrhosis
mortality [Saffer (1991)]. For example, a study by Saffer (1997) related alcohol adver-
tising messages to traffic fatalities, utilizing panel data on the 75 top television markets.
This study allowed for the possibility that advertising is endogenously linked to drinking
(and hence to alcohol-related crashes). The results suggested that a ban on broadcast ad-
vertising of beer and wine in the U.S. would reduce traffic fatalities by about 5 percent.
At present the methodological difficulties in studying the effects of commercial ad-
vertising on alcohol consumption and abuse have precluded a confident conclusion
about whether the regulation of commercial advertising is a potentially important policy
instrument.
5. Consequences of alcohol consumption and taxation
The health and social consequences of drinking render it an important problem for
nations where alcohol consumption is common, and an incipient problem for others.
Alcohol-related health problems include both the short-term consequences of intoxica-

tion and the long-term consequences of chronic heavy drinking. The former category
includes injuries sustained in traffic crashes, industrial accidents, drownings, and alco-
hol overdose, as well as from intentional violence. The latter, chronic category includes
damage to the liver and other organs, impaired cognition and immune-system func-
tion, and alcohol dependence. Alcohol is also teratogenic; drinking during gestation
may cause persistent abnormalities in physical and cognitive development of the child
[Larkby and Day (1997)].
The U.S. Centers for Disease Control (CDC) estimated "alcohol-related mortality"
for 1987 to be 105 thousand, 4.9 percent of all deaths in that year [CDC (1990)]. By this
estimate, alcohol is a smaller problem than smoking but a far larger problem than drug
abuse. Nearly half (46 percent) of the "alcohol problem" in the United States stemmed
from accidents and intentional violence. By another conventional public-health mea-
sure, "years of life lost before age 65", violent deaths constitute fully 80 percent of the
alcohol problem in the United States [CDC (1990)]. Thus the chronic effects of excess
consumption have a relatively small effect on life expectancy.
A number of other countries have still higher death rates from heavy drinking, partic-
ularly in central Europe and other nations where per capita consumption is high. Perhaps
most notable in this regard is the sharp drop in Russian male life expectancy during the
1990s, which has been attributed in part to the surge in alcohol consumption following
the end of the Gorbachev reforms. From 1990 to 1994 male life expectancy at birth fell
by 6.2 years to 57.6. One analysis concluded that "Although factors such as nutrition
and health services may be involved, the evidence is that substantial changes in alcohol
consumption over the period could plausibly explain the main features of the mortality
fluctuations observed" [Leon et al. (1997, p. 383)].
The health consequences of drinking alcoholic beverages are apparently not entirely
negative. Historically and even in modern times and places where the only available
water for drinking is contaminated, alcoholic beverages have offered a nearly germ-
free alternative [Vallee (1998)]. And the accumulating evidence suggests that moderate
drinking protects against coronary heart disease.
In this section, we review key studies on the relationship between alcohol consump-
tion, alcohol-control measures, and several health outcomes: motor-vehicle fatalities,
cirrhosis mortality, medical care, heart disease, and crime.
5.1. Motor-vehicle mortality
Several of the U.S. studies of alcohol-control policies and highway safety are summa-
rized in Table 1. The majority of these studies utilize state-year panel data on fatalities,
coupled with indicators of the minimum drinking age (MDA) and price or tax changes.
Per capita income is typically utilized as a control variable, along with fixed effects for
state and year, and in some cases autocorrelation corrections.' 1Most of the state-level
I There are exceptions, however. Cook and Tauchen (1984) and DuMouchel et al. (1987) do not control for
beverage price changes. Chaloupka et al. (1993) substitute demographic characteristics for state fixed effects.
studies, with the exception of Males (1986), conclude that highway fatalities decline
when the minimum purchase age or alcohol excise taxes are increased [a null find-
ing on price is reported in Sloan et al. (1995)]. Ruhm (1996) suggests that changes in
alcohol-control measures may be confounded with such omitted variables as grassroots
campaigns against drunk driving and state economic conditions.
This "reduced form" approach to studying highway fatalities and other remote effects
of alcohol-control measures requires some motivation [Cook (1981)]. As described in
Section 3 above, prices and other control measures influence outcome measures, if at
all, through their effect on drinking. Consider the following links:
Link 1: Increased excise taxes on alcoholic beverages reduce per capita consumption
of ethanol;
Link 2: A reduction in average consumption of ethanol is associated with reduced
prevalence of intoxication;
Link 3: A reduction in intoxication prevalence is associated with a reduced prevalence
of driving under the influence (DUI);
Link 4: A reduced DUI prevalence reduces the motor-vehicle-fatality rate.
One approach to estimating the effect of excise taxes on motor-vehicle fatalities is to
estimate each of these structural relationships, as in Equations (4) and (7). But even if we
were able to do so, the result would be less persuasive than the "reduced form" estimate.
For one thing, the intermediate variables, particularly the measures of drinking, are
subject to large errors in measurement. For another, these intermediate variables are not
precise enough to ensure that the links join into a single chain of argument. It is possible,
for example, that excise taxes reduce the frequency of intoxication at home but not away
from home, in which case the excise tax would not much affect DUI prevalence even
though links 1, 2, and 3 are each generally true.
Since first reported by Cook (1981) and Cook and Tauchen (1982), the reduced-form
estimates have become standard practice in exploring the effects of alcohol-control mea-
sures on a variety of outcome measures. An ancillary benefit of this approach has been
to help establish the causal importance of drinking with respect to certain outcomes. For
example, social scientists have tended to favor explanations for the observed association
between crime and drinking that deny the direct causal influence of drinking [Collins
(1989)]. But evidence that higher alcohol prices reduce some kinds of crime suggests
that alcohol is the culprit after all.
Returning to our review of drinking and motor-vehicle crashes, we see that two of the
studies in Table 1 utilize microdata on risk-taking behaviors to examine the effects of
availability restrictions on drunk driving. Kenkel (1993) analyzes the Health Promotion
and Disease Supplement to the 1985 Health Interview Survey, which contained infor-
mation on drinking and driving practices. He incorporates alcohol-control measures,
measures of the legal threat to drunk driving, and an indicator of health knowledge
based on individual awareness of connections between drinking and health risks. Binge
drinking is defined as the number of days in the past year with 5 or more drinks, and
1648 PJ.Cook and M.J. oore
Table 1
Economic studies of alcohol-related traffic mortality
Authors Data Outcome Policy Result
Cook and State-year Highway fatality MDA

Tauchen (1984) rates
Males (1986) State-year panel, Fatality rates MDA 0

1975-1983
*
DuMouchel et al. State-year panel, Fatalities, age MDA
(1987) 1975-1984 16-24
Saffer and State-year panel, Fatality rates, by Beer taxes

Grossman (1987a) 1975-1981 age (youth) MDA *
Chaloupka, State-year panel, Fatality rates: Beer tax rates

Saffer, and 1982-1988 total, night, & MDA
Grossman (1993) youth Laws, fines mixed
Kenkel (1993) 1985 National Self reports of Price -* (youths)

health interview drinking & MDA
survey driving, last 12 Knowledge
months Monopoly +*
Deterrents +* (mostly)
Sloan et al. Behavioral risk Self reports of Price

(1995) factor survey drinking & MDA
1984-1990 driving, last 30 Deterrents mostly negative
days Liability rules &insignificant
Ruhm (1996) State-year panel Total & night Beer tax rate
1982-1988 fatality rates MDA -* (18-20 only)
laws, fines, & mostly insignificant
penalties
* Statistically significant effect at the p = 0.10 level.

I Kenkel's results are shown for a heavy drinking equation. Heavy drinking significant increases drunk
driving in the second equation of his recursive framework.
drunk driving by responses to a survey question asking how many times in the past year
the sample member drove after having too much to drink.
Kenkel concludes that increases in price and health knowledge reduce the prevalence
of binge drinking, while a state monopoly in liquor sales is associated with an increase
in heavy drinking. Binge drinking in turn increases the prevalence of drunk driving.
Interestingly, the legal-threat variables tend to reduce drunk driving as well, and by
the same mechanism - the threat of punishment reduces binge drinking rather than
persuading people to separate their drinking and driving.
Sloan et al. (1995) also analyze microdata, in this case from the Behavioral Risk
Factor Surveys. In addition to the alcohol-control and deterrence effects, they incor-
porate indicators of the tort-liability rules to the binge drinking and drunk driving
models. While some effects of the criminal and legal variables have the expected ef-
fects on the two outcome variables, most are not statistically significant. The price and
MDA variables exert a strong effect on drunk driving, primarily through their effect on
binge drinking. Also interesting are the results on the incentive effects of compulsory-
insurance laws and experience rating. It appears that these policies, which tend to raise
the price of careless behavior, lead to significant declines in binge drinking.
5.2. Cirrhosis
Cirrhosis is characterized by a progressive replacement of healthy liver tissue with scar-

ring, leading to liver failure and death. While it has a variety of causes, alcohol accounts
for a majority of cases within population groups where drinking is widespread; indeed,
the cirrhosis-mortality rate has long been used as an indicator of the prevalence of al-
coholism in a population [Bruun et al. (1975)]. The likelihood of cirrhosis is closely
related to lifetime consumption: according to one review, an individual weighing 150
pounds who drank 21 ounces of 86 proof whiskey per day for 20 years would have a 50
percent chance of suffering from cirrhosis [Lelbach (1974)].
Following a long term increase from 1950 to 1973, cirrhosis-death rates in the U.S.
have declined steadily from a peak of 14.9 per 100,000 to 8.1 in 1992 [DeBakey et
al. (1995)]. There is substantial variation in cirrhosis-death rates across countries. As
reported by Edwards, et al. (1994), death rates from cirrhosis per 100,000 living in
Europe ranged from a high of 54.8 in Hungary in 1991 to a low of 2.9 in Ireland. In
the countries of the Western hemisphere, cirrhosis mortality rates are highest in Mexico
and Chile (48.6 and 46.2), and lowest in Canada (9.3).
There is considerable evidence that cirrhosis-death rates are sensitive to alcohol avail-
ability, suggesting that the group at risk for alcohol-related cirrhosis, long-term heavy
drinkers, is at least somewhat price sensitive. Notable cases occurred during and after
the First World War. Alcohol was diverted to military purposes during the war, and sev-
eral countries instituted prohibition. The results have been analyzed in Canada, Finland,
and the United States, and in each case the reduction in availability was accompanied
by a rapid and substantial drop in the cirrhosis-death rate. In France, which ordinar-
ily has a high cirrhosis mortality rate, sharp drops in availability occurred during both
World Wars with particularly dramatic effects on the mortality trend [Bruun et al. (1975,
p. 43)].12
12 Miron's (1997) econometric results lead him to challenge the belief that Prohibition had a large effect on
cirrhosis rates in the U.S. [Warburton (1932)]. A look at the data does suggest that the decline in cirrhosis
occurred before Prohibition: The age-adjusted cirrhosis mortality rate fell from 17.0 in 1911 to 8.9 in 1920
and remained at about that level through Prohibition and long after [DeBakey et al. (1995)]. On the other
hand, the influence of the Temperance movement was felt long before the 18th Amendment. A number of
states adopted prohibition before it became the nation's law, and these and other restrictive measures may
account for the early decline, which Prohibition then sustained.
1650 P.J.Cook and M.J. Moore
Table 2
Drinking status and medical care
Former drinker Abstainer Heavy drinkert
HIE outpatient care +* +* 0

NHIS outpatient care 0
0* +*
HIE inpatient care +* +* 0
NHIS inpatient care +* +* 0
Source: Adapted from Manning et al. (1991) Tables 5-7 and 5-9. See text for discussion.
* Statistically significant effect at the p < 0.10 level.
The authors include the log of monthly ethanol consumption and the square of this variable in their regres-
sions. For the regressions on HIE data, the estimated coefficients are not in any case discernibly different from
zero. For the NHIS data the authors report that for the regression on outpatient data that log consumption is
significantly negative and the squared term is significantly positive.
Cirrhosis mortality is also responsive to small changes in price. Cook (1981) and
Cook and Tauchen (1982), in a longitudinal study of state cirrhosis mortality rates, find
that increases in state liquor-excise taxes lead to an immediate (and statistically signifi-
cant) reduction. 13 While this disease takes years to develop, death rates respond quickly
because the progression of the disease (towards death) is slowed when drinking is cur-
tailed. Over the long run a reduction in heavy drinking will reduce cirrhosis mortality
still further, since the rate of initiation of cirrhosis will be reduced.
5.3. Medical care
While it may be true that changes in alcohol-control measures affect medical-care uti-
lization, there has been no analysis of this linkage. There have been several important
studies of the relationship between drinking and medical-care use, however.
Manning et al. (1989, 1991) examine the effects of heavy drinking on outpatient and
inpatient care using data from the RAND Health Insurance Experiment (HIE) and the
National Health Interview Survey of 1983. Indicators of drinking used as predictors
of health-care utilization include monthly volume of alcohol consumed together with
dummy variables for former drinker and abstainer. Four medical-care-utilization vari-
ables are analyzed, defined by whether they count inpatient or outpatient visits, and
whether they count all such visits or only those in which the diagnosis was likely to be
alcohol-related. Table 2 summarizes the results for the more comprehensive measure of
care that is not limited to alcohol-related diagnoses. (The results for the alcohol-related
diagnoses are very similar.)
The two data sets yield the same results for inpatient care. Former drinkers and
abstainers use significantly more medical care than current drinkers. Among current
drinkers, there is no discernible relationship between amount consumed and frequency
13 See Moore (1996) for a report of similar findings with a somewhat different specification.
of inpatient care. For outpatient care, on the other hand, the two data sets yield some-
what conflicting results.
One hypothesis to which the inpatient results conform is that abstinence and ces-
sation of drinking may reflect some underlying health condition that is also associated
with medical-resource use. In the model developed in Section 3, if there are correlations
among the unobservables in the regression equations for health, medical care, and drink-
ing equations, an exogenous adverse health shock might both increase the demand for
medical care and reduce the demand for alcohol. In this instance, estimation of a model
in which medical care is regressed on alcohol use could show a positive relationship
between drinking cessation and the use of medical care that is not entirely causal. 14
5.4. Heart disease
Interest in the beneficial effects of moderate alcohol consumption on the heart was stim-
ulated by publicity surrounding the so-called "French Paradox", that heart disease is
lower among the French than for a number of peoples (including Americans) despite
the French penchant for smoking and enjoying artery-clogging diets. The primary hy-
pothesis advanced to explain this phenomenon is that alcohol consumption, which is
also heavier among the French than in the U.S., is somehow responsible.
Most of what we know about alcohol and the heart comes from prospective studies
reported in the epidemiological and medical literature. 15 These studies have consis-
tently reported beneficial effects of moderate drinking, including the Honolulu Heart
Study [Yano et al. (1977)], the Nurse's Health Study [Stampfer et al. (1988)], the lipid
research clinics follow-up study [Criqui et al. (1987)], the British Regional Heart Study
[Shaper, et al. (1988)], the Kaiser-Permanente Study [Klatsky et al. (1990)], and the
Physicians' Health Study [Camargo et al. (1997)]. The typical result is that a U-shaped
relationship between drinking and the risk of coronary heart disease (CHD) is found,
with the beneficial effect maximized at about 2-3 drinks per day. Estimates of the ben-
eficial impact range from 25 to 50 percent reductions in CHD mortality for moderate
drinkers.
Of course these nonexperimental associations may have other explanations. Reverse
causation is certainly a problem. Many nondrinkers are former drinkers who have quit
for health reasons, and it should not be surprising that they are more likely to exhibit
heart problems than moderate drinkers. But that is not the whole answer since the rela-
tionship persists when those with prior conditions are excluded from the study. What we
do not know is if there are unobserved "third causes" which explain both the propensity
to abstain and the propensity to heart disease. In the next section we discuss the findings
on drinking and productivity, which also exhibit an unexplained penalty for abstainers.
14 As Manning et al. (1991) note, controlling for differences in health status diminishes the effects of absti-
nence and cessation by one-fourth to one-half. The differences remain significant.
15For reviews of this literature see Shaper (1990) and Lands and Zakhari (1991).
1652 P.J. Cook andM.J. Moore
Drinking has other effects on the circulatory system as well. It has been found to re-
duce clotting, thus reducing the risk of arterial blockages and ischemic strokes. Through
this mechanism, moderate alcohol use can have an immediate effect on mortality [Du-
four (1996)]. At the same time, this thinning effect increases the likelihood of hem-
orrhagic stroke. Hypertension, one of the primary risk factors in heart disease, is also
increased by regular alcohol consumption. Finally, rhythm disturbances leading to heart
attack are also more likely following spells of heavy drinking. The so-called "holiday
heart" syndrome, where the rate of heart attacks increases immediately following holi-
days such as New Years, is a manifestation of this phenomenon.
Research on the effects of alcohol availability on heart disease has lagged behind that
on cirrhosis and traffic accidents. Given the observed relationships between both light
and heavy drinking and taxes on one hand, and between light and heavy drinking and
heart disease on the other, we would expect taxes to play some role in mitigating certain
diseases of the circulatory system, and to exacerbate others.
Drinking may also have some effect on the risk of certain types of cancer. The Physi-
cians' Health Study, a prospective cohort study of 22 thousand healthy men ages 40
and over, found a U-shaped pattern between all-cause mortality rates and alcohol con-
sumption. The lowest mortality rate was for light drinkers (2-4 drinks per week) and
highest for the group drinking the most (2 or more drinks per day), with the upturn due
to cancer and cardiovascular disease [Camargo et al. (1997)].
5.5. Crime and suicide
Under the influence of alcohol, a parent may be provoked to strike an irritating child; a
college student may forcefully insist on having sex with his date; friends may escalate
an argument into a bloody fight; a robbery victim may foolishly attempt resistance in the
face of a loaded gun; soccer fans may riot in response to an unsatisfactory game. Some
individuals under certain circumstances are more prone to violence, or to provoking
violence, when drinking than when sober [Fagan (1990)].
Drinking affects violent behavior through a number of mechanisms. Drinking may
change the objective consequences of violence, since alcohol acts as an anesthetic and
also as an excuse. It may also act on information-processing capacity, making people
myopic and narrowing their repertoire of responses to a tense situation. It may also
cause self-management problems, in which impulse gets the better of long-term interests
[Cook and Moore (1993b)].
Economists have not contributed much to the empirical work on this subject. One
exception is Cook and Moore (1993b), in which we examine the effects of aggregate
drinking and alcohol taxation on four forms of violent crime in a state-year panel of
data for the years 1979-1988. Using a fixed-effects specification for state and year,
we find that per capita alcohol consumption has a significantly positive effect on rape,
aggravated assault, and robbery, and a negligible effect on criminal homicide rates. In
the reduced-form estimates, the state beer-excise tax rate has a strong and significantly
negative effect on rape and robbery, but not on homicide or assault.
There is scant economic research in the areas of domestic violence. Markowitz and
Grossman (2000) examine the effects of beer tax rates, illegal drug prices, and alcohol-
control measures on violent behavior towards children. Their findings include a signif-
icant reduction in the likelihood of any violent behavior to beer-tax increases. There is
a weak relationship between the number of licensed retail alcohol outlets and the like-
lihood of violence directed at children. More severe acts of violence appear to be more
responsive to tax increases than is violence in general, with elasticities equal to about
-0.25 across all specifications of the model. Estimated effects of the number of retail
outlets likewise indicate a positive effect on severe violence.
The propensity to commit suicide may also be influenced by heavy drinking. The
blood of suicide victims often contains a high percentage of alcohol [Hayward et al.
(1992)], and receiving treatment for alcoholism or alcohol abuse is a significant risk
factor for suicide [Draper (1994)]. Skog and Elekes (1993) examined the relationship
between alcohol consumption and suicides in Hungary, and found the two to be highly
positively correlated, with a lag of one year in alcohol consumption.
In an interesting report of a natural experiment, Wasserman et al. (1994) examined
the relationship between male suicides and alcohol consumption in the Soviet Union
during the period of Perestroika, 1985-1988. The early years of this period were char-
acterized by a very restrictive alcohol policy. Relative to the last year of the Brezhnev
regime (1984), suicides and violent deaths declined sharply in 1986, falling to 65 per-
cent of their 1984 level. By 1988, violent deaths were 72 percent of their 1984 level,
and suicides 61 percent. In 1990, the last year covered by the data, these rates of death
due to violence and suicide held at 85 and 68 percent. Meanwhile, total male mortality
had returned to its 1984 level by 1990.
Simple regression analyses of the Soviet data indicate that the suicide rate for men
falls by 1.3 per 100,000 for every 1 liter-per-year reduction in pure alcohol consumption
per capita. Violent death rates are more sensitive to alcohol consumption with estimated
effects ranging across provinces from 7 to 20 deaths per 100,000 per liter of alcohol.
6. Productivity
The belief that drinking impairs productivity has helped motivate a wide range of both
private and public responses, from workplace rules banning drinking on the job to alco-
hol regulations governing the armed forces. National estimates of alcohol-related social
costs are typically dominated by the value of lost productivity. Historically this concern
with the quality and quantity of work provided by the labor force was a major factor in
Nineteenth Century temperance movements in the United States and Europe [Roberts
(1984), Rumbarger (1989)]. Clark Warburton (1932) stated the argument concisely:
"Prohibition, if it actually resulted in the cessation of use of alcoholic beverages,
might be expected to affect the efficiency of industry in several ways. The prin-
cipal effect of alcohol is on the central nervous system, and experiments show
that a decrease in the consumption of alcohol during, or immediately preceding,
1654 P.. Cook and M.J. Moore
working hours is accompanied by greater skill at work. The finer co-ordination

made possible by the absence of alcohol tends to reduce the accident rate. The
elimination of drinking bouts should tend to eliminate absenteeism, especially on
Monday, and irregularity in reporting at work. The impossibility of drinking to
excess should result in less sickness and absence on account thereof, and in a
longer average working life." (pp. 195-196)
In recent times the belief that alcohol abuse reduces the productivity of some employ-
ees has persuaded the majority of large corporations in the United States to establish
occupational alcoholism programs or employee-assistance programs [Walsh (1982)].
Productivity concerns were also paramount in the Soviet Union in 1985, when Mikhail
Gorbachev decreed stringent new controls on production, availability, and price of al-
coholic beverages, which reduced consumption by about one-third in the next couple of
years. There is evidence that on-the-job drinking fell, which may have accounted in part
for the rapid economic growth of that period (The Economist 12/23/89, p. 50).
Curiously, however, the belief that heavy drinking impairs productivity does not re-
ceive unambiguous support from the econometric work on this subject. We begin our
review with the literature focusing on the direct effects of current drinking (or alcohol-
related health status) on earnings or some other indicator of productivity.
6.1. Direct effects of drinking on productivity
An early American effort to estimate the productivity costs of drinking is due to Irving
Fisher (1926). His view was that drinking slowed down the "human machine" (p. 118),
and he noted that "All of us know that industrial efficiency was one of the chief reasons
for Prohibition" (p. 158). He supported his claim of impaired productivity by citing a
number of experiments, which showed that drinking reduces proficiency or speed at
some task. In particular, he noted an experiment in which four typesetters were studied
over a four-day period; two of them were given drinks, and the other two were used as a
control group. The conclusion was that drinking three glasses of beer in a day reduced
productivity by about 10 percent. Fisher made a heroic extrapolation from this result,
projecting a 5 percent increase in national productivity as a result of reduced drinking
caused by Prohibition.
Modem scholars studying productivity effects have enjoyed larger sample sizes but
unlike Fisher have utilized nonexperimental data. The typical econometric study esti-
mates the productivity effects of drinking, utilizing survey data in which respondents are
asked about their drinking, work, income, and other items. The dependent variable is a
measure of earnings or hours worked, while the key independent variable is a measure
of the quantity or pattern of contemporaneous drinking, or alcohol-related psychiatric
disorder (alcohol dependence or abuse).
Contemporaneousdrinking. The most consistent finding is paradoxical, namely that

drinkers tend to earn more than nondrinkers. For example, based on the Quality of
Employment Survey (QES) data for full-time male American workers in 1972-1973,
lifetime abstainers and recent abstainers earned $9,000 and $8,500 respectively; for
drinkers, however, earnings were from $10,000-11,000 across the drinking spectrum
from 1 drink per month to 120 or more [Cook (1991)]. The pattern that "abstainers earn
less" holds up when other characteristics of the individual are controlled for in an OLS
regression, and appears to be true for women as well as men [Berger and Leigh (1988),
Bryant et al. (1992), Zarkin et al. (1998)]. Some studies find an inverted U-shape be-
tween earnings and drinking [French and Zarkin (1995)], but others confirm the QES
finding that there is a "drinking bonus" at all levels of self-reported alcohol consump-
tion.
Kenkel and Ribar (1994) provide one of the most thorough explorations of the re-
lationship between drinking and earnings, although they do not analyze the abstainers
as the special case, which they apparently are. The authors' data are from the National
Longitudinal Survey of Youth (NLSY) for 1989 (when the respondents were 24-32
years old). Among their measures of contemporaneous drinking are the number of days
in the past month in which the respondent drank ("days drinking") and the number of
days in which he or she consumed 6 or more drinks ("heavy drinking"). The effects of
"days drinking" on log earnings and log of hours worked was negligible for men and
small but discernibly positive for women, even when a long list of control variables
were included. The results for "heavy drinking" indicate little or no effect for women,
and a small negative effect for men.
The authors note three possible sources of bias in OLS results of this sort: omitted
variables that may influence both drinking and earnings; errors in self-reported drinking;
and reverse causation, whereby earnings influence alcohol consumption. They address
the latter two problems through a simultaneous-equations analysis in which the iden-
tifying variables are indicators of alcohol availability in the respondent's state and of
alcoholism in his or her family. This specification yields evidence that "heavy drinking"
and "days drinking" reduce male earnings, while they increase female hours and have
no discernible effect on earnings.
Finally, several studies have analyzed the effect of drinking on absenteeism, also
with mixed results. Manning et al. (1991) report results from two data sets, the Rand
Health Insurance Experiment (HIE) and the National Health Interview Study for 1983.
In neither do they find a relationship between quantity consumed by current drinkers
and absenteeism. (Using the HIE they find that "former drinkers" have 38 percent higher
absentee rates than others.) On the other hand, French and Zarkin (1995), using survey
data for workers at four large work sites, find that both overall drinking and frequency
of drunkenness are positively related to absenteeism.
Alcohol dependence and abuse. Figure 1 suggests that in addition to the effect of
current drinking on productivity, there may also be an effect of past drinking as mediated
by health status. Several studies have explored this linkage using data on two alcohol-
related conditions termed "alcohol dependence" and "alcohol abuse" in the Diagnostic
and Statistical Manual of Mental Disordersof the American Psychiatric Association.
"Alcohol dependence" is defined by symptoms indicating psychological and physical
1656 .J. Cook and M.lJ. Moore
dependence on alcohol and impaired functioning at work or home. "Alcohol abuse" is

signaled by problems and risky behavior associated with bouts of heavy drinking.
Mullahy and Sindelar (1993) utilize data from the Epidemiological Catchment Area
survey of the New Haven area, which includes diagnostic questions on mental illness.
Twenty percent of their primary sample (males age 30-59) had at some point in their
lives experienced the symptoms defining alcohol dependence or abuse. This group,
which the authors term "alcoholics," had a substantially lower employment rate than
the non-alcoholics, and lower earnings - results that held up after controlling for other
characteristics. Strangely, alcoholic males in their 20s and 60s actually had higher earn-
ings than nonalcoholics.
Kenkel and Ribar (1994) in their analysis of the NLSY data discussed above, estimate
the effect of current alcohol dependence and abuse. In their OLS estimates they find
no evidence that either of these conditions affect hours worked, but report that these
conditions suppress earnings for men and enhance them for women. The simultaneous-
equations analysis also finds reduced earnings for men but no effect (except longer
hours) for women.
Kenkel and Wang (1998) extend this analysis of NLSY data by comparing job at-
tributes of alcoholic and non-alcoholic men. They find that male alcoholics are less
likely to receive a variety of fringe benefits, more likely to be injured on the job, and
work for smaller firms. Alcoholics are less likely to be in white-collar occupations, but
those who are earn about as much as non-alcoholics. In blue-collar professions, how-
ever, the alcoholics earn about 15 percent less.
Summary and interpretation. The estimated relationship between self-reported drink-

ing and measures of the quality and quantity of labor supply (hours worked, earnings,
absenteeism) is not what we might expect. Abstainers clearly earn less, even when we
control for a wide range of other individual characteristics. The evidence on whether
the labor market penalizes heavy drinking (without regard to symptoms of dependence)
is at best inconsistent. Prime-age males who exhibit symptoms of alcohol dependence
or abuse earn less, but women in this category appear to earn more.
So were Irving Fisher and Mikhail Gorbachev wrong? Was it a mistake for the U.S.
military to crack down on heavy drinking, and for most corporations to establish em-
ployee assistance programs? We doubt it. More likely some combination of systematic
errors in self-reported drinking and specification error account for the anomalous find-
ings.
In the latter regard, we note the likelihood that the workplace social environment has
an important influence on individual drinking, a causal link that has not been captured
in the existing literature. 16 For example, a man who has a taste for working with his
hands may become a construction worker, an occupation that places him in a "wetter"
16 In fact there is a considerable social-science literature on the ways in which the job environment may en-
courage or discourage drinking. Some occupations have long been associated with heavy drinking, including
those in which alcohol is readily available (bartenders, brewers) and in which workers are often unsupervised
work environment than if he instead had chosen to become an accountant. A woman

who enters a male-dominated profession such as law may tend to take on male drinking
patterns, which compared to other women would make her a heavy drinker.
Still, the balance of the evidence favors the view that in moderate-drinking popula-
tions, such as prime-age adults in the U.S. the true direct effect of drinking on produc-
tivity is likely to be negative but small. One source of evidence on this matter is the
views of the drinkers themselves. A number of surveys have asked respondents whether
their drinking has caused them any problems [Room et al. (1994)]. For example, in a
national survey conducted in the United States in 1984, 2.9 percent of men stated that
their drinking had harmed their employment opportunities, and 0.8 percent reported
they had lost or nearly lost a job as a result of their drinking [Hilton and Clark (1987)].
The corresponding percentages for women were about half those of the men.
On the other hand, the direct effect of drinking on productivity is not the whole story.
As suggested by the lower loop in Figure 1, heavy drinking may have an indirect effect
on productivity by interfering with the process of accumulation of human capital. We
summarize the evidence on this matter below.
6.2. Drinking and human capital
In their analysis of ECA for New Haven, Mullahy and Sindelar (1989, 1991) found that
teenage alcohol dependence led to early termination of schooling that in turn reduced
subsequent income. First onset of alcoholism's symptoms before age 19 (as reported
retrospectively by adult respondents) was associated with an 11 percent reduction in
schooling attainment, controlling for several other characteristics.
Cook and Moore (1993c) suggest two possible mechanisms by which drinking and
schooling may interact for adolescents:
(1) Heavy drinking may interfere with learning and classroom performance, thereby
reducing the contribution to human capital of an additional year of schooling and
hence the incentive to continue;
(2) To the extent that higher education is rationed according to past scholastic perfor-
mance and reputation, heavy drinking may have consequences that increase the cost
of continuation (p. 414).
Thus, a forward-looking student would make drinking and schooling decisions together;
a myopic student would make them sequentially; but in either case alcoholic-beverage
prices and other determinants of high-school drinking are a determinant of school per-
sistence.
The authors utilize NLSY data for high-school seniors in 1982, which allow inclu-
sion of an extensive list of covariates. They find that the beer tax and minimum legal
(salespeople, farmworkers) [Trice and Sonnenstuhl (1988), Harford and Brooks (1992)]. The U.S. military
was a particularly "wet" environment fueled by tax-free alcohol and heavy-drinking traditions, until a more
stringent set of policies on drinking and drugs was introduced in the early 1980s [Bray, Marsden, Herbold,
and Peterson (1992)].
purchase age in the respondent's state have a direct effect on school persistence, as mea-
sured either by the number of years of college or the likelihood that the respondent will
eventually graduate from college.
Schooling is not the only dimension of human capital that may be affected by drink-
ing. Kenkel and Ribar (1994) find that the likelihood of marriage is negatively affected
by heavy drinking and alcoholism symptoms for both men and women, a finding that
holds up well across different specifications.
These results suggest that much of the effect of drinking on productivity may be
indirect, mediated by the accumulation of human capital. If so, controls on youthful
drinking become particularly important in influencing the course of the economy.
7. Evaluation of alcohol taxation and other alcohol-control measures
A number of alcohol-control measures have been subjected to systematic evaluation,

as discussed above and elsewhere [e.g., Edwards et al. (1994)]. Most of this research
has been limited to estimating the effects of such regulations on alcohol consumption
and abuse and the consequences thereof. But some economists have attempted to take
the evaluation effort further by assessing the costs of specific regulations and by plac-
ing a monetary value on the estimated benefits. This effort to determine by technical
means whether a policy innovation is beneficial on balance is controversial, given that it
requires placing a monetary value on life and disability, and also requires explicit judg-
ments concerning how much deference the public should accord revealed individual
preferences concerning drinking and alcohol-related behavior. While economists can-
not necessarily claim full wisdom on such matters, they can help provide structure to
the argument.
Probably the most pervasive government intervention in the alcohol market is the
imposition of specific taxes. The economics of alcohol excise taxes is well developed,
and will be the main focus here.
7.1. Background
Alcohol excise taxes vary widely across time and space. Historically, the very first in-
ternal revenue measure adopted by the U.S. Congress was an excise tax on whiskey [Hu
(1950)]; a subsequent increase in that tax from 9 to 25 cents per gallon engendered an
armed insurrection. The appropriate level for alcohol excise taxes remains a contentious
issue today in the United States at both the federal and state level. Generally the real
values of excise-tax rates have trended downward during the post-War period. They are
unit taxes, defined in terms of volume rather than value of the product, and legislated
increases have not kept up with inflation. For example, the federal tax on distilled spirits
in 1998 (the equivalent of 21 cents per ounce of ethanol) was about four times higher in
1951.
In the European Union alcohol excises have been one of the most difficult-to-resolve
issues in the tax-harmonization effort, since the tax rates differ widely. Nordic coun-
tries in particular have long used high taxes to restrict drinking, rather than simply as a
revenue measure, whereas the wine producing countries tend to have much lower taxes
[Kay and Keen (1986)].
Alcohol excises and duties have been an important source of public finance in certain
times and places. This has been particularly true in Russia and the old Soviet Union - in
the early 1980s, taxes on the liquor trade provided about 13 percent of the state budget
(The Economist 12/23/89, p. 50) - although in most advanced countries alcohol taxes
constitute less than one percent of the total. Still, there is widespread acceptance that
alcohol should be taxed more heavily than other commodities.
Public-finance theory provides a framework for evaluating alcohol-excise taxes. Ap-
plication of the standard criteria of economic efficiency and equity requires some ac-
count of the externalities and health effects of alcohol consumption. Also relevant in
practice have been historical comparisons and comparisons with tax rates in other juris-
dictions [Cook (1988)].
7.2. Equity criteria
"Equity" or "fairness" is a standard concern of normative tax theory [Stiglitz (1988),

Rosen (1988)]. Several general principles are widely accepted
* Equals should be treated equally (the "horizontal equity" criterion);
* Households with greater ability to pay should be taxed more heavily than households
with less (the "vertical equity" criterion).
* Households that receive greater benefit from government activities should be taxed
more heavily than households that benefit less (the "benefit" criterion).
Incidence. Implementing any of these principles requires at a minimum some knowl-

edge of the incidence of alcohol-excise taxes. These taxes are paid by wholesale dealers.
The evidence from the United States suggests that taxes are ultimately passed on to con-
sumers with a markup in the form of higher retail prices [Center for Science in the Public
Interest (CPSI) (1989)]. Given the imperfect competition that tends to characterize the
relevant markets, it is reasonable to suppose that markups vary widely, depending on
the circumstances of the industry when the tax increase is imposed.
Given that consumers end up paying 100 percent or more of a tax increase, it is
necessary in judging vertical equity that we know how alcohol sales are distributed
across income levels.17 In the United States expenditures on alcoholic beverages in-
crease roughly in proportion to income [Sammartino (1990)], but the volume of alcohol
17 It should be noted that in the United States, by one estimate 20 percent of all alcoholic beverage sales are
to businesses [Sammartino (1990, p. 76)]. No information is available on the incidence of this portion of sales.
1660 P.J. Cook and M. J. Moore
consumed increases little if at all with income. Hence excise taxes, which are imposed
by volume rather than value, are highly regressive. I8
This conclusion must be qualified, however. First, the incidence of a tax increase is
not determined solely by who buys the product, but also by how the tax increase affects
producers and sellers [Rosen (1988, p. 266)]. More generally, an increase in the excise
tax rate will tend to have effects on other markets, which should be taken into account
in calculating the incidence. For example, a tax increase on beer will reduce the traffic
accident rate, which in turn will reduce the cost of driving, both directly and indirectly
through reduced insurance rates.
Horizontal equity and user fees. With respect to horizontal equity, the fundamental
issue is whether otherwise-similar households should be taxed differently because some
purchase more alcohol than others. The household incidence of alcohol excises is highly
concentrated; by one estimate, 6.5 percent of U.S. adults consume half the total alcohol
sold [Malin et al. (1982)]. The equity justification for imposing the tax on drinkers is as
a sort of "user fee", which charges them for the negative externalities of their drinking.
The characterization of an alcohol excise tax as a user fee is also related to the bene-
fit principle of tax fairness. People who abuse alcohol benefit from certain government
programs more than nonabusers. For example, lifetime heavy drinkers have elevated
morbidity and disability, and hence claim a disproportionate share of government ex-
penditures on medical care and disability payments through the Social Security system
[Rice et al. (1990)]. Government revenues from alcohol taxes help defray these and
other alcohol-related public expenses.
Given that the bulk of alcohol-related costs are associated with rare events (most
notably traffic accidents), this "user fee" is akin to an insurance premium. Alcohol taxes
differ among individuals in direct relation to how much they drink, which is a strong
predictor of the likelihood of an alcohol-related problem [Moore and Gerstein (1981,
p. 45), Edwards et al. (1994)]. However excises do not discriminate with respect to other
correlates of alcohol-related problems, such as age, sex, prior history of drinking and
alcohol-related problems, or drinking pattern. A 21-year-old man who drinks 7 beers a
week in a single session and then attempts to drive home pays the same tax as a 40-year-
old woman who drinks one beer with dinner each night. From the actuarial viewpoint,
then, this tax is imperfect.
A fairly comprehensive study of the external costs of heavy drinking [Manning et
al. (1989, 1991)] found that most of these costs are not financed by government in the
U.S., but rather by private insurance companies or by innocent victims. The authors
concluded that the present value (using a 5-percent discount rate) of external costs per
ounce of ethanol consumed was about 48 cents, double the average state and federal tax
18 Lyon and Schwab (1995) suggest that the cross-section relationship between tax and income may in part
reflect differences with respect to location over the life cycle. But their calculations for the alcohol excise
taxes suggests that these taxes are highly regressive in a life-cycle framework as well.
per ounce, which suggests that the current "user fee" is not high enough. 19 However,
this judgment is based on a more inclusive standard than the benefit criterion, which
is limited to benefits provided by government. The implicit notion is that the drinker
should pay regardless of whether the cost is to a government program or to a private
organization or individual.
The analysis by Manning et al. attempted to sort out alcohol-related costs between
those that are borne by the drinker or his or her household (internal) and those that
are borne by those outside the household (external). By this reckoning, earnings lost as
a result of heavy drinking are internal costs; reductions in payroll taxes or claims on
Social Security benefits are external. Medical and disability costs that are reimbursed
by insurance are internal if the insurance premium is paid by the drinker and reflects his
true risk status, and external otherwise. Motor-vehicle injuries are also divided between
internal (where the injury is to the drunk driver himself) and external (where someone
else is injured in an accident involving a drunk driver).
The question of where the line should be drawn between internal and external is
especially difficult with respect to intrafamily effects. Family members have individual
interests that are sometimes in conflict, though these differences remain "internal" to
the extent that they can be negotiated within the family [Heien and Pittman (1993),
Heien (1995)]. Nonetheless there is a clear public stake in preventing alcohol-induced
family violence, child abuse and neglect, and fetal damage. The "fairness" problem with
including these costs in the justification for higher excise taxes is that the other family
members may end up paying twice if the drinker does not change his or her behavior -
the higher tax reduces money left over for other members of the household. 2 0
Another gray area between internal and external costs is with respect to injuries to
passengers of vehicles driven by drunk drivers. If the passengers are consenting adults,
then their choice reveals an ex ante judgment that accepting the ride is preferable to the
next-best alternative. This argument appears to generalize to all users of the road, who
presumably know there is some chance of encountering a drunk driver when they choose
to drive on it, and choose to accept that risk. But voluntary acceptance in this case does
not imply that there is no externality. If other users of the road were able to negotiate
directly with the drunk, they would likely find room for a Pareto-improving bargain that
kept the drunk from driving. That is less clear for the case of vehicle passengers, who
presumably do have the opportunity to negotiate with the drunk.
Finally, we note the interesting ambiguity concerning the "drinker should pay" ben-
efit criterion. Suppose that the alcohol excise tax was increased so that total collections
were equal to the external costs associated with drinking. The increase in these tax rates
would cause a reduction in tax collections from other sources. A standard assumption in
revenue estimates is that gross national product is fixed, and that a tax increase causes
19 Miller and Blincoe (1993) redo the estimate of motor-vehicle accidents to include nonfatal injuries. They
estimate that the external cost of drinking is $0.63 per ounce just to account for these accidents.
20 Interestingly, in the 19th century "dram-shop" liability provided the wife of an alcoholic a cause of action
against the saloon where her husband was drinking away the household means of sustenance.
1662 P.J. Cook and M. J. Moore
a reduction in factor incomes, which in turn will reduce income and payroll tax collec-
tions. Boyd and Seldon (1991), using a computable general equilibrium model, estimate
that an increase in alcohol and tobacco taxes will increase net government revenue by
only 60 percent of the increase in collections on those taxes. The "loss" of 40 percent
is the result of the tax consequences of the reallocation of economic activity induced by
the tax increases.
7.3. Economic efficiency
If there are negative externalities in the consumption of alcoholic beverages, then in the
absence of government action prices will be "too low" in the sense that at the margin
the value of drinks to consumers will be less than their cost. Taxes on alcohol can then
be justified as a mechanism for internalizing the external costs of alcohol abuse; ideally,
the tax on a drink should equal the expected value of the external cost of consuming
that drink.
There is a fundamental difference between this Pigovian principle, which is based
on an efficiency argument, and the equity principle that the "drinker should pay." The
total revenue generated by a Pigovian tax may well be greater than the total external
cost, because it reflects the external cost of the marginal drink rather than of the average
(inframarginal) drink.
The efficiency principle, unlike the fairness principle, requires that corrective taxation
alter consumers' behavior. If an increase in the tax (and price) of alcoholic beverages
had no effect on the prevalence of alcohol abuse and its external consequences, then it
would not improve economic efficiency. The evidence presented in section 5 above that
excise tax increases reduce motor-vehicle fatalities is particularly germane.
The application of the corrective tax principle to drinking must account for the fact
that the external cost of a drink differs depending on the personality of the drinker, the
time and place of drinking, how many drinks have been consumed already, and the type
of alcoholic beverage. 2 1 It is possible to institute some crude differentiation in tax rates
(e.g., on premise vs. off premise), but the problem remains. Diamond (1974) demon-
strated that when social costs differ with circumstances, then the value of that uniform
tax that maximizes social welfare (under certain assumptions) is equal to a weighted
average of the marginal external costs. Pogue and Sgontz (1989) applied this theory to
the case of alcoholic beverages, with a model that assumes all drinkers can be classified
21 Acan of beer, a glass of wine, and a shot of spirits all contain approximately the same amount of ethanol,
but tend to be taxed quite differently; inthe United States, for example, the federal excise tax on a shot of
spirits exceeds the tax on a can of beer by a factor of two and on a glass of wine by a factor of three. Anumber
of commentators have advocated that tax rates be made uniform, the same per unit of ethanol, regardless of
the type of beverage [CSPI (1989)]. But the arguments in support of this differentiation, based on claims
about both regressivity and external cost, are not supported by available evidence [Cook and Moore (1993a)];
inparticular, there is little difference inregressivity, and there is no basis for claiming that beer is the "drink
of moderation". For another perspective, see Saffer (1994).
as either "abusers" or "nonabusers." In that model the correct tax depends on the pro-
portion of the drinking population who are abusers, and their price elasticity of demand
relative to nonabusers. We suggest an alternative approach that does not require this
artificial dichotomy and takes better advantage of the empirical literature: the literature
provides estimates of the effects of a change in tax on alcohol consumption, and on
highway fatalities and other damages. The former is the basis for estimating incremen-
tal loss of consumers surplus, while the latter can be used to estimate the reduction in
social cost. Taxes should be increased so long as the latter exceeds the former.
A number of authors have suggested that the principle of corrective taxation be ex-
tended to account for the presumed fact that consumers tend to underestimate the inter-
nal costs of their drinking. If consumers tend to disregard certain costly consequences
of their drinking out of ignorance or myopia, then it is possible that a tax increase would
bring their drinking closer into line with their "true" preferences [Atkinson and Meade
(1974), Godfrey and Harrison (1990)]. The consequence is to increase the magnitude of
the proper corrective tax [Pogue and Sgontz (1989), Phelps (1988)].
This analysis of efficiency presumes that the price system functions well in other re-
spects. That is not the case. For example, taxes on income have the effect of reducing
the incentive to provide labor services to the market. Taxes reduce take-home pay and
may distort such choices as how much to work, how much effort to expend while work-
ing, and how much to invest in education and training [Rosen (1980)]. One approach
to correcting for the disincentive effects of income taxation is to impose special taxes
on commodities that are complements to leisure and substitutes for investment in hu-
man capital [Kay and Keen (1986, p. 88), Slemrod (1990, p. 159), Corlett and Hague
(1953)]. Alcohol is believed to be one such commodity (although note the mixed evi-
dence reported in Section 6).
Where the bulk of the external cost results from drunk driving, alcohol-induced vi-
olence, and other behavior that is subject to sanctions, then it might be more efficient
to increase the "price" imposed on violators than to increase the alcohol tax [Kenkel
(1996)]. For the case of drunk driving, that "price" may include legal and private penal-
ties if apprehended and convicted, as well as the expected cost of causing an accident,
which in the United States may include an increase in insurance-premium rates and civil
liability [Sloan and Githens (1994), Sloan, Reilly and Schenzler (1995)]. Still, impos-
ing stiffer sanctions is socially costly in itself, and constrained by various practical and
ethical considerations. 2 2 And no matter how stiff the sanctions, there will be some who
will ignore them and drive drunk or fight anyway. Given these concerns, there is a role
for an ex ante tax to preempt some of the drinking that would otherwise lead to trouble.
22 The violator may be judgment proof, in the sense of not being able to pay a fine as large as the cost to
the victim and society [Shavell (1986)]. Imposing other forms of punishment is socially costly. And if the
probability of being apprehended in less than one, the punishment must be greater than the harm to preserve
an appropriate deterrent. For a similar argument about gun control, see Cook and Leitzel (1996).
7.4. Total social cost
Closely related to the problem of computing the appropriate excise-tax rates is a prob-
lem that has been of somewhat less interest to economists, calculating the total cost of
alcohol abuse. Estimates of the monetized social burden of a disease or health-related
activity have become commonplace in the public-health literature because of their im-
portance in the policy process. Such estimates are a precursor to evaluation, since the
cost is an indication of the benefit should a comprehensive "cure" be found. A costly
disease, one that causes a large reduction in the overall standard of living, thus appears
to have a strong claim on the public fisc for research and prevention activities. But
economists have served more as critics than as producers of these estimates.
The most prominent estimates for alcohol abuse and other diseases follow the tem-
plate developed by a task force of the U.S. Public Health Service [Hodgson and Mein-
ers (1979, 1982)]. This "cost-of-illness" (COI) method is somewhat at odds with the
economists' normative accounting framework, primarily because the COI accounting
framework is structured around production rather than consumption.
Several estimates of the social costs of alcohol abuse have been widely disseminated
in the United States [Berry and Boland (1977), Harwood et al. (1984), Rice et al. (1990),
Harwood, Fountain, and Livermore (1998)]. The most recent of these estimates was
$148 billion (in 1993), including $99 billion for lost earnings of those who died prema-
turely or were disabled due to alcohol-related accidents or disease, and $19 billion for
medical care and rehabilitation. As is traditional with this approach, no account is taken
of the subjective value that individuals place on their life and health, and on the lives
of those they care about. Some authors have modified the COI approach to incorporate
a willingness-to-pay-based valuation for additional life years; Miller et al. (1998), for
example, estimate the "comprehensive" social cost of alcohol-involved crashes in the
United States as $115 billion in 1993, much of which stems from the subjective value
of lost life and ability.
Even with this addition of the "consumption" value of life, the COI framework does
not fit the standard economic framework because it fails to distinguish between internal
and external costs. As we saw in the discussion of a cost-based excise tax, this distinc-
tion is highly relevant, since otherwise the drinker would be asked to pay the internal
costs twice. However, in other applications it may be appropriate to compute the total
social cost, rather than only the external cost. For example, the value of a cure for liver
cirrhosis should include the cirrhosis-related costs currently bome by drinkers (and their
households), as well as those borne by the rest of society.
Finally, estimates of the cost of alcohol abuse are also subject to what might be called
"conceptual" uncertainty, a lack of clarity about the conditions under which the esti-
mated "cost" would be "saved". For example, estimates of productivity losses due to
excess drinking typically presume that the socioeconomic characteristics of the labor
force are not influenced by alcohol abuse. The implicit question answered by the es-
timated productivity loss is this: How much would the quality and quantity of labor
increase if some (magical) intervention were instituted that eliminated alcohol abuse
among the working-age population while having no effect on that population's socioe-
conomic characteristics? Yet according to the results reviewed in Section 5 above, the
primary mechanism by which alcohol abuse influences productivity is through workers'
marital status and formal education. The general problem here is that alcohol plays a
diverse and complex role in shaping everything from personal health-related habits and
occupational choice to family life and social intercourse. It seems likely that any inter-
vention that reduced or eliminated alcohol abuse would have a number of ramifications
for other aspects of life: What, then should be held constant in estimating the costs of
alcohol abuse?
A number of economists [Myrdal (1930), Osterberg (1983)] have suggested that it
would make more sense to estimate the costs and benefits of specific real-world inter-
ventions (such as a marginal increase in the excise tax rate), rather than the imaginary,
perfectly effective intervention that underlies the estimate of total costs. Knowing the
intervention helps guide the evaluators in deciding which causal mechanisms to explore
and which to ignore. While this perspective seems valid, there nonetheless remains a
considerable demand for estimates of the total costs of alcohol abuse for use in influ-
encing the political and policy agenda.
8. Concluding observations
Economists' contributions to the study of the health consequences of drinking have

included both theoretical and empirical advances, and have spanned both the positive
and normative realms. Arguably the most important contribution has been the repeated
empirical demonstration that consumers as a whole are responsive to the general price
level of alcoholic beverages. The demand curves for beer, wine and spirits are negatively
sloped; more surprisingly the consumption of ethanol is negatively related to the prices
of the beverages that contain it, despite the obvious opportunity for substitution among
beverage types. As a result, an increase in the price level of alcoholic beverages tends
to reduce the prevalence of alcohol abuse and its health-related consequences. Thus the
excise tax on alcoholic beverages is an effective alcohol-control measure that can be
used to promote the public health.
Economists have been somewhat at odds with the public-health community in arguing
that the effectiveness of a government alcohol-control measure is not sufficient justifi-
cation for imposing it. In this area, as in others, consumer sovereignty should be given
its due, and in setting policy the social costs of drinking should be balanced against the
pleasure to consumers. As we have seen, the normative literature has included a debate
over how fully to respect the revealed preferences of consumers and households, given
that consumers are not necessarily well informed about the consequences of drinking,
that ethanol is addictive, and that drinkers may take inadequate account of their depen-
dents' welfare.
Further economic research is needed to sort out the costs and benefits of drinking.
As we write this (in 1998), the empirical literature on productivity effects is particu-
larly unsatisfactory; how can we explain the persistent findings, at least for U.S. data,
that drinkers earn more than abstainers, or that for women who drink (but not men)
earnings increase with the amount of alcohol consumed? Economists have for the most
part simply ignored another important issue, the apparent health benefits of moderate
drinking.
Outside of the policy arena, alcohol remains interesting for economists because it
provides a clear example of a commodity whose consumption has important intertem-
poral effects, and which is subject to social influence. The theory of rational addiction
has provided one framework for exploring the first mechanism; there is also an oppor-
tunity here to collaborate with psychologists in an effort to reconcile the assumptions
of the model with experimental evidence on intertemporal decisionmaking. Meanwhile,
economists are just beginning the study of social influence in drinking, another area
which may reward cross-disciplinary collaboration.
The primary justification for public support of research on drinking and its conse-
quences is the importance to public health worldwide. Economists have played a rela-
tively small but important role in this research program, challenging some established
beliefs about the singularity of alcohol and offering a broader normative framework. We
believe that a closer collaboration between economists and other behavioral scientists
will pay off both in terms of scientific progress and policy influence.
References
Adams Business Media (1997), The Beer Handbook (Adams Business Media, Cleveland, OH).
Atkinson, A.B., and T.W. Meade (1974), "Methods and preliminary findings in assessing the economic and
health consequences of smoking with particular reference to lung cancer", Journal of the Royal Statistical
Society Series A 137:297-312.
Babor, T.F. (1985), "Alcohol, economics and the ecological fallacy: Toward an integration of experimental
and quasi-experimental research", in: E. Single and T. Storm, eds., Public Drinking and Public Policy
(Addiction Research Foundation, Toronto) 161-189.
Babor, T.F., J.H. Mendelson, I. Greenberg and J. Kuehnle (1978), "Experimental analysis of the "happy hour":
Effects of purchase price on alcohol consumption", Psychopharmacology 58:35-41.
Baumann, K.E., and S.T. Ennett (1996), "On the importance of peer influence for adolescent drug use: Com-
monly neglected considerations", Addiction 91(2):185-198.
Becker, G.S. (1992), "Habits, addictions, and traditions", Kyklos 45:327-346.
Becker, G.S. (1996), Accounting for Tastes (Harvard University Press, Cambridge).
Becker, G.S., M. Grossman and K.M. Murphy (1991), "Rational addiction and the effect of price on con-
sumption", American Economic Review 81(2):237-241.
96(4):675-700.
Ben-Porath, Y. (1967), "The production of human capital and the life cycle of earnings", Journal of Political
Economy 75:352-365.
Berger, M.C., and J.P. Leigh (1988), "The effect of alcohol use on wages", Applied Economics 20:1343-1351.
Berry, R.E., and J.P. Boland (1977), The Economic Cost of Alcohol Abuse (The Free Press, New York).
Boland, B., and R. Roizen (1973), "Sales slips and survey responses: New data on the reliability of survey
consumption measures", Drinking and Drug Practices Surveyor 8:5-10.
Boyd, R., and B.J. Seldon (1991), "Changes in sin taxes: A general equilibrium perspective", Land Economics
67(3):365-374.
Bray, R.M., M.E. Marsden, J.R. Harbold and M.R. Peterson (1992), "Progress toward eliminating drug and
alcohol abuse among U.S. military personnel", Armed Forces and Society 18(4):476-496.
Bruun, K., G. Edwards, M. Lumio, K. Makela, L. Pan, R.E. Popham, R. Room, W. Schmidt, O. Skog, P.
Sulkunen and E. Osterberg (1975), Alcohol Control Policies in Public Health Perspective (Aurasen Kir-
japaino, Forssa, Finland).
Bryant, R.R., V.A. Sumaranayake and A. Wilhite (1992), "Alcohol use and wages of young men: Whites vs.
nonwhites", International Review of Applied Economics 6(2): 184-202.
Camargo Jr., C.A., C.H. Hennekens, J.M. Gaziano, R.J. Glynn, J.E. Manson and M.J. Stampfer (1997),
"Prospective study of moderate alcohol consumption and mortality in US male physicians", Archives of
Internal Medicine 157(1):79-85.
Center for Science in the Public Interest (CSPI) (1989), Impact of Alcohol Excise Tax Increases on Federal
Revenues, Alcohol Consumption and Alcohol Problems (Center for Science in the Public Interest, Wash-
ington, DC).
Centers for Disease Control (1990), "Alcohol-related mortality and years of potential life lost - United States,
1987", Morbidity and Mortality Weekly Report 39(11), March 23.
Chaloupka, EJ., and M. Grossman (1994), Drinking and Addiction? Evidence from the First National Health
and Nutrition Examination Survey (University of Illinois, Chicago, IL).
Chaloupka, EJ., and H. Wechsler (1996), "Binge drinking in college: The impact of price", Contemporary
Economic Policy 14(4): 112-124.
Chaloupka, EJ., and A. Laixuthai (1997), "Do youths substitute alcohol and marijuana? Some econometric
evidence", Eastern Economic Journal 23(3):253-276.
Chaloupka, EJ., H. Saffer and M. Grossman (1993), "Alcohol control policies and motor vehicle fatalities",
Journal of Legal Studies 22(1):161-186.
Chaloupka, FJ., M. Grossman, G.S. Becker and K.M. Murphy (1993), "Alcohol addiction: An econometric
analysis", Presented at the Annual Meeting of the American Economic Association, Anaheim, CA, January.
Chamberlain, G. (1984), "Panel data", in: Z. Grilliches and M. Intrilligator, eds., Handbook of Econometrics
(North-Holland, New York).
Chesson, H.W., P. Harrison and W.J. Kassler (1997), "Alcohol, youth, and risky sex: The effect of beer taxes
and the drinking age on gonorrhea rates in teenagers and young adults", Working Paper.
Clements, K.W., W. Yang and S.W. Zheng (1997), "Is utility additive? The case of alcohol", Applied Eco-
nomics 29:1163-1167.
Coate, D., and M. Grossman (1988), "Effects of alcoholic beverage prices and legal drinking ages on youth
alcohol use", Journal of Law and Economics 31(1):145-171.
Collins, D., and H. Lapsley (1996), The Social Costs of Drug Abuse in Australia in 1988 and 1992 (Com-
monwealth Department of Human Services and Health, Canberra, Australia).
Collins Jr., J.J. (1989), "Alcohol and interpersonal violence: Less than meets the eye", in: N.A. Weiner and
M.E. Wolfgang, eds., Pathways to Criminal Violence (Sage Publications, Newbury Park, CA) 49-67.
Cook, P.J. (1981), "The effect of liquor taxes on drinking, cirrhosis, and auto fatalities", in: M. Moore and
D. Gerstein, eds., Alcohol and Public Policy: Beyond the Shadow of Prohibition (National Academy of
Sciences, Washington, DC) 255-285.
Cook, P.J. (1988), "Increasing the federal excise taxes on alcoholic beverages", Journal of Health Economics
(March):89-91.
Cook, P.J. (1991), "The social costs of drinking", in: A.G. Aasland, ed., The Expert Meeting on the Negative
Social Consequences of Alcohol Abuse (Norwegian Ministry of Health and Social Affairs, Oslo, Norway)
49-81.
Cook, P.J., and J. Leitzel (1996), "Perversity, futility, jeopardy: An economic analysis of the attack on gun
control", Law and Contemporary Problems 59(1):91-118.
Cook, P.J., and M.J. Moore (1993a), "Taxation of alcoholic beverages", in: M. Hilton and G. Bloss, eds., Eco-
nomic Research on the Prevention of Alcohol-Related Problems, NIH Publication No. 93-3513 (National
Institute on Alcohol Abuse and Alcoholism, Rockville, MD) 33-58.
Cook, P.J., and M.J. Moore (1993b), "Economic perspectives on alcohol-related violence", in: S.E. Martin,
ed., Alcohol-Related Violence: Interdisciplinary Perspectives and Research Directions, NIH Publication
No. 93-3496 (National Institute on Alcohol Abuse and Alcoholism, Rockville, MD) 193-212.
Cook, P.J., and M.J. Moore (1993c), "Drinking and schooling", Journal of Health Econonics 12:411-429.
Cook, P.J., and M.J. Moore (1994), "This tax's for you", National Tax Journal (September):559-573.
Cook, P.J., M.J. Moore, and R. Pacula (1993), Drinking by Young Adults, Part I: Demographics (Duke Uni-
versity, Durham, NC).
Cook, P.J., and G. Tauchen (1982), "The effect of liquor taxes on heavy drinking", Bell Journal of Economics
13(2):379-390.
Cook, P.J., and G. Tauchen (1984), "The effect of minimum drinking age legislation on youthful auto fatalities,
1970-1977", Journal of Legal Studies 13:169-190.
Corlett, W.J., and D.C. Hague (1953), "Complementarity and the excess burden of taxation", Review of
Economic Studies 21:21-30.
Criqui, M.H., L.D. Cowan, H.A. Tyroler, S. Bangdiwala, G. Heiss, R.B. Wallace and R. Cohn (1987),
"Lipoproteins as mediators for the effects of alcohol consumption and cigarette smoking on cardiovascu-
lar mortality: results from the lipid research clinics follow-up study", American Journal of Epidemiology
126:629-637.
DeBakey, S.F., ES. Stinson, B.F. Grant and M.C. Dufour (1995), Liver Cirrhosis Mortality in the United
States, 1970-1992 (National Institute on Alcohol Abuse and Alcoholism, Washington, DC).
Diamond, P. (1974), "Consumption externalities and imperfect corrective pricing", Bell Journal of Economics
5: 526-538.
DiNardo, J., and T. Lemieux (1992), "Alcohol, marijuana, and American youth: The unintended effects of
government regulation", Working Paper No. 4212, November (National Bureau of Economic Research).
Draper, B. (1994),"Suicidal behavior in the elderly", International Journal of Geriatric Psychiatry 9:655-661.
Duan, N., et al. (1984), "Choosing between the sample-selection model and the multi-part model", Journal of
Business and Economic Statistics 2(3):283-289.
Dufour, M.C. (1996) "Risks and benefits of alcohol use over the life span", Alcohol Health and Research
World 20(3):145-151.
DuMouchel, W.A., AF. Williams and P. Zador (1987), "Raising the alcohol purchase age: Its effect on fatal
motor crashes in 26 states", Journal of Legal Studies 16:249-266.
Edwards, G., et al. (1994), Alcohol Policy and the Public Good (Oxford University Press, New York).
Fagan, J. (1990). "Intoxication and aggression", in: M. Tonry and J.Q. Wilson, eds., Crime and Justice, Vol.
13 (University of Chicago Press, Chicago) 241-320.
Fisher, I. (1926), Prohibition at Its Worst (The MacMillan Company, New York).
French, M.T., and G.A. Zarkin (1995), "Is moderate alcohol use related to wages? Evidence from four work-
sites", Journal of Health Economics 14(3):319-344.
Gaviria, A., and S. Raphael (1997), "School-based peer effects and juvenile behavior", Working Paper (Uni-
versity of California, San Diego).
Gerstein, D.R. (1981), "Alcohol use and consequences", in: M.H. Moore and D.R. Gerstein, eds., Alcohol
and Public Policy: Beyond the Shadow of Prohibition (National Academy of Sciences, Washington, DC)
182-224.
Gius, M.P. (1996), "Using panel data to determine the effect of advertising on band-level distilled spirits
sales", Journal of Studies on Alcohol 57(1):73-76.
Godfrey, C., and L. Harrison (1990), "Preventive health objectives and tax policy options", in: A. Maynard
and P. Tether, eds., Preventing Alcohol and Tobacco Problems, Vol. 1 (Avebury, Gower Publishing Co.,
Hants, England) 54-74.
Grossman, M. (1972), "On the concept of health capital and the demand for health". Journal of Political
Economy 80:223-255.
Grossman, M. (1993), "The economic analysis of addictive behavior", in: M. Hilton and G. Bloss, eds., Eco-
nomics and the Prevention of Alcohol-Related Problems, NIH Publication No. 93-3513 (National Institutes
of Health, Washington, DC) 91-124.
Grossman, M., EJ. Chaloupka, H. Saffer and A. Laixuthai (1994), "Effects of alcohol price policy on youth:
A summary of economic research", Journal of Research on Adolescence 4:347-364.
Grossman, M., EJ. Chaloupka and I. Sirtalan (1998), "An empirical analysis of alcohol addiction: Results
from the monitoring the future panels", Economic Inquiry 36:39-48.
Grossman, M., D. Coate and G.M. Arluck (1987), "Price sensitivity of alcoholic beverages in the United
States: Youth alcohol consumption", in: H. Holder, ed., Control Issues in Alcohol Abuse Prevention: Strate-
gies for States and Communities (JAI Press, Greenwich) 169-198.
Grossman, M., F. Chaloupka, G. Becker and K. Murphy (1992), "Alcohol addiction: An econometric analy-
sis", Working Paper.
Grunewald, P.J., A.B. Millar and P. Roeper (1996), "Access to alcohol: Geography and prevention for local
communities", Alcohol Health and Research World 20(4):244-251.
Harford, T.C., and S.D. Brooks (1992), "Cirrhosis mortality and occupation", Journal of Studies on Alcohol
53:463-468.
Harwood, H., D. Fountain and G. Livermore (1998), The Economic Costs of Alcohol and Drug Abuse in the
United States 1992 (U.S. Government Printing Office, Washington, DC).
Harwood, H.J., D.M. Napolitano, P.L. Kristiansen and J.J. Collins, eds. (1984), Economic Costs to Society
of Alcohol and Drug Abuse and Mental Illness: 1980 (Research Triangle Institute, Research Triangle Park,
NC).
Hauge, R. (1988), "The effects of changes in availability of alcoholic beverages", in: M.D. Laurence, J.R.
Snortum and FE. Zimring, eds., Social Control of the Drinking Driver (University of Chicago Press,
Chicago) 169-188.
Hayward, L., S.R. Zubrick and S. Silburn (1992),"Blood alcohol levels in suicide cases", Journal of Epidemi-
ology and Community Health 46:256-260.
Heckman, J.J. (1981), "Heterogeneity and state dependence", in: S. Rosen, ed., Studies in Labor Markets
Heien, D.M., and D.J. Pittman (1993), "The external costs of alcohol abuse", Journal of Studies on Alcohol
54:302-307.
Heien, D.M. (1995), "The economic case against higher alcohol taxes", Journal of Economic Perspectives
9(1) Winter:207-209.
Hilton, M.E., and W.B. Clark (1987), "Changes in American drinking patterns and problems, 1967-1984",
Journal of Studies on Alcohol 48(6):515-522.
Hodgson, T.A., and M. Meiners (1979), Guidelines for Cost-of-Illness Studies in the Public Health Service
(Public Health Service Task Force on Cost-of-Illness Studies, Bethesda, MD).
Hodgson, T.A., and M. Meiners (1982), "Cost-of illness methodology: A guide to current practices and pro-
cedures", Milbank Memorial Fund Quarterly 60(3):429-462.
Hoyt, G.M., and F.J. Chaloupka (1994), "Effect of survey conditions of self-reported substance use", Con-
temporary Economic Policy 12(3):109-121.
Hu, T.-Y. (1950), The Liquor Tax in the United States, 1791-1947 (Columbia University Press, New York).
Kay, J., and M. Keen (1986), "Alcohol and tobacco taxes: Criteria for harmonisation", in: S. Cnossen, ed.,
Tax Coordination in the European Community (Kluwer Academic, Amsterdam) 85-111.
Keller, M. (1978), "Alcohol consumption", in: The New Encyclopaedia Britannica in 30 Volumes: Macropae-
dia, Vol. 1, 15th edn. (Encyclopedia Britannica, Chicago) 437-450.
Kenkel, D.S. (1993), "Drinking, driving, and deterrence: The effectiveness and social costs of alternative
policies", Journal of Law and Economics 36:877-913.
Kenkel, D.S. (1996), "New estimates of the optimal tax on alcohol", Economic Inquiry 34:296-319.
Kenkel, D.S. (1998), "Price and tax effects on demand and health outcomes", in: National Institute on Alcohol
Abuse and Alcoholism Tenth Special Report to the US Congress on Alcohol and Health, in press.
Kenkel, D.S., and D. Ribar (1994), "Alcohol consumption and young adults' socioeconomic status", in:
Brookings Papers on Economic Activity-Micro June: 119-161.
Kenkel, D.S., and P. Wang (1998), "Are alcoholics in bad jobs?", Working Paper 6401 (National Bureau of
Economic Research).
1670 PJ. Cook andM.J Moore
Klatsky, A.L., M.A. Armstrong and G.D. Friedman (1990),"Risk of cardiovascular mortality in alcohol
drinkers, ex-drinkers and nondrinkers", American Journal of Cardiology 66(17):1237-1242.
Lands, W.E.M., and S. Zakhari (1991), "Alcohol and cardiovascular disease", Alcohol Health and Research
World 14(4):304-312.
Larkby, C., and N. Day (1997), "The effects of prenatal alcohol exposure", Alcohol Health and Research
World 21(3):192-198.
Lelbach, K. (1974), "Organic pathology related to volume and pattern of alcohol use", in: R.J. Gibbins et al.,
eds., Research Advances in Alcohol and Drug Problems, Vol. 1 (John Wiley and Sons, New York).
Leon, D.A., L. Chenet, V.M. Shkolnikov, S. Zakharov, J. Shapiro, G. Rakhmanova, S. Vassin and M. McKee
(1997), "Huge variation in Russian mortality rates 1984-1994: Artifact, alcohol, or what?", The Lancet
350:383-388.
Leung, S., and C.E. Phelps (1993), "My kingdom for a drink ...? A review of estimates of the price sensitivity
of demand for alcoholic beverages", in: M.E. Hilton and G. Bloss, eds., Economics and the Prevention of
Alcohol-Related Problems. NIAAA Research Monograph No. 25, NIH Publication No. 93-3513 (National
Institutes of Health, Bethesda, MD) 1-31.
Lyon, A.B., and R.M. Schwab (1995), "Consumption taxes in a life-cycle framework: Are sin taxes regres-
sive?", Review of Economics and Statistics 77(3):389-406.
Males, M.A. (1986), "The minimum purchase age for alcohol and young-driver fatal crashes: A long-term
view", Journal of Legal Studies 15(1):181-211.
Malin, H., . Coakley, C. Kaelber, N. Munch and W. Holland (1982), "An epidemiologic perspective on
alcohol use and abuse in the United States", in: Alcohol Consumption and Related Problems (National
Institute of Alcohol Abuse and Alcoholism, Washington, DC).
Manning, W., L. Blumberg and L. Moulton (1995), "The demand for alcohol: The differential response to
price", Journal of Health Economics 14(2):123-148.
Manning, W.G., E.B. Keeler, J.P. Newhouse, E.M. Sloss and J. Wasserman (1989), "The taxes of sin: Do
smokers and drinkers pay their way?", Journal of the American Medical Association 261:1604-1609.
Manning, W.G., E.B. Keeler, J.P. Newhouse, E.M. Sloss and J. Wasserman, eds. (1991), The Costs of Poor
Health Habits (Harvard University Press, Cambridge).
Manning, W.G., and J. Mullahy (1998), "The effects of price on patterns of alcohol consumption, etc.", Speech
presented at the annual meeting of the Allied Social Science Association, Chicago, January 1998.
Manski, C.F. (1995), Identification Problems in the Social Sciences (Harvard University Press, Cambridge).
Markowitz, S., and M. Grossman (2000), "Alcohol regulation and violence towards children", Journal of
Health Economics, in press.
Mello, N.K. (1972), "Behavioral studies of alcoholism", in: B. Kissin and H. Begleiter, eds., The Biology of
Alcoholism, Vol. 2 (Plenum, New York).
Mello, N.K., H.B. McNamee and J.H. Mendelson (1968), "Drinking patterns of chronic alcoholics: Gambling
and motivation for alcohol", in: J.O. Cole, ed., Clinical Research in Alcoholism (American Psychiatric
Association, Washington, DC).
Midanik, L. (1982), "The validity of self-reported alcohol consumption and alcohol problems: A literature
review", British Journal of Addiction 77:357-382.
Miller, T.R., and L.J. Blincoe (1993), "Incidence and costs of alcohol-involved crashes in the United States",
Accident Analysis and Prevention 26:583-592.
Miller, T.R., D.C. Lestina and R.S. Spicer (1998), "Highway crash costs in the United States by driver age,
blood alcohol level, victim age, and restraint use", Accident Analysis and Prevention 30(2):137-150.
Miller-Tutzauer, C., K.E. Leonard and M. Windle (1991), "Marriage and alcohol use: A longitudinal study of
'maturing out' ", Journal of Studies on Alcohol 52(5):434-440.
Miron, J.A. (1997), "The effect of alcohol prohibition on alcohol consumption", Working Paper (Boston
University).
Miron, J.A., and J. Zwiebel (1991), "Alcohol consumption during prohibition", AEA Papers and Proceedings
81(2):242-247.
Moore, M.H., and D.R. Gerstein, eds. (1981), Alcohol and Public Policy: Beyond the Shadow of Prohibition
(National Academy of Sciences, Washington, DC).
Moore, M.J. (1996), "Death and tobacco taxes", RAND Journal of Economics 27(2):415-428.
Moore, M.J., and P.J. Cook (1995), "Habit and heterogeneity in the youthful demand for alcohol", Working
Mullahy, J., and J.L. Sindelar (1989), "Life cycle effects of alcoholism on education, earnings, and occupa-
tion", Inquiry 26:272-282.
Mullahy, J., and J.L. Sindelar (1991), "Gender differences in labor market effects of alcoholism", American
Mullahy, J., and J.L. Sindelar (1993), "Alcoholism, work, and income", Journal of Labor Economics
11(3):494-520.
Mullahy, J., and J.L. Sindelar (1996), "Employment, unemployment, and problem drinking", Journal of
Mullahy, J., and J.L. Sindelar (1997), "Women and work: tipplers and teetotalers", Health Economics
6(5):533-537.
Myrdal, G. (1930), "Alkoholens vinst - och Forlustkonto". Tirfling 24(4):106-109.
National Household Survey on Drug Abuse (1996), Online data from Prevline, Preven-
tion Online web site, from The National Clearinghouse for Alcohol and Drug Information
(www.health.org/pubs/nhsda/96hhs/tabtoc.htm).
Nelson, J.P., and J.R. Moran (1995), "Advertising and US alcoholic beverage demand: System-wide esti-
mates", Applied Economics 27(12): 1225-1236.
NorstrOm, T. (1987), "The abolition of the Swedish alcohol rationing system: Effects on consumption distri-
bution and cirrhosis mortality", British Journal of Addiction 82:633-642.
Norton, E.C., R.C. Lindrooth and S.T. Ennett (1998), "Controlling for the endogeneity of peer substance use
on adolescent alcohol and tobacco use", Health Economics 439-453.
NTC Publications (1996), World Drink Trends: International Beverage Alcohol Consumption and Production
(NTC Publications,Trends Henley-on-Thames, Oxfordshire, England).
Ohsfeldt, R.L., and M.A. Morrisey (1997), "Beer taxes, workers' compensation, and industrial injury", Re-
view of Economics and Statistics 79(1):155-160.
Osterberg, E. (1983), "Calculating the costs of alcohol: The Scandinavian experience", in: M. Grant, M. Plant
and A. Williams, eds., Economics and Alcohol: Consumption and Controls (Gardner Press, New York)
82-96.
Pacula, R.L. (1998), "Does increasing the beer tax reduce marijuana consumption?", Journal of Health Eco-
nomics 17(5):557-586.
Pernanen, K. (1991), Alcohol in Human Violence (Guilford Press, New York).
Phelps, C.E. (1988), "Death and taxes: an opportunity for substitution", Journal of Health Economic 7:1-24.
Pogue, TF., and L.G. Sgontz (1989), "Taxing to control social costs: The case of alcohol", American Eco-
Polich, M.J., and B.R. Orvis (1979), Alcohol Problems: Patterns and Prevalence in the U.S. Air Force (The
Rand Corporation, Santa Monica, CA).
Pollak, R.A. (1970), "Habit formation and dynamic demand functions", Journal of Political Economy 78:745-
763.
Popham, R.E., and W. Schmidt (1981), "Words and deeds: The validity of self-report data on alcohol con-
sumption", Journal of Studies on Alcohol 42(3):355-358.
Pudney, S. (1989), Modeling Individual Choice: The Econometrics of Comers, Kinks, and Holes (Basil Black-
well, Oxford).
Ramsey, EP. (1927), "A contribution to the theory of taxation", Economic Journal 37(1):47-61.
Rice, D.P., S. Kelman, L.S. Miller and S. Dunmeyer (1990), "The economic costs of alcohol and drug abuse
and mental illness: 1985", Report submitted to the Office of Financing and Coverage Policy of the Alcohol,
Drug Abuse, and Mental Health Administration (US Department of Health and Human Services, Institute
for Health and Aging, University of California, San Francisco).
Rimm, E.B., E.L. Giovannucci, W.C. Willett, G.A. Colditz, A. Ascherio, B. Rosner and M.J. Stampfer (1991),
"Prospective study of alcohol consumption and risk of coronary disease in men", The Lancet 338:464-468.
Roberts, J.S. (1984), Drink, Temperance and the Working Class in Nineteenth Century Germany (Allen &
Unwin, Boston).
Room, R., S. Bondy and J. Ferris (1994), "Drinking and risk of alcohol-related harm in a 1989 Canadian
national sample", Working Paper (Addiction Research Foundation, Toronto, Ontario).
Rosen, H.S. (1980), "What is labor supply and do taxes affect it?", American Economic Review Papers and
Proceedings 70(2):171-176.
Rosen, H.S. (1988), Public Finance, 2nd edn. (Irwin, Homewood, IL).
Ruhm, C.J. (1995), "Economic conditions and alcohol problems", Journal of Health Economics 14(5):583-
603.
Ruhm, C.J. (1996), "Alcohol policies and highway vehicle fatalities", Journal of Health Economics 15:435-
454.
Rumbarger, J.J. (1989), Profits, Power, and Prohibition: Alcohol Reform and the Industrializing of America
1800-1930 (State University of New York Press, Albany).
Saffer, H. (1991), "Alcohol advertising bans and alcohol abuse: An international perspective", Journal of
Saffer, H. (1994), "Alcohol tax equalization and social costs", Eastern Economic Journal 20(1) (Winter):33-
43.
Saffer, H. (1995), "Alcohol advertising and alcohol consumption: Econometric studies", in: S.E. Martin, ed.,
The Effects of the Mass Media on the Use and Abuse of Alcohol. NIAAA Research Monograph No.
28, NIH Publication No. 95-3743 (National Institutes of Health, National Institute on Alcohol Abuse and
Alcoholism, Bethesda, MD) 83-99.
Saffer, H. (1997), "Alcohol advertising and motor vehicle fatalities", Review of Economics and Statistics
79(3):431-442.
Saffer, H., and M. Grossman (1987a), "Beer taxes, the legal drinking age, and youth motor vehicle fatalities",
Journal of Legal Studies 16:351-374.
Saffer, H., and M. Grossman (1987b), "Drinking age laws and highway mortality rates: Cause and effect",
Economic Inquiry 25(3):403-417.
Sammartino, F. (1990), Federal Taxation of Tobacco, Alcoholic Beverages, and Motor Fuels (Congressional
Budget Office, Washington, DC).
Shaper, A.G. (1990), "Alcohol and mortality: A review of prospective studies", British Journal of Addiction
85:837-847.
Shaper, A.G., G. Wannametheee and M. Walker (1988), "Alcohol and mortality in British men: Explaining
the U-shaped curve", Lancet 2(8623):1267-1273.
Shavell, S. (1986), "The judgment-proof problem", International Review of Law and Economics 6:45-58.
Sindelar, J.L. (1993), "Measurement issues in alcohol survey data", in: M.E. Hilton and G. Bloss, eds., Eco-
nomics and the Prevention of Alcohol-Related Problems. NIH Publication No. 93-3513 (National Institute
on Alcohol Abuse and Alcoholism, Rockville, MD) 201-228.
Skog, O.-J. (1980), "Social interaction and the distribution of alcohol consumption", Journal of Drug Issues
10(Winter):71-92.
Skog, O.-J. (1985), "The collectivity of drinking cultures: A theory of the distribution of alcohol consump-
tion", British Journal of Addiction 80:83-99.
Skog, O.-J., and Z. Elekes (1993),"Alcohol and the 1950-1990 Hungarian suicide trend - Is there a causal
connection?", Acta Sociologica 36:33-46.
Slemrod, J. (1990), "Optimal taxation and optimal tax systems", Journal of Economic Perspectives 4(1):157-
178.
Sloan, F.A., and P.B. Githens (1994), "Drinking, driving, and the price of automobile insurance", Journal of
Risk and Insurance 61(1):33-58.
Sloan, F.A., B.A. Reilly and C. Schenzler (1995), "Effects of tort liability and insurance on heavy drinking
and drinking and driving", Journal of Law and Economics 38:49-77.
Stampfer, M.J., G.A. Colditz, W.C. Willett, EE. Speizer and C.H. Hennekens (1988), "Prospective study
of moderate alcohol consumption and the risk of coronary disease and stroke in women", New England
Stiglitz, J.E. (1988), Economics of the Public Sector, 2nd edn. (W.W. Norton, New York).
Stinson, ES., J.D. Lane, G.D. Williams and M.C. Dufour (1997), "U.S. apparent consumption of alcoholic
beverages based on state sales, taxation, or receipt data", in: U.S. Alcohol Epidemiologic Data Reference
Manual (National Institute of Alcohol Abuse and Alcoholism, Division of Biometry and Epidemiology,
Alcohol Epidemiologic Data System, Rockville, MD).
Sulkunen, P. (1983), "Alcohol consumption and the transformation of living conditions: A comparative study",
in: R.G. Smart, EB. Glaser, Y. Israel, H. Kalant, R.E. Popham and W. Schmidt, eds., Research Advances
in Alcohol and Drug Problems, Vol. 7 (Plenum, New York) 247-297.
Trice, H.M., and W.J. Sonnenstuhl (1988), "Drinking behavior and risk factors related to the work place:
Implications for research and prevention", The Journal of Applied Behavioral Science 24(4):327-346.
US Department of Health and Human Services (1997), Ninth Special Report to the US Congress on Alcohol
and Health (US Government Printing Office, Washington, DC).
Vallee, B.L. (1998), "Alcohol in the western world", Scientific American 278(6):80-85.
Wales, T.J., and A.D. Woodland (1983),"Estimation of consumer demand systems with binding non-negativity
constraints", Journal of Econometrics 21:437-468.
Walsh, D.C. (1982), "Employee assistance programs", Milbank Memorial Fund Quarterly/Health and Society
60(3):492-517.
Warburton, C. (1932), The Economic Results of Prohibition (Columbia University Press, New York).
Wasserman, D., A. Varnik and G. Eklund (1994),"Male suicides and alcohol consumption in the former
USSR", Acta Psychiatrica Scandinavica 89:306-313.
Williams, G.D., ES. Stinson, J.D. Lane, S.L. Tunson and M.C. Dufour (1996), "Apparent per capita alcohol
consumption: National, state, and regional trends, 1977-1994", Surveillance Report #39 (National Institute
on Alcohol Abuse and Alcoholism, Division of Biometry and Epidemiology, Alcohol Epidemiologic Data
System, Rockville, MD).
Yamada, T., M. Kendix and T. Yamada (1993), "The impact of alcohol consumption and marijuana use on
high school graduation", Working Paper No. 4497 (National Bureau of Economic Research).
Yano, K., G.G. Rhoads and A. Kagan (1977),"Coffee, alcohol and risk of coronary heart disease among
Japanese men living in Hawaii", New England Journal of Medicine 297:405-409.
Zarkin, G.A., et al. (1998), "Alcohol use and wages: New results from the national household survey on drug
abuse", Jounmal of Health Economics 17(1):53-68.
Chapter 31
PREVENTION
DONALD S. KENKEL
Departmentof Policy Analysis and Management, Cornell University, Ithaca, NY
Contents
Abstract 1676
Keywords 1676
1. What does prevention mean? 1677
1.1. Scope of the term "prevention" 1677
1.2. Scope and organization of this chapter 1679
2. What does prevention mean in health economics models? 1680
2.1. Human capital models 1680
2.2. Insurance models 1683
2.3. The supply of prevention 1684
2.4. Summary 1685
3. Will moral hazard lead to too little prevention? 1685
3.1. The moral hazard problem 1686
3.2. Solutions to the moral hazard problem 1687
3.3. The extent of the moral hazard problem 1691
3.4. Moral hazard in public sector insurance systems 1693
4. Will other market failures lead to too little prevention? 1694
4.1. Externalities 1694
4.2. Lack of consumer information 1696
4.3. Research and development of prevention as a public good 1701
5. Is prevention cheaper than cure? 1704
6. What policies can encourage prevention? 1706
6.1. Using taxes and subsidies to encourage prevention 1706
6.2. Improving access to clinical preventive services 1707
6.3. Developing countries 1709
6.4. Providing information to encourage prevention 1710
7. What does health economics offer prevention research? 1712
References 1714
Handbook of HealthEconomics, Volume 1, Edited by A.J. Culyer andJP Newhouse

1676 D.S. Kenkel
Abstract
Prevention ranges from medical decisions such as vaccinations and clinical preventive
services delivered during periodic health examinations to private health lifestyle deci-
sions such as regular exercise and non-smoking. The aim of this chapter is to provide
an overview of economic issues that cut across a variety of prevention decisions. After
discussing what prevention means, the chapter reviews some basic theoretical insights
about prevention from human capital models and insurance models. Consumer or house-
hold behavior receives most of the attention, partly because there is not an identifiable
industry that produces prevention viewed broadly. The chapter next explores market
failures that might lead to too little prevention from a societal perspective: ex ante moral
hazard from health insurance, externalities from vaccinations, lack of consumer infor-
mation, and the public good aspects of prevention-related research and development.
Health economics provides some conceptual and empirical arguments for policies to
encourage prevention. However, the economic perspective often remains quite differ-
ent from the perspective of many public health professionals who are strong advocates
of prevention. With that distinction in mind, the chapter then turns to policy-relevant
questions of whether prevention can reduce total medical expenditures, and the effec-
tiveness of policy interventions to encourage prevention. The chapter concludes with
some reflections on what economics has offered and can offer to prevention research.
Keywords
insurance, moral hazard, externalities, information, cost-effectiveness
JEL classification:Ill, 112, 118

Ch. 31: Prevention 1677
1. What does prevention mean?
1.1. Scope of the term "prevention"
Public health experts argue for a very broad view of what prevention means. A standard
approach identifies three categories of prevention. Primary prevention consists of ac-
tions that reduce the occurrence or incidence of disease. This category includes not only
vaccinations and other medical care but, perhaps more importantly, public sanitation
measures and health lifestyle decisions such as regular exercise and non-smoking. Sec-
ondary prevention consist of actions that reduce or eliminate the health consequences
of a disease given its occurrence. Many clinical preventive services delivered during pe-
riodic health examinations fall into this category. Screening for cardiovascular disease,
cancer, diabetes and other chronic illnesses allows early detection and treatment, pre-
sumably leading to better outcomes. Tertiary prevention consists of actions that reduce
disability associated with a chronic illness. Educating diabetic patients on foot care to
prevent complications is an example. Although originally discussed mainly in terms of
disease prevention, increasingly the concern is with preventing accidents, injuries, and
so on. For example, the U.S. Preventive Services Task Force (1996, p. xxx) suggests the
need for clinicians to re-order their priorities away from traditional tests such as routine
chest radiographs, and towards discussions with patients about prevention topics such
as the use of safety belts.
The broad view of prevention has led many to adopt new terminology emphasizing
"wellness" and "health promotion and disease prevention". In 1977 the U.S. federal
government established the Office of Disease Prevention and Health Promotion. This
was followed by a Surgeon General's Report on the subject and a companion docu-
ment setting out national health promotion and disease prevention objectives in eight
areas: (i) physical activity and fitness; (ii) nutrition; (iii) tobacco; (iv) alcohol and other
drugs; (v) family planning; (vi) mental health and mental disorders; (vii) violent and
abusive behavior; and (viii) educational and community-based programs. Similarly, the
OECD (1994) argues for "a broader approach to health policy, an approach which em-
phasizes the promotion of healthy lifestyles and the active consideration of the health
consequences of government policies across a range of policy sectors".
To a large extent, the prevalent threats to health influence what is meant by preven-
tion. In developed countries, the prevalence of serious communicable diseases is low
and conditions such as heart disease and cancer are the major causes of death. As a re-
sult of this so-called epidemiologic transition, prevention increasingly involves lifestyle
changes that reduce risk factors for these conditions. 1 McGinnis and Foege (1993) re-
view evidence on the relative contributions of various external factors linked to the lead-
ing causes of death in the U.S. They conclude that the three most prominent contributors
1 At present, HIV/AIDS might be considered the exception that proves the general rule. HIV infection is
a serious communicable disease and the 8th leading cause of death in the U.S. However, until and unless a
vaccine is developed, like many of the other leading causes of death prevention of HIV infection requires
lifestyle changes, not medical care.
1678 D.S. Kenkel
to mortality in 1990 were tobacco, diet and activity patterns, and alcohol, which together
accounted for almost 40 percent of all deaths. With an increased aged population, health
problems associated with chronic but not life-threatening conditions will become more
important. The priorities for prevention are likely to evolve accordingly: Fries (1997)
argues for the need for preventive gerontology that emphasizes morbidity compression
as much as life extension.
In contrast, prevention aimed at communicable diseases remains the highest priority
in the developing world. The World Bank (1993) estimates that in Sub-Saharan Africa
communicable diseases account for 71.3 percent of the burden of disease (measured
in disability-adjusted life years (DALYs) lost), compared to 9.7 percent of the burden
of disease in the established market economies. Vaccine-preventable childhood infec-
tions account for 9.6 percent of the loss of DALYs in sub-Saharan Africa, compared to
0.1 percent in established market economies. For countries in sub-Saharan Africa, and
to a lesser extent India and other Asian countries, prevention still means vaccines and
public sanitation measures. Proper nutrition is also an important aspect of prevention in
many developing countries, both to prevent nutritional deficiency diseases and because
malnutrition interacts with infectious disease to worsen health outcomes [World Bank
(1993), Strauss and Thomas (1998)]. Fogel (1994) reviews historical and contemporary
evidence on the importance of nutrition as prevention more broadly. The World Bank
(1993) cautions that because of the demographic and epidemiologic transitions in de-
veloping countries the burden from noncommunicable diseases will increase sharply by
2030. In the future, developing countries will face many of the same challenges for pre-
vention as in the established market economies, in addition to the continuing challenge
of controlling the communicable diseases of the young.
The priorities for prevention in the formerly socialist economies of Europe are some-
what different than in either the developed or the developing world. By the World Bank's
(1993) estimates the distribution of the burden of disease in this group of countries
in 1990 was similar to that in established market economies: communicable diseases
accounted for only 8.6 percent of DALYs lost, while noncommunicable diseases ac-
counted for 74.8 percent. For example, the Soviet health system was successful in con-
trolling infectious diseases, so the country appeared to experience an epidemiologic
transition towards noncommunicable diseases similar to the experience of the estab-
lished market economies. However, male life expectancy in the Russian Federation in
1990 was only 63.8 years, at or below the level of the 1960s, and declined further to
59 years from 1990 to 1993 [Tulchinsky and Varavikova (1996)]. Tulchinsky and Var-
avikova (1996, p. 319) argue the Soviet system failed to respond effectively to the epi-
demiologic transition and continued to emphasize routine medical checkups and a clin-
ical approach rather than adopting policies to change risk factors including tobacco use,
alcohol abuse, and high-fat diets. The upheavals that are accompanying the adoption of
more market-based economies also make it difficult to predict future trends in many of
these countries. After the collapse of the Soviet Union the Russian Federation has wit-
nessed outbreaks of communicable diseases once thought to have been under control,
including diphtheria and cholera [Tulchinsky and Varavikova (1996)]. Environmental
problems are also severe: estimates suggest that 40 percent of the urban population lives
under extremely high air pollution levels, and 50 percent of the Russian population uses
drinking water below current microbiologic and chemical standards [Tulchinsky and
Varavikova (1996)]. Prevention efforts in the Russian Federation and other formerly
socialist economies must address a very broad range of significant threats to health.
1.2. Scope and organizationof this chapter
Other chapters of this Handbook contain detailed analysis of specific prevention de-
cisions, such as those related to tobacco [Chaloupka and Warner (2000)] and alcohol
[Cook and Moore (2000)]. The aim of this chapter is to provide an overview of eco-
nomic issues that cut across a variety of prevention decisions. Because prevention in
developed countries is often so different than prevention in developing countries and in
the formerly socialist economies, it is important to note that a good deal of this chapter's
discussion is in the context of developed countries in general and (to a lesser extent)
the U.S. in particular. This chapter, however, contains some discussion of prevention
in developing countries. Because prevention covers such a broad range of activities, it
should also be noted that this chapter can not do justice to all economics research on pre-
vention broadly conceived. For example, environmental economics research on air and
water pollution and toxic wastes addresses issues in prevention, as does transportation
economics research on air and traffic safety. For the most part reviews of these related
research literatures are left to experts in these other specialty fields of economics. 2 Fi-
nally, researchers from many other behavioral, social science, and medical disciplines
work on prevention. While this chapter makes some references to their contributions,
by focusing on health economics research the discussion undoubtedly neglects many
relevant findings from other disciplines. 3
The next section of this chapter discusses basic insights about prevention from the-
oretical models in the health economics literature. The following two sections explore
whether moral hazard and other market failures lead to too little prevention. The next
sections address the policy-relevant questions of whether prevention is cheaper than
cure and about the effectiveness of policy interventions to encourage prevention. The
last section concludes with some reflections on what economics has offered and can
offer to prevention research.
2 See, for example, Johansson (1987) and Cropper and Oates (1992) on measuring health benefits from
environmental policies and Blomquist (1988) on the economics of traffic safety.
3 Put differently, this chapter focuses on a body of research by economists about prevention, which is not
the same as the body of prevention research economists need to know.
1680 D.S. Kenkel
2. What does prevention mean in health economics models?
2.1. Human capital models
A number of models that analyze individual health decisions can be thought of as mod-
els of consumer demand for prevention. One line of research emphasizes the similarities
between decisions to invest in health and decisions to invest in other forms of human
capital [Mushkin (1962), Grossman (1972)]. In Grossman's (1972) seminal work, the
individual's health capital stock determines a flow of utility directly and also determines
the amount of time available for market and nonmarket production. Gross investment
in health capital is produced by combining time, purchased medical care, and other
purchased goods according to a household production process. The demand for health
inputs, including medical care, is thus derived from the demand for health capital.
This economic model of a stock of health capital parallels the idea of "wellness" in
the public health field. Investment decisions to add to the health capital stock could be
described as prevention, because a higher health stock leads to less time spent ill. On
the other hand, because gross investment offsets health losses due to depreciation of
the health capital stock, these investments could also be described as curative care. The
problem is that Grossman's basic model does not have any uncertainty so it does not re-
ally allow for a useful distinction to be made between preventive and curative care. In an
extension, Grossman and Rand (1974) distinguish prevention and cure by assuming that
groups with low depreciation rates primarily demand preventive care while groups with
high depreciation rates primarily demand curative care. When preventive and curative
care are treated as separate inputs into the household health production function, they
assume that the relative productivity of curative care increases with the depreciation
rate. Although Grossman and Rand acknowledge that this treatment is not definitive,
they emphasize that it captures an essential tradeoff where consumers view prevention
and cure as substitutes.
The distinction between prevention and cure is not very common in subsequent the-
oretical studies that build on the Grossman model. Some, including Muurinen (1982)
and Ehrlich and Chuma (1990), also do not involve uncertainty. Others that involve
uncertainty simply do not emphasize the distinction between prevention and cure [Dar-
danoni and Wagstaff (1987, 1990), Selden (1993), Liljas (1998), Picone, Uribe and
Wilson (1998)]. Cropper (1977) is an important exception that explicitly introduces un-
certainty into the Grossman model and derives comparative statics for preventive care
demand.
Whether or not prevention is distinguished from cure, the health capital framework
provides useful insights into the roles schooling, time preference, initial health stock
and age play in preventive health demand. A complete discussion of theoretical and
empirical results about the comparative statics of health capital demand are beyond the
scope of this chapter, but some important results can be highlighted.4 One of the first is-
4 Grossman (2000), The Human Capital Model.

sues that arises is the possible interaction between health and schooling human capital.
Grossman (1972) hypothesizes that schooling increases the efficiency of the household
production of health. Viewing prevention choices as inputs into the household produc-
tion of health, Grossman's model yields ambiguous predictions about the relationship
between schooling and prevention. Schooling reduces the shadow price of health capital
which increases the demand for health capital, but the derived demand for health inputs
such as prevention only increases if the price elasticity of the demand for health capital
exceeds one.
Kaestner and Grossman (1997) provide an in-depth review of research on the rela-
tionship between schooling and health. Some examples illustrate the variety of contexts
in which a positive relationship between schooling and prevention appears empirically.
Rosenzweig and Schultz (1983) find that pregnant women in the U.S. with more school-
ing invest in more prenatal prevention, such as earlier prenatal care and less smoking.
Rosenzweig and Schultz (1988) confirm their earlier findings but note that the relation-
ship between maternal schooling and refraining from smoking while pregnant appears
to have grown stronger over time. The Cebu Study Team (1991) explores the relation-
ship between maternal schooling and the incidence of childhood diarrhea in the city of
Cebu, Philippines. They estimate that the net effect of a one-year increase in maternal
education is a reduction in the incidence of childhood diarrhea episodes by about 5 per-
cent, due to improved excreta disposal practices and an increase in calories given to the
child. These effects of maternal schooling are partly offset by a substantial increase in
diarrhea because maternal schooling is also estimated to reduce the number of mothers
who breast-feed.
Schooling is also related to prevention for adult health and safety. For example, Leigh
(1990) finds that people with more schooling are more likely to use seatbelts. Kenkel
(1991a, 199 b) finds that more schooling is associated with healthier lifestyle choices
regarding smoking, drinking, and exercise. Kenkel (1994) finds that schooling is an
important determinant of women's demand for preventive medical services designed
for the early detection of breast and cervical cancer. Mullahy (1999) finds that more
schooling is associated with a higher propensity to be immunized against the flu. The
relationship is estimated to be particularly strong for people over 65, where each year of
schooling is associated with a 1.5 percentage point increase in the probability of being
immunized. All of the empirical studies listed control for factors such as family income
and age, but differ in their ability to control for other differences that might explain the
schooling-prevention link.
One role for schooling in health production is to improve allocative efficiency, that
is, the choice of health inputs, by improving consumers' health knowledge. Kenkel's
(1991a, 1991b) empirical study estimates the separate effects of health knowledge and
schooling on the consumption of cigarettes, alcohol, and exercise. Both health knowl-
edge and schooling are found to decrease smoking and heavy drinking and to increase
exercise. While part of the relationship between schooling and the health behaviors is
explained by differences in health knowledge, most of schooling's effects remain after
differences in knowledge are controlled for. The results suggest that schooling improves
1682 D.S. Kenkel
allocative efficiency of health production, but that these effects are not the main reason
schooling is linked to prevention. Instead, the estimated relationships between schooling
and the health behaviors might be due to unobservable differences across individuals,
as Fuchs (1982) and Farrell and Fuchs (1982) suggest.
In many conceptual analyses the individual rate of time preference plays an important
role in prevention decisions. Fuchs (1982) and Farrell and Fuchs (1982) suggest it as a
candidate for the "hidden third variable" behind the link between schooling and health,
if people with low rates of time preference are more likely to invest in both schooling
and prevention. Although their argument is not made in the context of a formal model,
it is consistent with Ehrlich and Chuma's (1990) theoretical analysis that a lower rate
of time preference increases the demand for longevity and optimal health investment.
Individual time preference plays a related role in Becker and Murphy's (1988) model of
rational addiction. In this model the rational person recognizes the future consequences
of current consumption of beneficial and harmful addictive goods. The link between ac-
cumulating a stock of addictive capital and future consequences, possibly health, means
this model is potentially quite useful for analyzing many preventive health decisions.
Becker and Murphy show that an increase in the rate of preference for the present raises
the demand for harmful goods but lowers the demand for beneficial goods, so "drug ad-
dicts and alcoholics tend to be present-oriented, while religious individuals and joggers
tend to be future-oriented" (p. 685).
In the models discussed so far, time preference is a determinant of health. In Becker
and Mulligan's (1997, p. 740) model of endogenous time preference, the causality is in
the reverse direction: "differences in health cause differences in time preference because
greater health reduces mortality and raises future utility levels". It is difficult to explore
these different explanations empirically because of the difficulties in measuring time
preference. As described by Barsky et al. (1997) and Hurd and McGarry (1995) the
Health and Retirement Survey provides measures of time preference and subjective
survival probability estimates that may provide leverage for future empirical work on
this subject.
In another interesting comparative static exercise, Ehrlich and Chuma (1990) show
that the initial stock of health determines consumer health decisions. A higher endow-
ment of health initially increases demand for health investment, so differences in en-
dowed health are magnified in terms of attained longevity. One of their pieces of sup-
portive empirical evidence is that females appear to be more fit from birth, and that from
1901-1982 the age-specific rates of growth of females' life expectancies outpaced those
of males at every age category. The implication is that women invest in more prevention
because of their higher initial health stock. Whether this is the correct causal interpre-
tation or not, women tend to make healthier choices than men in many but not all areas
of prevention [see, for example, Sobal, Revicki and DeForge (1992)].
The last comparative static results from the health capital approach to be highlighted
concern the role of the consumer's age in preventive health decisions. Comparing the
models of Grossman (1972) and Cropper (1977), age may play distinctly different roles
in the demand for curative and preventive care. In Grossman's model, because health
capital is assumed to depreciate at a higher rate as people get older, if the price elasticity
of the demand for health is smaller than one the derived demand for curative medical
care increases with age. For preventive care Cropper shows that there is an effect work-
ing in the opposite direction. When the length of life is exogenous, because the pay-off
period for any human capital investment is shorter for older individuals, use of preven-
tive care tends to decline with age. When Cropper makes the length of life endogenous
and assumes the depreciation rate rises with age, she obtains the same results as in
Grossman's model. Empirically, Kenkel (1994) finds that annual use of preventive med-
ical services designed for the early detection of breast and cervical cancer decreases with
age. Although not the only plausible explanation, the results are consistent with women
rationally reducing their use of preventive care as the payoff period to the investment
shortens over the lifecycle. In general, the age patterns of the demand for prevention
depend on the specific intervention because the risks of different illnesses show differ-
ent lifecycle patterns. An extra consideration is that for some preventive interventions
(e.g., exercise) the health benefits are realized much more quickly by older people, and
so will be not as heavily discounted as when young people consider the intervention.
2.2. Insurance models
In addition to models that take the human capital approach, models of insurance and be-
havior under uncertainty also shed light on prevention decisions. A seminal paper along
these lines is Ehrlich and Becker (1972), who use an expected utility model where there
are three ways an individual can respond to uncertainty: purchase a market insurance
policy that provides income if a bad state of nature occurs; engage in self-protection
activities that reduce the probability of a bad state of nature occurring; or engage in
self-insurance activities that reduce the size of a loss if a bad state of nature occurs.
Applied to health, activities Ehrlich and Becker call self-protection fall into the cate-
gory of primary prevention, while self-insurance corresponds to secondary prevention.
For example, because obesity is associated with higher risks of breast cancer, maintain-
ing proper weight is a form of self-protection or primary prevention that reduces the
probability of this disease (bad state of nature) occurring.
As an example of self-insurance or secondary prevention, regular mammograms do
not reduce the probability of breast cancer, but instead reduce the health loss from breast
cancer. This follows from what Eddy (1980) terms the fundamental assumption of ill-
ness screening: prognosis is a function of the earliness of detection. To illustrate the
magnitude of the differences in health outcomes possible with secondary prevention, in
randomized controlled trials that included women 50 and older, mammography screen-
ing for breast cancer is estimated to reduce breast cancer mortality by 20 to 30 percent
[U.S. Preventive Services Task Force (1996, pp. 76-77)]. However, secondary preven-
tion does not always work: routine screening for lung cancer is not recommended partly
because by the time lung cancer is suspected on chest X-ray often it has already spread
(micrometastatic dissemination). Weak evidence for the efficacy of screening, combined
with the substantial costs of screening, leads the Preventive Services Task Force and
1684 D.S. Kenkel
other organizations to emphasize primary prevention of lung cancer rather than screen-
ing [U.S. Preventive Services Task Force (1996, pp. 135-139)].
Insurance models provide additional insights into the interactions between insurance
coverage for curative care and the demand for prevention. Phelps (1978) develops an
expected utility model that explicitly incorporates both preventive and curative med-
ical care. Preventive care changes the probability of illness (self-protection) and may
influence the marginal productivity of curative medical care (similar to self-insurance).
Phelps (1978) and Nordquist and Wu (1976) shows that an increase in the price of cu-
rative medical care increases the demand for prevention, so consumers view prevention
and cure as substitutes. Related to this result, Section 3 below discusses in more detail
the moral hazard problem where health insurance that lowers the out-of-pocket price of
curative care decreases consumer incentives to purchase preventive care. Perhaps not
surprisingly, Phelps (1978) also shows that an equiproportional change in the price of
both preventive and curative care, for example through a common health insurance plan,
has ambiguous effects on the demand for preventive care.
Kenkel's (1994) empirical results suggest that coverage for curative care may actually
encourage the use of two preventive medical services designed for the early detection
of breast and cervical cancer. Because early detection is valuable only if it can be fol-
lowed with early curative care, individuals with insurance for curative care may find
early detection more attractive. At the other extreme, if curative care is prohibitively
expensive for an uninsured individual, early detection, or secondary preventive medical
care, is pointless. This result that prevention and cure may be complements instead of
substitutes is specific to the secondary preventive services studied and will not generally
carry over to primary prevention.
2.3. The supply of prevention
The field of health economics has not developed explicit models of the supply of pre-
vention. In part this is because there is not an identifiable industry that produces pre-
vention, viewed broadly; instead there are industries that produce tobacco products, nu-
tritious and less-than-nutritious foods, and so on. Put differently, if these market goods
are seen as inputs into a household production function, the household itself is both the
supplier and demander of prevention. Viewed this way human capital models that incor-
porate household production [e.g., Grossman (1972), Rosenzweig and Schultz (1983)]
are models of the supply of prevention.
Separate analysis of the supply of preventive medical care could be more fruitful, de-
pending upon the extent to which this supply differs in important ways from the supply
of physician services more generally. In markets for curative care, a question that has
attracted a great deal of attention is whether the physician can exploit her informational
advantage and induce demand, resulting in overuse. In the context of preventive care an
equally interesting question is: Will the physician undersupply preventive care, that is,
will she provide less care than the consumer would demand if he had the same informa-
tion set as the physician? Intuitively, possible profits from providing curative care could
create an incentive for the physician to withhold prevention. A complete understanding

of the supply of preventive medical care would require analysis of the structure of the
physician services markets and the health insurance market, recognizing the multiple
agency relationships between the physician, consumer, and third party payer.
2.4. Summary
To sum up, theoretical models provide insights and frameworks for thinking about the
economics of prevention. Consumer or household behavior has received most of the
attention. These lines of research have led to some empirical studies with tight links
between the theoretical model and empirical specification. For example, Rosenzweig
and Schultz (1983) estimate structural health input demand functions and household
health production functions, where the output is infant health and many of the inputs are
preventive decisions. Other notable studies where there are tight links between theory
and empirics are the studies of rational addictions to tobacco and alcohol [Chaloupka
(1991), Becker, Murphy and Grossman (1994), and Grossman, Chaloupka and Sirta-
lan (1998)]. For many other studies however, theoretical models only provide general
guidance for the specification of empirical models of prevention decisions, for exam-
ple, in terms of explanatory variables to be included in a demand model. Economists
with different methodological perspectives may disagree on whether this is an important
weakness. But the unique contribution of the models of prevention could be questioned,
in that ad hoc empirical modeling approach could very well lead to same empirical
specifications.
3. Will moral hazard lead to too little prevention?

According to an estimate by the Centers for Disease Control and Prevention [CDC
(1992)] total spending on prevention in the U.S. in 1988 was $32.8 billion, or 0.7 percent
of the GNP. 5 That level of spending reflects the fact that prevention is not being used
at levels recommended by major professional organizations such as the U.S. Preventive
Services Task Force and the American Cancer Society. Russell (1994, p. 81) estimates
that if these recommendations were followed, screening for cervical cancer, prostate
cancer and cholesterol alone would cost as much as $80 billion. By way of comparison
in 1988 national health expenditures totaled $539.9 billion, or 11 percent of the GNP.
Public health professionals tend to approach the question of whether this represents
an appropriate balance between prevention and cure differently than do economists.
5 This estimate is based on Brown et al. (1991). This study uses a broad concept of prevention and includes
expenditures on a fairly wide range of prevention activities in the following major funding sectors: federal,
state, and local governments, voluntary organizations, worksite health promotion programs, and personal
prevention services. However, important categories of prevention are excluded, such as personal expenditures
on exercise and vitamins and health foods (p. 5). In addition, the report does not attempt to account for the
value of time used to produce prevention, or for the value of consumption or activities foregone to improve
health.
1686 D.S. Kenkel
With the objective of improving health, public health professionals commonly empha-
size the importance of prevention and encourage greater use. There is increasing recog-
nition of resource constraints, and the staff of the centers for Disease Control and Pre-
vention (CDC) recently published a book on the use of cost-effectiveness analysis to
evaluate prevention [Haddix et al. (1996)1. However, the economist's notion of an opti-
mal level of prevention where the marginal benefits equals the marginal costs remains
somewhat foreign and even controversial. This perspective should be kept in mind when
considering the influence, or lack thereof, of the economic approach.
From an economic efficiency standpoint, whether there is enough prevention is an-
swered by considering whether individual decisions and private markets lead to socially
optimal outcomes; or conversely by considering the importance of relevant market fail-
ures, including insurance moral hazard, other externalities, and information problems.
When normative economics considers issues of access and equity, it comes closer to the
public health perspective. This section focuses on the moral hazard problem. The next
section considers the impact of other market failures on the level of prevention.
3.1. The moral hazard problem

The possibility that health insurance for curative care reduces incentives for prevention
is a commonly used example of one type of insurance moral hazard [Pauly (1986)].
A useful terminology for this type is ex ante moral hazard, because it concerns the
effect of insurance on actions the individual takes before his or her state of health
is known [Zweifel and Breyer (1997)]. Or in the terminology of Ehrlich and Becker
(1972), ex ante moral hazard occurs if the purchase of market insurance decreases the
demand for self-protection. The ex ante moral hazard problem considered here is dis-
tinct from ex post moral hazard that occurs if the associated financial loss or cost of
treatment is a function of insurance. The problem of ex post moral hazard leading in-
sured consumers to purchase too much curative medical care has received a great deal
of attention in health economics. Relatively less attention has been directed towards the
problem of ex ante moral hazard leading to too little prevention.
Whether ex ante moral hazard occurs can be shown to depend crucially upon whether
the price of market insurance reflects self-protection or prevention activities [Ehrlich
and Becker (1972), Breyer and Zweifel (1997)]. If market insurance premiums are actu-
arially fair and reflect self-protection activities, the individual still has the correct incen-
tives to spend on self-protection because it lowers the price of market insurance. Ehrlich
and Becker show that in this case the optimal amount of self-protection that maximizes
expected income is purchased. In addition, the optimal amount of self-protection with
market insurance is not necessarily less than the optimal amount spent if market insur-
ance is not available. In contrast, if the price of market insurance does not reflect the
individual's spending on self-protection, the only effect of market insurance is to re-
duce the difference in income when healthy compared to income when sick. In this case
the availability of market insurance causes spending on self-protection to fall, creating
ex ante moral hazard. Whether the price of market insurance reflects self-protection in
turn hinges upon whether insurance companies can observe the level of self-protection.
As is true more generally, the moral hazard problem therefore ultimately stems from an
informational asymmetry, where the insurer can not observe some of the actions of the
insured.
The ex ante moral hazard associated with prevention is a type of externality, because
the insured consumer ignores the effect of his or her self-protection activities on the pre-
miums paid by other members of the insurance pool [Gravelle (1986)]. Manning et al.
(1991) attempt to quantify the extent to which personal decisions about smoking, heavy
drinking, and sedentary lifestyles create financial costs for others because of collec-
tively financed programs such as health insurance plans and retirement pensions. Since
typically the premiums or taxes used to finance these programs do not vary according
to a person's health habits, the individual does not see the full social cost of his or her
health choices. For example, Manning et al. (1991, Table 6-11) estimate that the seden-
tary person imposes $1,650 in discounted lifetime external costs (1986 dollars). Most
of the external costs stem are attributable to the increased use of most forms of medi-
cal care. However, because of the estimated reduction in life expectancy, Manning and
his colleagues estimate that sedentary people spend less over their lifetimes on nursing
home care. The external benefits due to this effect (and the fact that sedentary people
are expected to collect less from public and private pensions) are much smaller than
the categories of external costs of sedentary lifestyles. Nevertheless, it should be recog-
nized more generally that a consumer's prevention or self-protection activities may yield
either positive or negative externalities to other members of the insurance pool. Most
analyses implicitly assume negative externalities so that the moral hazard problem is
too little prevention.
3.2. Solutions to the moral hazardproblem

Shavell (1979) contains a general theoretical analysis of incomplete coverage against
loss as a partial solution to the moral hazard problem. When prevention is very costly,
full insurance is optimal despite moral hazard because it is not socially optimal to in-
vest in any prevention. But when prevention is less costly, partial coverage becomes a
desirable way to reduce moral hazard: the insured consumer has some (but not com-
plete) incentives to invest in prevention. In the context of health insurance, the ex ante
moral hazard problem may be small because common forms of health insurance in
fact offer very incomplete coverage. Even if the consumer has generous coverage for
the monetary components of the loss (medical expenditures and foregone earnings),
he will be uninsured for the utility losses. In many cases the uninsurable utility losses
from health risks far exceed the insurable monetary losses. For example, because of
the pain and suffering associated with illness estimates of willingness to pay for mor-
bidity and mortality improvements typically exceed by substantial margins estimates of
the costs of relevant medical expenditures and foregone earnings [Tolley, Kenkel and
Fabian (1994)]. This large uninsurable component suggests the moral hazard problem,
where health insurance distorts incentives to invest in prevention, is probably small. Of
course, some consumers will be at the margin where having insurance matters to their
prevention decisions.
1688 D.S. Kenkel
Recognizing that common forms of private and public health insurance provide in-
complete coverage for the losses due to ill health raises the question of why this is so.
One explanation is that health is an example of what Cook and Graham (1977) call
an "irreplaceable commodity". Cook and Graham extend Ehrlich and Becker's (1972)
model to consider state-dependent preferences, so that a dollar received contingent upon
good health is viewed as a different commodity than a dollar received contingent upon
poor health. That is, good health can not be replaced. When actuarially fair insurance
is available, even without moral hazard a risk-averse consumer will not fully insure a
normal irreplaceable commodity. Blind (1996) analyzes the demand for self-protection
and insurance of an irreplaceable commodity, although not in the context of health.
Intuitively, health is an irreplaceable commodity because of the incompleteness of the
technology of cure. Despite insurance for curative care, prevention remains attractive
because the choice is between completely preventing the illness or incompletely curing
it. But suppose, for example, that improvements in coronary artery bypass graft surgery
make the operation a nearly perfect cure for heart disease. Prevention and cure would
then be nearly perfect substitutes, and the ex ante moral hazard problem from insuring
curative care would be much larger.
Turning to other limits on the ex ante moral hazard problem, a market solution is to
offer risk-rated health insurance, where for example smokers pay higher premiums than
non-smokers. Assuming actuarially fair insurance, the premium surcharge for smoking
will equal the expected medical expenditures due to smoking. A smoker no longer im-
poses external costs on others in his or her insurance pool, although of course there may
be other externalities (e.g., second-hand smoke). The moral hazard problem is solved
because a consumer with risk-rated health insurance faces the same full-price of smok-
ing as someone without insurance. To date, however, in the context of smoking and
other health behaviors this solution to moral hazard appears to be relatively rare. In a
1992 survey of a nationally representative sample of worksites with more than 50 em-
ployees, only 10 percent of firms offering group health insurance to their employees
indicated that plans were available where the premiums varied according to individual
health behavior [Public Health Service (1992)].
An alternative market solution to the ex ante moral hazard problem is for the insurer to
invest directly in prevention. The financial benefits of prevention that are external to any
one consumer in the pool are internal to the insurer. Schlesinger and Venezian (1986)
analyze the joint production of insurance protection and loss prevention under a variety
of assumptions. When insurance protection and loss prevention can be fully bundled,
Schlesinger and Venezian show that under either monopolistic or competitive conditions
insurers will invest in loss prevention as long as the marginal cost of prevention does not
exceed the reduction in the actuarial value of the insured loss. The monopolistic insurer
captures the benefits of prevention as rents, while under competition the benefits of
prevention are passed along to the consumer in the form of reduced insurance premiums.
Schlesinger and Venezian (1986) also point out that the correct interpretation of in-
surance loading fees depends on whether insurers invest in loss prevention. When the
loading fee is high, only a small proportion of premiums collected are paid back to
consumers as claims. While a high loading fee is often viewed as evidence of high ad-
ministrative costs and inefficiency, it could also reflect a high level of socially efficient
loss prevention activities. In an insurance line where loss prevention activity is preva-
lent, boiler and machinery insurance, only 40 percent of premiums are paid back as
claims. Compared to such insurance lines, loading fees for health insurance are rela-
tively low, especially for large groups [Phelps (1992, p. 297)]. This suggests health in-
surers are not investing much in loss prevention. This is consistent with survey evidence
on whether health plans or physicians in one market area provided prevention programs
such as smoking control, stress management, weight control, cholesterol screening and
blood pressure screening [Schauffler and Rodriguez (1994)]. The most commonly of-
fered programs - cholesterol and blood pressure screening - were only offered to about
30 percent of the insured survey respondents; less than 20 percent report being offered
each of the other programs.
Dowd (1982) discusses a number of barriers that may limit insurers' investments in
loss prevention. His discussion is in the context of Health Maintenance Organizations
(HMOs), but the insights apply equally to other health insurance arrangements. Dowd
considers a primary prevention program, such as an education program about healthy
lifestyles, that both reduces health care costs and improves health outcomes. The insurer
receives financial benefits from the program while the insured consumer enjoys better
health. As a result, the prevention program is a public good shared by the insurer and
the insured consumer, so neither party has the correct incentives to invest in the optimal
amount of prevention.
When considering investments in prevention, the insurer also faces enrollment un-
certainty, because the consumer may withdraw from the plan before the full benefits
of prevention are realized [Dowd (1982)]. For example, health lifestyle education may
take decades to pay off in the form of lower rates of cancer and heart disease in later
life; while on average consumers stay enrolled in a health insurance plan for only a few
years. As Dowd points out, if the insured consumer could be certified as a participant in
the prevention program in a way that would be recognized by all insurers, the consumer
could capture both the financial benefits as well as the health care benefits. An analogy
is with automobile insurers offering lower premiums upon the completion of a drivers'
education course. This leads to risk-rated health insurance where the premiums reflect
the consumer's investments in self-protection so the insurer does not need to invest in
loss prevention.
Assuming a consumer's participation in the prevention program can not be credibly
certified to other insurers, Dowd suggests that a solution is for the consumer and the
insurer to share the enrollment risk. In one scenario discussed, the consumer makes the
initial investment in the prevention program. The insurer then returns the investment
to the consumer in the form of reduced premiums as the cost savings from prevention
accrue. In effect, the insurer uses its private information on who has participated in the
prevention program to offer risk-rated health insurance premiums to its own enrollees
only. Although the insurer does not really invest in loss prevention, the insurer offers
the program (at a charge) to its enrollees in order to gain information about program
1690 D.S. Kenkel
participation. In Dowd's second scenario, the insurer invests directly in the preven-
tion program but enters into a contract with its enrollees that requires reimbursement
to the insurer if the consumer disenrolls from the plan before enough benefits from
the prevention program have been realized. Under either scenario, a consumer who
leaves the insurance plan too early does not receive the full stream of benefits from the
prevention program. As a consequence, enrollment risk still leads to too little preven-
tion.
In addition to commercial insurers and HMOs investing in loss prevention, in the
U.S. many employers offer worksite health promotion programs. Many larger firms
self-insure, that is, the employers assume the risk for paying health care claims. Self-
insured firms, and firms that purchase experience-rated health insurance from commer-
cial insurers, have incentives to invest in prevention that are similar to the incentives of
insurance companies. They also face similar barriers. However, the additional benefits
of improving employee productivity and reducing absenteeism now become relevant.
These benefits will often be shared by the employee and the employer, if compensa-
tion does not perfectly reflect individual productivity or if the employer offers paid sick
leave. These additional benefits may help explain why firms seem to be more involved
in prevention than insurers. By 1992, 81 percent of worksites of 50 or more employ-
ees offered at least one health promotion activity, an increase from 66 percent in 1985
[Public Health Service (1992)]. The most common are smoking cessation activities, but
other examples include worksite programs on health risk assessment, blood pressure
control, weight control, and exercise. Over a quarter of the firms surveyed identified
reducing health insurance costs as one of the top two or three reasons for initiating their
worksite health promotion programs.
Empirical evidence on the determinants of firms' health promotion activities suggest
firms invest in the programs partly as loss prevention and partly for other reasons. In
a study of a variety of health promotion programs, Kenkel and Supina (1992) find that
firms are more likely to invest in the programs when more of their employees are el-
igible for employment-related group health insurance. This study also finds that high
employee turnover is associated with less worksite health promotion, consistent with
Dowd's (1982) argument that enrollment uncertainty is an important barrier to third-
party investment in prevention. Zarkin and Garfinkel (1994) and Kenkel (1997) find
some of these same patterns for firms' offerings of employee assistance programs for
the prevention and treatment of substance abuse and mental health problems. In addi-
tion, these two studies find that self-insured firms are more likely to invest in worksite
alcohol programs than are firms that purchase market insurance. This makes sense if
self-insured firms more fully internalize the benefits of loss prevention than can market-
insured firms. However, other patterns are harder to explain viewing worksite programs
purely as loss prevention. For example, Kenkel (1997) finds that unionized worksites are
more likely to offer alcohol programs. In addition to the investment aspects, this type of
finding suggests firms may also use worksite health promotion programs as attractive
fringe benefits to offer workers.
3.3. The extent of the moral hazardproblem
Although a commonly used example of ex ante moral hazard in theoretical discussions,

there appears to be relatively little direct empirical evidence that health insurance leads
to less prevention. To provide suggestive evidence on the extent of ex ante moral hazard,
Table 1 presents results from logit models showing the relationships between health
insurance and the probability of being sedentary, obese, smoking, heavy drinking (more
than 4 drinks on days with any drinking), drunk driving (in the past year), failing to
wear seatbelts regularly, and failing to have at least one smoke detector. The data are
Table I
Heath insurance status and prevention choices
The effect of health insurance Males Females

on the odds of:
Sedentary lifestyle 0.787

[0.711,0.871]
Obesity 1.211 0.941
[1.038,1.412] [0.837,1.058]
Smoking 0.626 0.729
[0.565,0.693] [0.661,0.804]
Heavy drinking 0.946 0.615
[0.818,1.094] [0.486,0.779]
Drunk driving 1.009 0.827
[0.872,1.167] [0.667,1.015]
No seatbelt use 0.800 0.702
[0.715,0.895] [0.623,0.792]
No smoke detector 0.566 0.566
[0.503,0.636] [0.505,0.635]
No Pap examination - 0.611
[0.554,0.674]
No breast examination by - 0.584
physician [0.529,0.644]
Do not know how to self- - 0.769
examine breasts [0.668,0.885]
No mammogram - 0.594
[0.532,0.662]
Odds ratios from a set of logit models are reported; 95% confidence inter-
vals are in brackets. Each model also included measures of income, educa-
tion, age, race, marital status, general health status, chronic conditions, and
worker type as additional explanatory variables.
1692 D.S. Kenkel
from the 1990 U.S. National Health Interview Survey. 6 For females the data provide
additional measures of the lack of preventive medical care - failing to have a Pap smear
or a breast examination within the past year, not knowing how to perform breast self-
examinations, and failing to have had a mammogram (for women over 35). To allow
for gender differences in health behaviors all models are estimated separately for men
and women, yielding sample sizes of about 13,000 men and 22,000 women. The models
include measures of income, education, age, race, marital status, general health status,
chronic conditions, and worker type as additional explanatory variables.
Odds ratios significantly less than one indicate that men and women with health insur-
ance are less likely to be sedentary, smoke, fail to use seatbelts, and fail to own a smoke
detector. For women, health insurance significantly reduces the odds of failing to have
the various types of preventive medical care. That is, most of the results in Table I sug-
gest that health insurance leads to healthier choices, and certainly provide very little
evidence for a moral hazard effect where insurance leads to less prevention. The only
exception is that men with health insurance appear to be more likely to be obese, but
this does not hold for women. These results should be viewed as descriptive, rather than
estimates of the causal (moral hazard) effect of health insurance on prevention choices.
If there is unobservable consumer heterogeneity in factors that jointly determine health
insurance and health behavior choices, the estimated relationships are biased estimates
of the causal effects.
One candidate for the source of unobservable heterogeneity is risk aversion, if more
risk averse consumers are more likely both to buy insurance and to invest in prevention.
This is similar to the argument and evidence that Hemenway (1990) presents for "propi-
tious selection" where risk avoiders tend to take physical precautions and seek financial
security, including health insurance. Barsky et al. (1997) use measures of risk tolerance
from the Health and Retirement Survey to explore this issue more directly. The risk tol-
erance measures are based on answers to hypothetical questions specifically designed
to elicit information about respondents' risk aversion. Barsky and his colleagues find
that risk tolerance is a significant predictor of smoking, drinking, and failing to have
insurance, although it explains only a small fraction of the variation of these behaviors.
The empirical challenge is to find variation in insurance status that allows the econome-
trician to identify the moral hazard effect of insurance on prevention despite propitious
selection. The idea of propitious selection also poses a conceptual challenge, because
Landsburg (1993) shows contrary to the intuition offered by Hemenway (1990), in any
separating equilibrium in insurance markets it should be the less risk averse group that
takes more precautions.
The available empirical evidence is not conclusive, but tends to suggest that ex ante
moral hazard is not a very strong force leading to insured consumers investing in less
6 The sample is restricted to survey respondents who did not have any public health insurance. Otherwise, it
becomes hard to disentangle whether consumers with public insurance through Medicaid and Medicare invest
differently in prevention because they are insured or because they are poorer (Medicaid) or older (Medicare)
than other respondents. For additional discussion of this empirical exercise, see Kenkel (1995).
prevention. This could be for several reasons. Market responses including risk rating
and insurer and employer investments in loss prevention partly address moral hazard,
but these responses also appear relatively uncommon. A better explanation may be that
the ex ante moral hazard problem is largely solved by health insurance that covers only
the financial and not the health losses of serious illness. A practical issue also deserves
attention. In principle prevention decisions are based on expected health insurance cov-
erage at the time health loss occurs. In practice this means for many chronic illnesses
there may be a very weak relationship between current health decisions and current
health insurance status. There is a great deal of movement in and out of uninsurance, so
many currently uninsured may reasonably expect to be insured in the future. In addition,
the U.S. Medicare program provides essentially universal coverage for conditions that
develop after the age of 65, a highly relevant age period for many serious illnesses.
3.4. Moral hazard in public sector insurance systems
With moral hazard, incentives for prevention are partly shifted away from the in-
sured consumers onto the providers of insurance. The discussion to this point has been
in the context of privately-provided health insurance, where the responses of profit-
maximizing insurance firms and employers may limit moral hazard. As just noted, in
the U.S. Medicare provides a public sector health insurance system for people over the
age of 65, which may have implications for the extent of the moral hazard problem.
And most other developed countries rely much more than the U.S. on public sector
health insurance. Besley and Gouveia (1994) review data from the OECD of the share
of public expenditures in total health expenditures from 1960 to 1991. After substantial
growth in most countries the public sector share has stabilized at fairly high levels. By
1991 public sector expenditures accounted for more than half total health expenditures
in 22 of the 24 OECD countries, with the only exceptions being the U.S. and Turkey.
Besley and Guoveia (1994) suggest that it is useful to categorize health care systems
into three basic types: Type I, private financing and delivery; Type II, public financing
and (substantial) private delivery; and Type III, public financing and delivery.
In any health insurance system with public financing, the public sector as a whole has
an incentive to encourage prevention. As opposed to a system with private financing, it is
not clear if this general incentive will be internalized by the relevant agents who are able
to influence consumer preventive behavior. In Type II systems the problem is to transmit
the public sector incentives for prevention to the agents in the private sector who deliver
medical care. In Type III systems the problem is to transmit within the public sector the
general incentives for prevention to the specific public sector employees who deliver
care. As an example of how this could work in a public system, a proposed reform to the
British National Health Service included payments to general practitioners for reaching
targets on child immunization and cervical cancer screening. The reforms were not well
received, however [OECD (1992)].
The possible difficulties of creating useful incentives for prevention in public sector
health insurance systems means that ex ante moral hazard may be more important in
1694 D.S. Kenkel
these systems than in the U.S. private system. This suggests an explanation for the casual
observation that cigarette smoking seems more prevalent in many European countries
than in the U.S., despite the relatively low taxes imposed on cigarettes in the U.S.:
in European countries widespread public sector health insurance reduces individual and
third-party incentives for prevention. 7 Kenkel (1996) is a preliminary empirical analysis
of the effects of public sector health insurance on prevention. Using data aggregated
at the national level, Kenkel tests whether public sector health insurance acts to shift
demand for alcohol and tobacco. The results of the main analysis are mixed, providing
some evidence that public sector health insurance shifts alcohol demand up, as predicted
by the moral hazard effect. A positive model of health-related commodity taxation could
also be developed to explain why many European countries tax cigarettes and alcohol
at much higher rates than the U.S.: these countries use taxes to reduce ex ante moral
hazard created by their public sector health insurance systems. Kenkel's (1996) analysis
of this issue shows a positive relationship between the public sector share of health
expenditures and the levels of cigarette and alcohol prices (as proxies for taxes).
4. Will other market failures lead to too little prevention?
4.1. Externalities
Externalities arising from vaccination choices are also a commonly mentioned ratio-
nale for government involvement in prevention. The externalities arise from the con-
cept of "herd immunity" where any given individual's chances of getting an infectious
disease fall when others in the society are immune because of previous vaccinations
[Phelps (1992)]. Consequently, societal marginal benefits of a vaccination exceed pri-
vate marginal benefits, and private vaccination decisions will result in a vaccination
rate that is less than the socially optimal rate. However, herd immunity also means that
the socially optimal vaccination rate is less than 100 percent: the marginal benefits of
vaccinating the last person are zero and so exceed any positive marginal cost.
Many countries make vaccinations against common childhood diseases compulsory,
apparently in efforts to achieve 100 percent vaccination rates. Brito et al. (1991) show
that while private decisions lead to too few vaccinations, when vaccinations are perfect
the compulsory outcome with everyone vaccinated is worse in terms of social welfare
than the private market outcome. People who would have chosen to be vaccinated any-
way are not better off under the compulsory vaccination policy, because with perfect
7 Comparative data suggest the casual observation is only partly accurate. Smoking prevalence is relatively
low in the U.S., ranked (in order of male smoking rates) 78 out of 87 countries with available data [WHO
(1997, Table 3)]. But the U.S. ranks much higher (11 out of 111 countries with available data) in terms of per
capita consumption of cigarettes [WHO (1997, Table 7)]. Part of the reason for this discrepancy in rankings is
that the U.S. ranks much higher (33 out of 87) in terms of smoking prevalence among women [WHO (1997,
Table 4)]. The comparisons also suggest the consumption of cigarettes per smoker is higher in the U.S. than
most other countries.
vaccinations they are unaffected by others' vaccination choices. By revealed preference,

people who would have chosen to be unvaccinated without compulsion are clearly made
worse off by the policy. Brito and his colleagues show that the competitive equilibrium
is characterized by a level of private cost (e.g., in time or pain) of vaccination 0* below
which individuals private choices lead to vaccinations. The social optimum is charac-
terized by a higher cost threshold 0**, implying that the private equilibrium leads to
too few vaccinations. The analysis assumes that the compulsory vaccination policy is
entirely effective, so that everyone is vaccinated under that policy. If people with high
vaccination costs avoid vaccinations even under the compulsion policy, it might more
closely approximate the social optimum. Brito and his colleagues note that the con-
clusion that compulsion is worse than the competitive outcomes also depends on the
assumption of perfect vaccinations. Unvaccinated people pose some infection risk and
thus create external costs for imperfectly vaccinated consumers.
The extent of the market failure in vaccinations depends in part upon the responsive-
ness of individual vaccination decisions to the prevalence of the preventable disease.
Empirically, Philipson (1996) provides important empirical evidence on the "prevalence
elasticity" of the demand for measles vaccinations. A measles epidemic that occurred in
the U.S. during 1989-1991 created great variation in the prevalence of measles across
states and over time. Using data from the 1991 National Health Interview Survey, Philip-
son estimates a proportional hazard model of the duration until a child receives her first
measles vaccination. The large and statistically significant estimated effects of measles
prevalence on the hazard (probability) of receiving a vaccination imply prevalence elas-
ticities ranging from 1.56 to 1.89. These are interpreted as the effects of a one percent
increase in measles prevalence on the hazard rate. A related study by Ahituv, Hotz
and Philipson (1996) finds that AIDS prevention (condom usage) responds to the lo-
cal prevalence of AIDS. Other evidence on this type of consumer response to disease
epidemics is discussed in Philipson (2000).
With prevalence elastic demand, as disease prevalence falls because of the preven-
tion decisions by others, some consumers free ride and fail to purchase prevention
themselves. However, individual optimizing behavior places a check on the extent of
the externality. As the number of unvaccinated people rises, disease prevalence rises
as well, increasing private incentives for vaccination. Francis (1997) shows that with
a strong form of prevalence elasticity, after a certain disease transmission threshold all
remaining uninfected consumers choose to become vaccinated. In this particular case
there is no externality associated with vaccination decisions. Philipson (1996) shows
that the prevalence elasticity of demand also limits the impact of public price subsidies
and other public sector interventions to correct the market failure.
Geoffard and Philipson (1997) demonstrate that the interaction between the preva-
lence of disease and the demand for vaccines can create forces that make the eradication
of diseases difficult. Price reductions, whether through competition in private markets
or through public sector subsidies, do not in general lead to disease eradication. The
problem is that as the disease becomes less prevalent the demand for vaccines falls,
allowing the disease to return. This is also why public sector interventions may have
1696 D.S. Kenkel
small aggregate effects for prevention, because they crowd out private efforts as they
reduce disease prevalence. The crowd-out is larger, and therefore public sector subsi-
dies are less effective, the larger is the prevalence elasticity of the demand for vaccines.
This poses a dilemma for designing policies to achieve the socially optimal amount of
prevention. Subsidies are most effective in encouraging prevention when prevention de-
cisions are prevalence-inelastic (for example, prevention of a non-infectious disease),
but in this situation the benefits of prevention decisions are private and there is no ef-
ficiency rationale based on externalities for public sector action. When prevention de-
cisions are highly prevalence-elastic, they generate large externalities but public sector
subsidies will be relatively ineffective in increasing aggregate prevention because of
their crowding out effect.
Public health officials express concern about current levels of vaccinations, but these
concerns are not necessarily based on judgements about the extent of market failures
or how current rates compare to the socially optimal rates. For example, the American
Public Health Association (1992) describes the 1989-1991 measles epidemic as one in-
dicator of the "failure to deliver vaccine to vulnerable preschool children on schedule"
and calls for all children to be fully immunized. Similarly, the U.S. Preventive Ser-
vices Task Force (1989, 1996) and other organizations recommend essentially univer-
sal delivery of vaccines for diphtheria-tetanus-pertussis, polio, measles-mumps-rubella,
haemophilus influenza type b, hepatitis B, and varicella.
The case of polio is especially interesting, because the Task Force notes that more
than a decade has passed since the last documented transmission of wild poliovirus in
the U.S. and that it appears to have been eradicated in the entire Western hemisphere.
Ironically, all U.S. cases of polio are now caused by vaccine-strain poliovirus infections.
Presumably, despite these infections the Task Force will continue to recommend polio
vaccines until it is eradicated world wide, as has already been achieved with smallpox.
From an economic efficiency standpoint disease eradication will not always be a rea-
sonable public policy goal, i.e., eradication will not always yield more social benefits
than costs. Even though disease eradication generates a stream of benefits for all future
generations, the discounted present value of that stream is finite and may be smaller than
the costs of total eradication. Geoffard and Philipson (1997) show in a formal model
that dynamic externalities to future generations can justify disease eradication, depend-
ing crucially upon the interest rate used to discount future benefits. They conclude that
a deficit-financed eradication program may improve social welfare because it allows for
the intergenerational transfers necessary to pay current generations to overvaccinate for
the benefit of future generations.
4.2. Lack of consumer information
Because consumers lack information about the health consequences of their choices,
they will fail to make optimal personal prevention choices. While the externalities dis-
cussed above arise in the context of primary prevention of communicable diseases, in-
formation problems arise in the context of a wide variety of both primary and secondary
prevention activities. Three main topics in consumer health information are: (1) mea-
suring the extent to which consumers lack information; (2) estimating the links between
consumers' information and their prevention choices; and (3) determining whether mar-
ket forces create or correct information problems.
In terms of the extent of the problem, health economics research suggests consumers
are imperfectly informed about many important prevention choices. As noted in section
1, it has been estimated that consumer choices about tobacco, diet and activity patterns,
and alcohol account for almost 40 percent of mortality in the U.S. [McGinnis and Foege
(1993)]. Kenkel (1991b) uses data from a supplement to the 1985 U.S. Health Interview
Survey to analyze the extent to which consumers lack knowledge about smoking, ex-
ercise, and drinking. For instance, respondents were asked whether smoking increases
the chances of a list of health problems. The respondents' answers were compared to
the answers currently presumed correct by Public Health Service agencies and scored
accordingly. Out of 7 questions about smoking the average number of correct responses
was 5.6, suggesting that people knew a great deal but not everything about the health
effects of smoking. People knew less about drinking and exercise. The patterns are not
surprising: nearly everyone (over 90%) knew that smoking increases the risk of lung
cancer and that heavy drinking increases the risk of cirrhosis of the liver, but most did
not realize that smoking increases the risk of bladder cancer or that heavy drinking in-
creases the risk of cancer of the mouth. People knew the least about the proper amount
of exercise, with the average being somewhat less than one correct answer to the three
questions.
For some health risks, most notably smoking, an additional concern is whether con-
sumers who are aware of the health effects have accurate perceptions of the magnitudes
of the risks. This is a difficult research question, but available evidence suggests that
some consumers overestimate the health risks of smoking, while others underestimate
these risks. In a series of studies, Viscusi (1990, 1991, 1992, 1998) presents evidence
that most people, including most smokers, overestimate the health risks of cigarette
smoking. Viscusi (1990, 1991) reports analysis of national survey data collected in
1985, where perceived lung cancer risk is measured based on answers to the question:
"Among 100 cigarette smokers, how many of them do you think will get lung cancer
because they smoke?". Viscusi (1998) reports additional analysis of survey data on per-
ceptions of the total mortality risk of smoking, based on answers to the question "Out of
every 100 cigarette smokers, how many of them do you think will die from lung cancer,
heart disease, throat cancer, or any other illness because they smoke cigarettes?". 8 Di-
viding the answers by 100 yields the perceived lung cancer and total mortality risk. In
8 Viscusi (1992) also reports the results of a series of smaller scale surveys conducted in 1990 and 1991
that explored the sensitivity of the risk responses to several variations in question formulation. For example,
he presents evidence that the perceived risk of mortality from lung cancer due to smoking is similar to the
perceived risk of the incidence of lung cancer due to smoking. Viscusi (1992, p. 83) concludes that "the
similarity of the responses for [six] different question formulations suggests that the empirical findings are
not an artifact tied to some specific question phrasing".
1698 D.S. Kenkel
the 1985 sample the average perceived lung cancer risk was 0.426, and when the sam-
ple is restricted to current smokers the average is 0.368. In the 1997 sample the average
perceived total mortality risk was 0.501 for the full sample, and 0.424 for the sample
restricted to current smokers.
Part of the difficulty in determining the accuracy of consumers' perceptions of smok-
ing risks is that the "true" risk level is unknown, and the state of scientific knowledge
of the risks changes over time. Based on the state of scientific knowledge at the time of
the surveys, Viscusi (1992, 1998) argues that reasonable scientific reference ranges are
from 0.05 to 0.10 for lung cancer risks and 0.18 to 0.36 for the total mortality risks of
smoking. Comparing the distribution of perceived risks in the samples to these ranges
leads to the conclusion that most people and most smokers overestimate the mortality
risks of smoking. Moreover, his evidence suggests that overestimation of risks remains
an important empirical phenomenon using any plausible estimate of the true risk of
death from smoking. For example, a little over 5 percent of both the full sample and
current smokers perceive a smoking mortality risk of 1, and 20.5 percent of the full
sample and 12.8 percent of smokers perceive a risk of 0.8 or above.
A limitation to Viscusi's approach is that the measures relate to the perceived risk in
a hypothetical population of smokers, but people may feel that the risks they face per-
sonally are lower. Comparisons of different measures of consumer information about
smoking risks are consistent with this argument [USDHSS (1989, pp. 221-222)]. For
example, from a 1975 survey approximately 90 percent of smokers believe that smok-
ing is harmful to health, but from a 1986 survey only 75 percent of smokers were con-
cerned about the effects of smoking on their own health. More recently, Schoenbaum
(1997) presents evidence that current heavy smokers overestimate their expectations of
reaching age 75. His study uses data from a data set many health and labor economists
use, the Health and Retirement Survey, a national sample of adults aged 50 to 62. Re-
spondents were asked, "Using any number from zero to ten, what do you think are the
chances that you will live to be 75 or more?" with zero labeled "no chance at all" and
10 labeled "absolutely certain". Dividing by ten provides an estimate of the subjective
probability of survival to age 75. Although respondents could view this question as an
ordinal ranking rather than a probability assessment, Hurd and McGarry (1995) present
additional evidence that responses to this question behave like probabilities. Schoen-
baum compares the subjective estimates to estimates of the "true" survival probabilities
from life tables. Using life table values the mean survival probability for men who are
current heavy smokers is 0.263, while the mean subjective value is 0.501. Similarly,
the life table mean survival probability for women who are current heavy smokers is
0.308, while the mean subjective value is 0.601. Among men who are never smokers,
former smokers, and current light smokers, subjective survival probabilities correspond
quite closely to the relevant life table values, but for these groups of women subjec-
tive survival probabilities are below the life table values. The pattern of results suggests
that heavy smokers overestimate their survival probability and hence must underesti-
mate the risks of smoking; other groups either appear to have accurate perceptions or
overestimate all risks (not just smoking risks).
The apparently different conclusions reached by Viscusi and Schoenbaum illustrate

the inherent difficulty of research into whether consumer risk perceptions are accurate.
However, it is also important not to overstate the degree of difference between these
studies. Some of the difference is to be expected given differences in the composition of
the samples studied. Schoenbaum's evidence suggests that older (aged 50 to 62) heavy
smokers underestimate the risks of smoking. The samples of smokers in Viscusi's stud-
ies are much younger, as they are representative of the population over 16. If smokers
with higher risk perceptions are more likely to quit, it is to be expected that people who
remain smokers at older ages have lower than average risk perceptions. In addition, Vis-
cusi (1991) finds evidence that younger people are particularly likely to overestimate the
risks of smoking, and suggests that it may result from their greater exposure to the pub-
lic anti-smoking campaign. 9 It should also be pointed out that Schoenbaum finds that
light smokers on average appear to have accurate risk perceptions, and by the definition
used light smokers make up the majority of smokers in the sample (59 percent of male
smokers and 77 percent of female smokers). Given the distribution around the average
subjective survival probabilities, there are many light smokers who underestimate their
survival probability in Schoenbaum's sample. 10 In sum, both Viscusi and Schoenbaum
present evidence consistent with some smokers overestimating the risks of smoking and
others underestimating the risks. There is more disagreement about the extent of the in-
formation problem and the relative frequencies of the two types of mistakes consumers
make.
In addition to measuring the extent to which consumers lack information, health eco-
nomics research also addresses the links between consumers' information and their
prevention choices. Viscusi (1990) and Kenkel (1991a) estimate consumer demand for
health-related goods using cross-sectional data with direct measures of consumer infor-
mation. Their findings that information is an important determinant of demand confirm
and extend earlier findings about cigarette demand. A series of econometric studies es-
timate that "information shocks", such as the 1964 Surgeon General's report on the
hazards of smoking and required cigarette warning labels, significantly decreased the
demand for cigarettes [for example, Lewit, Coate and Grossman (1981) and Schneider,
Klein and Murphy (1981)]. Because this line of research is discussed in more depth in
Chaloupka and Warner (2000) it need not be further reviewed here.
9 However, it has been suggested that young people act as if they feel invulnerable to many risks, which
suggests that young people's perceptions of hypothetical risks (measured by Viscusi) may diverge more from
their estimates of personal risks (measured by Schoenbaum). This issues could be explored by collecting data
on the personal risk perceptions similar to that used by Schoenbaum for people of a variety of ages.
10 Unfortunately, Schoenbaum (1997) does not present much information on the distribution of responses.
Because there are also many nonsmokers who underestimate their survival probabilities, it is difficult to de-
termine whether light smokers who underestimate their survival probability overestimate the risks of smoking
or overestimate other risks. Of course, they could have private information in which case their estimates may
be more accurate than the life table values as estimates of the risks they personally face. While focusing on
personal risks is a useful approach, these ambiguities in interpretation reveal some advantages to Viscusi's
approach.
1700 D.S. Kenkel
A related line of research identifies health information shocks related to diet and nu-
trition choices. Scientific evidence accumulated fairly steadily showing the potential
health benefits of diets high in dietary fiber and low in fats and cholesterol. While pub-
lic health organizations and the U.S. federal government attempted to disseminate this
information, it seems to have spread relatively slowly through the 1970s. But in the
mid-1980s, the Federal Trade Commission (FTC) and the Food and Drug Administra-
tion (FDA) changed the regulatory environment, making it easier for firms to advertise
the link between diet and disease. In a series of studies Ippolito and Mathios (1990,
1995, 1996) explore the impact of the resulting health information shocks. In the ce-
reals market, producer claims about the health benefits of adding dietary fiber ap-
pear to have been an important information source for consumers, leading to substan-
tial dietary improvements [Ippolito and Mathios (1990)]. Similarly, individual food
consumption data and food production data show that consumption of fats, saturated
fats, and cholesterol fell from 1977 to 1985, but fell more rapidly between 1985 and
1990 after producer health claims became more common [Ippolito and Mathios (1995,
1996)].
Two important lessons can be drawn from the apparently strong response of con-
sumer nutrition choices to advertised health claims. First, many consumers seem to be
imperfectly informed about diet and nutrition. But the second lesson is that producers'
incentives to use health claims in advertisements for their products creates a power-
ful force to improve consumer information. The evidence suggests that the pre-1985
FDA and FTC regulations were overly strict and served to block this source of informa-
tion, worsening rather than improving market performance. It should be noted that the
post-1985 environment was not a laissezfaire free market but reflected the normal reg-
ulatory rules that all advertising claims must be truthful and not deceptive. The policy
environment changed again with the Nutrition Labeling and Education Act of 1990, and
on-going research investigates the implications for consumer dietary choices [Ippolito
and Mathios (1993)].
More generally, a potentially useful line of health economics research is to focus more
on the demand and supply of health information as a commodity itself, and whether
market forces tend to create or correct information problems. The ex ante benefits of
information depend on the consumer's expectations about whether the information will
change actions that determine utility. This is based on Hirshleifer and Riley's (1979) dis-
cussion of the expected value of information in general. As an example of this approach,
Philipson and Posner (1994) provide a provocative discussion of consumer demand for
information about AIDS. They argue that people in the highest risk groups for AIDS
also have the strongest private incentives to become informed about AIDS prevention.
In support of this prediction, Philipson and Posner (1994, p. 28) review evidence that
available information allowed the San Francisco gay community to learn rapidly about
AIDS prevention and adopt safer-sex practices, as shown by a rapid decline in homo-
sexually transmitted gonorrhea cases beginning in 1981 as the first major news of the
AIDS epidemic became public.
In addition to the incentives of consumers to gather information, a more complete

model would consider the incentives of suppliers of prevention-related goods and ser-
vices to provide or manipulate information. The studies by Ippolito and Mathios (1990,
1995, 1996) discussed above are case studies where producer incentives led to improved
consumer information about diet and nutrition. Condom manufactures have similar
incentives to improve consumer information about safer-sex practices. As a counter-
example, while cigarette manufacturers can no longer make direct health claims for
their products as they once did, many observers argue that some advertising campaigns
are designed to provide misleading associations between smoking and healthy images.
Along the same lines, Warner, Slade and Sweanor (1997, p. 1088) review evidence that
in the past filter cigarettes and low tar cigarettes may have been " "health image" prod-
ucts, public relations gimmicks, rather than truly "health-oriented" products. Warner,
Slade and Sweanor (1997) discuss in more detail the emerging market in long-term
nicotine maintenance, including current products such as the nicotine gum and patch,
and potential future products such as cigarette-like nicotine delivery systems. Regu-
lation of these products, and regulation of producer-supplied information about their
health effects, will help determine the health impact of this emerging market.
In many cases, the usefulness of producer-supplied information is hard to determine
because of the complexities of how consumers process risk information. A study of
public information programs about the risks of radon provides an interesting exam-
ple. Radon is a naturally-occurring odorless and colorless gas that is thought to be the
second-leading cause (after smoking) of lung cancer. The health risks are a function of
the level of radon in living spaces, and many homeowners face minimal or zero risks.
Smith, Desvousges, Johnson and Fisher (1990) report that providing a simple and ac-
curate one-page fact sheet about radon similar to that used by many radon testing com-
panies created concerns about radon risks when it was not warranted. The unwarranted
reactions by respondents at low risks were strong enough that the research team revised
the study design for ethical reasons. When consumers overestimate radon's risks prof-
its in the radon mitigation industry increase, and in this case the provision of factually
accurate information appears to lead to such overestimates.
The general point of the preceding discussion is to make the case that imperfect infor-
mation about prevention should not be treated as an exogenous market failure. Instead,
there seems to be a great deal to learn from an economic approach of the demand and
supply of prevention-related information. Applying an economic approach to different
prevention choices could yield predictions about the extent of imperfect information
and the type of misinformation (for example, over- or under-estimates of risks) likely to
result from market forces.
4.3. Research and development ofprevention as a public good
The market failures discussed to this point tend to create forces leading to inadequate
consumer demand for prevention. Health sciences research and the development of new
1702 D.S. Kenkel
medical technologies are probably as least as important as consumer demand in deter-

mining long-run trends in health and in the cost and quality of medical care [Weisbrod
(1991)]. There will be an inadequate supply of prevention if market forces lead to un-
derinvestment in research and development (R&D) about prevention, either absolutely
or compared to R&D about cure. The levels of R&D spending on prevention and cure
are difficult to determine, but there is a widespread perception that the U.S. health care
system emphasizes "high-technology curative care" over prevention [Blank (1988)].
Medical R&D firms may be unwilling to invest in the development of many forms
of prevention because patent protection is unavailable and property rights are poorly
defined [Dranove (1998)]. For example, an R&D firm could not capture many finan-
cial benefits from scientific research establishing that regular exercise yields important
health benefits. Dranove argues that the problem is exacerbated by the U.S. Food and
Drug Administration's (FDA) regulations, which require substantial scientific agree-
ment about the efficacy and safety of preventive interventions such as nutritional sup-
plements. As discussed above, once safety and efficacy are established private industry
may play an important role in disseminating information and improving nutrition [Ip-
polito and Mathios (1996)]. But with poorly defined property rights industry will tend
to under-invest in the R&D necessary to establish safety and efficacy in the first place.
Dranove (1998) argues further that gathering the scientific evidence (e.g., through
clinical trials) to establish the safety and efficacy of prevention is likely to be very
costly, especially compared to the process for cure. The benefits of prevention tend to
be diffuse and take a long time to develop, so clinical trials of prevention require larger
samples and longer durations than clinical trials of cure. As an example, Dranove's cal-
culations suggest there may be underinvestment in additional clinical trials to establish
the safety and efficacy of vitamin E supplementation to prevent heart disease. A recent
controversy over folate supplementation to prevent cardiovascular disease is a similar
example. Hornberger (1998, p. 61) cites "well-publicized concerns of federal funding
agencies that the costs of such a trial may not outweigh its potential benefits, given other
funding priorities". To address this concern, Hornberger (1998) uses Bayesian statistical
and decision-making analytic techniques to conduct a cost-benefit and sample size anal-
ysis of a controlled trial of folate. His analysis suggests that a trial with an appropriate
sample size could provided expected societal net benefits of over $11 billion within 15
years.
Medical R&D firms' willingness to invest in the development of new childhood vac-
cines also reflects the incentives created by governments as major purchasers. In the
U.S. the Vaccines for Children Program, begun in the mid-1990s, expanded previous
programs to give Medicaid-eligible and uninsured children the ability to obtain free vac-
cines from any enrolled private practitioner. In the first full year of this program (1995),
publicly purchased doses of children's vaccines accounted for over half of the doses
distributed nationally [Grabowski and Vernon (1997, Table 1-2)]. At the same time, the
negotiated prices paid by the government for the vaccines are far below prices paid by
the private sector. Public sector prices for diphtheria-tetanus-pertussis vaccine, measles-
mumps-rubella vaccine, and oral polio vaccine were between 20 and 50 percent of the
private catalogue prices [Grabowski and Vernon (1997, Table 1-1). Grabowski and Ver-
non (1997) argue that the expansion in government purchases at low prices reduces the
availability of funds to invest in new vaccines and negatively affects expectations about
the returns of future R&D. As a result, "a government program designed to advance
children's health care could ultimately produce unintended consequences that would
lead to fewer opportunities for improved health care in future periods" [Grabowski and
Vernon (1997, p. 53)].
In addition to FDA regulations and the indirect effects on R & D of vaccine purchas-
ing decisions, in the U.S. and many other countries government intervention in medical
R&D markets also takes more direct forms. Intramural research at the U.S. National
Institutes of Health (NIH) is an example of government-produced research. Through
its extramural programs the NIH also provides financial support for medical R&D by
university and private-sector scientists. The National Cancer Institute's [NCI] (1997)
list of recent accomplishments illustrates the range of NIH R&D. The NCI supports
prevention research including studies of a vaccine to prevent human papillomaviruses,
which are associated with 90 percent of cervical cancers, and the Breast Cancer Pre-
vention Trial of the use of the drug tamoxifen. Advances in NCI-supported research for
treatment include studies of the use of chemotherapy prior to surgery for breast cancer,
and hormone therapy after radiation therapy for prostate cancer. 1
Given the roles of the public sector, the appropriate balance between prevention- and
cure-related R&D is partly a question of non-market failure as well as market failure
[Wolf (1979)]. From an economic efficiency standpoint, the NIH and other sources of
publicly financed R&D should attempt to invest where private markets fail, and in that
way strike the correct balance between supporting prevention- and cure-related R&D.
Actual practice may be quite different. For example, Wiseman and Mooney (1998) ar-
gue that the World Bank and the World Health Organization seem to rely too much on
cost of illness (COI) and burden of illness (BOI) studies in setting research priorities.
In general, COI and BOI estimates only provide information on the size of the health
problem and do not provide the information decision-makers need about the relative
effectiveness and costs of different interventions. The continued popularity of COI and
BOI estimates might be explainable in a positive model of public sector R&D priority-
setting, in terms of bureaucrats' incentives to address "big" problems regardless of likely
effectiveness and costs. Similarly, the public sector decision-making process may result
in low prices paid for childhood vaccines because the current benefits of providing free
vaccinations for children are given more weight than the possible future unintended
(and hard to observe) adverse consequences for R&D into new vaccines.
11 The NCI's more precise terminology describe these as advances in "treatment" rather than "cure" because
the concept of curing cancer is problematic. Successful treatment of cancer is measured according to 5-year
or longer-term survival rates, not "cures" achieved.
1704 D.S. Kenkel
5. Is prevention cheaper than cure?
The answer to this question is often mistakenly believed to be crucial to determining

whether there is enough prevention. But when only explicit medical care costs are con-
sidered, whether prevention is less expensive than cure is not particularly relevant to
the evaluation of prevention from a societal perspective. A cheap prevention policy that
involves few medical expenditures but large utility sacrifices may not be socially de-
sirable; an expensive prevention policy that increases medical expenditures but yields
large improvements in health may be desirable. Cost-effectiveness analysis from a so-
cietal perspective and social cost-benefit analysis provide answers to more meaningful
public policy questions.
Whether prevention can reduce total medical expenditures is relevant to the evalua-
tion of prevention activities from other perspectives, however. For example, adminis-
trators of a public health insurance system will be interested in the impact of covering
preventive care on public sector budgets. Prevention is often mentioned as a system-
wide cost-containment strategy. For example, when proposing U.S. national health care
reform President Clinton argued that "it's just common sense ... that] long-term costs
to the health system will be lower if we have comprehensive preventive services". The
"bottom-line" impact on health care costs also may be the major consideration for ad-
ministrators of private sector health insurance plans and employers [Warner (1987)]. It
has been suggested that compared to cure, many private and public health insurers use a
double standard when evaluating prevention and cure. In common practice, the adoption
of curative care seems to require evidence of safety and efficacy at the least, and cost-
effectiveness at the most; while adoption of preventive care seems to require evidence
of cost-effectiveness at the least, and adoption often requires evidence of substantial
cost-savings [Warner and Warner (1993)].
In an important analysis Russell (1986) suggests that for many cases, prevention will
not be cheaper than cure, although it still may be desirable social policy. A simple math-
ematical point about the economics of prevention illustrates her main point. Prevention
does not always reduce medical care costs, because often the intervention is delivered to
a large group, only a very small fraction of whom would get the disease without the in-
tervention and thus incur treatment costs. Suppose the intervention prevents all disease
in a group of N individuals at a cost of $P per individual. Then total medical costs with
the intervention are N x P. If the incidence rate is i without the intervention, and cu-
rative care costs C per person, then the total medical costs without the intervention are
i x N x C. So whether prevention is cheaper than cure depends on whether P < i C, or
P/C < i. Even when P/C is a small number, prevention will not be cheaper than cure
given the incidence rates of most serious illnesses. Based on her review of a set of case
studies, Russell (1986, p. 110) concludes that while there are exceptions, "prevention
usually adds to medical expenditures, contrary to the popular view that it reduces them".
As an example of an exception to this rule, Warner (1997) cites evidence that smoking
cessation programs eventually pay for themselves through reductions in later health care
Ch. 31. Prevention 1705
costs as the incidence of lung cancer, heart disease, and other smoking-related diseases
falls.
From a societal perspective a somewhat better question to address is whether pre-
vention is more cost-effective than cure. Prevention has been a popular subject for
cost-effectiveness analysis (CEA) [Russell (1987), Haddix et al. (1996)]. In fact, many
text examples and both of the worked examples (prenatal prevention of neural tube de-
fects and cholesterol reduction) of CEAs provided by the Panel of Cost Effectiveness
in Health and Medicine are examples of prevention [Gold et al. (1996)]. These CEAs
provide a series of cost-effectiveness ratios showing the cost of achieving one unit of
health outcome, often life years saved or quality-adjusted life years (QALYs) saved.
Using two examples cited by the Panel on Cost-Effectiveness: (i) screening women for
cervical cancer every year instead of every two years costs an additional $1.1 million
the per year of life saved [Gold et al. (1996, Table 1.2)]; (ii) making dietary changes to
reduce blood cholesterol levels for a high-risk 60 year-old man with a blood cholesterol
level of 300 costs $13,000 per year of life saved [Gold et al. (1996, Table 1.4)].
Returning to the comparison of the cost-effectiveness of prevention and cure, no gen-
eralizable pattern emerges. The evaluation of life-saving interventions by Tengs et al.
(1995) provides many examples where prevention is less cost-effective than cure and
vice versa. A striking and surprising example is that Tengs and her colleagues report
that heart transplantation for patients age 55 or younger and favorable prognosis in-
volves $3,600 of costs per life year saved, an order of magnitude less than regular leisure
time physical activity, such as jogging, in men age 35 ($38,000 per life year). Tengs and
Graham (1996) analyze the cost-effectiveness of 185 available life-saving interventions,
and find no apparent relationship between the cost-effectiveness of the intervention and
whether it has been implemented. Compared to the status quo, Tengs and Graham cal-
culate that policy makers could save twice as many lives at the same social cost by
investing in the most cost-effective interventions.
Studies of the cost-effectiveness of prevention face many of the methodological is-
sues that arise more generally for CEA of health and medical care. The general method-
ology of CEA is discussed in detail in Gold et al. (1996) and in Garber (2000). One
somewhat controversial issue that is important for the CEA of prevention is how to ac-
count for future medical costs. The Panel on Cost-Effectiveness did not reach a defini-
tive conclusion on this aspect of CEA methodology. However, Meltzer (1997) argues
that all future medical costs, even those unrelated to the initial intervention under con-
sideration, should be part of the cost accounting for the cost-effectiveness ratios. The
estimated cost-effectiveness of preventive interventions could be quite sensitive to how
future costs are treated, although as Garber (2000) points out for many preventive in-
terventions these costs will accrue far in the future and be heavily discounted. Russell
(1994) emphasizes another area of costs often neglected: the human costs associated
with falsely positive screening tests, including time, energy, anxiety and risks from un-
necessary treatments.
Based on the foundations of applied welfare economics, the cost-effectiveness ques-
tion is not as useful as the cost-benefit question: Are the social benefits of prevention
1706 D.S. Kenkel
greater than the social costs? Completing a cost-benefit analysis (CBA) of prevention
requires estimates of the dollar value of the health effects based on willingness to pay for
health improvements [Tolley, Kenkel and Fabian (1994)]. A completed CBA addresses
whether resources devoted to prevention are in their most highly valued use. In contrast,
CEA only addresses the relative cost-effectiveness of a given intervention compared to
other choices. In principle, the CBA methodology resolves other shortcomings of CEA
as well. For example, future medical costs for other diseases will be accounted for in
the estimates of willingness to pay for health improvements, and the human costs asso-
ciated with falsely positive examinations can be incorporated directly into the analysis.
Because of the limitations of CEA, evidence like that provided in Tengs et al. (1995)
is not really useful in choosing between prevention and cure. For example, the surpris-
ing result that heart transplantation is more cost-effective than exercise might not carry
over to a CBA, considering costs CEA may neglect such as the pain and suffering of the
heart disease victims and their families before the transplantation. In practice satisfacto-
rily addressing all of the measurement issues that arise in a CBA of prevention remains
difficult and poses a number of challenges, however.
6. What policies can encourage prevention?
As noted earlier, public health professionals are strong advocates of prevention. The
economic approach offers theoretical support for this advocacy, by identifying market
failures including ex ante moral hazard, vaccination externalities, lack of consumer in-
formation, and underinvestment in research and development about prevention. CEA
studies also provide evidence that some specific preventive interventions are at least
as cost-effective as other health interventions, suggesting that new prevention policies
warrant serious attention. Often, however, policy proposals seem to be made on the pre-
sumption that prevention needs to be encouraged. To date, health economics research
probably sheds more light on which policy tools may work to achieve the objective of
more prevention, and relatively less light on whether encouraging more prevention is a
reasonable policy goal.
6.1. Using taxes and subsidies to encourageprevention
Health economics provides a number of conceptual arguments for taxes and subsidies
as prevention policies. Several studies discuss the theory of optimal commodity taxa-
tion with moral hazard [Arnott and Stiglitz (1986), Gravelle (1986)]. In the typical case
the analysis implies unhealthy goods (cigarettes, alcohol) should be taxed and healthy
goods (exercise equipment, smoke detectors) should be subsidized [Arnott and Stiglitz
(1986)]. Brito et al. (1991) show that taxing unvaccinated people and/or subsidizing
vaccinated people is a better solution to contagion externalities than compulsory vac-
cination. Phelps (1988) suggests that higher beer taxes are a second-best solution to
incomplete information about the risks of drunk driving, creating an opportunity for
substitution between death and taxes.
In practice, most countries levy special taxes on alcohol and cigarettes. Current policy
has its origins in "sin taxes" and revenue concerns, but the externalities and other market
failures associated with the consumption of alcohol and cigarettes provide economic
efficiency rationales as well. In a seminal study, Manning et al. (1991) estimate that the
external costs of heavy drinking probably warrant higher alcohol taxes in the U.S., while
taxes on cigarettes appear to be at the about the right level given the estimated external
costs of smoking. Subsequent work extending and updating this study has not always
reached the same conclusions; this research is discussed in more detail in other chapters
of this Handbook [see Cook and Moore (2000) on alcohol taxation and Chaloupka and
Warner (2000) on cigarette taxation]. Special taxes are not commonly levied on health-
related consumer goods other than alcohol and tobacco, although similar public health
and externality arguments could be made.
Philipson and Posner (1995) investigate the effects of subsidies for testing for sexu-
ally transmitted diseases (STDs). Interpreting unprotected (risky) sex between two indi-
viduals as an economic trade, they argue that STD testing allows traders to learn about
quality. For curable STDs, testing is also a form of secondary prevention that provides
benefits to the individual through earlier treatment. At the time Philipson and Posner
were writing, however, early detection provided fairly small benefits to an individual
infected with HIV, leading to their argument that "the main beneficiary of the test is
therefore not the person testing, but rather his potential partner" [Philipson and Posner
(1995, p. 447)]. Recent advances in drug treatment of people who are HIV positive or
have full-blown AIDS mean the relative benefits to self and others have changed, but
do not change the main insights of this study. In particular, their analysis concludes that
subsidizing testing reduces incidence if the pretesting status quo is risky sex, but if the
pretesting status quo is safe sex subsidizing STD is likely to increase incidence. This
unintended consequence arises because not only does subsidizing testing increase the
total volume of unprotected sexual trades, but it also may increase the total volume of
such trades between individuals of different HIV status, which creates the potential for
new infections. Philipson and Posner's empirical analysis provides evidence that com-
pared to no-testing, a negative test result does not increase the propensity to engage in
safe sex, although a positive test result does. Overall, however, their empirical analysis
is inconclusive on whether testing increases or decreases disease incidence.
6.2. Improving access to clinicalpreventive services
Several policy-related trends in U.S. health care markets may help improve consumer
access to clinical preventive services. First, many states have passed legislation man-
dating that private insurance plans cover preventive services [Thompson et al. (1989)].
Second, HMOs and other managed care insurance plans that traditionally place a greater
emphasis on prevention are gaining substantially in market share. Third, public health
insurance through Medicaid and Medicare offer better coverage for preventive care now
than in the past. For example, Medicare coverage was extended to influenza vaccina-
tions based on evidence showing vaccinations to be cost-effective [Centers for Disease
Control (1994)]. In 1997 Medicare coverage of preventive services was expanded again.
1708 D.S. Kenkel
An additional impetus for prevention comes from the increasing number of Medicaid
and Medicare recipients enrolled in managed care plans. Finally, beginning in the mid-
1980s changes in federal requirements substantially expanded Medicaid eligibility of
children and pregnant women. 12
Whether improved access actually leads to more use depends on the responsiveness of
the demand for preventive care to insurance coverage, i.e. the price elasticity of demand.
Using cross-sectional data from the 1989 National Health Interview Survey, Hafner-
Eaton (1993) finds that well uninsured nonelderly persons were half as likely as their
insured counterparts to visit a physician. Kenkel (1994) analyzes women's demand for
two specific cancer screening preventive services, breast examinations and the Pap test
for cervical cancer. Having insurance is estimated to increase the probability of annual
use of these services by around 5 percentage points. Mullahy (1999) finds that among
people aged 25 to 64 having insurance increases the probability of being immunized
against the flu by over 3 percentage points. This is a very strong effect because the
vaccination rate in this sample is only 8.3 percent. Mullahy is careful to interpret the in-
surance variable in his study as a very broad measure of access to the health care system,
rather than a pure price effect of insurance coverage on the demand for preventive care,
and this is probably the correct interpretation of Hafner-Eaton's and Kenkel's findings
too. A study by Cherkin, Grothaus and Wagner (1990) provides evidence on the pure
price response of prevention demand, for a relatively small change in price however. In
an employed population of HMO enrollees, Cherkin, Grothaus and Wagner find that a
$5 office visit copayment on use of preventive care resulted in a decrease in physical
examinations but did not significantly affect immunization rates for young children or
cancer screening tests received by women.
Results from the RAND Health Insurance Experiment (HIE), where enrollees were
randomly assigned to different health insurance plans, provides evidence that differ-
ences in medical care usage are due to health insurance, and not some other differ-
ence between insured and uninsured persons [Newhouse and the Insurance Experiment
Group (1993)]. Specifically, Lillard et al. (1986, pp. 93-94) report that the proportion
of women receiving preventive care in any one year is somewhere between 3 and 7
percentage points higher in the group receiving free care than in the groups with less
complete insurance. It is reassuring that estimates based on experimental data in the
HIE are generally comparable to estimated effects in econometric studies using nonex-
perimental data.
As indicated by their name, Health Maintenance Organizations (HMOs) appear to
place an emphasis on prevention. In the past, HMOs provided more complete coverage
for well-care visits and preventive services while traditional fee for service insurance
plans often excluded coverage, but more recently differences in coverage are diminish-
ing. Whether HMOs in general take other steps to encourage prevention is not clear.
Results from the Rand Health Insurance Experiment indicate that increased HMO mem-
12 The Medicaid expansions, and the econometric studies of their effects, are discussed in more detail in
Currie (2000).
bership will increase use of preventive services [Newhouse and the Insurance Experi-
ment Group (1993)]. However, only one HMO was studied, the Group Health Cooper-
ative (GHC) of Puget Sound, and it may not be typical. Significantly, Bates and Winder
(1984, p. 212) note that the GHC has a comprehensive education program emphasiz-
ing preventive care; Thompson (1996) provides a more recent discussion of the GHC's
experiences with clinical preventive services. In a review of the earlier literature, Luft
(1978) concludes that when HMO members were found to use more preventive care than
people with conventional insurance, the important factor was whether the individual had
insurance coverage for preventive visits. Kenkel (1994) finds that HMO members are
actually less likely to use preventive care, controlling for insurance coverage.
Consistent with the magnitudes of the estimated insurance responses discussed above,
a common observation is that expanding health insurance coverage will not lead to sub-
stantial improvements in the receipt of clinical preventive services. One explanation is
that time costs and other components of the "full price" of medical care are important
barriers to access to preventive care. Frank et al. (1995) use data from the Baltimore Im-
munization Study to estimate a model of the demand for immunizations for preschool
children. In this study of low-income high-risk children, overall vaccination rates for
measles, mumps, and rubella (MMR) were almost 80 percent, compared to a rate of
diphtheria, tetanus, and pertussis (DTP) of about 70 percent. Convenience factors like
the availability of adult time and regular source of care emerge as important influences
on the demand for immunizations, while money prices and insurance were not signifi-
cant.
6.3. Developing countries

Improving access to preventive care in developing countries poses fundamentally dif-
ferent challenges than those discussed above in the context of U.S. health care markets.
In low-income countries health expenditures per capita are in the range of $2 to $40,
compared to per capita expenditures of $2,800 in the U.S. and from $400 to $2,500
in other established market economies [World Bank (1993, Table 5.2)]. Furthermore,
because of the lack of insurance in low income countries out-of-pocket payments ac-
count for more than half of health care expenditures. The World Bank (1993, Table 5.3)
estimates that providing a public health package and an essential clinical package, con-
sisting of prenatal and delivery care, family planning services, management of the sick
child, treatment of tuberculosis, and case management of sexually transmitted diseases,
would cost only an average of $12 per person year in low-income countries. Although
inexpensive by U.S. standards, this is quite substantial compared to per capita health
care expenditures and per capita income in these countries. However, the World Bank
also estimates that because of their high cost-effectiveness these packages could re-
duce the total burden of illness in developing countries by 32 percent, and argues that
the packages are "affordable in low-income countries if governments reallocate current
health expenditures and increase public spending and if they implement policies that en-
courage selective payments directly from better-off patients and from existing insurance
1710 D.S. Kenkel
schemes". In short, dramatic improvements in access to preventive care in developing

countries require fairly dramatic reforms of the health care systems.
6.4. Providing information to encourageprevention

Public policies that help improve consumer information about prevention include fund-
ing basic and applied health research, official statements and documents summarizing
the best scientific evidence on prevention, school health education, and mass media
campaigns. The impacts of these policies on consumer information and health-related
behaviors have been studied by researchers from a variety of behavioral and social sci-
ence disciplines, including economics. Health economics studies have made important
contributions to the evaluation of specific interventions and shed light on the role in-
formation plays more broadly. A further challenge for the field of health economics
is to identify research questions where the economic approach provides a comparative
advantage for producing useful insights.
Russell's (1994) study of medical screening shows the importance of a different per-
spective in forming prevention and information policy. Improving consumer informa-
tion about medical screening tests faces the fundamental challenge of accurately con-
veying all of the complexities and tradeoffs involved. Russell (1994, p. 2) uses the term
"pseudo-truth" to describe common recommendations, because they "leave out the com-
plexities and tradeoffs, the mixture of solid information and educated guesses, that have
gone into their development". Part of the problem is that the experts themselves are im-
perfectly informed. In addition, some of the tradeoffs involved require judgements that
will be different for people with different values and preferences. Viewed as a principal-
agent problem, the question is whether the choices recommended by experts are the
same as the consumers would make with the same information. Russell believes that
this is often not the case, and suggests a number of specific ways the process of reach-
ing screening recommendations could be improved. The fundamental problem of the
complexity of any recommendation is not easily solved, however.
While Russell's discussion is in the context of medical screening tests, similar is-
sues arise in the context of the public provision of information about health lifestyles as
prevention. For example, public information campaigns on the health and safety risks
of alcohol abuse face the challenge of incorporating into their recommendations recent
research showing possible beneficial effects from the moderate alcohol consumption.
After reviewing the evidence, the Deputy Director of the National Institute on Alco-
hol Abuse and Alcoholism (NIAAA) carefully recommends that "For certain groups
of people, alcohol consumption is associated with a net health risk ... Middle-aged and
older adults who do not fall into any of the above-mentioned exclusionary categories and
who enjoy consuming alcohol in moderation are likely to experience net health benefits"
[Dufour (1996)].13 Phillips and Zeckhauser (1996) suggest that presenting consumers
13 For obvious reasons, the wine industry is very interested in using this research in general and the NIAAA's
recommendations in particular in promoting their products. This is another example where there are clear
profit incentives in private markets to provide prevention-related information, but where it is less clear if
private provision would lead to well-informed consumers.
with information about the moderate alcohol consumption on life years saved is much
more effective than the ways such information is currently provided.
An example where health economists potentially have a comparative advantage in
studying information provision is in studying the role of physicians and other health
professionals as sources of consumer health information. The theoretical and empirical
literature in health economics has long recognized that the physician provides informa-
tion and advice, along with medical care. However, most studies focus on the physi-
cian's advice about the appropriate level of medical care [e.g., Dranove (1988), Kenkel
(1990)]. Only a few studies broaden the focus and study the information-providing role
of physician's advice about health-related but non-medical goods. Jones (1994) and
Jones and Yen (1994) include advice from either a physician or a family member as a
determinant of cigarette demand. Both studies find some counter-intuitive results, where
advice is associated with more smoking. An obvious problem is that the receipt of ad-
vice may be endogenous, for example if physicians target their heaviest smoking pa-
tients to receive advice. Kenkel and Terza (1997) develop an econometric approach that
takes into account the possibility that in nonexperimental data the receipt of physician
advice is potentially endogenous. They find that advice about drinking has a substantial
and significant impact on consumer demand for alcohol, and that failing to account for
the endogeneity of advice masks this result.
The limited research by health economists on physician advice builds on a substan-
tial body of research from other fields. Clinical trials provide strong evidence of the
efficacy of physician advice for smoking cessation and the prevention of alcohol abuse.
In these trials, patients are randomly assigned to a treatment group that receives ad-
vice or to a control group that does not. Differences in outcomes across the groups
provide estimates of the treatment effects. Smoking cessation rates twelve months af-
ter a medical intervention were an average of 6 percentage points higher in a meta-
analysis of 39 controlled trials, including 25 in which the principal intervention was
physician counseling or advice [Kottke et al. (1988)]. Similarly, a review of randomized
trials of physician advice on drinking placed "brief counseling among the most strongly
supported intervention modalities for alcohol problems ... " [Bien, Miller and Tonigan
(1993, p. 319)]. In a recently completed multinational trial the World Health Organi-
zation (WHO) Brief Intervention Study Group (1996) find that "at-risk drinkers who
received advice reported drinking approximately 17 percent less on average than those
in the control group".
The U.S. Preventive Services Task Force (1996) provides a set of recommendations
about physician counseling and advice on a wide range of prevention topics. On some
major topics, such as smoking cessation and alcohol abuse, there is solid evidence
demonstrating the effectiveness of interventions by physicians in changing patient be-
haviors. For another important prevention topic, there is solid evidence that nutritional
counseling can be effective in changing people's dietary habits, but most of this evi-
dence concerns counseling by nurses, nutritionists and other health professionals rather
than physicians. For most of the other topics, including physical activity, prevention
of motor vehicle accidents, prevention of household and recreational injuries, preven-
1712 D.S. Kenkel
tion of youth violence, prevention of dental and periodontal disease, prevention of HIV
and other sexually transmitted diseases, prevention of unintended pregnancies and the
prevention of gynecologic cancers, there is weak (or no) evidence that physician coun-
seling is an effective way to change the behavior in question. Nevertheless, the Task
Force recommends physician counseling on most of these topics on the grounds that the
behavior in question (e.g., physical activity) is clearly linked to better health even if the
link between physician advice and the behavior has not been adequately studied. The
implicit assumption seems to be that providing more information to patients through
physician advice will do no harm and perhaps do some good. Potentially, health eco-
nomics research could explore both parts of this assumption, but little work has been
done to date.
A common criticism is that the potential for the medical care sector to actively pro-
mote healthier lifestyle choices is largely unrealized. While high proportions of physi-
cians report counseling their patients about smoking [Wechsler et al. (1983), Wells et
al. (1984, 1986)], only 30 to 40 percent of smokers report having received the advice.
For other health choices like exercise and diet, even by physicians' reports advice is not
very common. An avenue for future research is to conduct economic analysis of physi-
cian's incentives in the context of the principal-agency relationship between patients
and physicians.
7. What does health economics offer prevention research?
Health economics research reflects the tension of recognizing the importance of health
care institutions and the state of scientific and medical knowledge without abandoning
the unique insights of the economic approach. As noted in the introduction, what is
meant by prevention depends on the context of a specific country at a specific time. For
example, it is certainly sensible that U.S. health economists' attention shifted over time
from polio [Weisbrod (1971)] to AIDS [Philipson and Posner (1994)]. Epidemiologic
and demographic transitions will continue to influence priorities for prevention, and can
also influence the questions about prevention that are most relevant to health economics
research. For example, reductions in mortality rates at older ages are contributing to
population growth of the U.S. elderly, with important implications for health care costs
and pension costs [Fogel (1994)]. Pickett and Bridges (1987, p. 76) argue, however, that
increased longevity due to primary and secondary prevention need not lead to "a gloomy
future in which large numbers of aged, infirm people will spend their waning years in
nursing homes or other institutions at great cost to themselves, their families, and pub-
licly supported medical programs, especially Medicare". Their illustrative calculations
suggest that prevention could lead to a scenario where the impact of a substantial in-
crease in the size of the Medicare beneficiary population could be partially offset by
an increasing proportion of healthier low-cost survivors in the elderly population. Con-
cerns about health and social policy for an aging population illustrate how the context
of prevention evolves. To stay relevant the research agenda of health economics must
evolve accordingly.
However, health economics can be most useful when its research agenda also reflects
the power of the economic approach. Health economists reframe issues and ask differ-
ent questions than those posed by other researchers. In terms of normative economics,
Russell's (1986) comparison of prevention and cure is an example. She makes a use-
ful point in her review of the evidence that prevention can add to medical expenditures
rather than reducing them. But a more fundamental point that emerges from her anal-
ysis is to question the question: "Is prevention cheaper than cure?" Russell shows that
this is not really the right question from the economic approach, even though to non-
economists it sounds like a quintessentially economics question. Instead, positive and
normative economic analysis focuses on the question: "Is prevention worthwhile?" As
is implicit in Warner's (1984) discussion, health economics research into prevention is
especially worthwhile when it raises new questions and provides new perspectives like
these.
In terms of positive economics, Becker's (1976) famous definition of the economic
approach continues to provide guidance for future directions in health economics: "The
combined assumptions of maximizing behavior, market equilibrium, and stable pref-
erences, used relentlessly and unflinchingly, form the heart of the economic approach
as I see it". It is interesting to consider the extent to which health economics research
on prevention uses (unflinchingly or not) these assumptions. For example, the theoret-
ical models of consumer demand for prevention, and the empirical studies prompted
by these models, exemplify the rich set of predictions that can be drawn out of the as-
sumption that individuals make their health choices to maximize lifetime utility. Profit-
maximizing behavior by insurance firms, providers of medical care, and providers of
health-related goods also plays prominent roles in many of the analyses reviewed in this
chapter. Some of the studies are also examples of Becker's (1976) point that models
that assume maximizing behavior can incorporate a rich set of values and preferences,
and that the formation of these values and preferences are suitable objects for economic
analysis [Becker and Murphy (1988), Becker and Mulligan (1997)].14
As a broad and perhaps unfair generalization, however, health economics research
on prevention seems to have made less use of the assumption of market equilibrium. 15
An exception to this generalization is the exciting recent work on economic epidemiol-
ogy. The concept of prevalence elasticity, briefly reviewed above and in more detail in
Philipson (2000), shows how in equilibrium one individual's preventive behavior deter-
mines and is determined by the choices of other individuals. Economic epidemiology
14 By making preference formation endogenous, in recent work Becker might seem to be moving away
from his own definition that the economic approach requires the assumption of stable preferences. However,
his recent work seems fully consistent with his main argument that "The assumption of stable preferences
provides a stable foundation for generating predictions about responses to various changes, and prevents the
analyst from succumbing to the temptation of simply postulating the required shift in preferences to 'explain'
all apparent contradictions to his predictions" [Becker (1976, p. 5)].
15 I am more confident that this statement is an accurate self-criticism of my own research agenda than I am
that it is a reasonable generalization about health economics research on prevention.
1714 D.S. Kenlkel
considers the equilibrium behavior of consumers in interaction with other consumers.

The equilibrium behavior of consumers in interaction with the other economic agents
involved in prevention is a promising direction for future research. As noted above, there
is room for more theoretical and empirical analysis of physicians as providers of clinical
preventive services and as advisers about health lifestyles. The market for prevention-
related information also deserves systematic study, where depending upon the context
profit-maximizing producers of health-related commodities may help or hinder the flow
of accurate information to consumers.
In many cases the equilibrium analysis of topics in prevention will be complicated be-
cause so many classes of economic agents are involved. For example, as for other types
of medical care, the physician as a provider of preventive care can be described as a dou-
ble agent trying to serve both the consumer and the insurer. Other cases involve inter-
actions between providers of health-related goods and consumers, interactions between
consumers of prevention and their employers, or even interactions between pharma-
ceutical companies as suppliers of prevention-related drugs, physicians as prescribers,
insurers, employers, and of course the consumers. It should be stressed that the need for
more consideration of market equilibrium in these cases is crucial for empirical work
and is not solely or even primarily a concern about theoretical models of prevention.
Useful empirical analysis is not always tightly linked to a theoretical model. However,
the power of the economic approach is still evident in less structural empirical studies
that are well-informed by the relevant theoretical framework.
References
Ahituv A., V.J. Hotz and T. Philipson (1996), "The responsiveness of the demand for condoms to the local
prevalence of AIDS", Journal of Human Resources 31(4):869-897.
American Public Health Association (1992), "Childhood immunizations: Easy access versus requirement for
essential services". Policy Statement 9102 Adopted by the Governing Council of the American Public
Health Association. American Journal of Public Health 82(3):477-478.
Arnott, R., and J. Stiglitz (1986), "Moral hazard and optimal commodity taxation", Journal of Public Eco-
nomics 29:1-24.
Barsky, R.B., F.T. Juster, M.S. Kimball and M.D. Shapiro (1997), "Preference parameters and behavioral het-
erogeneity: An experimental approach in the health and retirement study", Quarterly Journal of Economics
112(2):537-579.
Bates, I., and A. Winder (1984), Introduction to Health Education (MayField, Palo Alto, CA).
Becker, G.S. (1976), "The economic approach to human behavior", in: The Economic Approach to Human
Behavior (University of Chicago Press, Chicago).
Becker, G.S., and C.B. Mulligan (1997), "The endogenous determination of time preference", Quarterly Jour-
nal of Economics 112(3):729-758.
96:675-700.
Becker, G.S., M. Grossman and K.M. Murphy (1994), "An empirical analysis of cigarette addiction", Ameri-
can Economic Review 81(2):237-241.
Besley, T., and M. Gouveia (1994), "Alternative systems of health care provision", Economic Policy
(October):200-258.
Bien, T., W. Miller and J.S. Tonigan (1993), "Brief interventions for alcohol problems: A review", Addiction
88:315-336.
Blank, R.H. (1988), Rationing Medicine (Columbia University Press, New York).
Blind, K.V.K. (1996), "Self protection and insurance of irreplaceable commodities", Jahrbucher fur Nation-
aloknomie und Statistik 216(2):194-208.
Blomquist, G.C. (1988), The Regulation of Motor Vehicle and Traffic Safety (Kluwer Academic Publishers,
Boston).
Brito, D.L., E. Sheshinksi and M.D. Intriligator (1991), "Externalities and compulsory vaccinations", Journal
of Public Economics 45:69-90.
Brown, R.E., A. Elixhauser, J. Corea, B.R. Luce and S. Sheingold (1991), National Expenditures for Health
Promotion and Disease Prevention Activities in the United States (The Medical Technology Assessment
and Policy Research Center, Battelle, Washington, DC).
Cebu Study Team (1991), "Underlying and proximate determinants of child health: The Cebu longitudinal
health and nutrition study", American Journal of Epidemiology 133(2):185-201.
Center for Disease Control and Prevention (1992), "Estimated national spending on prevention - United
States, 1988", Mortality and Morbidity Weekly Review 41(29):529-531.
Center for Disease Control and Prevention (1994), "Implementation of the medicare influenza vaccination
benefit - United States, 1993", Mortality and Morbidity Weekly Review 43(42):771-772.
Chaloupka, F.J. (1991), "Rational addictive behavior and cigarette smoking", Journal of Political Economy
99(4):722-742.
Chaloupka, F, and K. Warner (2000), "The economics of smoking", in: A.J. Culyer and J.P. Newhouse, eds.,
Handbook of Health Economics (Elsevier, Amsterdam) Chapter 29.
Cherkin, D.C., L. Grothaus and E.H. Wagner (1990), "The effect of office visit copayments on preventive care
services in an HMO", Inquiry 27:24-38.
Cook, PJ., and D. Graham (1977), "The demand for insurance and protection: The case of irreplaceable
commodities", Quarterly Journal of Economics 91(1):143-156.
Cook, P.J., and M. Moore (2000), "Alcohol", in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health
Cropper, M.L. (1977), "Health, investment in health, and occupational choice", Journal of Political Economy
85:1273-94.
Cropper, M.L. and W.E. Oates (1992), "Environmental economics: A survey", Journal of Economic Literature
30(2):675-740.
Currie, J. (2000), "Child health in developed countries," in: A.J. Culyer and J.P, Newhouse, eds., Handbook
of Health Economics (Elsevier, Amsterdam) Chapter 19.
Dardanoni, V., and A. Wagstaff (1987), "Uncertainty, inequalities in health, and the demand for health",
Dardanoni, V., and A. Wagstaff (1990), "Uncertainty and the demand for medical care", Journal of Health
Economics 9:23-38.
Dowd, B.E. (1982), "Financing preventive care in HMOs: A theoretical analysis", Inquiry 19:68-78.
Dranove, D. (1988), "Demand inducement and the physician/patient relationship", Economic Inguiry
26(2):281-298.
Dranove, D. (1998), "Is there underinvestment in R&D about prevention?" Journal of Health Economics
17(1):117-127.
Dufour, M.C. (1996), "Risks and benefits of alcohol use of the life span", Alcohol Health &Research World
20(3):145-151.
Eddy, D. (1980), Screening for Cancer: Theory, Analysis, and Design (Prentice-Hall, Englewood Cliffs, NJ).
Ehrlich, I., and G. Becker (1972), "Market insurance, self-insurance, and self-protection", Journal of Political
Economy 80:623-49.
Ehrlich, I., and H. Chuma (1990), "A model of the demand for longevity and the value of life extension",
Journal of Political Economy 98(4):761-782.
1716 D.S. Kenkel
Farrell, P., and V.R. Fuchs (1982). "Schooling and health: The cigarette connection", Journal of Health Eco-
nomics 1:217-230.
Fogel, R.W. (1994), "Economic growth, population theory, and physiology: The bearing of long-term pro-
cesses on the making of economic policy", American Economic Review 84(3):369-395.
Francis, P.J. (1997), "Dynamic epidemiology and the market for vaccinations", Journal of Public Economics
63:383-406.
Frank, R.G., C.S. Dewa, E. Holt, N. Hughart, D. Strobino and B. Guyer (1995), "The demand for childhood
immunizations: Results from the Baltimore immunization study", Inquiry 32:164-173.
Fries, J.F. (1997), "Editorial: Can preventive gerontology be on the way?" American Journal of Public Health
87(10):1591-1593.
Fuchs, V.R. (1982), "Time preference and health: An exploratory study", in: V.R. Fuchs, ed., Economic As-
pects of Health (University of Chicago Press for NBER, Chicago, IL).
Garber, A. (2000), "Advances in cost-effectivenes analysis of health interventions", in: A.J. Culyer and J.P.
Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 4.
Geoffard, P.-Y., and T. Philipson (1997), "Disease eradication: Private versus public vaccination", American
Gold, M.R., J.E. Siegel, L.B. Russell and M.C. Weinstein, eds. (1996), Cost-Effectiveness Analysis in Health
and Medicine (Oxford University Press, New York).
Grabowski, H.G., and J.M. Vernon (1997), The Search for New Vaccines: The Effects of the Vaccines for
Children Program (The American Enterprise Institute Press, Washington, DC).
Gravelle, H.S.E. (1986), "Insurance and corrective taxes in the health care market", Journal of Economics
(Zeitschrift fur Nationalokonomie) (Supplement 5):99-120.
Grossman, M. (1972), "On the concept of health capital and the demand for health", Journal of Political
Economy 80(2):223-255.
Grossman, M., E Chaloupka and I. Sirtalan (1998), "An empirical analysis of alcohol addiction: Results from
the Monitoring the Future panels", Economic Inquiry 36(1):39-48.
Grossman, M., and R. Kaestner (1997), "Effects of education on health", in: J.R. Behrman and N. Stacey,
eds., The Social Benefits of Education (University of Michigan Press, Ann Arbor, MI).
Grossman, M., and E. Rand (1974), "Consumer incentives for health services in chronic illness", in: S.J.
Mushkin, ed., Consumer Incentives for Health Care (Milbank Memorial Fund, New York) 114-151.
Haddix, A.C., S.M. Teutsch, P.A. Shaffer and D.O. Dunet, eds. (1996), Prevention Effectiveness: A Guide to
Decision Analysis and Economic Evaluation (Oxford University Press, New York).
Hafner-Eaton, C. (1993), "Physician utilization disparities between the uninsured and insured. Comparisons
of the chronically ill, acutely ill, and well nonelderly populations". Journal of the American Medical Asso-
ciation 269(6):787-792.
Hemenway, D. (1990), "Propitious selection", Quarterly Journal of Economics 105:1063-1069.
Hirshleifer, J., and J. Riley (1979), "Economics of uncertainty and information: A survey", Journal of Eco-
nomic Literature 17(4):1375-1421.
Homberger, J. (1998), "A cost-benefit analysis of a cardiovascular disease prevention trial, using folate sup-
plementation as an example", American Journal of Public Health 88(1):61-67.
Hurd, M.D., and K. McGarry (1995), "Evaluation of the subjective probabilities of survival in the health and
retirement study", Journal of Human Resources 30(Supplement):S268-S292.
Ippolito, P., and A. Mathios (1990), "Information, advertising and health: A study of the cereal market", Rand
Journal of Economics 21(3):459-480.
Ippolito, P., and A. Mathios (1995), "Information and advertising: The case of fat consumption in the United
States", American Economic Review: Papers and Proceedings 85(2):91-95.
Ippolito, P., and A. Mathios (1993), "New food labeling regulations and the flow of nutrition information to
consumers", Journal of Public Policy and Marketing 12:188-205.
Ippolito, P., and A. Mathios (1996), Information and Advertising Policy: A Study of Fat and Cholesterol Con-
sumption in the United States, 1977-1990 (Bureau of Economics Staff Report, Federal Trade Commission,
Washington, DC).
Johansson, P.-O. (1987), The Economic Theory and Measurement of Environmental Benefits (Cambridge
University Press, Cambridge).
Jones, A.M. (1994), "Health, Addiction, Social Interaction and the Decision to Quit Smoking", Journal of
Jones, A.M., and S.T. Yen (1994), "Individual cigarette consumption and addiction: A flexible limited depen-
dent variables approach", Working Paper (University of Antwerp).
Kenkel, D.S. (1990), "Consumer health information and the demand for medical care", The Review of Eco-
nomics and Statistics 72(4):587-595.
Kenkel, D.S. (1991a), "Health behavior, health knowledge, and schooling", Journal of Political Economy
99(2):287-305.
Kenkel, D.S. (1991b), "What you don't know really won't hurt you", Journal of Policy Analysis and Manage-
ment 10(2):304-309.
Kenkel, D.S. (1994), "The demand for preventive medical care", Applied Economics 26:313-325.
Kenkel, D.S. (1995), "Health insurance and health promotion", Working Paper (Cornell University).
Kenkel, D.S. (1996), "Health promotion and public sector health insurance: International evidence", Working
paper (Cornell University).
Kenkel, D.S. (1997), "Self-insurance and worksite alcohol programs: An econometric analysis", Journal of
Studies on Alcohol 58(2):211-219.
Kenkel, D.S., and D. Supina (1992), "Determinants of worksite health promotion", Economics Letters
40:345-351.
Kenkel, D.S., and J.V. Terza (1997), "Estimating the effect of physician advice on alcohol consumption: Count
data regression with an endogenous treatment effect", Working Paper (Comell University).
Kottke, T.E., R.N. Battista, G.H. DeFriese, and M.L. Brekke (1988), "Attributes of successful smoking cessa-
tion interventions in medical practice", Journal of the American Medical Association 259(19):2883-2889.
Landsburg, S.E. (1993), "Selection in insurance markets", Working Paper (University of Rochester).
Lawrence, R.S. (1990), "Diffusion of the U.S. Preventive Services Task Force recommendations into prac-
tice", Journal of General Internal Medicine 5(5 Supplement):S99-S103.
Leigh, J.P., (1990), "Schooling and seatbelt use", Southern Economic Journal 57:195-210.
Lewit, E.M., D. Coate and M. Grossman (1981), "The effects of government regulation on teenage smoking",
Liljas, B. (1998), 'The demand for health with uncertainty and insurance", Journal of Health Economics
17(2):153-170.
Lillard, L., W. Manning, C. Peterson, N. Lurie, G. Goldberg and C. Phelps (1986), Preventive Medical Care:
Standards, Usage, and Efficacy (Rand Corporation, Santa Monica, CA).
Luft, H. (1978), "Why do HMOs seem to provide more health maintenance services?", Milbank Fund Quar-
terly/Health and Society 56:140-168.
Manning, W.G., E.B. Keeler, J.P. Newhouse, E.M. Sloss and J. Wasserman (1991), The Costs of Poor Health
Habits (Harvard University Press).
McGinnis, J.M., and W.H. Foege (1993), "Actual causes of death in the United States", JAMA 270(18):2207-
2212.
Meltzer, D. (1997), "Accounting for future costs in medical cost-effectiveness analysis", Journal of Health
Mullahy, J. (1994), "Medicaid and the timing of preventive care for young children", Paper presented at the
NBER.
Mullahy, J. (1999), "It'll only hurt a second? Microeconomic determinants of who gets flu shots", Health
Mushkin, S.J. (1962), "Health as an investment", Journal of Political Economy 70(2, Supplement): 129-157.
1718 D.S. Kenkel
Muurinen, J.M. (1982), "Demand for health: A generalized Grossman model", Joumal of Health Economics
1:5-28.
National Cancer Institute (1997), The Nation's Investment in Cancer Research: A Budget Proposal for Fiscal
Year 1999 (National Cancer Institute, National Institutes of Health).
Newhouse, J., and the Insurance Experiment Group (1993), Free for All? Lessons from the RAND Health
Insurance Experiment (Harvard University Press, Cambridge, MA).
Nordquist, G., and S.Y. Wu (1976), "The joint demand for health insurance and preventive medicine", in:
R. Rosett, ed., The Role of Health Insurance in the Health Services Sector (National Bureau of Economic
Research, New York).
Organization for Economic Co-operation and Development [OECD] (1992), "The reform of the health sys-
tem in the United Kingdom", in: The Reform of Health Care: A Comparative Analysis of Seven OECD
Countries (OECD Publications, Paris, France).
Organization for Economic Co-operation and Development [OECD] (1994), "New orientations for social
policy", OECD Social Policy Studies No. 12 (OECD Publications, Paris, France).
Pauly, M. (1986), "Taxation, health insurance, and market failure in the medical economy", Journal of Eco-
nomic Literature 24:629-675.
Phelps, C.E. (1978), "Illness prevention and medical insurance", Journal of Human Resources 13:183-207.
Phelps, C.E. (1988), "Death and taxes - An opportunity for substitution", Journal of Health Economics
7(1):1-24.
Phelps, C.E. (1992), Health Economics (Harper-Collins, New York, NY).
Phillips, C.V., and R. Zeckhauser (1996), "Communicating the health effects of consumer products: The
case of moderate alcohol consumption and coronary heart disease", Managerial and Decision Economics
17:459-470.
Philipson, T. (1996), "Private vaccination and public health: An empirical examination for U.S. measles",
Journal of Human Resources 31:611-630.
Philipson, T. (2000) "Economic epidemiology and infectious diseases", in: A.J. Culyer and J.P. Newhouse,
Philipson, T.J., and R.A. Posner (1994), "Public spending on AIDS education: An economic analysis", Journal
of Law and Economics 37:17-38.
Philipson, T.J., and R.A. Posner (1995), "A theoretical and empirical investigation of the effects of public
health subsidies for STD testing", Quarterly Journal of Economics 110(2):445-474.
Pickett, G., and W.F. Bridgers (1987), "Prevention, declining mortality rates, and the cost of Medicare",
American Journal of Preventive Medicine 3(2):76-80.
Picone, G., M. Uribe and R.M. Wilson (1998), "The effect of uncertainty on the demand for medical care,
health capital, and wealth", Journal of Health Economics 17(2):171-186.
Public Health Service (1990), Healthy People 2000: National Health Promotion and Disease Prevention Ob-
jectives (US Department of Health and Human Services, Washington, DC).
Public Health Service, and US Department of Health and Human Services (1992), "1992 National survey of
worksite health promotion activities: Final report", Prepared by Prospect Associates and Response Anal-
ysis Corporation, PB93-100204 (US Department of Commerce National Technical Information Service,
Springfield, VA).
Rosenzweig, M.R., and T.P. Schultz (1983), "Estimating a household production function: Heterogeneity, the
demand for health inputs, and their effects on birth weight", Journal of Political Economy 91(5):723-46.
Rosenzweig, M.R., and T.P. Schultz (1988), "The stability of household production technology: A replica-
tion", Journal of Human Resources 23(4):535-549.
Russell, L. (1986), Is Prevention Better than Cure? (The Brookings Institution, Washington, DC).
Russell, L. (1987), Evaluating Preventive Care: Report on a Workshop (The Brookings Institution, Washing-
ton, DC).
Russell, L. (1994), Educated Guesses: Making Policy about Medical Screening Tests (University of California
Press, Berkeley and Los Angeles, CA).
Schauffler, H.H., and T. Rodriguez (1994), "Availability and utilization of health promotion programs and
satisfaction with health plan", Medical Care 32(12):1182-1196.
Schlesinger, H., and E. Venezian (1986), "Insurance markets with loss-prevention activity: Profits, market
structure, and consumer welfare", Rand Journal of Economics 17(2):227-238.
Schneider, L., B. Klein and K.M. Murphy (1981), "Governmental regulation of cigarette health information",
Schoenbaum, M. (1997), "Do smokers understand the mortality effects of smoking? Evidence from the health
and retirement survey", American Journal of Public Health 87(5):755-759.
Selden, Th. (1993), "Uncertainty and health care spending by the poor: The health capital model revisited",
Shavell, S. (1979), "On moral hazard and insurance", Quarterly Journal of Economics 93(4):541-562.
Short, P., and D. Lefkowitz (1992), "Encouraging preventive services for low-income children: The effect of
expanding Medicaid", Medical Care 30:76-80.
Smith, V.K., W.H. Desvousges, FR. Johnson and A. Fisher (1990), "Can public information programs affect
risk perceptions?", Journal of Policy Analysis and Management 9(1):41-59.
Sobal, J., D. Revicki and B.R. DeForge (1992), "Patterns of interrelationships among health-promotion be-
haviors", American Journal of Preventive Medicine 8(6):351-359.
Strauss, J., and D. Thomas (1998), "Health, nutrition, and economic development", Journal of Economic
Tengs, T.O., et al. (1995), "Five-hundred life-saving interventions and their cost-effectiveness", Risk Analysis
15(3):369-390.
Tengs, T.O., and J.D. Graham (1996), "The opportunity costs of haphazard social investments in life-saving",
in: R.W. Hahn, ed., Risks, Costs, and Lives Saved: Getting Better Results from Regulation (Oxford Uni-
versity Press, New York) 167-182.
Thompson, G.B., L.G. Kessler and L. Boss (1989), "Breast cancer screening legislation in the United States:
A commentary", American Journal of Public Health 79(11):1541-1543.
Thompson, R.S. (1996), "What have HMOs learned about clinical prevention services: An examination of the
experience at Group Health Cooperative of Puget Sound", The Milbank Quarterly 74(4):469-509.
Tolley, G.S., D.S. Kenkel and R. Fabian (1994), Valuing Health for Policy: An Economic Approach (Univer-
sity of Chicago Press, Chicago).
Tulchinsky, T.H., and E.A. Varavikova (1996), "Addressing the epidemiologic transition in the former Soviet
Union: Strategies for health system and public health reform in Russia", American Journal of Public Health
86(3):313-320.
US Department of Health and Human Services [USDHHS] (1989), "Reducing the health consequences of
smoking: 25 years of progress", A Report of the Surgeon General, DHHS Publication No. (CDC) 89-8411
(US Department of Health and Human Services, Public Health Service, Centers for Disease Control, Center
for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health).
US Preventive Services Task Force (1989), The Guide to Clinical Preventive Services: An Assessment of the
Effectiveness of 169 Interventions (Williams and Wilkins, Baltimore, MD)
US Preventive Services Task Force (1996), The Guide to Clinical Preventive Services, 2nd edn. (US Depart-
ment of Health and Human Services, Office of Public Health and Science, Office of Disease Prevention
and Health Promotion).
Viscusi, W.K. (1990), "Do smokers underestimate risks?", Journal of Political Economy 98(5):1253-1269.
Viscusi, W.K. (1991), "Age variations in risk perceptions and smoking decisions", Review of Economics and
Viscusi, W.K. (1992), Smoking: Making the Risky Decision (Oxford University Press, New York)
Viscusi, W.K. (1998), "Public perception of smoking risks", Paper prepared for The International Conference
on the Social Costs of Tobacco, Lausanne, Switzerland, August 21-22, 1998.
Warner, K.E. (1984), "Publicity, price and puffing: A health economics contribution to disease prevention",
1720 D.S. Kenkel
Warner, K.E. (1987), "Selling health promotion to corporate America: Uses and abuses of the economic
argument", Health Education Quarterly 14:39-55.
Warner, K.E. (1990), "Wellness at the Worksite", Health Affairs 9:63-79.
Warner, K.E. (1997), "Cost effectiveness of smoking-cessation therapies: Interpretation of the evidence and
implications for coverage", Pharmacoeconomics 11(6):538-549.
Warner, K.E., T.M. Wickizer, R.A. Wolfe, J.E. Schildroth and M.H. Samuelson (1988), "Economic implica-
tions of workplace health promotion programs: Review of the literature", Journal of Occupational Medicine
30:106-112.
Warner, K.E., J. Slade and D.T. Sweanor (1997), "The emerging market for long-term nicotine maintenance",
Warner, K.E., and P.A. Warner (1993), "Is an ounce of prevention worth a pound of cure? Disease prevention
in health care reform", Journal of Ambulatory Care Management 16(4):38-49.
Wechsler, H., S. Levine, R.K. Idelson, M. Rohnnan and J.O. Taylor (1983), "The physician's role in health
promotion - A survey of primary care physicians", New England Journal of Medicine 308(2):97-100.
Wells, K.B., C.E. Lewis, B. Leake and J.E. Ware (1984), "Do physicians preach what they practice? A
study of physicians' health habits and counseling practices", Journal of the American Medical Associa-
tion 252(20):2846-2848.
Wells, K.B., C.E. Lewis, B. Leake, M.K. Schleiter and R.H. Brook (1986), "The practices of general and
subspecialty internists in counseling about smoking and exercise", American Journal of Public Health
76(8):1009-1013.
Weisbrod, B.A. (1971), "Costs and benefits of medical research: A case study of poliomyelitis", Journal of
Weisbrod, B.A. (1991), "The health care quadrilemma", Journal of Economic Literature 29(2):523-552.
Wiseman, V., and G. Mooney (1998), "Burden of illness estimates for priority setting: A debate revisited",
Health Policy 43:243-251.
Wolf, C. (1979), "A theory of non-market failures", Journal of Law and Economics 22(1):107-139.
World Bank (1993), World Development Report 1993: Investing in Health (Oxford University Press, New
York).
World Health Organization (1997), Tobacco or Health: A Global Status Report (The World Health Organiza-
tion, Geneva).
World Health Organization Brief Intervention Study Group (1996), "A cross-national trial of brief interven-
tions with heavy drinkers", American Journal of Public Health 86(7):948-955.
Zarkin, G.A., and S.A. Garfinkel (1994), "The relationship between employer health insurance characteristics
and the provision of employee assistance programs", Inquiry 31:102-114.
Zweifel, P., and F. Breyer (1997), Health Economics (Oxford University Press, New York).
Chapter 32
THE MEASUREMENT OF HEALTH-RELATED QUALITY

OF LIFE FOR USE IN RESOURCE ALLOCATION
DECISIONS IN HEALTH CARE
PAUL DOLAN
Sheffield Health Economics Group andDepartment of Economics, University of Sheffield, Sheffield
Contents
Abstract 1724
Keywords 1724
1.1. How are health outcomes to be measured? 1725
1.2. What is required of an outcome measure? 1727
2. Theoretical considerations 1729
2.1. What is to be valued? 1729
2.2. How is it to be described? 1731
2.3. How is it to be valued? 1732
2.4. Who is to value it? 1738
2.5. How are values for all health states to be generated? 1739
2.6. How are valuations to be aggregated? 1740
3. Empirical considerations 1740
3.1. How valid are the assumptions of the QALY model? 1740
3.1.1. Utility independence 1740
3.1.2. Constant proportional trade-off 1741
3.1.3. Risk attitude over life years 1741
3.1.4. A zero rate of time preference 1742
3.1.5. Stability of lifetime preferences 1742
3.1.6. Additive separability 1743
3.2. What is the "best" outcome measure? 1743
3.2.1. Criteria for assessing a health outcome measure 1743
3.2.2. Evaluating descriptive systems 1744
3.2.3. Evaluating valuation methods 1745
3.3. Do valuations differ according to respondent characteristics? 1747
3.4. How have values for all health states been generated? 1748
3.5. How have valuations been aggregated? 1748
1724 P Dolan
4. Discussion 1749
4.1. A resume 1749
4.1.1. The QALY model 1749
4.1.2. The description of health 1749
4.1.3. The valuation of health 1750
4.1.4. The source of values 1750
4.1.5. The calculation of values for all health states 1750
4.1.6. The aggregation of values 1750
4.2. A research agenda 1751
4.2.1. Valuing profiles or states 1751
4.2.2. Valuing health or health gain 1751
4.2.3. Time preference 1751
4.2.4. "Economic" validity 1751
4.2.5. Harmonisation 1752
4.2.6. The implications of using different values 1752
4.3. The process of eliciting valuations 1752
4.4. Concluding remarks 1754
References 1755
Abstract
An important consideration when establishing priorities in health care is the likely effect
that alternative allocations will have on the health-related quality of life (HRQoL) of the
relevant population. This chapter considers some of the important issues surrounding
the description and valuation of HRQoL. It discusses six main questions that need to be
addressed when measuring HRQoL: What is to be valued?; How is it to be described?;
How is it to be valued?; Who is to value it?; How are values for all health states to
be generated?; and How are valuations to be aggregated? Since it is difficult to answer
many of these questions on theoretical grounds alone, the chapter considers whether the
existing empirical evidence can provide more definitive answers. Many important yet
unresolved issues emerge and directions for future research are suggested. It is argued
that this research agenda should have the gathering and analysis of qualitative data at its
forefront.
Keywords
cost-utility analysis, health-related quality-of-life, utility measurement, health state

valuation, quality-adjusted life-years (QALYs), healthy-years equivalents (HYEs)
Ch. 32: The Measurement of Health-Related Quality of Life 1725
1. Introduction
In all countries throughout the world, there is a considerable degree of government

involvement in the market for health care. If governments are going to deploy the re-
sources at their disposal where they will be of greatest benefit, then information on the
benefits associated with alternative allocations is required. This raises questions about
how benefit is to be defined and measured and how it is to be allocated amongst the
population. This chapter addresses the first of these questions only, concentrating on the
benefits an individual derives from health care.
For the purposes of this chapter, the benefit that a patient derives from a particular
health care intervention is defined according to enhanced quality and/or length of life.
This is defined as a health outcome. By defining a health outcome in this way, this chap-
ter can properly be regarded as contributing to the methodology of cost-utility analysis
rather than to that of cost-benefit analysis which, by requiring the monetary valuation
of all of the benefits derived by all of the members of the community affected by an
intervention, can in principle include the option and non-use values associated with that
intervention. Of course, whether it is appropriate to include such values is another ques-
tion which is beyond the scope of this chapter. This definition of a health outcome may
also ignore the additional benefit that a patient derives from being given the opportu-
nity to participate in decisions about her treatment [Mooney (1994)]. There is debate in
the literature about whether this could be incorporated into a health outcome measure
but if a health care programme produces benefits in these or other ways, they could in
principle be included as a reduction in the costs of the programme.
The structure of this chapter is as follows. It begins by asking how health outcomes
are to be measured and then, in the context of determining priorities in health care,
considers what is required of an outcome measure. Section 2 then discusses the six
main methodological questions that need to be addressed when measuring the health-
related quality of life associated with a health outcome; namely, What is to be valued?,
How is it to be described?, How is it to be valued?, Who is to value it?, How are values
for all health states to be generated?, and How are valuations to be aggregated? The
message from this section is that it is very difficult to answer many of these questions
on theoretical grounds alone, and so Section 3 considers whether the existing empirical
evidence can provide more definitive answers. Many important yet unresolved issues
emerge during the course of this discussion and Section 4 discusses those that the author
considers to be the most pressing.
1.1. How are health outcomes to be measured?
Traditionally the impact of health care has been measured in terms of its effect on mor-
tality [McKeown (1979)]. However, in developed countries, the complete elimination of
mortality before the age of 50 would increase life expectancy by only about three years
[Mathers (1997)]. And, of course, health is much more than merely being alive and there
is clearly a need to say something about the health of the majority of people who do not
1726 P Dolan
suffer premature death. There now exist many different ways in which morbidity can be
defined and measured but morbidity data are also of limited value since these data say
nothing about what the severity and priority weights given to one condition relative to
another ought to be. Therefore, it is necessary not only to say what conditions people
are suffering from but also to say something about the impact that these conditions have
on their lives. This is what the measurement of health-related quality-of-life (HRQoL)
aims to achieve.
It is important to distinguish HRQoL from more clinically focused measures, such as
the extent to which tumour size, blood pressure or cholesterol level is reduced through
treatment. The distinction is made clear in a study by Jachuk et al. (1982) in which
they show that doctors and patients have very different ideas about whether or not anti-
hypertensive treatment results in improvements in health. The doctors (who considered
treatment to be successful in all cases) were thinking about whether the patient's blood
pressure was reduced or not (and the likely effect that this would have on their future
health) whilst the patients (half of whom felt no change in health status and some of
whom actually felt worse with treatment) were concerned with the effect that the treat-
ment had on their current HRQoL. The study also highlighted the fact that patients
and their partners can also have different perceptions about the impact of treatment on
HRQoL itself: all except one of the partners thought that the patient's health had got
worse as a result of treatment. This finding raises the question, which is discussed in
Section 2.4, of whose perspective HRQoL is to be measured from?
To capture the health effects of policies and programmes, it is necessary to say some-
thing about their effects on life expectancy and HRQoL. There now exists a number of
approaches which try to combine these two attributes into a single composite measure.
The Disability Adjusted Life Year (DALY) has been developed in order to calculate the
loss, expressed in terms of years of life in full health, associated with premature mortal-
ity and morbidity. Premature mortality has been calculated using life tables with values
of 80 years at birth for males and 82.5 years at birth for females whilst the HRQcL
(morbidity) weights were calculated by asking a panel of health care providers to assign
a value to each of 22 health states [Murray (1996)].
The Quality Adjusted Life Year (QALY) approach "assigns to each period of time
a weight, ranging from 0 to 1, corresponding to the health-related quality-of-life dur-
ing that period, where a weight of 1 corresponds to optimal health, and a weight of 0
corresponds to a health state judged to be equivalent to death" [Weinstein and Stason
(1977)]. The QALYs relating to a health outcome are then expressed as the value given
to a particular health state multiplied by the length of time spent in that state. Therefore,
QALYs measure health outcomes in terms of gains in health as compared to DALYs
which measure them as losses from normative benchmark. As a result, QALYs can be
used to compare the benefits of medical interventions in clinical trials (i.e. within diag-
nostic groups) and, through the calculation of cost-per-QALY ratios, can also be used to
determine priorities between diagnostic groups. For these reasons, the QALY approach
is the preferred one for the purposes of this chapter. The assumptions of the QALY
model are discussed in some detail in Sections 2 and 3.
1.2. What is requiredof an outcome measure?
Whilst there are methodological problems associated with the estimation of life ex-
pectancy, it is the question of how to attach weights to different levels of HRQoL that
is the central concern here and the many theoretical and empirical issues raised are dis-
cussed in some detail in the next two sections. Before this discussion can take place, it is
necessary to discuss the required properties of a HRQoL measure are. This will depend
upon what the measure is designed to do.
There are essentially four ways in which information on HRQoL can be used: (i) in
randomised controlled trials to determine whether a particular intervention produces
greater gains in terms of HRQoL or fewer losses in terms of adverse side effects; (ii) in
individual patient care to improve clinical practice; (iii) in describing, monitoring and
predicting overall levels of population heath; and (iv) in informing resource allocation
decisions in health care. It is the last of these uses that is of interest here. This means
that any measure of HRQoL must enable comparisons between different health care
programmes which may impact upon different aspects of HRQoL in different ways.
When determining priorities in any area of public policy, it is important to be clear
about the objectives of the policy. In very general terms, resources will be allocated to
achieve both an efficient and an equitable outcome but, since neither objective can be
fully satisfied, there will be a trade-off between them. However, before this trade-off
can be determined, it is necessary to define what is meant by efficiency and equity in
the context of health and health care. Economists are divided on the definition of both
concepts.
"Welfarists" believe that the output of health care should be judged according to the
extent to which it contributes to overall welfare, as determined by individual preferences
over health relative to other arguments in the utility function. Welfarists often stress the
importance of equality of access to effective health care, so that, whilst individuals
face the same opportunities to use health care, the exercising of individual choice is a
legitimate reason for differences in health [Herrero (1996)]. "Extra-welfarists" define
the output of health care according to its contribution to health itself. This school of
economists will typically focus attention on equality of health, whereby it is justifiable,
at the margin, to override individual preferences in the interests of improvements in
the overall level or distribution of health. This is justified on the grounds that health
is fundamental to an individual's capacity to "flourish" as a human being [Culyer and
Wagstaff (1993a)].
Against this background, in assessing the effectiveness or otherwise of the health
care system, it is sufficient for welfarism to focus attention on people's behaviour
whilst extra-welfarism requires psychological and philosophical investigation into peo-
ple's well-being. However, and crucial in the context of this chapter, what constitutes
a health improvement is to some extent defined according to individual preferences by
welfarists and extra-welfarists alike. Whilst the role that individual preferences should
play in determining priorities in health and elsewhere is a matter of intense debate,
there is general agreement that preferences should have some role. Therefore, for the
1728 P Dolan
purposes of informing allocation decisions, measures of HRQoL must, at least in part,

be preference-based.
In considering efficiency, the debate between welfarists and extra-welfarists is more
about how to use individual preferences once they have been elicited. Specifically, when
comparing alternatives, it is about whether or not the same health state should be given
the same value across all individuals. Welfarists take each individual to be the source of
their own health state value and would therefore suggest not. Because it has been shown
that many people in poor health are able successfully to compensate for it [Meyerowitz
(1983) and Cassileth et al. (1984)], this approach may have damaging limitations in
the context of interpersonal comparisons [Sen (1987)]. For example, we might choose
to return to full health a less seriously ill person over a more seriously ill one on the
grounds that the latter has learnt to cope with her condition better than the former.
Because of this problem, extra-welfarists assume that a particular health state has the
same effect on all individuals. But, of course, this raises interpersonal comparability
problems of its own. For example, the condition of paraplegia is likely to have a much
more profound effect on the life of a builder than on the life of an academic yet assigning
the same value to that condition would imply that the utility loss was the same for
both people. There appears to be no simple solution to this problem and it remains a
normative judgement about whether and in what circumstances to adopt a welfarist or
an extra-welfarist approach.
Issues relating to the definition and measurement of equity are discussed by Williams
and Cookson (2000). The important point for the purposes of this chapter is that taking
account of equity and distributional considerations is not inconsistent with the measure-
ment of individual utility, nor is it inconsistent with the interpersonal comparison of
individual utilities. Indeed, the measurement of individual utilities provides the flexibil-
ity to formulate (and subsequently revise) any number of possible health-related social
welfare functions (HRSWFs). And the process of collapsing individual utilities into an
overall HRSWF makes explicit the assumptions and philosophical basis upon which
such aggregation is based [Dolan (1998)].
An additional consideration relates to externalities: in the context of this discussion,
the extent to which one person's health status might affect another person's utility. At
one level, such information is included in a preference-based HRQoL measure since,
when expressing their own preferences over different health states, people are allowed
to take into account any factors they consider to be relevant; including the effect that
the health states might have on other people. However, to the extent that they disregard
considerations others may have about their health, then such externalities are ignored.
This raises many issues concerning the meaning, role and relevance of altruism in health
care decision-making; issues which are beyond the scope of the present chapter. How-
ever, information about an individual's preferences over different levels of HRQoL is
essential if such issues are to be addressed adequately.
To recap, then, in order to inform resource allocation decisions, any measure of
HRQoL must satisfy two conditions; it must enable comparisons between different
health care programmes which may impact upon different aspects of HRQoL in dif-
ferent ways and it must reflect individual preferences.
2. Theoretical considerations
2.1. What is to be valued?
In very general terms, the answer to this question is a simple one; it is the alternative
types and levels of HRQoL that an individual might experience over the course of a
lifetime. This profile of health clearly consists of two components; namely, the different
states of health that an individual might be in and the different lengths of time spent in
each state. One approach, then, would be to construct profiles for each possible life path
and then to elicit individual preferences over them. Such profiles might also contain
probabilistic information on the likelihood of experiencing alternative "sub-profiles" at
various points in the overall life path.
This approach has the advantage that it places few restrictions on individual prefer-
ences. For example, each individual is allowed to determine the relative weights they
wish to attach to the order or timing of particular health states, or to the effect that dif-
ferent lengths of time in any given state would have on them. The valuation of profiles
of health is the key feature of the healthy-years equivalent (HYE) approach which asks
individuals to state the number of years in perfect health that are considered equivalent
to a particular profile [Mehrez and Gafni (1989)]. In order for the number of HYEs
to be a valid representation of an individual's preferences over different profiles, it is
necessary to assume that the individual is risk neutral with respect to (discounted or
undiscounted) years in full health (i.e. if utility on the y-axis is plotted against length of
life on the x-axis, then the resulting utility function is linear).
Less restrictive still is the "ex ante HYE" in which the health profile is framed in
terms of uncertainty (i.e. a probability distribution), which only makes the assumption
that preferences are monotonic with respect to years of life in full health [Johannesson
(1995); see Cook et al. (1994) for one of the few empirical investigations into valuing a
health profile containing probabilistic information].
But the problem with the HYE approach is that in most contexts there will be a
large number of possible profiles of health, each of which would require preference
measurement. To allow greater generalisability, an alternative approach would be to
elicit preferences for one health state (of a specified duration) at a time. The value
of any given profile of health could then be estimated by taking the (discounted or
undiscounted) weighted average of the value for each of the health states in that profile
multiplied by the time spent in each state. This is the approach adopted in the calculation
of QALYs. Of course, this places greater restrictions on individual preferences since a
number of assumptions have to be made when calculating this weighted average.
Pliskin et al. (1980) established three conditions which they claimed must hold for
the QALY model to represent individual preferences over a health profile of constant
1730 P. Dolan
quality; utility independence, constant proportional trade-off, and risk neutrality over
life years. Utility independence means that the utility, U, of a particular health profile,
consisting of a given health state, Q, and a given number of years, T, can be calculated
by multiplying the utility of Q by the utility of T, i.e.
U(Q, T) = U(Q)U(T), (1)
where U(Q) depends only on the health state irrespective of the number of life years
and U(T) depends only on the number of life years irrespective of the health state.
Constant proportional trade-off entails that an individual is willing to sacrifice a constant
proportion of remaining life years in order to achieve a given improvement in HRQoL,
irrespective of the number of life years that remain. Risk neutrality over life years means
that an individual is risk neutral with respect to gambles over life years for all health
states. If these assumptions hold then all life years have the same utility:
U(T 1 ) = U(T2 ) = U(T'). (2)
Thus, the value of a health state is now linearly related to the time spent in that state:
U(Q, T) = U(Q)T. (3)
Miyamoto and Eraker (1985) developed a less restrictive model which allows the indi-
vidual to have any risk posture with respect to gambles over life years, represented by
the parameter r:
U(Q, T) = U(Q)Tr. (4)
This still requires r to be independent of the health state and for the individual to exhibit
constant proportional risk attitude: if the gamble outcomes (in terms of years of life) are
doubled then the certainty equivalent number of years of life should also double.
The model developed by Bleichrodt et al. (1997) is less restrictive still. They show
that if the value of all health states is the same when the time spent in those states is zero
years (which they refer to as the "zero-condition" and consider to be "unobjectionable
in the medical context"), then risk neutrality is all that is required for the QALY model
to hold. Although, strictly speaking, risk neutrality on life years requires only that an
individual exhibits risk neutrality over discounted years of life, most authors define and
measure risk posture according to the assumption that each year of life yields a constant
marginal utility. Therefore, whilst attitudes towards time are often (and mistakenly)
subsumed within attitudes towards risk, it is important to be clear that the standard
QALY model assumes both risk neutrality and neutrality towards the timing of events.
In addition, it is often assumed that preferences are stable over a lifetime, i.e. that
the utility of Q is constant across all single-period utility functions [Bleichrodt (1995)].
Finally, for health profiles characterised by changes in HRQoL, it is necessary for one
additional assumption to hold in the QALY model; namely, additive separability. This
means that each individual's utility function is strongly separable on the time dimension,
i.e. when Q changes over time, the utility derived from the whole profile is equal to the
sum of the utility derived from each health state. In other words, the value attached to a
particular health state is independent of the state(s) that precede or follow it i.e.,
U(Q1 , Q2... Q; T1 , T 2 .. T n)
= U(Q1 )T 1
+ U(Q2 )T2 +,, + U(Q)T n. (5)
Thus, the enhanced generalisability that comes from constructing an almost infinite
number of profile scores (QALYs) from valuations of a finite number of composite
states comes at a price; namely, a number of restrictive assumptions about individual
preferences. The validity of these assumptions is discussed in Section 3 below. Without
pre-empting this discussion two important observations can be made. First, although
fewer assumptions are made in deriving HYEs than in deriving QALYs, this does not
necessarily mean that the estimated number of HYEs will differ from the estimated
number of QALYs. Second, it does not mean that the HYE approach is necessarily a
better representation of individual preferences [Culyer and Wagstaff (1993b)].
2.2. How is it to be described?
With the exception of the profile in which the individual spends the rest of a life in
their current state of health, whichever approach is adopted a decision is required about
the way in which health is to be described. One method was a scenario in which a
relatively detailed written description of a state of health is presented to respondents.
This approach suffers from similar generalisability problems to the HYE and has been
used primarily in the context of choosing between treatments for the same patient where
it is important to be able to detect relatively small changes in HRQoL.
More often, states of health have been described using a health state descriptive sys-
tem, and there now exists a number of different types, each designed for a particular
purpose. For example, condition-specific instruments are designed to measure HRQoL
within a particular condition or disease group. They typically contain descriptions of a
limited number (often only one) of the dimensions of health, since they are designed
to be sensitive to small changes within the dimension(s) relevant to the particular con-
dition. Attempts have been made to establish an "exchange rate" between the different
condition-specific measures so as to facilitate comparisons across disease groups but
this has proved problematic given the diverse nature q dimensions across measures
[Cairns et al. (1991)].
Generic measures have been developed which measure health status across a range
of different dimensions and, as a result, these are typically less sensitive than condition-
specific measures. Most of these measures consist of a generic health profile (which is
to be distinguished from the use of the term in the foregoing discussion) which allow
1732 P Dolan
a comparison of health within each dimension independently but do not allow the dif-
ferent dimensions to be combined to form an overall single index. This is because the
scores within each dimension are not comparable with one another and/or because the
relative weight attached to each dimension is not known. And for many generic profiles
it would be impractical to try to generate overall scores for all health states because
the combination of the various levels of the different dimensions would typically gen-
erate a universe of states that is too large to elicit indices for; e.g., the SF-36 [Ware and
Sherbourne (1992)] generates over ten million possible health states.
Therefore, neither condition-specific nor profile measures are suitable for use in in-
forming resource allocation decisions across a range of diverse interventions. Those that
are suitable include the Quality of Well-Being Scale [Patrick et al. (1973)], the McMas-
ter Health Utility Index [Feeny et al. (1995)], the EQ-5D [Brooks (1996)], and the 15-D
[Sintonen (1994)]. The relative merits of these measures are discussed in Section 3.2.2.
2.3. How is it to be valued?
If changes in HRQoL are to be quantified, once a health state descriptive system has
been developed, the next question is how best to determine values attached to the health
states. In determining these values, there are three broad strategies: (i) use expert judge-
ment, (ii) use indices obtained from relevant literature, or (iii) use direct measurement
of the preferences of an appropriate population [Torrance (1986)]. There are a number
of potential sources of bias associated with the first two strategies; for example, experts
may focus on different attributes than patients [Jachuk et al. (1982)] or valuations re-
ported in one study may be inappropriate for use in another study. Therefore, the third
strategy is generally seen as the most appropriate, particularly since it accords with
standard welfare economics in which individuals are taken to be the best judges of their
own well-being and which suggests that resource allocation decisions should reflect the
preferences of those who will be affected by these decisions.
How valuations should be derived raises the central question of which valuation
method(s) should be used. An important consideration when answering this question
is the level of measurement that is required. When considering resource allocation deci-
sions in health care, information is required not only on the ordering of preferences but
also on their intensity [Ng (1992)]. By going beyond ordinal preferences and requiring
cardinal measurement, problems associated with Arrow's General Possibility Theorem
[Arrow (1951)] are avoided [Keeny (1976)]. For most purposes, including cost-utility
analysis, it is necessary for states to be expressed on an interval scale [Lipscomb (1982)]
which provides information on how far apart states are in terms of severity. Therefore,
only those methods that in principle generate such a scale will be considered here.
Typically, the aim is to represent health on a scale whereby death and full health are
assigned values of 0 and 1 respectively. Therefore, states rated as better than dead have
values between 0 and 1 and states rated as worse than dead have negative scores. These
negative scores are usually expressed in a ratio scale which in principle bounded by
negative infinity. The three methods that have been widely used to generate valuations
with internal state properties for states ruled as better than dead are the visual analogue
scale (VAS), the standard gamble (SG) and the time trade-off (TTO). A fourth method,
the person trade-off (PTO), is increasingly being used to generate health state valua-
tions [Nord (1995)]. By asking respondents to compare two treatments that will benefit
different numbers of people, it elicits, at least in part, preferences over the distribution
of benefits and is considered by Williams and Cookson (2000).
The VAS requires respondents to rate health states on a scale (typically represented
by a vertical "thermometer-type" line) with "worst" and "best" endpoints, usually rep-
resented by 0 and 100, respectively. The labels of these endpoints are often death and
full health although, to allow for the possibility that some states may be rated as worse
than dead, some researchers (for example, those using the EQ-5D) have labelled these
endpoints "worst imaginable health state" and "best imaginable health state". Accord-
ing to psychometric theory, VAS valuations will have interval scale properties if they
are elicited using the "bisection" method [Stevens (1971)]. This requires respondents
first to rate the states they consider to be the best and the worst, then to rate the state
whose value they consider to be roughly halfway between the values assigned to the two
extreme states, then to rate the state whose value is roughly halfway between this mid-
state and the best state, then to rate the state whose value is roughly halfway between
the mid-state and the worst state, and so on.
The SG asks the respondent to choose between the certainty of an intermediate health
state and the uncertainty of a treatment with two possible outcomes, one of which is
better than the certain outcome and one of which is worse. For a chronic state, Qb,
rated as better than dead, the intermediate state is Qb and the treatment outcomes are
full health and death, respectively. For a chronic state, Q, rated as worse than death,
the intermediate state is death and the treatment outcomes are full health and Qw. In
both cases, the object is to find the probability, p, attached to the better of the two
uncertain outcomes at which the respondent is indifferent between the certain and the
uncertain alternatives. The value for Qb is given by p, whilst the value for Qw is given
by -p/(l - p).
The TTO asks the respondent to choose between two alternatives. For a chronic state,
Qb, rated as better than dead, the first alternative is to live for a defined period of time,
t, in Qb and then die. The second alternative is to live for a shorter period of time in full
health and then die. For a chronic state, Q, rated as worse than dead, the first alternative
is to die immediately and the second alternative is a number of years in Qw followed by
a number of years in full health (which combined sum to t). In both cases, the time in
full health, x, is varied until the respondent is indifferent between the two alternatives.
The value for Qb is given by x/t and the value for Qw is given by -x/(x - t). More
detailed descriptions of both the SG and TTO methods, including the procedures for
valuing temporary health states, can be found in Torrance (1986).
Because valuations from the VAS are elicited in a choiceless context, and thus do not
require people to make trade-offs between different arguments in their utility function,
the method is commonly regarded by economists as theoretically inferior to the choice-
based SG and TTO methods. An exception is Broome (1993) who regards a method
1734 P Dolan
he describes as identical to the VAS (although he does not use the term himself) as
"uncontaminated" by factors which he considers to be irrelevant to the measurement of
"goodness"; like risk attitude in the SG and discounting of future utilities in the TTO.
The SG and TTO methods both start from the premise that, given that health is an im-
portant argument in an individual's utility function, we can estimate the welfare change
associated with a change in health if we can determine the compensating change in
one of the remaining arguments in an individual's utility function that leaves utility un-
changed. By assuming that health improvements are a negative function of risk in the
case of the SG and a positive function of longevity in the case of the TTO, both methods
can be viewed as sharing a common theoretical background.
Another important argument in an individual's utility function is wealth and in prin-
ciple it would be possible to measure the extent to which an individual is willing to
sacrifice wealth in order to experience one health profile relative to another. Or, analo-
gous to the discussion in Section 2.1, to elicit their willingness to pay (WTP) for discrete
states of health relative to one another and then to estimate the value of different profiles
by taking the weighted average of the WTP for each of the health states multiplied by
the time spent in each state.
Using WTP to value health profiles or health states has the advantage that it allows
preferences for health to be considered alongside other non-health attributes that the
individual values. In contrast, for the SG and TTO methods to represent utility fully,
preferences over health and non-health attributes must be assumed to be independent.
There is some evidence that this assumption does not hold [Viscusi and Evans (1990)].
However, in valuing a health state, the extent to which a respondent will consider the
consequent non-health effects is likely to be limited. In any event, it is not clear whether
it is appropriate to take account of non-health preferences when valuing health outcomes
(see the brief discussion of the debate between the welfarists and extra-welfarists in
Section 1.2). And, whilst the WTP method is increasingly being used in the health
care context [Olsen and Smith (1998)], it is typically used to value the overall benefits
of health interventions rather than the individual health effects of those interventions
[Donaldson (1990)].
Focusing on the measurement of individual utility, von Neumann and Morgenstern
(1953) in developing Expected Utility Theory (EUT), showed that if a cardinal utility
could be expressed as equivalent to a gamble, under certain assumptions (including tran-
sitivity and independence), it would be a linear function of the risk involved in the gam-
ble. The level of risk involved in SG questions is then linear in utility, thus generating
a scale with internal properties. This has led many to regard the SG as the "gold stan-
dard" for health status measurement [Torrance and Feeny (1989), Gafni (1994)]. How-
ever, there is evidence that people systematically violate the axioms of EUT [Llewellyn-
Thomas et al. (1982), Camerer (1993)]. Indeed, Schoemaker (1982) concludes that the
SG yields only ordinal preference rankings. Thus, much of the appeal of the SG is lost
since it will only be an accurate measure of utility if the axioms of EUT apply [Richard-
son (1994)].
The literature often distinguishes between utility, which results from decisions un-
der uncertainty (as measured by the SG, for example), and value, which results from
decisions based on certainty [Gafni et al. (1993)]. In the TTO both of the alternatives
presented to the respondents have outcomes that are known with certainty, so it is said
to produce a value, not a utility, function [Dyer and Sarin (1982) and Bennett et al.
(1991)]. However, this view is based on a very narrow definition of utility, one that has
arisen as a direct result of EUT [Richardson (1994)]. In its broader sense, and one which
is perhaps more relevant to the measurement of HRQoL, utility is defined as a cardinal
index of strength of preference which can be expressed as an internal on a ratio scale.
It is possible to measure this under conditions of uncertainty or certainty (although, of
course, the utility of a given state of health may differ according to whether attitudes
towards risk are being measured or not).
The SG is also advocated on the grounds that almost all decisions about health care
are made under conditions of uncertainty [Mehrez and Gafni (1991)]. Whilst this is
indeed the case, most of the uncertainty is not of the stark "life or death" form found in
most SG questions. More importantly, the appropriateness or otherwise of a valuation
method is determined by its ability to act as a proxy for utility and not by its capacity to
model the situation being valued [Buckingham and Drummond (1993)]. In this respect,
the TTO may be considered more appropriate since, by definition, it gives the number of
years in full health which are valued equally to a (longer) period in the health profile (or
state) being measured. In other words, it collapses the relationship between the health
profile (or state), its duration and its value into one single measure [Richardson (1994)].
Nevertheless, Dolan and Jones-Lee (1997) have shown that for a response to a TTO
question to provide a direct and unbiased estimate of health state value, it is necessary
that: (i) there is no reallocation of lifetime consumption; and (ii) there is no discounting
of future utilities. It turns out that the effect of lifetime reallocation is likely to be very
small but that discounting can produce significantly biased estimates. Of course, time
preference will play a part in all valuation methods since all profiles and health states
are for a specified duration. But if utility independence holds, SG values will not be
affected by discounting since the time spent in full health and the health state being
valued is the same [Johannesson et al. (1994)]. And since time is used as the means of
valuation itself in the TTO, attitudes towards time will be a more important determinant
of value using this method. It is difficult, then, to choose between SG and TTO on
theoretical grounds since valuations from neither method can automatically be assumed
to map directly onto utility. This is important since it implies rejecting the idea that the
SG should be regarded as the "gold standard" for measuring health state values.
Both the SG and TTO methods aim to produce valuations for states of health, so it is
worth considering the extent to which in theory the methods should produce the same
valuations. At one level, since both methods are used to value the same thing; namely, a
particular state of health, then the SG and TTO should produce the same results. If this
were the case, then much of the heat would be taken out of the debate about which is
the most appropriate method.
1736 P Dolan
However, given two assumptions about individual preferences, it might be expected

that the methods will normally produce different results. The first is that people are risk
averse. This implies that people will be less willing to accept the gamble outcomes in
the SG and more willing to accept the certain outcome. The second assumption is that
people have positive time preference valuing years of life in the near future more highly
than years more distant years. This implies that people will be more willing to give up
years of life at the end of a profile, as in the TTO. Thus, both assumptions imply that,
for the same health states, SG values will be higher than TTO values.
As well as the on-going debate about the relative merits of the SG and TTO, there has
also been a controversy of a different kind. In proposing the HYE, Mehrez and Gafni
(1989) also suggest how the number of healthy years considered equivalent to a given
profile should be calculated. This involves a two-stage SG procedure. In the first stage
the individual is offered a choice between the certainty of (Qi, T) and the uncertainty of
a gamble with a probability, p, of living T years in full health, QFH, and a probability,
1 - p, of dying immediately, QD. The individual then sets the value for p (denoted by
p*) such that
(Q, T) p(QF, T) + ( - p*)(Q, T). (6)
In the second stage, the individual is asked how many years for certain, H*, they would
need to live in QFH in order to be indifferent between this certainty and the uncertainty
of the gamble shown on the right hand side of Equation (6), i.e. to set H* such that
(QFH, H*) _ p*(QFH, T) + (1 _ *)(QD T). (7)
A number of authors [Buckingham (1993), Culyer and Wagstaff (1993b)] have noted
that, since the right hand sides of (6) and (7) are the same, transitivity entails indifference
between both left-hand sides, i.e.
(Q T) (QFH, H*). (8)
A TTO question yields the number of years in QFH, denoted by X*, which is indifferent
to (Qi, T), i.e.
(Qi T) - (QFH, X*) (9)
And so, again by transitivity, there is indifference between both right-hand sides:
(QFH, H*) (QFH, X*). (10)
Assuming strict monotonicity, H* = X*. For an individual with transitive and mono-
tonic preferences, then, the two-stage SG procedure will produce exactly the same result
as that elicited by a TTO question.
This conclusion is refuted by Gafni et al. (1993) who, using the distinction between
value and utility alluded to above, claim that the two-stage SG procedure elicits a utility
function
u(Q, T) = u(QFH, H), (11)
whilst the TTO elicits a value function
v(Q', T) = v(QFH, X*). (12)
Since the value function, v(.), is determined only by an individual's strength of pref-
erence whilst the utility function, u(.), is determined by strength of preference and
attitude towards risk [Dyer and Sarin (1982)], and since most people are assumed to
exhibit relative risk aversion, Gafni et al. argue that v(.) and u(.) will differ. But Dyer
and Sarin define u[v(x)] = u(x) and, according to Loomes (1995), the meaning of this
is clear: "for every value v(.) there is a corresponding (unique) value of u(.), so that
all outcomes which have the same v(-) as each other necessarily have the same u() as
each other". Therefore, if (Qi, T) and (QFH, X*) have the same value and if (Qi, T)
and (QFH, H*) have the same utility, then
u(QFH, X*) = u(QFH, H*), (13)
which requires that X* = H*, i.e. that the two-stage SG and the TTO are equivalent.
In addition, if the two-stage SG procedure will in principle produce the same results
as those elicited by a riskless TTO question, then the former cannot incorporate risk at-
titude. Thus, the risk attitude that is inherent in the first stage of the two-stage procedure
must be exactly cancelled out in the second stage [Wakker (1996), Morrison (1997),
Rittenhouse (1997)]. Therefore, although asking respondents to value health profiles
might more accurately represent their preferences than asking them to value discrete
states that make up those profiles (see the discussion in Section 2.1 above), valuing pro-
files using a two-stage SG procedure cannot "fully represent individual preferences" as
claimed by Mehrez and Gafni (1989).
If differences are observed between responses to two-stage SG and to TTO questions,
then this could be explained by violation of one or both of the two axioms upon which
equivalence is based; namely, transitivity and/or monotonicity. But, as Morrison (1997)
notes, if these assumptions do not hold (i.e. if the utility function is not EU), then it
is inappropriate to use the probability equivalence method (the first stage SG question)
and/or the certainty equivalence method (the second stage SG question). By definition,
then, the two-stage SG procedure assumes that individuals satisfy the axioms of EU.
Therefore, since various authors have shown in various ways that a two-stage SG ques-
tion is equivalent in theory to a TTO question, it is difficult to see what advantages there
are in asking such questions.
1738 PRDolan
2.4. Who is to value it?
Much of the debate in the literature concerns whether the preferences of actual patients
or those of the general public (in their role as potential patients and tax and premium
payers) should be given the most weight when determining the value attached to a par-
ticular health profile or state. But in reality the distinction between these two groups
is very blurred. Even in supposedly "healthy" populations (for example, the general
public), there is a substantial degree of ill health; many currently "healthy" people have
experienced ill health at some time in their lives and many have relatives or close friends
who are currently experiencing ill health. Moreover, the extent to which people are able
to imagine the impact that certain health states will have on them is likely to vary accord-
ing to the likelihood of them experiencing those states; people who expect to experience
certain health states are likely to have given some thought to the possible consequences
of those states [Dolan (1999)].
Maintaining the dichotomy between patients and the public for the sake of exposition,
many consider that it is most appropriate to elicit valuations from those people who are
currently experiencing the health states for which values are sought. It is argued that
these are the only people who know what it is really like to be in those states whilst
those without direct experience of the health states cannot accurately predict the impact
that the states will have on their quality of life. It was noted in Section 1.2 that there is a
direct positive link between the time spent in ill health and adaptation to that ill health,
either through "genuine" adaptation or through "cognitive dissonance", whereby people
adjust their expectations in the light of their changed circumstances [Festinger (1957)].
A question that arises, then, is whether or not adaptation should be taken into account
when allocating resources which will deal with the treatment of prospective patients.
This question is akin to the debate about whether social welfare should be measured
ex ante or ex post. The ex ante approach views social welfare as a function of the ex-
pected levels of utility attained by different individuals. The ex post approach means
that utility is calculated conditional on everybody experiencing the same state of the
world, and then to arrive at the overall level of social welfare, the utility of all the possi-
ble states of the world is weighted by the probability that these states occur. The ex ante
approach allows subjective probabilities to enter the calculus just as eliciting the pref-
erences of the general public allows the perceived future value of given health states to
enter the calculus. Whilst the ex post approach has its proponents, most notably Broome
(1991), most economists adopt the ex ante approach. This reflects the view that, since
resource allocation decisions affect future patients, it is legitimate to give weight to the
preferences of potential patients.
If, however, the extent to which people are able to cope with different conditions is
considered relevant when measuring social welfare then, since adaptation is a gradual
and continuous process, there is still the question of when to ask patients for their prefer-
ences. At one extreme, a value could be elicited immediately after a patient has entered
a given health state, when adaptation is likely to be negligible; at the other extreme, it
could be elicited only after the individual had been in that health state for a consider-
able length of time, after all "coping mechanisms" are in place. The literature has been
almost completely silent on this issue.
There are at least three other questions which need to be addressed when deciding
whether preferences should be elicited from patients or from the general public. First,
to what extent are valuations from different groups susceptible to strategic bias? There
is very little evidence on this subject but bias is potentially most likely amongst patients
who may feel that their treatment will be directly affected by their responses. Second,
is one of the purposes of the health care system to give reassurance to the public? If it
is, then resources should in part be allocated so as to reassure the public that treatment
is available to alleviate the health states they fear the most, even if this fear is in some
way misplaced [Edgar et al. (1998)]. Third, how accountable should resource allocation
decisions be? If they should be accountable to the ultimate numbers then, since the
general public pay for health care, their preferences should be used in the resource
allocation process.
It is common for the preferences of the whole population to be considered the most
relevant when comparing interventions that affect different population sub-groups [Gold
et al. (1996)]. However, when comparing interventions for the same condition, it might
be more appropriate to use the values of the patients experiencing that condition. And
where there exists a pre-determined budget for a particular population sub-group (for
example, the elderly), then it might be appropriate to use only the values of those within
that sub-group. In a similar vein, Daniels (1991) argues that the allocation of resources
to Medicaid patients should depend only on the preferences of Medicaid recipients. In
any event, it is important that valuations are elicited from as many different population
sub-groups as possible since evidence on inter-rater differences will highlight whether
the debate about whose values should count in an empirically unmark one or not.
2.5. How are valuesfor all health states to be generated?
Since for most health state classification systems, it is not feasible to elicit direct val-
uations for all the health states it generates, it is necessary to interpolate some of the
values. There are essentially two different approaches that can be adopted here: the
decomposed approach and the composite approach. The former involves asking the re-
spondent to value each level within a particular dimension assuming that the levels of
all other dimensions are held constant [Torrance et al. (1982)]. This approach therefore
requires few (and in some cases no) valuations of composite health states, which can
subsequently be generated by specifying a multi-attribute function (MAUF). The prob-
lem with the decomposed approach is that the conditions that the MAUF must satisfy
are stringent; the least-restrictive model (in which the MAUF is multilinear) requires
that preferences for various levels of each dimension do not depend upon the particular
levels at which the other dimensions are fixed.
The composite approach, on the other hand, requires respondents to value a subset
of composite health states. An important consideration when choosing these states (and
when choosing a larger subset from which to sample if the number that each respondent
1740 P.Dolan
can value is deemed to be too small) is that they should be widely spread over the val-
uation space so as to include as many combinations of levels across the dimensions as
possible. This is subject to the constraint that the states are likely to be considered plau-
sible by respondents [Gudex et al. (1997)]. The next step, using appropriate regression
or statistical techniques, is to estimate a model which allows valuations for all the health
states to be interpolated from direct valuations on a subset of these [Dolan (1997a)]. The
advantage of this approach is that fewer restrictions need to be placed on the resultant
model.
2.6. How are valuations to be aggregated?
Whilst the issue of whose values should count has received a great deal of attention
in the literature, the question of how these individual responses should be aggregated
has received much less attention. Many economists would argue that the theoretically
correct way to aggregate individual preferences is to calculate the mean from any given
distribution, irrespective of the skewness of that distribution. This is based on the prin-
ciples of standard welfare economics which take account of the strength of each indi-
vidual's preferences. However, an alternative view is that, in the realm of public policy,
each person's valuation should be treated as equal in a voting context and so group
preferences should be expressed in terms of the median.
Because of the lower and, particularly, upper bound on health state valuations, it
is anticipated that valuations for less severe health states would be negatively-skewed
whilst those for more severe states would be positively-skewed. Thus, the transition
from a more severe state to a less severe one would be valued less according to the
mean than the median and hence the choice of the measure of central tendency may
have important implications for resource allocation decisions.
Whichever measure of central tendency is advocated, the starting point is the same;
namely, individual utilities. That some people might disagree with the policy implica-
tions that result when individual utilities are aggregated is not surprising since neither
the mean nor median view is an accurate representation of the preferences of each and
every individual member of the group from whom individual utilities were elicited. But
each individual counts for only one.
3. Empirical considerations
3.1. How valid are the assumptions of the QALY model?
3.1.1. Utility independence
Miyamoto and Eraker (1988) tested the assumption that the utility of length of life is
independent of the utility of quality of life in a sample of 46 coronary artery disease pa-
tients, and found some support for it. However, in a study of 172 students which tested
the converse (that quality of life is utility independent from length of life), Bleichrodt
and Johannesson (1996) found that less than one in seven respondents satisfied this con-
dition. Of those violating utility independence, almost three quarters assigned a lower
value to a given health state when it lasted for 10 years as opposed to 30 years.
3.1.2. Constantproportionaltrade-off
In early work, there was strong evidence to suggest an increasing rather than a con-
stant proportional trade-off, i.e. people will trade-off a larger fraction of their remaining
years of life as the number of these years increases [Sackett and Torrance (1978), Pliskin
et al. (1980), McNeil et al. (1981)]. However, two recent studies have produced some-
what different results. Bleichrodt and Johannesson (1996) found that increasing propor-
tional trade-off was no more common than decreasing proportional trade-off and, as a
result, the mean values for states lasting 10 years and for 30 years were almost identical.
Stalmeier et al. (1996) asked 71 students to value metastatic breast cancer for durations
of 5, 10, 25 and 50 years. The valuation for the 5 year duration was significantly lower
than for the other three durations but the values for 10, 25 and 50 years were similar.
The authors suggest that respondents are using a proportional heuristic (i.e. are trading
off a constant proportion of remaining life expectancy) which, if correct, would mean
that U(Q)T is a good approximation of U(Q, T).
This rather encouraging conclusion holds only, however, for those states for which
more time is always preferred to less. Sutherland et al. (1982), having found from a
sample of 20 colleagues that the proportion preferring death to varying durations in
each of five health states increased as the duration of the states increased, postulate
that for some states there exists a "maximum endurable time" beyond which people do
not wish to live. This concept is reinforced by the results from a much larger general
population study which showed that most states here much more likely to be rated as
nurse than dead if they lasted for ten years as opposed to one year [Dolan (1996a)]. Of
the 50 respondents in the Stalmeier et al. study who stated a preference for 25 years
over 50 years with metastatic breast cancer, only 13 went on to give a lower TTO score
to the longer duration.
3.1.3. Risk attitude over life years
McNeil et al. (1978) presented bronchogenic carcinoma patients with a gamble involv-
ing a 50% chance of full health for 25 years and a 50% chance of immediate death.
The mean certainty equivalent number of years was 5 which, assuming no discount-
ing, translates into a risk coefficient for the power utility curve of 0.43 (where for risk
neutrality this figure should be 1), thus indicating moderate risk aversion. In a similar
study of the general population, McNeil et al. (1981) found similar results. Stiggelbout
et al. (1994) in a study of young men with testicular cancer found mildly risk averse
preferences: the corresponding risk coefficient was 0.74. Verhoef et al. (1994) found
a similar coefficient (of 0.80) amongst healthy women but there was evidence of risk-
seeking preferences over gambles involving short durations. Conversely, Mehrez and
1742 P. Dolan
Gafni (1987) found that risk-seeking behaviour arose more often when the length of
time over which utility was assessed increased.
3.1.4. A zero rate of time preference
The results from those studies which have reported increasing proportional trade-offs
(Section 3.1.2) are consistent with a positive rate of time preference. As explained in
Section 2.1, implied risk aversion over gambles involving life years (Section 3.1.3) is
also consistent with a declining value of T over time. There have been studies which
have tried to measure individual rates of time preference directly. On the whole, the
results suggest that the rate of time preference is zero at the aggregate level, lending
support to the studies by Bleichrodt and Johannesson (1996) and Stalmeier et al. (1996)
discussed in Section 3.1.2. In a study of 29 economics undergraduates, Cairns (1992)
found that the timing of a health state did not appear to matter as much as the timing of
identical levels of wealth. Redelmeier and Heller (1993), in a study of time preference
rates over acute health states, observed discount rates of zero in 62% of cases. And
Dolan and Gudex (1995) found that the median discount rate was zero across six EQ-
5D states of health.
However, all studies have shown that there is wide variation in time preference rates
at the individual level. Moreover, in the Dolan and Gudex (1995) significantly more
responses implied negative rates of time preference than positive ones, suggesting that
instead of wanting to postpone poor health (as most discounting models would predict),
more people want to get it out of the way. The possibility that people may have negative
discount rates (at least within some specified time period) is now recognised in the
literature on time preference. For example, Loewenstein (1987) found that people were
willing to pay more to avoid receiving a fleeting unpleasantness that was delayed for
three days than they were to avoid the event immediately.
3.1.5. Stability of lifetime preferences
To make comparisons of health states during the course of an individual's life requires
preferences over such states to be stable over their lifetime. To the author's knowledge
there is no direct evidence on whether this is so since no cohort study has yet been con-
ducted. There are, however, cross-sectional data which suggest that preferences over
states of health are not entirely independent of the respondent's age. Sackett and Tor-
rance (1978) found that TTO valuations increased with age, suggesting that people be-
come more tolerant of poor health as they get older, possibly through adapting to a gen-
eral deterioration in health. Dolan et al. (1996a) found a somewhat different pattern in
that, although valuations rose up to about 40 years of age, there was a general decrease
from about 40 to 60 and then a much sharper fall in later years. This suggests that, as
people's life expectancy shortens, they see less reason to tolerate suffering during their
remaining years.
Whatever the precise relationship between health state valuations and age, it seems
likely that the same person will place a different value on health at different stages of
life. For example, they may place less weight on full health when young than when
raising a young family. The weight may then decline again when the children have
grown up, but rise again when they retire [Loomes and McKenzie (1989)]. Williams
(1988) asked 377 individuals to choose and rank the 3 "life-stages" from a set of 12
when they thought full health was most important. The results suggest that people may
indeed value some later stages in life more highly than some earlier ones.
3.1.6. Additive separability
There has been some empirical investigation into the extent to which the value at-
tached to a particular health state is independent of the state(s) that precede or follow
it. Richardson et al. (1996) asked 63 female volunteer workers first to value three dis-
crete breast cancer states and then a profile made up of the same three states using the
VAS, SG and TTO methods. The systematic pattern across respondents and methods
was that the profile value was significantly lower than the value that would be implied
by combining the scores for the discrete health states. Kupperman et al. (1997) elicited
valuations using the VAS and SG from 121 women in the early stages of pregnancy
for profiles relating to pregnancy outcomes and the remainder of the woman's life, and
for discrete states after decomposing the profiles into three pregnancy periods and the
remainder of the woman's life. The mean values for the profiles were again lower than
that implied by combining the valuations for the discrete health states. It appeared that
there was a reasonably good relationship between the two sets of valuations at the ag-
gregate level, particularly for the VAS, but at the individual level profile scores were not
very well predicted by valuations for discrete states.
The results from both studies suggest that respondents focus more on future than on
current health states. In Richardson et al. (1996) the profile ends with suffering and
then death, the knowledge of which "casts a shadow over, or devalues, the enjoyment of
earlier life years". This is consistent with the assertion made by Loewenstein and Prelec
(1991) that, because "people tend to assimilate to ongoing stimuli and to evaluate new
stimuli relative to their assimilation level", when separate events are seen as part of a
package they will dislike profiles of decreasing utility. And in Kupperman et al. (1997)
the valuation for the remainder of the woman's life was the most significant variable in
explaining the profile score.
3.2. What is the "best" outcome measure?
3.2.1. Criteriafor assessing a health outcome measure
Empirical assessment of instruments designed to measure or value HRQoL involves

considerations of feasibility, reliability, responsiveness and validity [McDowell and
Newell (1987), Streiner and Norman (1989)]. Feasibility means that the instrument must
be capable of being used in practice and must be acceptable to respondents. These con-
siderations require that the instrument should be cost-effective to use, should not require
1744 P Dolan
too long for respondents to complete the task(s) and should not place too great a cogni-
tive burden on them. One measure of the feasibility of an instrument is the proportion
of completed responses.
Reliability refers to the stability of responses when all pertinent conditions remain
unchanged and can be investigated in two ways; (i) split-test reliability which assesses
an individual respondent's consistency when a question is presented more than once
within the same administration, and (ii) test-retest reliability which assesses the stability
of responses over short periods of time. The reliability of an instrument is typically
measured in a relative way; if the variability in responses between respondents is greater
than the variability within respondents, then it is deemed to be reliable, but this, of
course, says nothing about its absolute performance.
Responsiveness refers to the extent to which a measure is able to detect important
changes in HRQoL. It is usually expressed in terms of "effect size", whereby the mean
change in score is divided by either the standard deviation at baseline or the standard
deviation of the change.
An instrument is valid if it accurately reflects the concept or phenomenon it claims to
measure. There are a number of different types of validity discussed in the psychometric
literature [Streiner and Norman (1989)]. Perhaps the most rigorous is construct valid-
ity which is assessed by examining; (i) the extent to which a measure correlates with
other measures (convergent validity), and (ii) the extent to which predicted relationships
between the measure and respondent characteristics are upheld (discriminant validity).
In the absence of a gold standard, both tests are problematic. The test for convergent
validity says nothing about which measure is the more valid if the different measures
produce different results, nor whether all or none of the measures are valid if they yield
similar results. In the case of discriminant validity, if the construct is not supported it
does not necessarily invalidate a measure since it may be that the construct itself is mis-
specified. However, if one measure yields very different results from a number of other
measures, or if one measure does not find differences according to wide differences in
severity of illness, then doubt will be cast on its validity.
Economists typically discuss validity in terms of the extent to which people's hypo-
thetical (or stated) preferences accord with their revealed preferences. For example, a set
of health state valuations may imply that a patient has a strict preference for one treat-
ment over another. If the patient's actual treatment decision accords with this implied
preference then the valuations are said to be valid (at least in that particular context).
But interpreting revealed preference data is itself problematic (not least because in the
absence of full information an individual's manifest preferences may not reflect her un-
derlying or "true" values) and this test of validity has rarely been undertaken in the
context of health care [Brazier and Deverill (1999)].
3.2.2. Evaluating descriptive systems
Brazier et al. (1996) provide a summary of the relative performance of the five mea-
sures that have been designed to produce preference-based single index scores for
each of the health states they generate - the Quality of Well-Being Scale (QWB), the
McMaster Health Utility Index (HUI II, designed for use with children, and HUI III),
the EQ-5D and the 15-D. From their review, the evidence regarding feasibility and re-
liability appears encouraging and there would appear to be little to choose between the
measures on these grounds alone [Sintonen (1994), Brazier et al. (1996)]. Evidence on
the responsiveness of one measure relative to another is in short supply.
There is more evidence regarding the construct validity of the various measures but it
is somewhat mixed. The QWB has been found to correlate well with other measures of
functional health [Bombardier et al. (1986)] but less well with psychological measures
[Andresen et al. (1995)]. The HUI II has been shown to discriminate between children
being treated and those no longer being treated for a brain tumour [Feeny et al. (1993)]
but in a comparison of low birth weight children and a random sample of children a
large proportion was found to have no problems [Barr et al. (1994)]. The EQ-5D and
15-D have been found to correlate well with one another [Sintonen (1995)] and also with
generic profile measures [Brazier et al. (1993), Sintonen (1994)]. However, the EQ-SD
was not able to distinguish between COPD patients with and without a co-morbidity
[Harper et al. (1997)]. There are too few studies to make an assessment of the HUI III
at the present time.
3.2.3. Evaluating valuation methods
It would appear that both the SG and the TTO are feasible in that they have both
been widely used in practice and most studies have reported high response rates and
even higher levels of complete data [Froberg and Kane (1989a)]. However, in a within-
respondent comparison of the SG and TTO in a sample of the UK general population,
Dolan et al. (1996b) found that a variant of the TTO which used a specially designed
board produced fewer missing values than the analogous version of the SG and fewer
missing values than variants of the methods which used a self-completion booklet.
Many studies have also found little to choose between the two methods in terms of
reliability. Torrance (1976) found the SG and TTO to have similar split-test correlation
coefficients (between 0.80 and 0.90). Reed et al. (1993) found the test-retest reliability
(as measured by the correlation coefficient) to be higher for the SG (r = 0.82) than for
the TTO (r = 0.74) but this dependence was not statistically significantly. Dolan et al.
(1996b) found that the "board-based" variant of the TTO performed best, producing a
correlation coefficient of 0.81 but this was not significantly higher than the correlation
coefficient of 0.71 for the self-administered variant of the SG.
In order to test construct validity, it is necessary to look at the constructs that health
state valuations are hypothesised to be associated with. The evidence currently available
suggests that variation in population subgroups is not explained by the different demo-
graphic characteristics of respondents, such as sex or socio-economic status. However,
Froberg and Kane (1989b) note that "We have seen that patients with a particular condi-
tion often assign a higher utility than do patients without the condition". In Dolan et al.
(1996b) both SG and TTO valuations were unaffected by the gender and employment
1746 P Dolan
status of the respondent and there was tentative support for the hypothesis that higher
valuations would be elicited from respondents with experience of illness. Therefore,
there would seem to be little to choose between the methods in terms of this particular
test of construct validity.
Bleichrodt and Johannesson (1997a) compared the direct rankings of a number of dif-
ferent health profiles with the rankings implied by calculating QALYs from valuations
elicited from the same respondents using the SG and TTO. Without discounting, the
correlation between the direct ranking and the implied ranking was significantly higher
for TTO than for SG. The rank correlation coefficient was positively related to the dis-
count rate in the case of the SG but negatively related to the discount rate for the TTO,
but the relationship between direct and implied rankings was still stronger for the TTO
than for the SG up to a discount rate of about 9%.
Overall, then, there would appear to be little compelling empirical evidence to favour
one method over the other, although the "benefit of the doubt" might be given to the
TTO. It is encouraging that in a study of over 3000 members of the UK general popula-
tion, Dolan et al. (1996a) show that the TTO produces data that are near complete and
highly consistent.
Of course, it is also important to consider whether in practice the SG and TTO pro-
duce similar results, or whether, as suggested in Section 2.3, SG values are higher than
TTO ones. Whereas empirical evidence on the ordinal relationship between SG and
TTO values is mixed, most studies to date have shown that the two methods do yield
different valuations from the same respondents for identical descriptions of health. Tor-
rance (1976) and Read et al. (1984) found correlations of 0.65 between the two methods
which led the former to conclude that the SG and TTO are equivalent. However, Read
et al. emphasised that high correlations can coexist with systematic differences between
sets of scale values. Wolfson et al. (1982), Stiggelbout et al. (1994), Bleichrodt and Jo-
hannesson (1997a) and Lenert et al. (1998) also found differences between the methods.
The ordering of such differences was the same in all six studies: SG values were higher
than TTO in accordance with the theoretical predictions of Section 2.2.
However, not all studies have found SG values to be higher than TTO ones. Krabbe
et al. (1996) elicited valuations from individuals in the standard way and also median
values from groups of people using a voting-type mechanism. No significant differences
between SG and TTO valuations were found in either case. Dolan et al. (1996b) elicited
health state valuations using either specially designed boards or a self-completion book-
let and found few significant differences between SG and TTO valuations although TTO
valuations were in general slightly higher than SG ones for both variants. Although all of
these studies invoked hypothetical scenarios, Hornberger et al. (1992) elicited patients'
valuations of their own health and also found that the TTO produced higher values that
the SG.
It was shown in Section 2.2 that the two-stage SG and TTO methods should produce
the same results. It is important also to consider the extent to which this is borne out by
evidence, but there is very little empirical evidence on this issue. Some evidence exists
from gambles involving monetary outcomes that people exhibit greater risk aversion in
Ch. 32: The Measurement ofHealth-Related Quality of Life 1747
probability equivalence questions (i.e. stage one of the SG) than in certainty equivalence
questions (i.e. stage two of the SG) [Hershey and Schoemaker (1985)]. This suggests
that responses to two-stage SG questions will produce systematically biased estimates
of TTO values.
Even if differences between the valuation methods do exist, and a choice between
them is difficult to make, it is possible that a systematic relationship exists between
the methods. As noted in Section 2.2, although the VAS is commonly regarded by
economists as theoretically inferior to the SG and TTO methods, it has the practical
advantages of being simpler to complete and cheaper to administer than either of the
other methods. Consequently, it is widely used in clinical and evaluative studies. There-
fore, if an algorithm can be found which maps VAS values onto SG and/or TTO rather,
then it might be possible to "convert" valuations elicited via the (cheap and simple)
VAS into (theoretically superior) SG and/or TTO values. In a comparison of mean VAS
and TTO values, Torrance (1976) concluded that "the two techniques exhibit a system-
atic relationship [that] can be approximated by ... a logarithmic function and a power
function" (p. 134). Since then, a number of authors have used a power function to esti-
mate SG and TTO valuations from VAS ones [Loomes (1993), van Busschbach (1994),
Stiggelbout et al. (1996)].
However, there a number of reasons to be cautious about such findings. First, the
power coefficients differ across studies (for example, a VAS score of 0.10 would convert
into a TTO score of 0.23 in the Stiggelbout et al. study and 0.34 in the van Busschbach
study). Second, the analyses were performed on aggregate rather than individual data,
thus making the choice between competing models more difficult as well as making
inefficient use of the data. Third, the models presented by Torrance did not hold at the
individual level [this is confirmed in studies reported in Dolan and Sutton (1997) and
Bleichrodt and Johannesson (1975b)]. Fourth, van Busschbach (1994) found that the
power model offered no improvement over a linear one [in a comparison of a number
of different models, Dolan and Sutton (1997) conclude that the linear model performs
best]. Given the current empirical evidence, then, it would seem that VAS valuations
cannot be converted into SG or TTO ones with any degree of confidence.
3.3. Do valuationsdiffer according to respondent characteristics?
The decision about whose values to use when measuring HRQoL is not one that can be
resolved by empirical evidence (rather it is a philosophical or political issue) but knowl-
edge about the extent to which valuations differ according to the background charac-
teristics of the respondent will make clear the implications of using valuations from
different population sub-groups. Most studies suggest that variation among population
subgroups is not explained by the different demographic characteristics of respondents
[Froberg and Kane (1989b) provide an extensive review of the literature]. The only no-
table exception to this is reported by Dolan et al. (1996a), who found that, on average,
valuations from women were lower than from men, particularly for more severe health
states. The large sample (of over 3000 respondents) in this study relative to some of
1748 P. Dolan
these earlier studies may be the reason why differences between population subgroups
were detected: as Froberg and Kane (1989b) note "low statistical power may be obscur-
ing differences".
There is some evidence to suggest that experience of illness may influence respon-
dents' valuations of health states. In early studies in this area, Rosser and Kind (1978)
found significant differences between medical patients and doctors and between medical
patients and psychiatric patients, and Sackett and Torrance (1978) reported that home
dialysis patients assigned higher values to kidney dialysis than did the general public.
More recently, Dolan (1996b) found that current health status has an important effect
on valuations with those in poorer health generally giving higher values. The possibil-
ity that valuations differ according to illness experience is consistent with the notion
that those in poor health successfully compensate for it (as mentioned in Section 1.2).
However, this conclusion is slightly tempered by Llewellyn-Thomas et al. (1984) who
found that respondents' own health status did not influence health state valuations, and
by Daly et al. (1993) who found that valuations given to menopausal symptoms did not
differ across subgroups of women who were divided on the basis of whether they had
experienced these symptoms.
3.4. How have valuesfor all health states been generated?
There is a lack of consensus about how to generate values for all health states in a partic-
ular descriptive system from valuations of a subset of states. Valuations of health states
defined by the HUI and 15-D measures have been elicited so as to facilitate estimation
using the decomposed approach and a set of scores for EQ-5D health states has been
generated using the composite approach [Dolan (1997a)]. There has been no direct em-
pirical comparison of the predictive ability of the two approaches in the health state
valuation context although, in a study of job choice, Currim and Sarin (1984) found that
the latter approach outperformed the former.
3.5. How have valuations been aggregated?
There are remarkably few data on the extent to which valuations differ according to
the measure of central tendency chosen. Many of the published studies present mean
scores, usually after a number of "outliers" (however defined) have been excluded.
Dolan (1997b), using data from a large-scale general population study, showed that (as
a result of the lower and upper bound on valuations) scores for less severe health states
were negatively-skewed whilst those for more severe states were positively-skewed. As
a result, EQ-5D "tariff" values estimated from median scores were higher than from
an individual-level model (which approximated mean values) for the least severe two-
thirds of states (reaching a maximum difference of 0.21), and lower for the most severe
one-third of states (reaching a maximum difference of 0.25). Such differences mean that
resource allocation decisions based on median values might differ markedly from those
based on mean values.
Problems of (mis)representing group preferences may exist even when mean and me-
dian values are similar. For example, identical mean and median values might emerge
from a group of people who give very polarised valuations, but neither measure of cen-
tral tendency will approximate the value of any respondent.
4. Discussion
4.1. A resume
4.1.1. The QALYmodel
The QALY has attracted the most attention in the literature to date, and therefore this
chapter has considered in some detail the assumptions that underlie it. The evidence can
be summarised as follows: (i) utility independence - the limited evidence suggests that
the probability equivalent value from a SG question increases with duration; (ii) con-
stant proportional trade-off - the evidence suggests that there is either increasing pro-
portional trade-off or that constant proportional trade-off holds; (iii) risk attitude over
life years - there is evidence of moderate risk aversion but the results do not appear to
be independent of the number of years involved in the gamble and could also be ex-
plained by risk neutrality and a positive rate of time preference in the region of 3%;
(iv) a zero rate of time preference - this holds at the aggregate level but there is great
variation at the individual level; (v) stability of preferences - the evidence is mixed but
the different stages of life appear to be valued differently; and (vi) additive separability
- the evidence suggests that this assumption does not hold; rather, people tend to pay
more attention to the state that comes at the end of a profile than would be assumed by
eliciting its independent valuation.
According to Broome (1993), additive separability is "the most dubious condition"
and it certainly appears, albeit from very limited evidence, that people's preferences
are less in accordance with this assumption than with any other. However, it does not
follow that it is appropriate in all circumstances to consider profile scores to be the "gold
standard" by which to judge valuations for discrete states. Since the evidence currently
available suggests that respondents focus more on future than on current health states,
then, if a future health state is not known with certainty, it might be more appropriate to
value the previous state independently so that the future state does not contaminate its
value. Therefore, the choice between valuing profiles or states should be made according
to whether or not such contamination is considered appropriate.
4.1.2. The description of health
The five generic measures have been shown to be feasible and, to a lesser extent, reliable
but there is very little evidence for their validity. Therefore, it is difficult to make a
choice between the measures at the moment and what is really required are within-
respondent comparisons of the methods.
1750 P.Dolan
4.1.3. The valuation of health
With regard to the two main methods that can be used to value health profiles or health
states, there appears to be little (either in theory and in practice) to choose between the
SG and the TTO. In order to interpret responses from both methods in the way that
health economists would like, it is necessary to assume that people treat risk linearly
in the SG and time linearly in the TTO; in other words, that the axioms of EU hold in
the former case and that there is a zero rate of time preference in the latter case. Whilst
there is evidence to suggest that neither assumption holds, if people treat risk and time
in a systematically non-linear way, then preference weights can still be estimated from
responses to the two methods. There are a number of theories which suggest how SG re-
sponses might be more accurately converted into meaningful preference weights [Tver-
sky and Kahneman (1992), Wakker and Stiggelbout (1995)] but there is little evidence
about the most appropriate transformation function(s) for responses to TTO questions.
4.1.4. The source of values
Most studies suggest that variation among population subgroups is not explained by the
different demographic characteristics of respondents but there is some evidence that ex-
perience of illness may influence respondents' valuations of health states. This suggests
that the debate about whether to use valuations derived from the general public or those
elicited from patients is an important one which may have very real implications for the
allocation of resources.
4.1.5. The calculation of valuesfor all health states
Valuations of health states defined by the different descriptive systems have been esti-
mated using either a decomposed or a composite approach. The choice would appear to
depend largely on the extent to which a theoretical construct matters; the decomposed
approach is driven more by theory and the composite one driven more by data.
4.1.6. The aggregationof values
Much of the published literature reports mean or median values and the implications
of using one measure of central tendency compared to the other is rarely discussed.
Since the range of possible values is constrained by "floor" and, particularly, "ceiling"
effects, mean and median values are likely to differ quite substantially especially those
close to the "floor" or the "ceiling" on the spectrum of severity. The differential effect of
the measure of central tendency chosen will be magnified when movements take place
between very mild and very severe states.
Ch. 32: The Measurement of Health-Related Qualiiy of Life 1751
4.2. A research agenda
4.2.1. Valuing profiles or states
There has been much debate in the literature about the pros and cons of breaking a given
health profile up into a series discrete health states but there has been remarkably little
investigation into the extent to which the sum of the parts provides a good approximation
of the whole. Whilst there is some evidence to suggest that this matters empirically
(Section 3.1.6), there is the need to examine whether the combined value for a series of
discrete states and the valuation of an entire profile are related in any systematic way.
Additionally, since the valuation of the latter cannot be regarded as the "gold standard"
in all circumstances (Section 4.1.1), it is important that criteria are established by which
a choice between the value of a whole profile and the combined value of different states
could be made if differences between the two are observed.
4.2.2. Valuing health or health gain
Ultimately it is the value of changes in profiles or states that is important if QALYs

or HYEs are to be used to inform resource allocation decisions [Fitzpatrick (1996)].
If valuations lie on an interval scale then the value of a move from state i to state j
can be calculated as the value given to state j minus the value given to state i. But it is
important to show the extent to which this is the case and to date such evidence is almost
non-existent. Testing interval scale properties will enable a more informed choice to be
made between profiles and discrete states and between the SG and TTO.
4.2.3. Time preference
There now exists a large body of theoretical and empirical literature relating to how in-
dividuals ought to and do make decisions under risk and uncertainty. There is relatively
very little literature relating to decisions over time although the imbalance is being re-
dressed, both at the theoretical level, where the normative appeal of the exponential
discount model is contested [Weinstein (1993), Bleichrodt and Gafni (1996)], and at
the empirical level where exponential discounting appears to be descriptively flawed
[Loewenstein (1987), Cairns (1992)]. Such work is vital since, as Bleichrodt and Gafni
point out, "health outcomes have a time dimension inextricably bound to them, [so] we
cannot ignore individual intertemporal preferences in valuing them".
4.2.4. "Economic" validity
Much of the empirical research that has been undertaken into establishing the validity
(or otherwise) of measures designed to describe and value HRQoL has been conducted
within the psychometric tradition, largely in terms of construct validity (Section 3.2).
Economists, on the other hand, are concerned that hypothetical decisions are validated
1752 P Dolan
against actual decisions. This has attracted a great deal of attention in the WTP literature
where the question is whether an individual really would be willing to pay a stated WTP,
and no more or no less [Mitchell and Carson (1989)]. It is important to that, wherever
possible, similar tests of validity are undertaken in the HRQoL field. For example, it
should be possible to compare the choices implied by QALY or HYE calculations with
decisions made in direct choices between the same alternatives [Loomes and McKenzie
(1989)].
4.2.5. Harmonisation
Studies have used many different descriptive systems, valuation methods and sources of
values, thus making comparisons across studies almost impossible. Against this back-
ground, a US panel on cost-effectiveness in health and medicine [Gold et al. (1996)]
recommended a "reference case" for use in cost-effectiveness analyses. They suggested
that the health state descriptive system should be generic, that the valuation method
should be preference-based and that the source of values should be a representative
sample of the general population. Whilst these recommendations have attracted few
dissenting voices amongst health economists, a study by Neumann et al. (1997) high-
lights how much current practice has to change to meet the recommendations. In look-
ing at studies between 1975 and 1995 in which effectiveness was measured in terms
of QALYs, Neuman et al. found that only about one-fifth used a generic measure, or
had elicited values using SG or TTO, or used values from the general population. Most
used a disease specific measure in which the authors own judgement was used to attach
values to the states. To facilitate comparability between studies, all future studies should
use the Gold et al. "reference-case".
4.2.6. The implicationsof using different values
An important question in the context of real world decision-making is the difference

that using one set of values as opposed to another actually makes in practice. Remark-
ably few valuation studies have been subjected to "real-world" sensitivity analysis. This
may be partly attributed to many of the earlier studies being seen as essentially method-
ological research, aimed at enhancing our understanding of health status measurement
rather than at contributing to real resource allocation decisions. There are, however, an
increasing number of studies which provide good information from large samples on
the extent to which values differ according to some of the factors cited above [Dolan
et al. (1996a)], so it should be possible in future to consider whether and how resource
allocation decisions would be affected by the choice of valuations.
4.3. The process of eliciting valuations
Given that much of this chapter has focused on theoretical and empirical issues relating
to the measurement of individual preferences, it is striking that so few studies have con-
sidered the nature of these preferences. The received wisdom amongst economists is that
individuals have well-defined preference functions. Whilst many economists are scep-
tical of the stated preference methodology [preferring instead "the confusions resulting
from external observation"; McCloskey (1985)], those who accept the methodology as-
sume that these preference functions can be "tapped into" by appropriate questions. This
is referred to by Fischhoff (1991) as the philosophy of articulated values. For example,
Lenert et al. (1997) argue that to use preferences to inform decisions "there must be
substantial evidence to show that the results of applied preference measures reflect, as
nearly as possible, the "true" values of patients".
This begs the question whether people really have true underlying preferences, par-
ticularly regarding health? There are plenty of examples in Section 3 which suggest
otherwise. Rather, seemingly subtle changes in problem structure, question format, or
other aspects of the assessment process, appear to change the stated preferences of re-
spondents. For example, both Dolan et al. (1996b) and Lenert et al. (1998) found that
whether the probability of success from treatment, in the case of the SG, or years lived in
full health, in the case of the TTO, was presented in a "ping-pong" fashion (i.e. oscillat-
ing between "good" and "bad" outcomes) or in a "top-down" fashion (i.e. in descending
order from "good" to "bad" outcomes) influenced the resultant valuations as much as
the valuation method itself.
Findings of this kind can be accounted for by an alternative paradigm [referred to by
Fischhoff (1991) as the philosophy of basic values] which asserts that "preferences are
not simply read off some master list but are constructed on the spot by an adaptive deci-
sion maker" [Slovic (1995)]. Thus, if responses are affected by theoretically irrelevant
framing effects, then respondents must not have "true" values; rather, the elicitation pro-
cedure is a major force in shaping them. But there are examples where this paradigm is
itself questionable. For example, in the Dolan et al. (1996b) study, the ordinalrankings
of health states were robust to changes in value elicitation procedure.
A philosophy of partial perspectives lies somewhere between the extremes of ar-
ticulated and basic values. This viewpoint holds that, whilst preferences, particularly
regarding health, do not come as fully fledged and instantly accessible as economists
typically believe, people in very general terms do have what Fischhoff refers to as "sta-
ble values of moderate complexity". If this is so elicitation procedures can help to shape
preferences and, after deliberation and reflection, respondents are able to give answers
to questions that enable something to be inferred about their "true" preferences. The
spectrum that the partial perspectives philosophy covers is very wide but since "many
respondents cannot attach values to the quality of life entailed by states of health un-
familiar to them" [Loomes (1994)], perhaps future studies should begin by adopting a
partial perspective that is closer to the philosophy of basic rather than articulated values;
certainly much closer than in most studies hitherto.
This might involve presenting respondents with a summary of all of their responses
at the end of an interview and allowing them to revise any of their answers in the light
of this "overview", or even confronting them with any apparent inconsistencies and
again giving them the opportunity to revise their responses. Whatever their precise pro-
tocols, future studies should be much more interactive than those conducted previously
1754 P. Dolan
which "is useful not only in eliciting preferences for health outcomes but also in help-
ing to clarify the values which underpin those preferences" [Shiell (1997)]. Such studies
would be much more resource intensive per respondent but there should be a willing-
ness (rather a reluctance) to trade-off quantitative data for the more detailed qualitative
data that intensive questioning could generate.
These qualitative data should provide insights into the cognitive processes that re-
spondents use in order to arrive at their responses, thus enabling researchers to get a
better understanding of why valuations differ in addition to how they differ. Many of
the studies referred to in this chapter have been written in ways typical of economists;
namely, to postulate a null hypothesis, then to collect quantitative data that test the hy-
pothesis, and finally to engage in considerable "post-hoc" theorising when the results, as
invariably happens, do not conform with the null hypothesis. Rather than "second guess-
ing" respondents, the collection of qualitative data "straight from the horse's mouth"
appears a more appropriate strategy in this context. In addition, the data might direct
us towards the optimal trade-off between theoretical accuracy and cognitive burden (for
example, in some contexts it might in principle be appropriate to present people with
many different profiles with many different probabilistic outcomes but this might in
practice represent more information than any one individual could meaningfully pro-
cess).
Therefore, the message from this section is this: Let's not get hung up on the validity
or otherwise of the restrictions imposed on individual preferences before we've first had
a debate about the nature of those preferences.
4.4. Concluding remarks
Very few health care interventions have no effect on HRQoL. So when it comes to allo-
cating resources, it is vital that changes in HRQoL are taken into account. Some readers
may have initially been optimistic about our ability to do this but, in the light of the
arguments developed in this chapter, with its emphasis on the theoretical and empirical
problems associated with the measurement of health outcomes, may have become in-
creasingly disillusioned with the whole enterprise. This would be unfortunate because
facing up to the violations of certain axioms and the many unanswered questions is bet-
ter than the alternative of disregarding HRQoL altogether. This negative response would
also ignore the considerable methodological advances that have been made in the field,
particularly in the last twenty years. Moreover, many of the issues (how health is de-
scribed, who is to value it, and so on) are issues that are faced by any measure of health
outcome - it is just that they are made more explicit when measuring HRQoL.
It is also important to remember that the violation of certain assumptions (for exam-
ple, those in the QALY model) does not mean that the models concerned should neces-
sarily be abandoned (for example, in favour of something like an HYE approach). Most
assumptions can only be satisfied approximately and thus a judgement will ultimately
have to be made about the extent to which the loss of realism (e.g. of more general
QALY-type models) are compensated for by their greater tractability (e.g. compared to
less general HYE-type approaches). When making resource allocation decisions across
many health care programmes, tractability may be a more significant consideration than
when making decisions at the level of the individual patient where accurately reflecting
each patient's preferences might be considered to be of paramount importance.
Finally, the validity of valuations for health profiles or states does not rest on there
being a precise answer to the question of how many QALYs or HYEs a particular pro-
gramme generates. In many cases, it is likely that the use of different values will make no
difference to the ordinalconclusions reached about which programme generates more
QALYs or HYEs than which. As Lockwood (1988) has argued, "only a very radical
scepticism, according to which one could not even, with any confidence, set numeri-
cal limits in such comparisons, would have the effect of rendering the QALY approach
wholly useless ... such wholesale scepticism would ... be very difficult convincingly
to sustain".
References
Andresen, E.M., D.L. Patrick, W.B. Carter and J.A. Malmgren (1995), "Comparing the performance of health
status measures for healthy older adults", Journal of the American Geriatric Society 43:1030-1034.
Arrow, K.J. (1951), Social Choice and Individual Values (Wiley, New York).
Barr, R.D., W. Furlong, S. Dawson, A.C. Whitton, I. Strautmanis, M. Pal, D. Feeny and G.W. Torrance (1994),
"A multi-attribute approach to health status in survivors of acute lymphoblastic leukemia in childhood",
American Journal of Paediatric Hematological Oncology 15:284-290.
Bennet, K., G.W. Torrance and P. Tugwell (1991), "Methodologic challenges in the development of utility
measures of health related quality of life in rheumatoid arthritis?", Controlled Clinical Trials 12.
Bleichrodt, H. (1995), "QALYs and HYEs: under what conditions are they equivalent?", Journal of Health
Economics 14:17-37.
Bleichrodt, H., and A. Gafni (1996), "Time preference, the discounted utility model and health", Journal of
Bleichrodt, H., and M. Johannesson (1996), "The validity of QALYs: an experimental test of constant pro-
portional trade-off and utility independence", Medical Decision Making 17:21-32.
Bleichrodt, H., and M. Johannesson (1997a), "Standard gamble, time trade-off and rating scale: Experimental
results on the ranking properties of QALYs", Journal of Health Economics 16:155-175.
Bleichrodt, H., and M. Johannesson (1997b), "An experimental test of a theoretical foundation for rating scale
valuations", Medical Decision Making 17:208-216.
Bleichrodt, H., P. Wakker and M. Johannesson (1997), "Characterizing QALYs by risk neutrality", Journal of
Risk and Uncertainty 15:107-114.
Bombardier, C., J. Ware, I. Russell, M.G. Larson, A. Chalmers and J. Leighton-Read (1986), "Auranofin
therapy and quality of life in patients with rheumatoid arthritis", American Journal of Medicine 81:565-
578.
Brazier, J., and M. Deverill (1998), "A comparison of five multi-attribute scales", HESG Conference,
Sheffield.
Brazier, J., and M. Deverill (1999), "A checklist for judging preference-based measures of health related
quality of life: Learning from psychometrics", Health Economics 8:41-52.
Brazier, J., N. Jones and P. Kind (1993), "Testing the validity of the Euroqol and comparing it with the SF36
health survey questionnaire", Quality of Life Research 2:169-180.
Brazier, J., S.J. Walters, J.P. Nicholl and B. Kohler (1996), "Using the SF-36 and EuroQol on an elderly
population", Quality of Life Research 5:195-204.
1756 P Dolan
Brooks, R. (1996), "EuroQol: The current state of play", Health Policy 37:53-72.
Broome, J. (1991), Weighing Goods (Basil Blackwell, Oxford).
Broome, J. (1993), "Qalys", Journal of Public Economics 50:149-167.
Buckingham, K. (1993), "A note on HYE (healthy year equivalent)", Journal of Health Economics 12:301-
309.
Buckingham, K., and M. Drummond (1993), "A theoretical and empirical classification of health valuation
techniques", HESG Conference, Strathclyde.
van Busschbach, J. (1994), "The validity of QALYs", PhD Thesis (Erasmus University, Rotterdam).
Cairns, J. (1992), "Future discounting: health, wealth and time preference", Project Appraisal 7(1):31-40.
Cairns, J., K. Johnston and L. McKenzie (1991), "Developing QALYs from condition-specific outcome mea-
sures", HERU Working Paper 14 (University of Aberdeen).
Camerer, C. (1993) "Individual decision-making", in: J. Kagel and A. Roth, eds., Handbook of Experimental
Economics (Princeton University Press).
Cassileth, B.R., E.J. Lusk, T.B. Strouse, D.S. Miller, L.L. Brown, P.A. Cross and A.N. Tenaglia (1984),
"Psychosocial status in chronic illness: A comparative analysis of six diagnostic groups", New England
Cook, J., J. Richardson and A. Street (1994), "A cost utility analysis of treatment options for gallstone disease:
Methodological issues and results", Health Economics 3:157-168.
Culyer, A.J., and A. Wagstaff (1993a), "Equity and equality in health and health care", Journal of Health
Culyer, A.J., and A. Wagstaff (1993b), "QALYs versus HYEs", Journal of Health Economics 12:311-324.
Culyer, A.J., and A. Wagstaff (1995), "QALYs versus HYEs: A reply to Gafni", Journal of Health Economics
14:39-45.
Currim, I.S., and R.K. Sarin (1984), "A comparative evaluation of multi-attribute consumer preference mod-
els", Management Science 30:543-561.
Daly, E., A. Gray, D. Barlow, K. McPherson, M. Roche and M. Vessey (1993), "Measuring the impact of
menopausal symptoms on quality of life", British Medical Journal 307:836-840.
Daniels, N. (1991), "Is the Oregon rationing plan fair?", Journal of the American Medical Association
265:2232.
Dolan, P. (1996a), "Modelling valuations for health states: The effect of duration", Health Policy 38:189-203.
Dolan, P. (1996b), "The effect of experience of illness on health state valuations", Journal of Clinical Epi-
demiology 49:551-564.
Dolan, P. (1997a), "Modelling valuations for EuroQol health states", Medical Care 11:1095-1108.
Dolan, P. (1997b), "Aggregating health state valuations", Journal of Health Services Research and Policy
2:160-165.
Dolan, P. (1998), "The measurement of individual utility and social welfare", Journal of Health Economics
17:39-52.
Dolan, P. (1999), "Whose values count?", Medical Decision Making 19:482-486.
Dolan, P., and C. Gudex (1995), "Time preference, duration and health state valuations", Health Economics
4:289-299.
Dolan, P., C. Gudex, P. Kind and A. Williams (1996a), "The time trade-off method: Results from a general
population study", Health Economics 5:141-154.
Dolan, P., C. Gudex, P. Kind and A. Williams (1996b), "Valuing health states: A comparison of methods",
Dolan, P., and M. Jones-Lee (1997), "The time trade-off: A note on the effect of lifetime reallocation of
consumption and discounting", Journal of Health Economics 16:731-739.
Dolan, P., and M. Sutton (1997), "Mapping visual analogue scale scores onto time trade-off and standard
gamble utilities", Social Science and Medicine 44(10):1519-1530.
Donaldson, C. (1990), "Willingness to pay for publicly-provided goods: A possible measure of benefit",
Dyer, J.S., and R.K. Sarin (1982), "Relative risk aversion", Management Science 28(8):857-886.
Edgar, A., S. Salek, D. Shickle and D. Cohen (1998), "The ethical QALY: Ethical issues in healthcare resource
allocations", Euromed Communications.
Feeny, D., W. Furlong, M. Boyle and G.W. Torrance (1995), "Multi-attribute health status classification sys-
tems: Health utilities index", PharmacoEconomics 7:490-502.
Feeny, D., A. Leiper, R.D. Barr, W. Furlong, G.W. Torrance, P. Rosenbaum and S. Weitzman (1993), "The
comprehensive assessment of health status in survivors of childhood cancer: Application to high-risk acute
lymphoblastic leukaemia", British Journal of Cancer 67:1047-1052.
Festinger, L. (1957), A Theory of Cognitive Dissonance (Stanford University Press).
Fischoff, B. (1991), "Value elicitation: Is there anything there?", American Psychologist 46:835-847.
Fitzpatrick, R. (1996), "A pragmatic defence of health status measures", Health Care Analysis 4:265-272.
Froberg, D.G., and R.L. Kane (1989a), "Methodology for measuring health state preferences II: Scaling meth-
ods", Journal of Clinical Epidemiology 42(5).
Froberg, D.G., and R.L. Kane (1989b), "Methodology for measuring health state preferences III: Population
and context effects", Journal of Clinical Epidemiology 42:585-592.
Gafni, A. (1994), "The standard gamble method: What is being measured and how is it interpreted?", Health
Gafni, A., S. Birch and A. Mehrez (1993), "Economics, health and health economics: HYEs versus QALYs",
Gold, M., J.E. Siegal, L.B. Russell and M.C. Weinstein (1996), Cost-Effectiveness in Health and Medicine
(Oxford University Press).
Gudex, C., P. Dolan, P. Kind and A. Williams (1997), "Valuing health states: Interviews with the general
public", European Journal of Public Health 7:441-448.
Harper, R., J.E. Brazier, J.C. Waterhouse, S.J. Walters, N.M.B. Jones and P. Howard (1997), "Comparison of
outcome measures for patients with chronic obstructine pulmonary disease in an outpatient setting", Thoras
52:879-887.
Herrero, C. (1996), "Capabilities and utilities", Economic Design 2:69-88.
Hershey, J.C., and P.J.H. Schoemaker (1985), "Probability versus certainty equivalence methods in utility
measurement: Are they equivalent?", Management Science 31:1213-1231.
Homberger, J.C., D.A. Redelmeier and J. Petersen (1992), "Variability among methods to assess patients well-
being and consequent effect on a cost-effectiveness analysis", Journal of Clinical Epidemiology 45(5):505-
512.
Jachuk, S.J., H. Brierley, S. Jachuk and P.M. Wilcox (1982), "The effect of hypertensive drugs on the quality
of life", Journal of the Royal College of General Practitioners 32(235):103-105.
Johannesson, M., J.S. Pliskin and M.C. Weinstein (1994), "A note on QALYs, time trade-off and discounting",
Medical Decision Making 14:142-149.
Johannesson, M. (1995), "Quality-adjusted life-years versus healthy-years equivalents: A comment", Journal
Keeny, R.L. (1976), "A group preference axiomisation with cardinal utility", Management Science 23:140-
145.
Krabbe, P.EM., M.L. Essink-Bot and G.J. Bonsel (1996), "On the equivalence of collectively and individually
collected responses", Medical Decision Making 16:120-132.
Kupperman, M., S. Shiboski, D. Feeny, E.P. Elkin and A.E. Washington (1997), "Can preference scores for
discrete states be used to derive preference scores for an entire path of events?", Medical Decision Making
17:42-55.
Lenert, L.A., S. Morris, M.K. Goldstein, M.R. Bergen, W.O. Faustman and A.M. Garber (1997), "Measure-
ment of the validity of utility elicitations performed by computerized interview", Medical Care 9:915-920.
Lenert, L.A., D.J. Cher, M.K. Goldstein, M.R. Bergen and A.M. Garber (1998), "The effect of search proce-
dures on utility elicitations", Medical Decision Making 18:76-83.
Lipscomb, J. (1982), "Value preferences for health: Meaning measurement and use in program evaluation",
in: R.L. Kane and R.A. Kane, eds., Values and Long Term Care (Lexington Books).
1758 P. Dolan
Llewellyn-Thomas, H., H.J. Sutherland, R. Tibshirani, A. Ciampi, J.E. Till and N.F. Boyd (1982), "The mea-
surement of patients' values in medicine", Medical Decision-Making 2:449-462.
Llewellyn-Thomas, H., H.J. Sutherland, R. Tibshirani, A. Ciampi, J.E. Till and N.F. Boyd (1984), "Describing
health states: Methodologic issues in obtaining values for health states", Medical Care 22:543-552.
Lockwood (1988), "Quality of life and resource allocation", in: M. Bell and S. Mendus, eds., Philosophy and
Medical Welfare (Cambridge University Press, Cambridge).
Loewenstein, G. (1987). "Anticipation and the valuation of delayed consumption", The Economics Journal
97:666-684.
Loewenstein, G., and D. Prelec (1991), "Negative time preference", American Economic Review 81:347-352.
Loomes, G. (1993), "Disparities between health state measures: Is there a rational explanation?" in: W. Ger-
rard, ed., The Economics of Rationality (Routledge, London).
Loomes, G. (1994), "Valuing health and safety: Some economic and psychological issues", mimeo (University
of York, York).
Loomes, G. (1995), "The myth of the HYE", Journal of Health Economics 14:1-7.
Loomes, G., and L. McKenzie (1989), "The scope and limitations of QALY measures", Social Science and
Medicine 28:299-308.
Mathers, C. (1997), "Health expectancies and DALYs: Towards consistency of measures", Room document
No. 4, December.
McCloskey, D.N. (1985), The Applied Theory of Price (Macmillan, London).
McDowell, I., and C. Newell (1987), Measuring Health: A Guide to Rating Scales and Questionnaires (Oxford
University Press).
McKeown, T. (1979), "The direction of medical research", Lancet 2(8155):1281-1284.
McNeil, B.J., R. Weichselbaum and S.G. Pauker (1978), "Fallacy of the five-year survival in lung cancer",
New England Journal of Medicine 299:1397-1401.
McNeil, B.J., R. Weichselbaum and S.G. Pauker (1981), "Speech and survival: Trade-offs between quality
and quantity of life in laryngeal cancer", New England Journal of Medicine 305:982-987.
Mehrez, A., and A. Gafni (1987), "An empirical evaluation of two assessment methods for utility measure-
ment for life years", Socio-Economic Planning Sciences 21:371-375.
Mehrez, A., and A. Gafni (1989), "Quality-Adjusted Life Years, Utility Theory, and Health-Years Equiva-
lents", Medical Decision Making 9:142-149.
Mehrez, A., and A. Gafni (1991), "The health-years equivalents: How to measure them using the standard
gamble approach", Medical Decision Making 11:140-146.
Meyerowitz, B.E. (1983), "Postmastectomy coping strategies and quality of life", Health Psychology 2:117-
132.
Mitchell, R.C., and R.T. Carson (1989), Using Surveys to Value Public Goods: The Contingent Valuation
Method, Resources for the Future (Washington).
Miyamoto, J.M., and S.A. Eraker (1985), "Parameter estimates for a QALY utility model", Medical Decision
Making 5:191-213.
Miyamoto, J.M., and S.A. Eraker (1988), "A multiplicative model of the utility of survival duration and health
quality", Journal of Experimental Psychology 117:3-20.
Mooney, G. (1994), Key Issues in Health Economics (Harvester Wheatsheaf).
Morrison, G.C. (1997), "HYE and TTO: What is the difference?", Journal of Health Economics:563-578.
Murray, C.J.L. (1996), "Rethinking DALYs", in: C.J.L. Murray and A.D. Lopez, eds., The Global Burden of
Disease (Harvard University Press).
von Neumann, J., and 0. Morgenstern (1953), Theory of Games and Economic Behaviour (Wiley, New York).
Neumann, P.J., D.E. Zinner and J.C. Wright (1997), "Are methods for estimating QALYs in cost-effectiveness
analyses improving?", Medical Decision Making 17:402-408.
Ng, Y.K. (1992), Welfare Economics (Macmillan).
Nord, E. (1995), "The person trade-off approach to valuing health care programs", Medical Decision Making
15:201-208.
Olsen, J.A., and R. Smith (1998), "Who have been asked to value what? A review of 54 WTP-based surveys
on health and health care", HESG Conference, Galway.
Patrick, D.L., J.W. Bush and M.M. Chen (1973), "Methods for measuring levels of well-being for a health
status index", Health Services Research 8:228-245.
Pliskin, J.S., D.S. Shepard and M.C. Weinstein (1980), "Utility functions for life years and health status",
Operations Research 28:206-244.
Read, J.L., R.J. Quinn, D.M. Berrick, H.V. Fineberg and M.L. Weinstein (1984), "Preferences for health
outcomes: Comparison of assessment methods", Medical Decision Making 4(3):315-329.
Redelmeier, D.A., and D.N. Heller (1993), "Time preference in medical decision-making and cost-
effectiveness analysis", Medical Decision Making 13:212-217.
Reed, W.W., J.E. Herbers and G.L. Noel (1993), "Cholesterol lowering therapy: What patients expect in
return", Journal of General Internal Medicine 8:591-596.
Richardson, J. (1994), "Cost-utility analysis: What should be measured?", Social Science and Medicine
39(1):7-21.
Richardson, J., J. Hall and G. Salkfeld (1996), "The measurement of utility in multiphase health states",
International Journal of Technology Assessment in Health Care 12:151-162.
Rittenhouse, B.E. (1997), "Healthy years equivalents versus time trade-off: Ambiguity on certainty and un-
certainty", International Journal of Technology Assessment in Health Care 13:35-48.
Rosser, R., and P. Kind (1978), "A scale of valuations of states of illness: Is there a social consensus?",
International Journal of Epidemiology 7:347-358.
Sackett, D.L., and G.W. Torrance (1978), "The utility of different health states as perceived by the general
public", Journal of Chronic Diseases 31:697-704.
Schoemaker, P.J.H. (1982), "The expected utility model: Its variants, purposes, evidence and limitations",
Journal of Economic Literature 20:529-563.
Sen, A. (1987), On Ethics and Economics (Basil Blackwell).
Shiell, A. (1997), "Health outcomes are about choices and values: An economic perspective on the health
outcome movement!", Health Policy 39:5-15.
Sintonen, H. (1994), "The 15D measure of HRQoL: Reliability, validity and the sensitivity of its health state
descriptive system", NCFPE Working Paper 41 (Monash University, Melbourne).
Sintonen, H. (1995), "EuroQol versus 15-D: A comparison of two health-related quality of life measures", in:
M. O'Hanlon and M. Buxton, eds., HERG Research Report 20, EuroQol Conference Proceedings (Brunel
University).
Slovic, P. (1995), "The construction of preferences", American Psychologist 50(5):364-371.
Stalmeier, P.F.M., T.G.G. Bezembinder and I.J. Unic (1996), "Proportional heuristics in time trade-off and
conjoint measurement", Medical Decision Making 16:36-44.
Stevens, S.S. (1971), "Issues in psychological measurement", Psychological Review 78:426-450.
Stiggelbout, A.M., M.J.C. Eijkemans, G.M. Kiebert, J. Kievit, J.W.H. Leer and J.C.J.M. de Haes (1996), "The
"utility" of the visual analog scale in medical decision making and technology assessment: Is it an alterna-
tive to the time trade-off?", International Journal of Technology Assessment in Health Care 2:291-298.
Stiggelbout, A.M., G.M. Kiebert, J. Kievit, J.W.H. Leer, G. Stoter and J.C.J.M. de Haes (1994), "Utility as-
sessments in cancer patients: Adjustment of time trade-off scores for the utility of life years and comparison
with standard gamble scores", Medical Decision Making 14:82-90.
Streiner, D.L., and G.R. Norman (1989), Health Measurement Scales: A Practical Guide to their Development
and Use (Oxford University Press).
Sutherland, H.J., H. Llewellyn-Thomas, N.E. Boyd and J.E. Till (1982), "Attitude toward quality of survival:
The concept of maximal endurable time", Medical Decision Making 2:299-309.
Torrance, G.W. (1976), "Social preferences for health states: An empirical evaluation of three measurement
techniques", Socio-economic Planning Sciences 10:129-136.
Torrance, G.W. (1982), "Preferences for health states: A review of measurement methods", Mead Johnson
Symposium 20:37-45.
1760 P.Dolan
Torrance, G.W. (1986). "Measurement of health state utilities for economic appraisal", Journal of Health
Economics 5:1-30.
Torrance, G.W., H.H. Boyle and S.P. Horwood (1982), "Application of multi-attribute utility theory to measure
social preferences for health states", Operational Research 30:1043-1069.
Torrance, G.W., and D. Feeny (1989), "Utilities and quality-adjusted life years", International Journal of
Technology Assessment in Health Care 5:559-575.
Tversky, A., and D. Kahneman (1992), "Advances in prospect theory: Cumulative representations of uncer-
tainty", Journal of Risk and Uncertainty 5:297-323.
Verhoef, L.C.G., A.FJ. de Haan and W.A.J. van Daal (1994), "Risk attitude in gambles with years of life:
empirical support for prospect theory", Medical Decision Making 14:194-200.
Viscusi, W.K., and W.N. Evans (1990), "Utility functions that depend on health status: Estimates and eco-
nomic implications", American Economic Review 80:353-374.
Wakker, P. (1996), "A criticism of healthy-year equivalents", Medical Decision Making 16:207-214.
Wakker, P., and A.M. Stiggelbout (1995), "Explaining distortions in utility elicitation through the rank-
dependent model for risky choices", Medical Decision Making 15:180-186.
Ware, J.E., and C.D. Sherbourne (1992), "The MOS 36-item short form health survey (SF36): Conceptual
framework and item selection", Medical Care 30(6):473-483.
Weinstein, M.C. (1993), "'Time preference studies in the health care context", Medical Decision Making
13:218-219.
Weinstein, M.C., and W.B. Stason (1977), "Foundations of cost-effectiveness analysis for health and medical
practices", New England Journal of Medicine 296:716.
Williams, A. (1988), "Economics and the rational use of medical technology", in: EF.H. Rutten and S.J.
Reiser, eds., The Economics of Medical Technology (Springer, Berlin).
Williams, A., and R. Cookson (2000), "Equity in health", in: A.J. Culyer and J.P. Newhouse, eds., Handbook
Wolfson, A.D., A.J. Sinclair, C. Bombardier and A. McGeer (1982), "Preference measurements for functional
status in stroke patients: Interrater and intertechnique comparisons", in: R.L. Kane and R.A. Kane, eds.,
Values and Long Term Care (Lexington Books, Lexington, MA).
Chapter 33
ECONOMIC EPIDEMIOLOGY AND INFECTIOUS

DISEASES*
TOMAS PHILIPSON
University of Chicago
Contents
Abstract 1762
Keywords 1762
2. Predictions about disease occurrence 1767
2.1. The behavior of rational epidemics 1767
2.2. Implications for private disease eradication 1768
2.3. Rational disease dynamics of epidemics 1769
2.4. The positive effect of prevalence on assortative matching 1773
2.5. The effect of immunity on the prevalence of a disease 1776
3. Rational epidemics and public health interventions 1777
3.1. Public price subsidies 1777
3.2. Mandatory vaccination 1780
3.3. The Pareto-optimal timing of epidemic interventions 1781
3.4. The dynamic welfare effects of eradication 1784
3.5. Public intervention into allocations of information 1785
4. The welfare loss of disease and medical R&D 1787
5. The evidence on prevalence elastic behavior 1789
6. Concluding remarks 1795
References 1797
'I am thankful to participants at the Handbook Conference at the University of Chicago for comments, espe-
cially Charles Phelps and Peter Zweifel, as well as seminar participants at Duke University and the University
of North Carolina at Chapel Hill. Financial support is acknowledged from The National Science Foundation,
The National Institutes of Health, and The Research Fellows Program of The Alfred P. Sloan Foundation.
Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.R Newhouse
1762 T Philipson
Abstract
Infectious diseases are is currently the main cause of mortality in the world and have
been even more important historically. This paper reviews recent research in economic
epidemiology. Specifically, it discusses the occurrence of infectious diseases and the ef-
fects of public health interventions designed to control them. Several key points include:
differences in the predictions regarding short- and long-run disease occurrence between
rational and epidemiological epidemics, the nonstandard effects of interventions when
epidemics are rational, the desirability and possibility of eradicating infectious diseases,
as well as the components of the welfare loss induced by infectious diseases.
Keywords
epidemiology, public health, infectious disease, AIDS
Ch. 33: Economic Epidemiology and Infectious Diseases 1763
1. Introduction
According to the 1997 WHO World Health Report, in 1996 there were about 52 million
deaths world-wide. Infectious diseases caused about one-third of all deaths and repre-
sented the primary cause of mortality. Historically, the share of world-wide mortality
due to infectious diseases has been even greater, although data tend to be less reliable
for historically earlier periods. Morbidity and mortality from infectious diseases such as
tuberculosis, malaria, and acute respiratory infection have always been at the forefront
of public policy in developing countries, where infectious diseases accounted for almost
one half (45%) of mortality in 1996. Worldwide concern about infectious diseases has
once again peaked with the onset of the most feared such disease of the twentieth cen-
tury - the Human Immunodeficiency Virus (HIV) that causes Acquired Immune Defi-
ciency Syndrome (AIDS). 2 Like most communicable diseases, especially those that are
potentially fatal, HIV has incited an extensive governmental response, which has con-
sisted of regulatory measures, subsidies for research, education, treatment, testing, and
counseling. The broad scope of these and similar public interventions, and the private
behavior they aim to change, make the use of economic analysis in the study of their
effects and desirability important.
Even though infectious diseases represent the primary cause of mortality worldwide,
health economists are just beginning to understand their behavior and evaluate the many
policies aimed at controlling their impacts. Indeed, currently, most research on the pub-
lic control of infectious diseases is conducted outside economics in the field of epidemi-
ology. 3 However, the evaluation of public health measures from an economic perspec-
tive is particularly important, since economic analysis separates the health effects of
public policies from those of private decision making. In particular, it is often argued
that the public, rather than private, control of infectious diseases is one of the main
achievements of modern public health. Indeed, relying on standard arguments about the
positive external effects of disease prevention, economists often echo such arguments
for an active public role in the prevention of infectious diseases, such as AIDS. 4 How-
ever, economists have rarely attempted to explain patterns of disease occurrence or to
evaluate public interventions in the context of a society with individuals who do the
best they can given their constraints. Such recent analysis in the rapidly growing field
1 The second leading cause at 28% was circulatory diseases, the third at 11% was cancers, and the fourth at
6% was respiratory diseases.
2 The Global Programme on AIDS of WHO, as well as Mann et al. (1992), summarizes the cumulative
evidence on the prevalence of AIDS, and the mortality it has induced in the world. See also Bongaarts (1996).
Bloom and Carliner (1988) discusses the financial impact of the epidemic in the US.
3 An early economic treatment of public health issues may be found in Weisbrod (1961). The dominant form
of epidemiological analysis is exemplified by the treatments and references contained in the works by Bailey
(1975), Anderson and May (1991), Castillo-Chavez (1989), Brandeau and Kaplan (1994), and Geoffard and
Philipson (1995).
4 See for instance Stiglitz (1997, p. 15).
1764 T Plzhilipson
of economic epidemiology has cast considerably doubt on the old textbook arguments
by economists, on both theoretical and empirical grounds.
This chapter outlines the contributions made so far by economic epidemiology in
explaining the occurrence of infectious diseases and helping to understand the effects of
public health interventions. The discussion will focus on three general questions raised
by this analysis:
1. How do economic and biological epidemiology differ in their predictions about the
short- and long-run behavior of infectious diseases?
2. How do they differ in their predictions concerning the effects of public health inter-
ventions?
3. How do they differ in determining the welfare loss of a disease, and thus in the
priorities for eradication R&D, and control which should be assigned to different
diseases?
Section 2 begins with addressing the first question and considers the behavior of ra-
tional epidemics. It stresses the central interaction between the extent of disease, which
is decreased by the demand for prevention, and the demand for prevention itself, which
is increased by the extent of disease. At the heart of the analysis lies that an increase
in the prevalence an infectious disease, i.e., the share of the population infected, in-
duces growth in private prevention. Although epidemiological analysis surely discusses
how various patterns of behavior affect disease occurrence, it does not analyze the im-
plications of how behavior changes in response to the new incentives created by the
growth of a disease nor does it analyze the effects these changes have on the desir-
ability of public health measures. Central to the study of rational epidemics is thus the
prevalence-elasticityof private demand for prevention against disease. It represents the
degree to which prevention rises in response to a disease outbreak and may differ in its
form across diseases. For example, the elasticity for vaccine-preventable diseases may
represent the number of additional vaccinations induced by each new infection, while
that for sexually transmitted diseases it may represent the increase in the matching of
sex partners who have the same infection status.
This type of prevalence elastic behavior has two major implications: first, growth of
infectious disease is self-limiting because it induces preventive behavior; second, since
the decline of a disease discourages prevention, initially successful public health ef-
forts actually make it progressively harder to eradicate infectious diseases. We discuss
a very general result concerning the inability of private markets to eradicate diseases
when demand is prevalence-elastic. This robust result does not depend on the market
structure under which vaccines are produced or on how expectations are formed about
future levels of prevalence. The result stems from the existence of barriers to disease
eradication on the demand side, as the disease disappears, so does the demand for vac-
cines; the subsequent decline in vaccinations allows the disease to return. Traddition, on
the supply side, a patent-protected producer of vaccines has a special dynamic incentive
to increase mark-ups: if the vaccine eradicates the disease, the demand for the monop-
olist's product is eradicated as well. Put simply, if there were fortunes to be made in
disease eradication, we would have more of them.
By considering the changes in private incentives for preventive behavior induced by

disease growth, the economic approach provides different predictions than that of epi-
demiology for the behavior of epidemics. In particular, the two approaches predict a dif-
ferent relationship between the hazard rate into infection and the prevalence of a disease.
In epidemiological analysis, this hazard rate is an increasingfunction of prevalence. In
other words, the larger the fraction of infected people in the population, the larger is the
fraction of uninfected people who become infected in the next period. A higher preva-
lence increases the chance that a susceptible individual will meet an infected individual.
Economic incentives imply that the hazard rate into infection should be a decreasing
function of the prevalence of the disease, when private demand for prevention is preva-
lence elastic. As the stock of infected individuals grows, uninfected individuals face a
larger risk of infection and hence raise their demand for prevention. The sharp reduc-
tions in infection hazards which often accompany outbreaks seem to provide empirical
support for economic, rather than epidemiological, epidemics.
Section 3 addresses the second question by analyzing the impact of public health poli-
cies in general and their ability to eradicate infectious diseases in particular. From the
perspective of the current population only, eradication is never Pareto-optimal, because
the benefits of lowering an almost extinct disease sooner or later fall below the cost of
vaccinating more people. However, the missing market is dynamic: future generations
cannot pay vaccine producers for the benefit they derive from the producers' product.
Public interventions like price subsidies and mandatory vaccination may complete this
missing market, but such measures have often failed to eradicate infectious diseases.
Our analysis explains why eradication through such policies remains extremely difficult.
Price subsidies alone will not bring about eradication for the same reasons that price re-
ductions through increased competition will not. Both price subsidies and mandatory
vaccination programs are limited in their ability to achieve eradication, because higher
vaccination rates for individuals covered by any public program lower the incentive of
those outside the program to become vaccinated. The prevalence-elasticity of demand
lowers the price-elasticity of demand in the case of subsidies and thus lowers the total
demand effect of increased mandatory coverage. As demand rises for those who are
subsidized, demand falls among those who are not. In the extreme cases total demand
is inelastic to subsidies. This implies that classic economic justifications of Pigovian
subsidies aimed at solving the private under-provision of vaccines may be highly in-
effective due to these dynamic effects. In addition, the prevalence-elasticity of demand
does not only affect the long-run results of such interventions but also their short-run re-
sults and thereby the Pareto-optimal timing of public subsidy programs. The prevalence
of a disease competes in time with public interventions in inducing protective activity,
and this makes the timing of public intervention a crucial factor in determining its eco-
nomic efficiency. If the subsidy is not fast enough, the growth in prevalence has already
induced protection; the public sector thus ends up paying for behavior that would have
been undertaken by the private sector of its own accord.
Section 4 discusses the third question concerning the welfare loss induced by a dis-
ease and the welfare effects of medical R&D to develop new methods of prevention or
1766 T Philipson
treatment. The implications for subsidizing of R&D are quite different from those in-
volved in subsidizing prevention as discussed above. In setting priorities among control
efforts for many separate types of diseases, a major question faced by public health au-
thorities is the welfare loss inflicted upon a population by a given disease. The orthodox
approach toward assessing disease burden has employed several cost-of-illness ("COI")
measures, each of which is a product of prevalence and (possibly quality-adjusted) per-
case severity of a disease. This approach has the seemingly self-evident implication that
the more morbidity or mortality inflicted by a disease, the larger its welfare loss. In con-
trast, we argue that the more prevalence-elastic the demand for prevention is, the more
this measure understates the total welfare loss. We interpret a disease as a random "tax"
on behavior which risks exposure, a tax which will distort individuals' consumption of
risky behavior by inducing them to forego that otherwise valuable activity. Standard tax
analysis argues that a tax imposes a burden in excess of the revenues collected by the
public treasury if costly tax avoidance occurs. Similarly, if costly disease-avoidance oc-
curs, a randomly collected disease tax on exposure imposes a burden beyond the case
reports of disease incidence collected by the public health authority. However, cost-of-
illness measures of the disease-induced loss, and indeed the measures used by public
health authorities, such as The World Health Organization (WHO) or The Center for
Disease Control and Prevention (CDC), are implicitly "revenue-focused" in that they
consider only the losses from morbidity and mortality and ignore the excess burden of
disease prevention. The major point we make is that the standard cost-of-illness mea-
sures do not constitute a relatively large fraction of the total welfare loss when preven-
tion is prevalence elastic. This is for the same reason that tax-revenue does not make up
the major loss when tax-avoidance is elastic. For example, almost all loss inflicted by
vaccine preventable diseases is from the excess burden. The case is similar for AIDS
where the excess burden consists of sexual consumption foregone from fear of infec-
tion. Many economists have argued that research expenditures on AIDS are excessive
given its relatively small case load. However, few diseases have caused as much be-
havioral change as AIDS in terms of foregone sex which, if according to biologists,
is perhaps the most valued human activity. Large research expenditures to eliminate
low-prevalence but behavioral diseases such as AIDS may be justified because their to-
tal welfare loss, the case-load revenue and the excess burden, is larger than for more
common diseases.
Section 5 discusses existing empirical analysis of the type of prevalence-elastic be-
havior that underlies the theoretical analysis reviewed. We consider evidence for in-
fluenza, AIDS, and measles and attempt to estimate the prevalence-elasticity of de-
mand for prevention. We review the results of a particular US study that tracked the
AIDS-preventive behavior of young individuals during the 1980s, which saw a great
deal of variation across states in the growth of the disease, using panel data from the
US National Longitudinal Survey of Youth (NLSY). We report on similar studies using
state-variation in measles outbreaks in the late 1980s. The prevalence elastic behavior
documented by these and other studies suggests the need to incorporate such responses
into the formulation of public health policy aimed at limiting the occurrence of disease.
Finally, Section 6 concludes by outlining several directions for future research.
It should be noted that this chapter discusses a set of questions subjectively selected
from the recent literature in this area. Naturally, a single review chapter cannot claim to
cover everything that has been and is being done in a expanding area of research, and
this chapter is no exception. 5 This is a deliberate choice, since we find it more useful
to focus on a few ideas central to a research agenda, rather than provide a disjointed
discussion of an exhaustive reference list.
2. Predictions about disease occurrence
This chapter first discusses the implications incentives have for the behavior of both
short-run epidemics and the long-run occurrence of disease. Economic and biological
epidemiology make different predictions about disease occurrence mainly due to their
different predictions about the relationship between prevention and prevalence.
2.1. The behavior of rationalepidemics
Consider individuals who are classified into four health categories at a given time t: sus-
ceptible St, infected It, immune through recovery Rt, and outside the system. Normal-
izing the total population to unity, we refer to the fraction infected in the population, It,
as the prevalence of the disease. A future path of prevalence is denoted Lt {II,; s t}
and a future path of prices is denoted pt _ {Ps; s t}, with the instantaneous demand
for vaccines at time t for two such paths denoted D(It, pt). We denote by b and m the
birth and mortality rates into and out of the system, respectively, by w the rate at which
infected individuals are withdrawn naturally from infection into immunity, and by t3 the
probability of transmission conditional on exposure to an infected person. The changes
over time in the health of the population are determined by:
dS = h[l- D(I t t)] - Sti t -mSt,

dt
d - ( + m), (2.1)
dRt=bD(it pt) + wt - mRt.
The change in the fraction of susceptible individuals is due to the entry of newborn in-
dividuals who do not vaccinate. Exits are due to new infections and non-disease-related
5 For example, we do not discuss the important literature on economic growth and AIDS: see, for instance,
Cuddington (1993a, 1993b) and Bloom (1997). Another area omitted is the statistical literature by economists
on AIDS forecasting (see, e.g., Hay and Wolak (1990, 1994)). For alternative reviews on markets for vaccines,
see Weisbrod and Huston (1987) and Pauly (1994).
1768 T: Philipson
mortality. The change in prevalence is due to the entry of new infections, while exits
are due to immunity and infection-related mortality. New infections are caused by con-
tact between susceptible and infected individuals under random matching, conveyed by
in the term ,BSt It. The change in fraction of recovered immune individuals is due to
the entry of newborn individuals who vaccinate, as well as those individuals recovering
from infection, and the exits of agents through non-disease-related mortality.
It follows directly that the prevalence rises over time whenever
dlt > 1
d_ O X_ W Bst 1. (2.2)
dt w+m
The factor BSt is the rate at which infected individuals infect susceptible individuals
with whom they come into contact, and the factor l/(w + m) is the average time of
infection. For the infected stock to grow, the average number of secondary infections
by an infected individual must be above unity, so that an infected individual more than
replaces himself by the time he or she exits the infected population. When there are only
susceptible individuals, which is the relevant case when a disease is to be eradicated, the
secondary infections generated by a single new infection is denoted p =B/(w + m), so
that the disease can take off in a completely susceptible population only when this ratio
is above unity.
2.2. Implicationsfor private disease eradication
A major technology aimed at limiting such diseases has been vaccines. Although the
introduction of a vaccine usually produces a sharp drop in the occurrence of a disease,
the eradication of vaccine preventable diseases predicted by many at the time of these
inventions has not been achieved except for smallpox. 6 Of the roughly forty vaccines
on the market, only the smallpox vaccine has eradicated its target disease. Diseases
such as measles, tuberculosis, and different types of influenza persist, despite explicit
governmental efforts to eradicate them, and recent attempts to develop a vaccine against
HIV or AIDS raise important questions about the causes behind these difficulties.
The prevalence-elasticity in private markets, coupled with rational demand for vac-
cines, represent powerful forces which make it difficult for private markets to achieve
eradication. We call the demand for vaccines prevalence-dependentif, when prices are
positive in the future, demand vanishes for low enough prevalence. That is, for any
strictly positive price path pt, there is a prevalence path I (pt) below which demand
vanishes: D(t , pt) = 0 for all Pt < t(pt). It can be shown that prevalence depen-
dent demand requires the simple condition that the benefits of vaccination not be large
"enough" when prevalence levels are low enough. If demand is prevalence-dependent
and if the prevalence goes to zero for any future prices, there must be a time to after
6 See, for instance, Plotkin and Mortimer (1988), and The World Bank (1993).
which the prevalence is driven down to a level which generates small enough demand.
As fewer individuals vaccinate after to, the population becomes increasingly suscepti-
ble. However, when an infection can regenerate itself in a susceptible population, which
occurs when p > 1, this implies that the prevalence increases again, making eradication
infeasible. In other words, the disease cannot be eradicated under positive prices when
p 1.
Since this argument holds for any price, it implies that, regardless of the market
structure in which vaccines are produced, the disease is not eradicated, since prices
are presumably above costs in the long run for any feasible market structure. In par-
ticular, although a vaccine monopolist is faced with a problem similar to that of using
an exhaustible resource, the resource (prevalence) will never be exhausted. Naturally, if
competition drives prices down to minimum average costs of production, then eradica-
tion is not achieved under this market structure either. Interestingly, this argument is not
only robust to the type of market structure, but also to many forms of expectations: it is
true under myopic as well as rationally formed expectations. The general difficulty with
eradication thus comes from the demand side of the vaccine market, rather than from
the supply side. 7
2.3. Rationaldisease dynamics of epidemics
So far, we have not specifically investigated the demand for prevention and have merely
outlined the implications of various properties of this demand. Geoffard and Philipson
(1996) discuss an environment in which the rational protection behavior of an individual
in an epidemic can be traced out simply. 8 A version of this model is also discussed in
Auld (1997). Consider a utility function u(h, d) over a binary demand for protection
(d = 1) and the state variable h representing the susceptible (s) or infected (i) health
state. Proceeding heuristically to illustrate the main ideas, the value function evaluated
in the susceptible state may be written as
V(s) = max{u(s, 1)+ aV(s), u(s, O) + oa[,ltV(i) + (1 -/BIt)V(s)]}, (2.3)
where a is the discount rate and /Bis the transmission parameter. This says that con-
tinued protection today implies susceptibility tomorrow, but if the individual does not
protect, he risks becoming infected with a probability which increases in prevalence.
This directly implies that the individual remains exposed as long as the current benefit
of exposed activity outweighs the expected loss in the future due to risk of infection:
d= u(s, O)- u(s, 1) aBIt[V(s ) - V(i)]. (2.4)
7 For a contrary view see, for instance, General Accounting Office (1994) for a discussion of why monopoly
production of vaccines makes eradication infeasible.
8 This analysis abstracts from the joint decision making treat may take place for certain diseases such as
STDs.
1770 T Philipson
Protection in an epidemic is then characterized by a simple rule under which the

individual engages in protection only after a reservation prevalence, denoted by K, has
been reached. Solving for the value function, this reservation prevalence can be shown
to satisfy
d=0 It < K _ [u(s, )-U (s0)] (2.5)

l [U(s, 1) - U(i, 0)
In other words, there is a threshold prevalence below which an agent engages in trans-
missive behavior and above which he engages in protection. Intuitively, this reservation
prevalence rises with the instantaneous cost of protection and the discount rate, and falls
with the cost of infection and the probability of transmission conditional on exposure.
The dynamics for a closed population, which no agents enter or exit, can then be
traced out by assuming that the reservation prevalence levels are distributed according to
the cumulative distribution function F(K). The epidemic starts with a prevalence level
denoted I (K, 0) in the group with reservation level K and then takes off by infecting in
a given period those who have not started to protect yet and have not been infected in
the past
I -
j I (K, 0) dF,
I fI )d (2.6)
It = BItQt G(It),
where G(It) -,< t [1 - I (K, 0)] dF is the susceptible population choosing to engage
in transmissive activity at prevalence It and Qt - exp[- fot BI ds] represents the share
not infected in the past even though they were exposed. This function G summarizes the
behavioral response of the population to the growth in prevalence. For an epidemiolog-
ical model in which behavior is exogenous or prevalence inelastic, the function would
be constant, that is, an increase in prevalence would not cause any change in protective
behavior. The degree to which the population's protective behavior responds to an in-
creasing prevalence of a disease determines the epidemic behavior of the disease. More
precisely, Geoffard and Philipson (1996) outline the conditions under which the growth
of the epidemic reveals the preferences of the population, in the sense that the risk atti-
tudes of the population implicit in F can be identified through the time path {It; t > 01
of prevalence.
This characterization of behavioral response demonstrates how economic models
generate implications observably different from those generated by epidemiological
models. The key implication concerns the hazard rate into infection from susceptibil-
ity and its relationship to the prevalence of a disease. The hazard rate measures the
propensity to be infected conditional on not being infected yet. With the inelastic be-
havior assumed by epidemiological analysis, this hazard rate is an increasingfunction
of prevalence. In other words, the larger the fraction of infected people in the popu-
lation, the larger is the fraction of uninfected people who become infected in the next
period. As prevalence rises, so does the chance that a susceptible individual meets an
infected individual. This is true across a wide variety of epidemiological models, since
they all share the feature that the demand for exposure does not respond to prevalence.
More precisely, if the demand for exposure is prevalence inelastic (G = 1), then the
hazard function h (It) satisfies:
h(It)= 1 I - Qt =It. (2.7)

1- Qt
Therefore, the prevalence has a positive effect on the hazard rate of infection ah/aI > 0.
In contrast, the incentives of epidemics imply that the hazard rate into infection may
be a decreasing function of the prevalence of the disease, because an increase in risk
provides a larger incentive for susceptible individuals to increase protective behavior. In
other words, although more and more people may be infected as the epidemic grows,
the share of still uninfected that become infected falls. The hazard as a function of
prevalence is now:
h(t)= -it - ,BihQG(I)

((2.8)
1-It 1- I
It follows that the relationship of prevalence on the hazard rate is now given by
I ah 1 G' 1
--~+ --I - - , (2.9)
h aI 1-I G 1-I
where l is the absolute value of the elasticity of G. It follows that
ah 1
al e*0 1 (2.10)
If the elasticity of the prevalence response is large enough, the hazard decreases as a
function prevalence. In particular, the elasticity needs to be larger the larger is preva-
lence, since the counteracting positive effect is more pronounced. The more the behavior
responds to the new risk, the lower the hazard rate into infection is among susceptible
people who have not already been infected.
Figure 1 illustrates empirically the way hazard rates often fall and eventually level
off as a result of a rising prevalence [see Geoffard and Philipson (1996) and Auld
(1997)]. The curves are estimated using data from the San Francisco Men's Health
Study (SFMHS). 9 The SFMHS, a longitudinal epidemiological survey study, consists
9 See Geoffard and Philipson (1995) for details on the estimation and "Sampling Methods and Wave 1 Field
Results of the San Francisco Men's Health Study", Survey Research Center Technical Report, University of
California, Berkeley, for the details on the sample design.
1772 T Philipson
Percentage of Hazard Rate into

Cohort Infected Infection
U,44
1 0,04
0,035
0, 43 i-
-0,03
0, 42 -i-
0,025
0, 41 - .
0,02
0,015
0,4 -i-
- 0,01
0,39
0,005
0,38 - I I l l ! l l l l l I
l
, 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Wave of Survey
Figure 1. Estimated prevalence and hazard rate.
of 15 half-year waves of data for individuals living in San Francisco during the period
1983-1992, and is designed to yield data on the natural history of AIDS. The respon-
dents were interviewed to obtain information about behaviors, attitudes, and beliefs rel-
evant to AIDS. The sample population consisted of English-speaking, unmarried men
aged 25-54. The survey sample was a stratified two-stage sample of all households
within the designated census tracts. All eligible persons in each selected housing unit
were included in the sample.
In the SFMHS, the duration of susceptibility is defined as the number of waves for
which the individual remains HIV-negative, that is, the wave at which he or she becomes
infected. Such durations were observed under both left and right censoring. In other
words, HIV-positive individuals entered the first wave and HIV-negative individuals
were present at the end of the last wave.
Figure 1 displays the prevalence rate and the hazard rate as functions of the waves
constructed from these data. The horizontal axis indicates the waves at which the rates
were measured. The estimated prevalence rates, It for wave t, are plotted on the increas-
ing (dotted) line and depict the percentage of individuals in the survey who were HIV-
positive at each wave. These prevalence rates were estimated by computing, for each
wave, the fraction of individuals present throughout sampling who were HIV-positive
in the sample. The hazard rates are plotted on the generally decreasing (solid) line and
depict, for each wave t, the fraction of individuals present throughout sampling who
were HIV-negative in wave t - 1, but turned HIV-positive during wave t. 10 Due to the
attrition of subjects, the observed prevalence rates are not equivalent to the observed
hazard rates. The figure displays a negative relationship between the estimated hazard
rates and prevalence levels: the estimated hazard function is decreasing and the cohort
prevalence is increasing. The displayed relationship thus provides suggestive evidence
for prevalence-elastic rational epidemics. Moreover, the eventual flattening of the haz-
ard rate around the 10th wave of the survey corresponds to a complete reduction of new
infections. This common pattern of a sharp reduction in hazard rates into infection as
epidemics grow is suggestive of the type of prevalence elastic behavior discussed.
2.4. The positive effect of prevalence on assortativematching
In the case of sexually transmitted diseases (STDs) such as AIDS, one form of pro-
tective behavior may be the choice of safer partners. In the extreme case of perfect
"assortative matching" on infection status, i.e., when all HIV-positives match with other
HIV-positives and HIV-negatives with other HIV-negatives, the growth of the disease is
zero. Therefore, the degree to which incentives lead to growth in infection-dependent
matching of partners determines the growth of a sexually transmitted epidemic.
Consider an environment which contains different classes of individuals whose risk
of HIV infection is known, and where risk is defined as the percentage of the class
infected by the AIDS virus. Such classes may be subpopulations stratified on gender,
appearance, weight, race, or other observable demographic characteristics. Given this
environment, determining who will engage in sexual activity with whom, and what type
of activity they will engage in determines the growth of the disease. Specifically, it de-
termines the demand for unprotected sex by pairs of individuals, each of whom has a
different infection status. The matching of infected individuals with uninfected individ-
uals is necessary to generate new cases of the disease. The general theory of matching
markets predicts that among traders of different quality levels, high-quality traders will
match among themselves and low-quality traders will match among themselves [Becker
(1991)]. In other words, low-risk individuals match up with other low-risk individu-
als, and high-risk individuals with other high-risk individuals. The complementarity in
10 By focusing on the propensity of HIV-negatives to become infected, the hazard rate is different from the
so-called incidence rate, which is the flow of new cases divided by the whole population size.
11 However, it is well known that such negatively sloped hazard functions may also be due to unobserved
differences among individuals [see, e.g., Heckman and Singer (1984)].
1774 T Philipson
health status that generates this assortative matching stems from the fact that low risk in-
dividuals have more to gain by the choice of low-risk partners than high risk individuals
do. This has the important implication that disease growth is slower than in the random
matching case considered by epidemiological analysis. Since the disease grows due to
sexual partnerships between negative and positive individuals, the economic matching
incentives slow disease growth. Such matching changes the matching patterns over the
disease, inducing more dependence across the statuses of partners as the disease grows.
For the case of HIV, Dow and Philipson (1996) estimated the extent of both such as-
sortative matching, as well as the extent to which such matching reduces HIV incidence
relative to the random matching assumed by epidemiological models. They estimate that
on average HIV-positive individuals are more than twice as likely to have HIV-positive
partners, and that this assortative matching reduced HIV incidence by about one-third.
They use the San Francisco Home Health Study (SFHHS), 12 which is detailed enough
to allow estimation of the joint infection status of a pair of partners. The fraction of
matches with a given infection status at a point in time may be represented by a 2 x 2
table T of the form:
T [P00 P ,] (2.11)
pio Pi l
Pij is the fraction of couples of infection status (i, j), where the respondent's infec-
tion status is given by i, while the respondent's partner has status j. If i or j is 0, the
individual is uninfected. The prevalence of the disease among respondents, p, or part-
ners, q, is then defined as the marginal probabilities of the table T: p =-pm + Pl and
q P01o + Pl . The degree to which there is assortative matching on infection status is
measured by the degree of positive dependence in the table. It is measured by the ratio
of the conditional probabilities of being with a positive partner for positive respondents
versus negative respondents,
P11/P
6a / (2.12)
po0/(l - P)
In other words, the dependence is measured by the fraction of positive partners of

positive respondents, relative to the fraction of positive partners of negative respon-
dents. Such matching is important for infection incidence since new cases are generated
only by unprotected contact between individuals of different infection statuses. More
precisely, the incidence in the table T [denoted (T)] is given by
((T) =/,[Poi + Plo], (2.13)
12 The SFHHS is an epidemiological study designed to yield data on the prevalence of HIV and related risk
factors in multicultural neighborhoods, including information about behavior, attitudes, and beliefs relevant
to HIV. The sampled population included persons currently unmarried, aged 20-44, and residing in San Fran-
cisco census tracts with substantial proportions of blacks and Hispanics.
where B, as before, is the probability of transmission and represents the likelihood of

an infection conditional on a match between an infected and a susceptible individual.
For example, when there is no assortative matching as in epidemiological analysis,
= 1. The incidence is then given by (p, q, = 1) = [p(l - q) + q(l - p)]. The
matching independence assumption is inconsistent with the assortative matching gen-
erated by the incentive of a non-infected individual to avoid infection. These incentives
imply that S > 1. As is well known, assortative matching is not an assumption; rather it
is implied by the fact that, ceterisparibus, everyone prefers their partner to be negative
rather than positive. Moreover, the greater the morbidity or mortality of a disease, the
stronger the incentive of negatives to search for and match with other negatives.
Figure 2 illustrates the differences in the incidence for different levels of assortative
matching consistent with a given prevalence level for the simplest possible case, in
which the prevalence levels of the respondents and partners coincide: p = q.
The top triangle depicts the disease incidence under the minimum assortative match-
ing consistent with the prevalence, the lower line depicts the incidence under maxi-
mum assortative matching, and the oval shaped line depicts incidence under infection-
independent matching. In other words, the top triangle represents the largest number of
matches between individuals of different infection status consistent with a given preva-
lence level. The bottom x-axis represents the lowest number, zero, since all negatives
of one group can match with negatives of the second group when the two groups have
the same prevalence levels. In the intermediate case, matches are infection-independent.
For the SFHHS of gays in San Francisco, the HIV-prevalence in the sample was roughly
one-half. Since this rate is high, the figure implies that incidence predictions will be very
sensitive to different levels of assortative matching. Dow and Philipson (1996) estimated
0/2
Figure 2. Assortive matching and incidence under same prevalence.

1776 T Philipson
that an HIV-positive individual is more than twice as likely as an HIV-negative individ-

ual to have a HIV-positive partner; > 2. The incidence reduction implied by such
matching varies between 40 and 25 percent so that in the figure the estimated incidence
line is about a third of the way from the independence line. 13
2.5. The effect of immunity on the prevalence of a disease
With many infectious diseases, an individual can only be infected once and is there-
after immune to future infections. In the case of such an immunity-bearing infection,
an infection may be a good which one is willing to pay for, rather than a "bad" which
one has to be paid to accept. This occurs if the infection is more severe when caught
in adulthood, so that a "front-loaded" early infection may be optimal. Many childhood
infections have this feature, and indeed the "get-it-over-with" attitude of parents reflects
the value of early infection. Absent manufactured artificial vaccinations, contracting and
surviving a disease represents the only means of "vaccinating" one's self, although at
a higher cost. Clearly, if there is no immunity conferred, front-loaded infections fail to
be optimal. Moreover, if the conditional risk of mortality remains constant over an indi-
vidual's lifetime, there is no incentive to receive immunity through infection. However,
consider such immunity-bearing infectious diseases as chicken pox, measles, rubella,
and mumps - often called child diseases, because of the low average age of infection.
Before vaccination was available, immunity represented the only source of protection.
Cohort studies of immunity-bearing diseases conducted before vaccine development
invariably showed that the fraction ever-infected or currently infected in a given age co-
hort rises so rapidly that, as the cohort reaches its late teenage years, around 95% of its
members have been infected. Nearly all of the remaining 5% share of the cohort escapes
infection for the remainder of their lives, so that the growth rate of cohort prevalence
goes to zero after the teenage years. This pattern held true for all low-cost child diseases
such as mumps, rubella, and chicken pox. For more severe diseases, however, such as
polio and diphtheria, cohort prevalence never escalates to nearly universal seropositiv-
ity. Since infection confers immunity at the cost of experiencing the disease, it may be
interpreted as a purchase of immunity similar to the purchase of medically manufac-
tured vaccines. For low-risk childhood diseases, it appears that the demand for such
immunity starts out as high for young children, but goes to zero after the late teenage
years. This property is consistent with the following two factors: first, as expected fu-
ture lifetime falls, the benefits of lifetime immunity fall as well, so benefits are highest
for young children; second, the price of immunity (the morbidity cost of infection) in-
creases in age. For example, the risk of dying from childhood diseases rises with age.
Both factors make early infection preferable to later infection. Indeed, consumers may
13 Observed assortative matching may be infection-induced, in that infection may transform a pair of part-
ners with initially different infection status into a pair of partners with the same status. The paper discusses
how to isolate this infection-induced assortative matching from incentive-induced assortative matching. Such
couples, however, represent a small share of the SFHHS.
be willing to "pay" for their children to be infected when vaccines are unavailable, as
evidenced by the common practice of having children sleep in the same bedroom with
a sick sibling, or so-called "measles parties". The apparent desire of parents for early
immunity, whether through infection or vaccination, reflects the fall of immunity's ben-
efits and the rise of infection cost with age. This incentive explains both the high rate
of infection for young people, and the extremely low rate of infection for individuals
past the late teenage years, after which the cost of infection increases steeply. Epidemi-
ologists explain such patterns in the age structure of disease through school mixing
patterns. Since infected children are more likely to meet susceptible children when they
are in school, the hazard rate into infection increases in early school age. This assumes
that the likelihood of infection conditional on exposure remains constant across ages,
but that exposure patterns differ across ages. However, mixing is a choice governed by
its costs and benefits. Parents let their children go to school in the potential presence
of a disease when the purchase of immunity through infection is on balance valuable.
Parents react more negatively when the infection bears no benefit as for diseases such
as AIDS; indeed, children with AIDS have faced pressure not to attend school.
3. Rational epidemics and public health interventions
The previous sections discussed some of the positive implications of prevalence-elastic

demand for disease occurrence. We will now discuss and evaluate several types of public
health interventions in light of such behavior of epidemics and long-run prevalence. 14
3.1. Publicprice subsidies
Economists have long offered qualitative arguments concerning the positive external
benefits of vaccination, effects which result in the private under-provision of vaccines.
This has provided a motivation for Pigovian subsidies aimed at correcting the under-
provision. However, we find that the steady state price elasticity of demand for vaccines
is reduced under prevalence-elastic demand, so that such subsidies become ineffective.
Therefore, the Pigovian subsidies traditionally seen as resolving the under-provision
problem of vaccines can be short-run, or out-of-steady-state, arguments.' 5 Under a
14 In predicting the effects of public health measures on disease occurrence, the economic approach differs
from the epidemiological mainly in assuming that there is demand in the private sector for disease prevention,
which may or may not be advanced by efforts undertaken in the public sector. The epidemiological approach
fails to consider both the possibility of privately provided disease prevention and the possibility that public
interventions may be rendered ineffective by private responses. Consequently, epidemiological models tend to
credit reductions in disease prevalence entirely to public interventions. Philipson and Posner (1993) provide
a basic qualitative discussion of the impact of private incentives on the AIDS epidemic and the reduced role
of public intervention. For a critique of this view, see Kremer (1995).
15 Most standard treatments of public finance or health economics discuss these Pigovian subsidies for so
called under-provided vaccines. See, for instance, Stiglitz (1978), Fuchs (1989), and Phelps (1992) and Pauly
1778 T Philipson
prevalence-responsive demand, the relatively low price elasticity may limit the efficacy
of Pigovian subsidies; in the extreme case, such subsidies may not raise total demand at
all.
In the steady state, the fraction of individuals in each of the three health states re-
mains constant over time at levels (S, I, R), now denoted without time indices. Denote
by D(I, p) the demand for vaccination under a constant future prevalence path at level
I, where Dp < 0 and DI > 0. The positive sign of DI we refer to as the prevalence
response of demand. The benefit of vaccination rises with prevalence. It can also be
shown that for each stationary price, there is a unique steady state prevalence denoted
I(p), which is increasing in price and locally stable. 16 Therefore, unlike many other
dynamic economic systems, our model will not exhibit cycles, even with myopic de-
mand: vaccination-induced cycles can only occur with a lagged prevalence response of
demand.
Using the unique and positive relationship between prevalence and price, vaccine
demand can be written as a function of price alone, as D(p) = D(I(p), p). The total
effect of a price increase on this demand then consists of not only the standard direct
negative effect, but also the indirect and positive effect through increased prevalence:
Dp = Dp + DIIp. (3.1)
The indirect positive effect depends on the degree to which prevalence rises with price.
This effect, in turn, falls with the prevalence response in demand DI, as can be seen in
the following steady state relationship: 17
1
DI + (w + m)/b D (3.2)
As price increases, demand decreases, causing prevalence to increase. The rise in preva-
lence creates a counteracting feedback by causing demand to increase. This feedback
limits the impact of price on prevalence. This counteracting effect becomes larger as
demand becomes more prevalence responsive. Consequently, the total effect of price on
demand falls with the prevalence response of demand DI, as can be seen by substituting
Ip into the expression for Dp:
Dp = b Dp. (3.3)
1+ -DI
W+m
(1994). Brito et al. (1991) argue that a 100 percent vaccination rate is not Pareto optimal. However, this does
not seem to be unique to the externality of vaccines - Pigovian subsidies may improve efficiency, but banning
behavior altogether may carry efforts too far. Recent analysis by Kremer (1996), Francis (1997), Xu (1998),
and Hsu (1998) analyses dynamics with heterogeneity including the role of subsidies in a dynamic setting.
Ainsworth and Over (1997) discuss the unique problems of AIDS in developing countries.
16 See Geoffard and Philipson (1997). We ignore the steady state (S, I, R) = (1,0, 0) with zero prevalence.
17 See Geoffard and Philipson (1997).
Ch. 33.' Economic Epidemiology and Infectious Diseases 1779
The total price effect turns out to be the partial effect discounted by a factor which falls
with the prevalence response of demand. In sum, the larger the prevalence response of
demand is, the less it responds to price.
When public subsidies vary with prevalence, as denoted by s(I), we call them
counter-cyclical if s'(I) < 0, or pro-cyclical if s'(I) > 0. Without a doubt, the majority
of public sector subsidy programs, whether international, national, or sub-national, are
pro-cyclical. Once we consider prevalence dependent subsidies, the total steady-state
demand function becomes
Ds(I, p) = D(I, p - s(l)). (3.4)
The total prevalence response of demand is then
dDs
dI = D - sDp (3.5)
An increase in prevalence affects subsidized demand directly, and indirectly by raising

the level of subsidies under a pro-cyclical policy, or lowering the level under a counter-
cyclical policy. An increase in subsidization induces a decrease in price and vice-versa.
One can also show that the larger the prevalence response of demand, the lower the
steady state prevalence I; formally, I is lower whenever DI is uniformly larger. 18 This
directly implies, therefore, that a pro-cyclical subsidy policy lowers steady state preva-
lence more than a counter-cyclical policy. Furthermore, since demand varies less with
price when the prevalence response of demand is larger, pro-cyclical subsidies mitigate
the effects on quantity of a less competitive market structure.
For a competitive market facing a stationary subsidy s, the market eradicates the
disease only if the subsidy covers the minimum average cost of production. On the other
hand, a monopoly producer of vaccines faces an unconventional incentive to keep the
disease alive: if the disease is eradicated, so is the demand for the monopolist's product.
Consider a monopolist who faces constant marginal costs c. Clearly, the monopoly price
can never fall below the subsidy level since demand is completely price inelastic at
such a price. If the monopoly price equals the subsidy, the vaccine is free to consumers
who demand it universally, as in D(O) = 1. Universal demand is assumed to lead to
eradication after a length of time T. Given a discount rate a, the profits gained from
eradication equal:
i = (s - c)D(O) e- at
dt, (3.6)
18 For a proof, see Geoffard and Philipson (1997).

1780 T Philipson
When the monopoly price p exceeds the subsidy s, we do not have universal demand.
In this case, profits are given by:
IN = , (p-c)D(p - s) e - t dt, (3.7)
If there exists a price p such that IN > rlE, the monopolist will not eradicate the
disease. This condition is equivalent to (s - c)D(O)(1 - e - T) > (p - c)D(p - s). This
condition demonstrates the crucial point that eradication is less likely to be profitable the
more prevalence-elastic demand is. High prevalence elasticity implies that Dp will be
low, and thus D(p - s) will not be much lower than D(O). A monopolist facing inelastic
demand will never find it profitable to eradicate, because the short-term increase in
quantity offered by eradication will not be large enough to compensate for the zero
future profits implied by eradication.
The monopolist chooses eradication provided current profits from universal vaccina-
tion exceed the loss in future profits from the elimination of the disease. However, the
important point here is that eradication is less likely to be profitable the more respon-
sive is demand to prevalence. This is so because eradication is less profitable the less
demand responds to price and, as discussed, DP falls with D1 . When demand is price
inelastic, it never pays to eradicate because the monopolist earns a loss both before and
after the disease is eradicated: a loss after eradication (as discussed) because his product
is valueless, and a loss before eradication because raising the price will increase current
profits. In addition, if future profits are not heavily discounted, so that a is high, the
cost of eradication is higher. In sum, if demand is highly responsive to prevalence or if
discounting is moderate, subsidized eradication is not profitable. 19
3.2. Mandatory vaccination
Virtually all observed mandatory vaccination programs are partial - they do not cover
whole populations or even whole age groups. Therefore, private decisions to vaccinate
outside of public programs remain an important component of the total demand for vac-
cination. The total demand when a public program covers a fraction f of the population
is given by
DT(I, p, f) = f + (1 - f)D(I, p). (3.8)
19 Subsidized suppliers, rather than demanders, may not be prevalence-elastic under so called supplier-
induced demand. Supplier subsidization was undertaken in England in 1990 when general practitioners
received bonuses if they achieved prespecified immunization targets for their patients. The fraction who
achieved the targets increased from 55% at the start to 85% at the end of the program in 1992 (Principal
Medical Officer, Department of Health, England, 1994). This policy raises the interesting question of whether
health activities with positive external effects, such as those in many areas of public health, may be efficiently
provided by supplier-induced demand.
The first term is the mandatory demand in the program, while the second term is the
private demand outside the program. Partial mandatory programs crowd out the private
demand for vaccination outside the program, in the sense that some individuals would
vaccinate in the absence of the program, but will not vaccinate in its presence. 2 0 More
precisely, the marginal effect on demand of increased public coverage is
dDT
-= (1- D) + (1 - f)DIIf. (3.9)
df
The first term is the direct positive effect resulting from the increased public coverage of
individuals who otherwise would not have vaccinated. The second term is the indirect
negative effect on private demand by individuals not covered by the public program. It
reflects the negative effect on demand exerted by a decrease in prevalence. This term
increases in the prevalence response of demand. Therefore, the higher the prevalence
elasticity, the less effective mandatory vaccination programs are at raising total demand.
3.3. The Pareto-optimaltiming of epidemic interventions
Critics of the public response to the AIDS epidemic commonly charge that public health
interventions occur "too late" into the epidemic. Table 1 below summarizes the speed
at which state governments in the U.S. responded to the AIDS epidemic, as measured
by expenditures on prevention programs. The table is computed using data on the time
elapsed before states implemented programs for education about AIDS prevention, the
main type of public preventive expenditure program in the U.S. A row of the table cor-
responds to a given fiscal year. The first column (Total left) refers to the number of
state governments that had not implemented an education program at the start of the
fiscal year; this represents the survival rate of non-implementation. The second column
(Starters) indicates the number of governments that started an education program dur-
ing that fiscal year; this represents the exit rate from non-implementation. The third
column (Survival) indicates the fraction who survived, or the fraction of state govern-
ments which had not implemented an education program by the start of the next fiscal
year. Lastly, the 95% confidence intervals are reported for this survival curve.
The table reveals the frequently discussed slowness of the public sector's response
[see, e.g., Shilts (1987)]. This is illustrated by the fact that, even in 1988, well into the
national epidemic, more than half of the states had not established prevention programs.
The common focus on the calendar timing of the public response is misleading, how-
ever, since it is disease prevalence, not disease duration, that determines incentives for
protection in the private sector. The rapidity of public-sector response should be judged
in relation to prevalence levels, since the latter drive private incentives for protection.
20 This crowding out effect is one possible interpretation of the relatively low pre-school vaccination rates in
the U.S., given the mandatory vaccination required in public schools.
1782 T: Philipson
Table 1
Duration in years before start of main AIDS prevention program
Fiscal year Total left Starters Survival [95% Conf. Interval]
1986/87 51 12 0.7647 0.623 0.859

1987/88 39 7 0.6275 0.480 0.744
1988/89 32 5 0.5294 0.385 0.655
1989/90 27 18 0.1765 0.087 0.292
1990/91 9 1 0.1569 0.073 0.269
1992/93 8 0 0.1569 0.073 0.269
Source: Computed using data from the AIDS Policy Center, Intergov-
ernmental Health Policy Project, George Washington University.
Figure 3 below depicts the resulting relationship between state implementation of ed-
ucation programs and state prevalence levels. Specifically, the figure shows the share of
US state governments as a function of disease prevalence in the states that had not yet
adopted an AIDS education expenditure program. It is the survival curve of the dura-
tion until program adoption where the duration is measured in terms of the prevalence
level. 21
The figure illustrates that disease prevalence increases adoption of public programs,
and that the rate of adoption rises with prevalence, as evidenced by the convexity of
the survival curve. While calendar response times do not affect private incentives, the
slow calendar response and the rapid prevalence response of governments suggest that
calendar response times may also be uncorrelated with the public sector incentives. 2 2
What determines the Pareto timing of an intervention into an epidemic caused by a
population that is heterogeneous in the willingness to bear risk? To gain insight into the
efficient timing of a subsidy program, consider first the effect of a complete subsidy to
a single individual, which will not affect overall prevalence It. The impact of the sub-
sidy on the behavior of this single individual depends on whether the prevalence has
already reached that individual's threshold prevalence K. If the prevalence has reached
this threshold prevalence, the subsidy has no effect on that individual. This implies
that a subsidy to a heterogeneous population affects only those individuals with a larger
threshold prevalence. More precisely, the prevalence and the public subsidy program are
competing incentives influencing the individual demand for protection, which implies
that subsidy programs may not be Pareto-optimal if they are undertaken when preva-
lence is too high. If they come in too late, the subsidies may be irrelevant for a large
share of the population for whom prevalence itself has already induced protection. This
21 We measure AIDS prevalence as AIDS cases reported to the Centers for Disease Control, per 100,000
state residents.
22 Indeed, the unconditional correlation between calendar duration and prevalence duration turns out to be
close to zero. For example, the District of Columbia responded first in terms of calendar time, but last in terms
of prevalence.
1.0
0.8
0.6
0.4
0.2
0.0
0 20 40 60 80 100
Prevalence duration (cases/1 00k)
Figure 3. Survical in state government spending.
may appear counter-intuitive, since it suggests that the larger the case load, the smaller
is the role for government intervention through subsidies. However, as prevalence rises,
the private incentive to undertake protective conduct also rises, thereby lowering the
effect of government subsidies.
Consider the aggregate effect of a typical subsidy program which is financed by non-
eligible individuals. Let Gn(p, I) and Ge(p, I) denote the share exposed in the dynamic
model for non-eligible and eligible individuals of the program, respectively. For a pro-
gram involving a percentage subsidy of s for protective activity, there is a potential for
Pareto improvement only if non-eligible individuals are willing to pay eligible indi-
viduals to demand protection, and they are willing to pay an amount at least equal to
the program cost. The following proposition shows that this can never be the case if
the subsidy program starts "too late", in terms of disease prevalence. More precisely, it
can be shown 2 3 thatfor every subsidy level s, there exists a limitprevalence level I (s),
such that the subsidy program is not Pareto-improving if it starts after prevalence has
reached I (s). Moreover the limitprevalence level I (s) decreases with s. The higher the
proposed subsidy, the larger must be the benefit of the program. Therefore, the program
23 See Geoffard and Philipson (1996).

1784 T Philipson
must be undertaken at lower prevalence levels in order to have greater effects. Interest-
ingly, at the later stages of an epidemic, lower subsidization levels outperform higher
levels.
The revenue for such a subsidy program would come from non-eligible individuals.
Self-interested individuals who have already been induced by disease prevalence to en-
gage in protective behavior, however, will pay nothing for other individuals to engage in
protective behavior, since they have nothing to gain from others' protective activity. 2 4
If a public program is not fast enough, the growth in prevalence will thus have elim-
inated the willingness of some individuals to pay for others to engage in protective
activity.
3.4. The dynamic welfare effects of eradication
Any evaluation of a disease eradication policy will depend crucially on the way dy-
namic, rather than static, externalities are considered. This is so because a main benefit
of an eradication program eliminates the costs of disease prevention in the future. For
the current population, eradication is never optimal, because the marginal costs of vac-
cination eventually outweigh the benefits of further decreases in disease prevalence.
However, there is a missing intergenerational market: future generations cannot pay
vaccine producers for the benefit of eradication, although collectively they benefit most
from an eradication program. 25
To consider the dynamic value of an eradication programs, let the lower bound B on
the willingness to pay for eradication by future generations be given by the net-present
value of these future prevention expenditures; using the demand for vaccination D, we
can write this as:
- at
pD (p) e 1
B D(p)p
- e- t dt= ) (3.10)
t=T a
On the other hand, the cost of any efficient eradication scheme cannot exceed the upper
bound B, the cost of universal vaccination for T periods (where T is the length of time
necessary to achieve eradication under universal vaccination). The quantity B is given
by:
rT P(1 - e - T)
B- pe-Wt dt - (3.11)
Jt= a
24 This cap on the external benefit from protective behavior may be particularly relevant to explicitly choice-
based diseases such as HIV, since many individuals may find protective behavior to be costless, or even prefer-
able. For example, monogamous married couples would have to be compensated to engage in transmissive
activity with high-risk groups.
25 There are other potential benefits of eradication not discussed here such as the elimination of a mutation
or drug-resistant strain of a virus.
Ch. 33: Economic Epidemiology andInfectious Diseases 1785
The upper bound must be the cost of subsidizing everyone, since vaccination demand
goes to zero as prevalence goes to zero. The prevention expenditures avoided are thus
larger than the required subsidies whenever
B
B B D(p) eT - 1. (3.12)
The discount factor a crucially determines the desirability of eradication. If the dis-
count factor is zero then eradication is always desirable for all demand functions, since
the value of the future prevention eliminated always exceeds the current cost of the
eradication program. On the other hand, if there is full discounting, and a becomes very
large, eradication is always dominated by the free-choice equilibrium. In addition, a
longer eradication program naturally makes eradication relatively more costly, as does
a low level of demand caused by high prices or other features of the disease.
It is useful to compare this dynamic social problem to the dynamic problem of a
subsidized monopolist. In the social problem, the benefits of eradication accrue in the
future, but in the monopolist's problem, eradication sends future profits to zero. The
discount factor thus has different effects in the two problems: less discounting makes
the monopolist less likely to eradicate, although it is more likely to improve welfare.
Thus, a deficit-financed eradication program, which spends beyond tax revenues dur-
ing its operation but recoups the deficit in future generations, may improve welfare
when discounting is at current market rates. Such a program would allow for the inter-
generational transfers necessary to pay current generations to over-vaccinate for the
benefit of future generations, who will not purchase vaccines from their manufacturers.
If eradication is to be achieved, there is still a choice between the regulatory approach
of mandating demand and the fiscal approach of subsidizing it. Given the difficulty of
subsidies relative to mandates, the fact that larger distortionary taxes are needed for the
latter only highlights the superiority of the regulatory approach. However, dependent on
the cost of enforcing the regulations, and the distortionary taxes needed for that purpose
as well, there may well be a trade-off between the two methods.
3.5. Public intervention into allocationsof information
Many public health interventions involve the dissemination or regulation of information.

This is particularly true for sexually transmitted diseases such as HIV where screening,
partner notification, education, confidentiality legislation, or surveillance reporting to
the population are all part of the public arsenal. The effect of changes in the allocation of
information, of "who knows what and when", cannot be studied using epidemiological
analysis, because changes in the allocation of information do not change behavior in
epidemiological analysis. This excludes explicit analysis of the effects and desirability
of the bulk of public AIDS-prevention spending in the United States and abroad, on
education, antibody testing, and counseling services for individuals at risk of acquiring
or transmitting HIV. Virtually all fiscal institutions which have responded to the HIV-
1786 T Philipson
epidemic, whether international, national or sub-national, have included HIV education

and testing in their prevention efforts. 26
To evaluate public education programs, we must recognize that the information struc-
ture of a disease determines its growth and prevalence. We may regard the case of an
"asymptomatic" disease such as HIV as a case of imperfect information - a person
may harbor HIV even in the absence of any symptoms observable to his or her partner.
Specifically, if we regard symptomatic disease as providing perfect information, asymp-
tomatic disease corresponds to the case of asymmetric information. The classic paper
by Akerlof (1970) first studied the effects of such information problems on the volume
of trade in economic markets. If one interprets sex as an economic trade in the sense
of being an activity which is ex ante beneficial to both parties involved, the problem of
AIDS can be viewed as a problem of quality uncertainty among traders. In such a mar-
ket, traders may learn their quality status through HIV testing. Incomplete information
will inefficiently restrict the volume of trade when traders forego mutually beneficial
trades, which would take place under perfect information, for fear of trading with a bad
partner. Similarly, HIV risk may cause pairs of susceptible individuals not to engage in
mutually beneficial unprotected sex for fear of infection. In the extreme case of perfect
information about infection status, this information theory predicts that infected individ-
uals will engage in sexual activity only with other infected individuals and noninfected
individuals will engage in it only with noninfected individuals - the degree of infor-
mation determines the degree of assortative matching in the population. The privately
determined disease growth rate would be much smaller under perfect information, zero
under full assortative matching, indeed so small as to wipe out the disease.
Philipson and Posner (1995) and Boozer and Philipson (2000) analyze the private
demand for information in this market and the effects of public subsidies on informa-
tion acquisition. The theoretical analysis implies that mainly low-risk HIV-positive and
high-risk HIV-negative individuals will alter their beliefs and behavior due to public
information subsidies, and that the aggregate response to such a program may be small,
because the effects across such risk groups are offsetting. This is analyzed empirically
using a longitudinal survey that imitated a public HIV testing program by actually ad-
ministering an HIV test as part of the survey - the San Francisco Home Health Study
(SFHHS), collected by the AIDS Prevention Center at the University of California at
San Francisco (UCSF) during 1988-1989. The survey recorded the beliefs and sexual
activity of individuals before and after the testing. Using this direct evidence on the
effects of providing traders with private information, the analysis assessed the longitu-
dinal impact of such a public information intervention. Consistent with the theoretical
26 Bloom and Glied (1991) first considered the incentives of private employment based HIV-screening, but
were not concerned with the three main questions of this survey. They considered the lack of private demand
for employer HIV-tests of employees for the purpose of reducing the costs of employment-based health insur-
ance. They calculate that due to the low prevalence of the disease, the costs outweigh the benefits relatively
more for smaller firms. Also see Mechoulan (1999) and Tremblay (1999) on the impact of testing on disease
prevalence.
discussion, the study found that knowledge of one's HIV status increased the volume
of sexual contact by 16 percent for high-risk HIV-negative respondents and had little
effect on high-risk HIV-positives. Mechoulan (1999) provides an interesting analysis
of the aggregate implications of these incentives for disease prevalence. He shows that
not only does prevalence rise with subsidization of testing, but testing also is likely to
welfare.
The responsiveness of high-risk negative individuals and low-risk positive individu-
als calls into question the rationale for public education and testing programs. If such
programs target high-risk groups, they do little to alter behavior favorably. Only those
agents who actually learn something modify their behavior - the positive low-risk and
negative high-risk traders. Indeed, the high-risk negatives subsequently engage in sexual
intercourse with a greater number of partners, potentially placing them at greater risk
of a new infection. The theoretical analysis and empirical results suggest that a public
testing program can have unintended effects, particularly when focused on high-risk
populations.
4. The welfare loss of disease and medical R&D
In setting priorities among control efforts across many separate types of diseases, a ma-
jor question facing public health authorities is the welfare loss inflicted upon a popula-
tion by a given disease. The orthodox approach has employed several cost-of-illness
("COI") measures, each of which is a product of prevalence and (possibly quality-
adjusted) severity of a disease. This approach has the seemingly self-evident implication
that the more morbidity or mortality inflicted by a disease, the larger its welfare loss.
This section discusses the economic welfare loss incurred due to infectious diseases and
then compares it to such common public health measures of this loss. The analysis is
founded on the idea that standard welfare analysis of taxation may applied to the welfare
effects of regulating diseases when interpreting exposure to disease as a valuable good,
and the expected cost of incurring the disease, here becoming infected, as a "random
tax" on the consumption of this good. 27 The principal result is that the cost-of-illness
measures cannot comprise a relatively large fraction of the total welfare loss when pri-
vate prevention is prevalence-elastic; the main welfare loss of a disease will then consist
of the distorted behavior the disease induces.
If a disease is interpreted as a tax on the consumption of exposure to the disease, the
disease "distorts" consumption to involve preventive activity that would not have been
undertaken in absence of the disease. This "excess burden" of the disease tax represents
the main difference between public health measures of disease cost, which consider
only the direct cost imposed on those infected by a disease, and an economic measure
of the disease, which includes the cost of the distorted preventive behavior. Since not
27 See Philipson (1995).

1788 T Philipson
all individuals who consume the good (expose themselves to the disease) contract the
illness, the tax is paid randomly. Public health measures of the loss inflicted by a disease,
such as cost-of illness measures, are invariably comprised of the product of the total
number of cases and the cost of each case. This represents only the "revenue" part of
the disease tax, thereby ignoring the excess burden due to the distorted behavior. The
case of polio illustrates this point - there are currently no infections in the United States
but each child has to be vaccinated. According to a revenue-focused measure of welfare
loss from disease, polio induces zero loss in welfare, although were it not for polio
expenditures of time and money on its prevention could be used more productively.
To consider the welfare loss of a disease let 0 u(s, O)- u(i, 0) denote its per-
period cost of infection as discussed in the individual decision problem in previous
sections, Technological medical advances may reduce this cost of infection through
new treatments. Let 1(0) be the prevalence under a given cost and denote by D(O)
D( (0), 0) the steady state demand for prevention. If L denotes the steady-state welfare
loss it was shown to be given by:
L D(O)C + [1 - D(O)]I(0)0, (4.1)
where C is the cost of prevention. Here, the first term reflects the excess burden the
disease imposes on individuals who do not consume exposure given the implicit disease-
tax on such activity. The second term reflects the loss suffered by those who engage in
exposure despite this tax, that is, individuals who will eventually make up the case load.
This is the revenue part of the total welfare loss of a disease. If the disease does not
induce harm the welfare loss is zero; 0 = 0 implies D(0) = 0 and hence L = 0. Figure
4 shows the differences between the two types of welfare losses as a function of the cost
of infection.
weiu& Lo
L (H)=LF
Cast
Figure 4. Disease cost and welfare losses.

The COI measures of the welfare loss of disease, usually a product of the prevalence
and the quality-adjusted per-capita cost of a disease, focuses solely on the second term
of this loss by ignoring the distortions induced by the disease. The COI measures of
public health correspond to the revenues or "Laffer"-curve LpH of the disease. When
demand is prevalence elastic, then the COI measure may take on the inverted U-shaped
feature because a rise in the cost reduces prevalence so that their product may not be
monotonic; the COI measure may claim that a more costly disease is welfare improving.
The effect of a reduced cost of infection through improved treatments, that is, reduc-
tions in the parameter 0, are determined by
dL
--- = Do[I - C]- (1 - D)[O + I]. (4.2)
dO
The first term is reduction in the welfare loss due to the reduced distortions and the
second term is the reduction due to the lowered average cost on those who choose to
consume exposure. The special case of COI analysis applies when demand is completely
inelastic. Then the reduction in the welfare loss would be the reduction in the COI
measure by the corresponding reduction in the per-capita cost of infection
dL
- = -(1 - D)I. (4.3)
dO
However, this highlights the difference in welfare effects of public efforts to change
the cost of infection 0 dependent on whether behavior is prevalence elastic. If new
medical technology lowers the consequences of disease, more consumption of exposure
will take place thereby partly offsetting the public health achievements. For example,
for AIDS the recent new treatments of protease inhibitors will reduce distorted behavior,
here safe sex thereby having an offsetting effect on prevalence. AIDS causes an excess
burden in terms of foregone sexual consumption. 2 8 The existence of this excess burden
is indeed why the case load is so small and the disease is self-limiting. Nonetheless, most
public health authorities (e.g., the World Health Organization (WHO) and the Centers
for Disease Control and Prevention (CDC)) remain focused entirely on the direct cost of
the illness and ignore the excess burden. However, when behavior is prevalence elastic
the excess burden will dominate the total welfare loss of disease for serious diseases.
5. The evidence on prevalence elastic behavior
Many of the discussed implications for the behavior of epidemics as well as for the
impact of public health interventions stem from the positive relationship between pre-
ventive behavior and disease prevalence. This section discusses empirical estimates of
this relationship.
28 The asymptomatic nature of HIV may make this burden especially large, because, as discussed above, even
two uninfected people may inefficiently choose to abstain from unprotected sex.
1790 T. Philipson
Ahituv et al. (1996) investigates the degree to which the local prevalence of AIDS
increases the demand for disease-preventing methods of contraception among young
adults. Using data from the National Longitudinal Survey of Youth (NLSY), they find
the use of condoms to be quite responsive to the prevalence of AIDS in one's state
of residence, and they find this responsiveness to have grown over time. Using both
cross-sectional and longitudinal evidence, they find that a one percent increase in the
prevalence of AIDS in one's home state increases the propensity to use a condom by up
to 50 percent. Small levels of growth in AIDS prevalence have thus resulted in enormous
growth in the demand for condoms.
In the U.S., AIDS has been spread primarily through unprotected sexual intercourse,
especially, although not exclusively, among homosexual males. Sexual activity ac-
counted for transmission in over 70 percent of all U.S. AIDS cases diagnosed through
June 1992; of these cases, 90.6 percent resulted from homosexual contact. Over the last
several years, however, the number of new AIDS cases in which transmission occurred
by a means other than sexual contact, such as intravenous drug use, has grown more
rapidly than the number of new cases due to sexual transmission. In addition, transmis-
sion via heterosexual contact has increased over time. Ahituv et al. examine the extent to
which one form of protection from STDs, condom use, has responded to the increased
prevalence of AIDS among young adults in the US. They focus on the demand for con-
doms particularly because of its suggested importance in stanching the spread of AIDS.
Using data from several large cohorts of youth over the second-half of the 1980s, they
present evidence on how local AIDS prevalence affects the demand for contraception
or, more precisely, they estimate the elasticity of demand for condoms with respect to
AIDS prevalence. While many other studies had examined the nature of sexual activity
under STD risk, none had attempted to estimate prevalence elasticities. In addition, un-
like those of prior studies, the data set used in this study spanned the entire 1980s, so
that behavior before and after the epidemic could be compared. This is a major advan-
tage of the NLSY over other sources of data on AIDS. The paper analyzes a subsampe
of the NLSY, namely individuals aged 25 through 27. For the sample years 1984, 1986,
and 1990, 8924 respondents fall into this age category, and these respondents generate
a total of 11,351 person-year observations.
In sum, it is found that condom demand rose substantially among 25 through 27
year-old people over this period. In particular, while not finding differences in condom
demand across regions in 1984, before AIDS was very prevalent, they find that condom
demand became geographically heterogeneous as the AIDS epidemic spread, with a
higher growth of condom use in states with higher AIDS prevalence rates. Moreover,
the highest rates of increase occurred among sexually active, single men and especially
those living in urban areas, all of whom are generally expected to have higher exposure
to the risk of HIV infection. Although other factors might account for this regional
change, these patterns strongly suggest that young adults altered their level of protection
against STDs in response to increased risks.
The use of personal and state-level characteristics as controls for regional demand
does not alter, and sometimes strengthens, the conclusions of this analysis. Specifically,
young adults exhibit a prevalence-elastic demand for condoms, and the elasticity rises
among men, blacks, unmarried people, urban residents, and more sexually active indi-
viduals who face a greater risk of HIV exposure. Moreover, prevalence-elasticity has
increased substantially over time: condom demand does not differ across states before
the epidemic, but as the epidemic progresses, the interstate differences grow. Finally,
the authors find that increases in the prevalence of AIDS in one's state of residence ex-
plain more than half of the dramatic rise in condom use which took place among young
adults in the second half of the 1980s. The prior analysis treats the NLSY as a set of
repeated cross-sections. Exploiting its longitudinal features, however, the rate of con-
dom adoption is still found to be significantly higher for individuals living in areas with
higher growth rates of AIDS prevalence.
Table 2 shows the primafacie evidence for the relation between differences in con-
dom demand and differences in AIDS prevalence. It compares, within a given year,
the demand for condoms broken down by quartiles of the AIDS prevalence rate in the
respondent's current state of residence. These results are presented in Table 2, which
displays the proportion of individuals using condoms by quartile as well as the p-value
associated with the null hypothesis that condom use in the first quartile equals that in the
fourth. Condom demand does not differ across quartiles of AIDS prevalence in 1984 for
any of the demographic groupings. However, in subsequent years, the quartiles begin to
diverge, with individuals in higher prevalence states using condoms more frequently.
In both 1988 and 1990, condom use consistently increases across quartiles, and these
increases are statistically significant.
Analysis of the longitudinal NLSY data reveals the same trend: the growth rate in
condom use differs across quartiles of the growth rate of AIDS prevalence between
1986 and 1988, as well as between 1988 and 1990, but it does not differ across quartiles
between 1984 and 1986. The higher the growth of risk, the more likely the growth of
preventive behavior. In addition, condom use by married individuals does not respond
to movements in AIDS prevalence. This is consistent with the theory, since such indi-
viduals represent a control group which does not face increased risk of infection from
increased rates of prevalence.
In order to investigate the robustness of these patterns, they presented results for four
alternative empirical specifications of individual-level, reduced-form demand functions
for condoms. In each, they control, in various ways, for personal and regional charac-
teristics other than the state-level AIDS prevalence rates in order to better isolate the
effects of regional and temporal changes in the full price of unprotected sex. They con-
trol for the pecuniary market price for condoms by using year and state dummies, so that
any uncaptured differences in price must come from possible but unlikely within state
and year variation in condom pricing. Table 3 shows the estimated prevalence elastici-
ties broken down by year and by various demographic subgroups, for the most general
empirical specification considered.
The significantly negative prevalence elasticities in 1984 and 1986 appear to be con-
sistent with the claim of public health officials that low condom use contributed to the
spread of the epidemic. However, these significantly negative coefficients become trans-
1792 T Philipson
Table 2
Condom use (proportion) by quartiles of state-of-residence prevalence per 100,000 population for 25-27
year olds [Data source: NLSY, selected years]
Quartile/ Men and Men Single Single men Single sexually Married Married men in
p-values women only men in urban areas active men men urban areas
1984
First 0.085 0.105 0.106 0.103 0.145 0.104 0.132
Second 0.082 0.108 0.074 0.048 0.085 0.146 0.150
Third 0.075 0.095 0.104 0.082 0.121 0.080 0.051
Fourth 0.098 0.118 0.092 0.106 0.122 0.166 0.165
p-values 0.481 0.676 0.714 0.934 0.667 0.232 0.586
0.651 0.893 0.787 0.415 0.709 0.317 0.176
1986
First 0.075 0.067 0.055 0.076 0.078 0.078 0.087
Second 0.080 0.099 0.088 0.103 0.117 0.112 0.095
Third 0.067 0.072 0.078 0.087 0.109 0.064 0.073
Fourth 0.072 0.105 0.085 0.092 0.101 0.136 0.155
p-values 0.823 0.100 0.296 0.632 0.577 0.123 0.126
0.833 0.217 0.680 0.884 0.806 0.196 0.223
1988
First 0.103 0.128 0.163 0.178 0.184 0.081 0.077
Second 0.112 0.146 0.158 0.118 0.161 0.127 0.151
Third 0.131 0.174 0.222 0.228 0.256 0.086 0.076
Fourth 0.158 0.191 0.225 0.248 0.228 0.118 0.113
p-values 0.001 0.013 0.067 0.064 0.329 0.310 0.431
0.004 0.061 0.073 0.002 0.117 0.478 0.229
1990
First 0.120 0.152 0.213 0.222 0.253 0.082 0.132
Second 0.159 0.214 0.244 0.245 0.268 0.174 0.177
Third 0.178 0.232 0.309 0.334 0.293 0.129 0.121
Fourth 0.194 0.266 0.305 0.325 0.341 0.208 0.220
p-values 0.000 0.000 0.024 0.025 0.103 0.001 0.074
0.001 0.001 0.058 0.038 0.367 0.009 0.143
Notes: The categories of states represent the quartile of the annual population-weighted distribution of the
state prevalence-per-100,000-population. For each quartile, the entry in the proportion of observations who
used condoms last month. The p-value entries take the from Pl; P2, where p is the p-value associated with
the hypothesis that the incidence of condom use in the first quartile is equal to that in the fourth, and P2 is for
the test of the null hypothesis of no difference in condom use across the four quartiles. All results are based
on weighted statistics from the NLSY.
formed into significantly positive coefficients for almost all demographic groups except
married men, as claimed.
Although this piece of evidence relates to the demand for transitory protection against
infectious disease, the same type of patterns characterizes the demand for permanent
protection through vaccination. Using data from the 1991 National Health Interview
Ch. 33: Economic Epidemiology andInfectious Diseases 1793
U
0)
0 ******** *
01 vfm O N N
I0
C~~~~Nf-
.;a N NC U f N 0 N - O N _
0) I
0. 00 '0' - N i Nfl0
N. ON
5
0)
NS
B
GO ***Go * *
nf ~t000ON '0N N
N v,
N
*t
Cm -)
0.s '0
c9
O
N
ON
N ^
666o6666660
0
0
-0 0
.0~ EO
0
0) 0)
C - t m r- noO
0)
0. 00 O C OD '
0
oON
t N ON
1z
0)
11
GO
N 000 l CIN 0 '000
i
.0
0 U
Q
0
0 CG C)N 0CD N N
0) 0 G
'C I I I I I I I I I I I
00
-a 0) ON .0
0) 0
* * * G * * * * * * *
.0
0 =I
,0
0)
0,
0)
. OG
"
cnN 00 /) 0
0.0.0. 00
0. 0. 0 0.
0 f )'
.0.0.0 00
. 000.
'
0.0.00
00
f
O
0 . 0.
-N N0
0 C M
N C nN ON 0 . 0 .-
- a
I : I
0 0U * G G G G G G G G G G
Ot
*0
0)
0)
GO
GO
8= t t t 7 E
0) G
0) GO-)
G
O GO G
GO
G m
O O
O GO
G GO O
O GO G
G G O
O G
G
v , a
* * *
0.
t
0
0
Q0 '000 0 ' N - Nm
0 C*e= ,
0
PI
1794 T Philipson
,
0)
0
: .25-
1-
a_
0-
25 45 65 85
Age
Figure 5. Flu shot propensity age profile. Jagged: Sample mean by age; Smooth: Predictions from quartic
regression.
Survey, Mullahy (1997) provides evidence on the prevalence-elasticity of influenza vac-

cine demand. Each year in the US, anywhere from 10 to 50 million individuals contract
influenza. In a typical year, about 20,000 of these individuals, 90% of whom are elderly,
die from complications of the disease. During influenza season, the hospitalization of
elderly patients reaches an annual high, while influenza accounts for more than 10% of
visits to doctors' offices during peak flu season. Mullahy finds a high correlation be-
tween the demand for vaccines and mortality risk, calculated as the product of disease
prevalence and the probability of mortality conditional on infection; high-risk groups
such as health care workers, individuals in areas with more recent outbreaks, or the el-
derly demand vaccines at relatively larger rates. This relationship is illustrated for the
elderly in Figure 5, which gives the age-profile of flu vaccine demand.
Less than 5 percent of people below age 50 demand the vaccine, but demand rises
sharply for individuals older than that, until about half the population over 70 demands
vaccination. The much higher risk of mortality faced by the elderly explains this dra-
matic change in demand; the risk of mortality conditional on infection rises very sharply
in age and in fact resembles the age profile of demand, as one would expect.
Also investigating the demand for vaccines, Philipson (1996) shows that the preva-
lence elasticity of demand for the child combination vaccine against measles, mumps
and rubella (MMR) parallels the prevalence elasticity of HIV and influenza vaccina-
tion. 29 The paper used data from the National Health Interview Survey (NHIS) on US
29 See also Goldstein et al. (1996).

measles vaccination between 1984 and 1991, as well as a child health supplement from
the 1991 NHIS; these data are combined with measles case loads in the respondents'
states of residence, as reported by CDC's Morbidity and Mortality Weekly. Due to a
measles outbreak between 1989 and 1991, the data display sufficient variation across
states and time to permit the estimation of prevalence-elasticity for the MMR vaccine.
The paper identifies the main dependent variable of interest as the duration represented
by the age in months at which a child obtains her first MMR vaccination. The study
finds that regional case load, a time varying covariate, has a large positive impact on
the hazard rate into vaccination. Using a proportional hazards model, the effects on du-
ration of a large set of control variables is investigated. These controls include several
measures of public health policies aimed at stimulating demand in the respondent's state
of residence. Prevalence exerts a large and highly significant positive effect on the haz-
ard rate into vaccination. Indeed, disease prevalence affects the hazard rate more than
any other determinants of vaccination completion, and this relationship persists across a
variety of specifications. The results of this study have also been confirmed by Conroy
and Fische (1996), who also find strong support for a prevalence-elastic demand using
the same type of data from NHIS.
We have summarized a wide variety of evidence for the prevalence elasticity of de-
mand for prevention of infectious diseases. A major drawback of the studies is that
they do not identify why demand is prevalence elastic. In particular, empirically it is
not known how information gets transmitted during the growth of a disease. Part of the
transmission occurs in the private sector through mass media, but part of it also is at-
tributable to public efforts at surveillance. Future empirical work should better attempt
to assess the process by which information is disseminated. If such prevalence-elastic
demand is identified in future studies, our arguments demonstrate that it has many im-
portant implications for both the behavior of infectious diseases as well as the effects of
public health interventions aimed at controlling their spread.
6. Concluding remarks
This paper has reviewed recent economic analysis of public health policies and infec-
tious disease. The discussion focused on three general questions which reveal how the
economic approach offers unique insights into the behavior of disease and optimal pub-
lic health interventions, insights which epidemiology cannot provide. We first exam-
ined the difference between the disease behavior predicted by the economic approach
and that predicted by epidemiology. We concluded that incentives for prevention make
epidemics self-limiting, because the prevalence of a disease raises the incentives for pre-
ventive behavior. Second, we examined the different implications for public health pol-
icy offered by the economic approach and the epidemiological approach. The economic
approach yields the insight that public intervention often provides less benefit than pre-
dicted by epidemiology, because private incentives counteract its effects. Lastly, we
compared the different ways in which economics and epidemiology evaluate the cost
1796 T Philipson
of a disease. Here we found that epidemiology focuses only on part of the cost, and
misses the sizable costs of disease-avoidance. This focus causes epidemiology to stress
the prevalence of a disease as the key determinant of its cost. We found this focus mis-
placed, because the total welfare loss imposed by a disease has more to do with the
severity of the disease and the steps people take to avoid it.
The growth in economic analysis of infectious disease has been substantial, but many
important areas remain open for future work. For example, future work could examine
how the endogenous matching of infected and uninfected individuals would change
the predictions generated by dynamic models of disease growth. Second, researchers
should more fully explore how asymmetric information about the disease status of sex
partners reduces the volume of sexual trades and thereby limits the growth of STDs
such as AIDS. Third, economic analysis should be brought to bear on the unique issues
surrounding disease control in low-income countries. Empirically, much more work
is needed on the sources and determinants of the prevalence-elasticity of demand for
prevention; such analysis could help reveal why certain subpopulations engage in less
prevention than others as seems to be the case in many African countries. One major
difference between rich and poor countries is that in poor countries, AIDS is spread
though markets, namely prostitutes. Therefore, sexual behavior may be less prevalence-
elastic if the market responds more in price than in quantity to the lowered quality of
supply implied by more infected prostitutes.
More generally, the burgeoning field of economic epidemiology offers several
promising directions for future research. Of course, epidemiology, despite its name, 3 0
comprises more than just the study of infectious diseases, and economic analysis of
other areas within the field seem useful avenues for future research. One such area is
the design and analysis of clinical trials. 3 1 Although there exists a substantial literature
on evaluation of social programs through methods of random assignments of treatments
in econometrics, the special features of clinical trials are plentiful and have not been ad-
dressed. A second area is the examination of the normative approach to disease control
advocated by epidemiologists, for infectious and non-infectious diseases. This approach
consists of discovering and estimating the most important covariates of disease 32 and
then, in a great leap of faith, recommending private or public intervention aimed at
controlling the most empirically important covariates. Economic analysis is biased to
suggest that there is little role for public prevention of non-communicable diseases, and
that the crowding out of prevention, discussed here, limits the benefits for communi-
cable diseases as well. It seems natural to inquire into the conditions under which this
canonical approach of epidemiology makes sense from the standpoint of Pareto opti-
mality. It seems clear that high-risk targeting often favored by such analysis may not
30 Epidemiology stems from two Greek words; the word for epidemics (epidemia) and the word for the study
of (logos). I thank Charles Phelps for this extension of my Greek vocabulary.
31 See, e.g., Philipson and DeSimone (1997) and Philipson and Hedges (1998).
32 So called "risk-factors" which consist of the significant regressors in equations with disease occurrence as
the dependent variable.
be efficient since a high risk group, by definition, has high costs of prevention relative
to benefits. A primary example of this is the continued advocacy of partner reduction
programs for the African prostitutes who make up the so called "core groups" of the
HIV epidemic in those countries. Since earnings are proportional to partners for these
workers, such programs are the equivalent of grant-reduction programs for academics,
which presumably would have a similarly low take-up rate.
To end on an optimistic note, it seems that epidemiology is an area where economics
as a social science may successfully compete with other approaches, coming out of the
natural sciences. Despite its recent growth, the field of economic epidemiology remains
in its infancy relative to its possibilities. The global importance of world-wide mortality
caused by infectious disease ensures that such research will pay valuable dividends
by improving the understanding of the way infectious diseases spread, and the ways
individuals and institutions can control them.
References
Ahituv, A., V.Hotz and T. Philipson (1996), "Is AIDS self-limiting? Evidence on the prevalence elasticity of
the demand for condoms", Journal of Human Resources 31(4):869-898.
Ainsworth, M., and A.M. Over (1997), Confronting AIDS: Public Priorities in a Global Epidemic (World
Bank, Oxford University Press).
Akerlof, G. (1970), "The market for lemons: Quality uncertainty and the market mechanism", Quarterly
Anderson, R., and R. May (1991), Infectious Diseases of Humans: Dynamics and Control (Oxford University
Press, London).
Auld, M.C. (1996), "Choices, beliefs, and infectious disease dynamics", Working Paper 938 (Department of
Economics, Queen's University).
Auld, M.C. (1997), "Behavioral response to the AIDS epidemic: Structural estimates from panel data", Work-
ing Paper (Department of Economics, Queen's University).
Bailey, N. (1975), The Mathematical Theory of Infectious Diseases and Its Applications, 2nd edn. (Hafner
Press).
Becker, G. (1991), A Treatise on The Family (Harvard University Press, Cambridge, MA).
Bloom, D. (1997), "AIDS and growth", Working Paper (National Bureau of Economic Research, Cambridge).
Bloom, D.E., and G. Carliner (1988), '"The economic impact of AIDS in the United States", Science 604:239-
244.
Bloom, D.E., and S. Glied (1991), "Benefits and costs of HIV testing", Science 252:1798-1804.
Bongaarts, J. (1996), "Global trends in AIDS mortality", Population and Development Review 22(1):21-45.
Boozer, M., and T. Philipson (2000), "Public intervention into markets of asymmetric information: The Case
of HIV", Journal of Human Resources, forthcoming.
Brandeau, M., and E. Kaplan (1994), Modelling the AIDS Epidemic: Measurement, Planning, and Policy
(Raven Press).
Brito, D., E. Sheshinski and M. Intriligator (1991), "Externalities and compulsory vaccinations", Journal of
Public Economics 45:69-90.
Castillo-Chavez, C. (1989), Mathematical and Statistical Approaches to Aids Epidemiology (Springer-Verlag,
Heidelberg).
Conroy, P., and R.P.H. Fishe (1996), "An economic analysis of vaccination rates for preschool children",
Working Paper (Department of Economics, University of Miami).
1798 T Philipson
Cuddington, J.T. (1993a), "Further results on the macroeconomic effects of AIDS: The dualistic, labor-surplus
economy", The World Bank Economic Review 7(3):403-417.
Cuddington, J.T. (1993b), "Modeling the macroeconomic effects of AIDS, with an application to Tanzania",
The World Bank Economic Review 7(2):187-189.
Cuddington, J.T., and J.D. Hancock (1994), "Assessing the impact of AIDS on the growth path of the Malaw-
ian economy", Journal of Development Economics 43:363-368.
Cuddington, J.T., J.D. Hancock and C.A. Rogers (1994), "A dynamic aggregative model of the AIDS epidemic
with possible policy interventions", Journal of Policy Making 16(5):473-496.
Dow, W., and T. Philipson (1996), "The implications of assortative matching for the incidence of HIV",
Francis, P.J. (1997), "Dynamic epidemiology and the market for vaccinations", Journal of Public Economics
63:383-406.
Fuchs, V. (1989), "Health economics", in: J. Eatwell, M. Milgate and P. Newman, eds., The New Palgrave: A
Dictionary of Economics (Norton, New York).
General Accounting Office (1989), GAO/PEMD 89-13 AIDS Forecasting: Undercounts Weaken Estimates
(GAO, Washington, DC).
General Accounting Office (1994), Vaccines for Children: Critical Issues in Design and Implementation (US
Government Printing Office, Washington, DC).
Geoffard, P., and T. Philipson (1995), "The empirical content of canonical models of infectious diseases",
Biometrika 82(1):101-114.
Geoffard, P., and T. Philipson (1996), "Rational epidemics and their public control", International Economic
Review 37(3):603-624.
Geoffard, P., and T. Philipson (1997), "Disease eradication: Public vs. private vaccination", American Eco-
Goldstein, K.P., T. Philipson, H. Joo and R. Daum (1996), "The effect of epidemic measles on immunization
rates", Journal of the American Medical Association 276(1):56-58.
Hay, J.W., and F. Wolak (1990), "Bootstrapping HIV/AIDS projection models: Back calculation with linear
inequality-constrained regression", Stanford Hoover Institute Working Paper in Economics: E-90-5.
Hay, J.W., and E Wolak (1994), "A procedure for estimating the unconditional cumulative incidence curve
and its variability for the human immunodeficiency virus", Applied Statistics 43(4):599-624.
Heckman, J., and B. Singer (1984), "Econometric duration analysis", Journal of Econometrics 24:63-132.
Hsu, H.-P. (1998), A Theory of Vaccination (Department of Economics, University of California, Berkeley).
Kremer, M. (1995), AIDS: The Economic Rationale for Public Intervention (World Bank, Washington, DC).
Kremer, M. (1996), "Integrating behavioral choice into epidemiological models of AIDS", Quarterly Journal
of Economics (May):549-573.
Mann, J., D. Tarantola and N. Netter (1992), AIDS in The World (Harvard University Press, Cambridge).
Mechoulan, S. (1999), "HIV-testing: A Trojan horse?", mimeo (Department of Economics, Northwestern
University).
Mullahy, J., "It'll only hurt a second? microeconomic determinants of the demand for flu vaccines", Health
Economics, forthcoming.
New Zealand, Ministry of Transport (1996), Land Transport Pricing Study: Safety Externalities, May 1996.
Pauly, M. (1994), "A study of the economic underpinnings of vaccine supply", Contract No. 282-92-0044
(Department of Human Health and Services).
Phelps, C. (1992), Health Economics (Harper-Collins, New York).
Philipson, T. (1995), "The welfare loss of disease and the theory of taxation", Journal of Health Economics
14:386-396.
Philipson, T. (1996), "Private vaccination and public health: An empirical examination for US measles",
Journal of Human Resources 31(3):611-630.
Philipson, T., and J. DeSimone (1997), "Experiments and subject sampling", Biometrika 84(3):618-632.
Philipson, T., and L. Hedges (1998), "Subject evaluation in social experiments", Econometrica 66(2):381-
409.
Philipson, T., and R. Posner (1993), Private Choices and Public Health: The AIDS Epidemic in an Economic
Perspective (Harvard University Press, Cambridge and London).
Philipson, T., and R. Posner (1995), "A theoretical and empirical investigation of the effects of public health
subsidies for STD testing", Quarterly Journal of Economics 110:445-474.
Philipson, T, and R. Posner (1996), "The economic epidemiology of crime", Journal of Law and Economics
39(2):405-436.
Plotkin, S., and E. Mortimer (1988), Vaccines (Saunders, Philadelphia).
Shilts, R. (1987), And The Band Played On (Random House, New York).
Stiglitz, J. (1978), The Economics of the Public Sector, 2nd edn. (Norton, New York).
Stiglitz, J. (1997), Introduction in Confronting AIDS: Public Priorities in a Global Epidemic, World Bank
Policy Research Report (Oxford University Press).
Tremblay, C.H. (1999), "AIDS education, condom demand, and the behavior of STD-infected youth", Work-
ing Paper (Department of Economics, Oregon State University).
Weisbrod, B. (1961), Economics of Public Health: Measuring the Economic Impact of Disease (University of
Pennsylvania Press, Philadelphia, PA).
Weisbrod, B., and J. Huston (1987), "Benefits and costs of human vaccine programs: Assessing the literature",
Working Paper (Department of Economics, Northwestern University).
World Bank (1993), "Investing in health", World Bank Development Report (Oxford University Press, Lon-
don).
Xu, X. (1998), "Technological improvements in vaccine efficacy, individual incentives to vaccinate, and eco-
nomic welfare", Working Paper (Department of Economics, University of California, Berkeley).
Chapter 34
EQUITY IN HEALTH CARE FINANCE AND DELIVERY*

ADAM WAGSTAFF
School of Social Sciences, University of Sussex, Brighton, United Kingdom
The World Bank, Washington DC, USA
EDDY VAN DOORSLAER

Departmentof Health Policy and Management, Erasmus University, Rotterdam, The Netherlands
Contents
Abstract 1804
Keywords 1804
2. What is equity? 1806
2.1. What is equity? A first pass 1806
2.1.1. Equity versus altruism 1807
2.1.2. Equity, social justice and ideology 1808
2.1.3. Equity, ideology and health care systems 1809
2.1.4. Ideology and the empirical literature on equity in health care 1810
2.1.5. Some interim conclusions 1811
2.2. What is equity? A closer look 1811
2.2.1. Sorting out definitions 1812
2.2.2. Conflicts between equity principles 1813
2.2.3. Justifying equity principles in the delivery of health care 1814
2.2.4. Justifying the ability to pay principle 1817
2.2.5. Where does this leave equity? 1818
3. Equity in health care finance 1819
3.1. Health care financing typologies 1819
3.2. Vertical equity and progressivity of health care finance 1822
3.2.1. Kakwani's progressivity index 1822
3.2.2. Empirical work on progressivity and health care finance 1824
3.3. Horizontal equity and income redistribution 1828
*We are grateful to Xander Koolman and Alessandro Marini for assistance with literature searching and
compiling the bibliography, and to Bengt Jinsson, Joe Newhouse and Carol Propper for comments on earlier
drafts. We alone are responsible for any errors or omissions. The findings, interpretations and conclusions
expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the
World Bank, its Executive Directors, or the countries they represent.
Handbook ofHealth Economics, Volume 1, Edited by A.J. Culyer and J.P. Newhouse
1804 A. Wagstaff and E. van Doorslaer
3.3.1. Measuring horizontal equity 1829

3.3.2. Empirical work on horizontal equity 1832
4. Equity in health care utilization 1833
4.1. Regression-based tests of inequity in health care utilization 1835
4.1.1. Testing for inequity 1835
4.1.2. Empirical tests of inequity 1839
4.2. Measuring inequity 1841
4.2.1. A direct standardization-based index 1842
4.2.2. Empirical work using the direct-standardization-based index 1843
4.2.3. An indirect standardization-based index 1844
4.2.4. Empirical work using the indirect standardization-based index 1845
5. Equality of health 1846
5.1. Pure inequalities in health 1847
5.1.1. Measurement of pure health inequality 1847
5.1.2. Empirical work on pure health inequalities 1848
5.2. Socioeconomic inequalities in health 1849
5.2.1. Measuring socioeconomic inequalities in health 1849
5.2.2. Empirical work on socioeconomic inequalities in health 1852
6. Conclusions 1855
References 1857
Abstract
The paper surveys the economics literature on equity in health care financing and deliv-
ery. The focus is, for the most part, on empirical work, especially that involving intema-
tional and temporal comparisons. There is, however, some discussion of the concept and
definition of equity. The empirical sections cover the literature on equity in health care
financing (progressivity and horizontal equity of health care financing arrangements),
equity in health care delivery (horizontal equity in the sense of treating persons in equal
need similarly), and equality of health.
Keywords
equity, inequality, equality, health care financing, health care delivery, health
JEL classification:D30, D31, D63, H23, H51, I110, I112

Ch. 34: Equity in Health Care Finance and Delivery 1805
1. Introduction
Equity is usually recognized by economists to be an important policy objective in the

health care field. Indeed, some go so far as to suggest that amongst the population
at large, equity takes precedence over other objectives, even efficiency.1 The interest
shown by economists in the equity issue seems to vary from one country to the next
and over time within countries. These variations and changes no doubt reflect in part
the variations and changes in the attitudes of policy-makers. The relatively low level of
interest amongst economists in the US in equity issues, at least on a per capita basis,
probably reflects the fact that policy-makers in the US have, on the whole, attached less
importance to equity than their counterparts have in most other OECD countries. The
recent growth of studies of equity in the US probably also reflects the commitment -
at least in its early days - of the Clinton administration to improving equity in health
care in the US. The views of policy-makers are probably only one factor that accounts
for the cross-country and temporal variation in the quantity of work by economists
on equity. Another is the variation in views amongst economists about the status of
research on equity. There was until fairly recently, and in some countries there still
is, a perception amongst many economists that research on equity must necessarily
be normative in character. Many seemed to shy away from the area because of this.
Fortunately, it is increasingly being recognized that while the question of what equity is
all about is indeed a normative question, the questions of whether equity, defined in a
specific sense, has been achieved, or has increased, or tends to be higher in one type of
health care system than other, lie firmly within the realm of positive economics. This,
along with the changing political climate, may well have accounted for the upswing in
research on equity. 2
Whatever the reason, there is now a good deal of interest, amongst both policy-makers
and economists, in equity in health care financing and delivery, as well as a good deal
of literature. This chapter provides a review of this literature. The review is not com-
prehensive in a number of respects. First, it excludes work by non-economists on - and
only on - the grounds that this is a chapter in a Handbook of Health Economics. This
decision inevitably results in a partial coverage of the field and we have almost certainly
lost more material through this decision than we would have done had we been writing
a chapter on, say, the health insurance market. Non-economists have written on almost
all the areas we touch on in our review: political and medical philosophers have written
extensively about the question of what equity is in the health field; medical sociologists
and epidemiologists have written extensively about equity in access to and the receipt
of health care, as well as about health inequalities. The "market share" of economists
! See e.g. MacLachlan and Maynard (1982) and Mooney (1986).

2 Another factor undoubtedly is the revolution in computer hardware and software over the last 15 years.
This has made distributional analyses of large household surveys far easier than was hitherto the case.
varies from sector to sector within the equity "industry", as does the degree of differen-
tiation between the products of economists and of non-economists. 3 The partial nature
of our review should be kept in mind. The second respect in which our review is partial
is that it is heavily oriented towards empirical work, and comparative empirical work to
boot. We do offer a discussion of the nature of equity, but this is not a comprehensive
overview of the field and is intended simply to provide something of a philosophical
backdrop to the empirical material. Our focus on comparative empirical work stems
from a belief that studies examining a system in isolation are less helpful for policy
purposes than comparative studies, unless a study of a single system manages to unpack
the factors contributing to any observed inequity, which is typically not the case. The
third respect in which our review is partial is that its focus is on industrialized countries
- the study by economists of equity in the health sectors of developing countries is still
in its infancy, though a good deal of work is currently underway. 4 The fourth respect
in which our review is partial is that we ignore completely the issue of geographical
equity. This is a large literature to which economists have contributed, though often in
conjunction with non-economists. Finally, we have tended to restrict our attention to
published studies or forthcoming articles or books in English. 5
We start in Section 2 with a discussion of the nature of equity. As indicated above, this
is not a comprehensive overview and aims simply to provide a philosophical backdrop
to the empirical material in the following sections. Our discussion is different from
and complements that of Williams and Cookson (2000). Sections 3, 4 and 5 concern
empirical material on equity in the financing of health care, the delivery of health care
and health inequalities. Section 6 offers some conclusions.
2. What is equity?
2.1. What is equity? A first pass
On the face of it, there would appear to be a good deal of agreement as to what equity in
health care entails. Of the various theories of social justice that might be brought to bear
3 The economists' market share is probably largest in the area of equity in health care finance, and probably
smallest in the area of inequalities in health. The product differentiation is probably least in the area of eq-
uity in access to and the delivery of health care, but even here is sizeable (not least because of the different
quantitative techniques used). In the area of the nature of equity, there is an appreciable degree of product dif-
ferentiation, partly in analytical techniques (philosophers are prone to use the case approach whilst economists
are keener on developing the general story), as there is in the area of health inequalities (economists are more
likely to rank people by income, whilst non-economists prefer to rank by occupation and education).
4 Studies in books or journal articles include those of Baker and van der Gaag (1993), Deolaalikar (1995),
Hammer, Nabi and Cercone (1995), Pannarunothai and Mills (1997), and van de Walle (1995).
5 Literature searches for the empirical material were undertaken in Econlit and BIDS (based on the Social
Science Index). Not all the material unearthed in these searches was included in the survey.
Ch. 34: Equity in Health Care Financeand Delivery 1807
on the issue of equity in health care, it has been argued by one eminent medical ethi-
cist [Gillon (1986)] that some have a greater applicability and acceptability than others.
Moreover, examining the policy statements on equity in several OECD countries sug-
gests that policy-makers are in broad agreement over what they mean by equity [OECD
(1992), Wagstaff and van Doorslaer (1993)]. Finally, in empirical work, researchers
from countries with such different health care systems as Britain and the United States
have adopted much the same notions of equity in their analysis.
2.1.1. Equity versus altruism
Before going into each of these areas, however, it is important to be clear that distri-
butional objectives in health care, and in social policy generally, can arise from two
sources: equity or social justice, on the one hand, and altruism or caring, on the other.
The concepts of equity and altruism are often confused. However, they are, as Culyer
(1980) and Goodin and Le Grand (1987) emphasize, quite distinct and have quite dif-
ferent implications for health policy.
Caring and altruism are matters of preference. In the context of health care a caring
individual might be one who derives utility - i.e. an external benefit - from seeing an-
other person receiving health care [Culyer (1980)]. In this case, the caring individual
prefers that the person in question receives health care and is prepared to sacrifice re-
sources to ensure that the person actually obtains treatment. Quite how much they are
prepared to sacrifice will depend on how much they care (which will depend on, inter
alia, their income) and on the cost of providing health care. Alternatively a caring indi-
vidual might be one who derives utility from the act of providing health care for others
[Mooney (1986)]. Quite how much of their income the individual will be prepared to
sacrifice to provide health care for others will depend on the utility they derive from
the act of providing medical care (which again will depend on their income) and on the
cost of providing health care. With caring preferences of either type, therefore, "costs
and benefits are balanced at the margin and ... the level of provision is ... determined
by the wealth of the community" [Culyer (1980, p. 70)]. The language of caring is
thus, as Culyer (1989) notes, the language of efficiency. Hence the term "Pareto optimal
redistribution" [Hochman and Rodgers (1969)].
Social justice (or equity), on the other hand, is not a matter of preference. As Culyer
(1980) puts it: ". .. the source of value for making judgements about equity lies outside,
or is extrinsic to, preferences... The whole point of making a judgement about justice
is so to frame it that it is (and can be seen to be) a judgement made independently of the
interests of the individual making it" (p. 60). Social justice thus derives from a set of
principles concerning what a person ought to have as of right. The difficulty, of course,
is how to obtain views about social justice in a way that ensures that they are not con-
taminated by the interests of the individuals concerned. A straightforward survey, for
example, whilst sometimes proposed, seems unlikely to elicit responses that pass this
test. One ingenious device that has been used to ensure that principles of justice are
genuinely impartial is the "veil of ignorance" [Rawls (1971)]. This puts self-interested
individuals in an "original position" where they are ignorant about the positions they
will occupy in society. The rules of justice agreed upon by individuals in these circum-
stances are argued to be genuinely impartial. The "veil of ignorance" is not, however,
the only means of arriving at a set of just rules. 6 Barry (1989) has argued that justice
can more simply be construed as the set of rules that can be justified on an impartial
basis.
The different motivations behind equity and caring have at least three important im-
plications for health care policy. First, distributional decisions regarding health care pro-
vision prompted by considerations of social justice ought not to be influenced by cost:
justice requires that an equitable pattern of provision be ensured, irrespectiveof the sac-
rifice to the rest of society [Culyer (1980, pp. 69-70)]. Second, there is scope for con-
flict between efficiency and equity: an efficient redistributional programme prompted by
caring preferences need not be equitable, and vice versa [Culyer (1980, p. 98)]. Third,
the distributional "rules" derived from the two approaches are likely to be different.
Indeed, differences emerge even within the two approaches, depending on the precise
stance adopted. For example, different rules emerge in the caring approach, depend-
ing on whether caring is postulated to relate to a person's absolute level of medical
care consumption [Culyer (197 1)], to the deviation of their consumption from the mean
[Lindsay (1969)], or to health itself [Culyer (1980)].
2.1.2. Equity, socialjustice and ideology
The upshot of the foregoing is that, when studying equity, one would like to analyse
equity objectives independently of any distributional objectives that are motivated by al-
truism. The philosophy literature contains, at least on the face of it, some useful pointers
in this respect.
Gillon (1986) provides a helpful summary of the various theories of social justice and
discusses their applicability to health care. 7 Libertarians, he notes, emphasize a respect
for natural rights, focusing in particular on two of Locke's natural rights - the rights
to life (i.e. not to be unjustly killed) and to possessions. Providing people acquire and
transfer their "holdings" without violating others' rights, their holdings are regarded
by libertarians as just. Hence Nozick's (1974) claim that taxation is warranted only to
maintain a "minimal state". Utilitarians, by contrast, aim at maximizing the sum of in-
dividual utilities or welfare, though some utilitarian writers have incorporated a concern
for individual autonomy into this maximand. Marxists emphasize "needs". Hence the
principle of "distribution according to need". In Marxist writings, this principle is of-
ten coupled with the principle of "from each according to his ability", which, in the
present context, can be interpreted as "from each according to his ability to pay". Rawls
6 Indeed, there is some debate about precisely what would be agreed behind such a veil and how far strategic
behaviour might limit the usefulness of the approach.
7 See also the annexe to Williams and Cookson's paper in this Volume.
(1971) proposes two principles of social justice, namely that individuals should have
the maximal liberty compatible with the same degree of liberty for everyone and that
deliberate inequalities are unjust unless they work to the advantage of the least well off.
Yet another view of social justice is that justice should reward merit.
Which of these theories of justice appear to command the greatest support in the
context of medical care? Gillon suggests that "allocation of medical resources on the
basis of non-medical merits is widely regarded as repugnant" (p. 97) but argues that the
principle of "distribution according to need" commands widespread support amongst
physicians and others working in the medical field. He challenges the extreme liber-
tarian position, pointing out that if Locke's right to health were to be included in the
libertarian list of natural rights, writers like Nozick would be forced to accept the legit-
imacy of taxation to benefit the poor and sick. Gillon also notes that utilitarianism, with
its emphasis on maximizing the sum of welfare, has much in common with the notion
of efficiency as allocating resources according to the likelihood of medical success.
2.1.3. Equity, ideology and health care systems
The two theories of justice most frequently encountered in the philosophy literature in
the context of medical care are, in fact, the libertarian and the Marxist approaches [Don-
abedian (1971)]. As Gillon notes, however, the principle of "distribution according to
need" is not exclusively Marxist. Indeed, it is a key component of 20th century egali-
tarianism [cf. Sugden (1983)]. Williams (1993) compares and contrasts the libertarian
and egalitarian positions. 8 He notes that in the egalitarian view, "access to health care is
every citizen's right (like access to the ballot box or to courts of justice), and this ought
not to be influenced by income and wealth" (p. 291). In the libertarian view, by contrast,
access to health care is viewed as "part of society's reward system". As Williams puts
it, "at the margin at least, people should be able to use their income and wealth to get
more or better health care than their fellow citizens if they so wish".
The egalitarian and libertarian viewpoints point, as Williams (1993) notes, towards
quite different health care systems. The egalitarian viewpoint suggests that a publicly
financed system should predominate, with health care being distributed according to
"need" and financed according to "ability to pay". The libertarian viewpoint, by con-
trast, points towards a mainly privately financed health care sector, with health care
being rationed primarily according to willingness (and ability) to pay. Public involve-
ment should be minimal and limited to providing a minimum standard of care for the
poor. In practice, in most countries health care is financed and delivered by a mixture
of systems and there are traces of both ideologies in policy-making, with the emphasis
often changing with changes of government.
Broadly-speaking, however, policy-makers in Europe give the impression of being
much more inclined towards the egalitarian viewpoint in health care matters than the
8 Cf. Culyer et al. (1981), Maynard and Williams (1984).

libertarian.9 There appears to be less agreement amongst American policy-makers about

equity objectives in the health care field. There appears to be broad agreement amongst
policy-makers in several European countries that payments towards health care should
be related to ability to pay rather than to use of medical facilities.10 The commitment
to linking health care payments to ability to pay is much less in evidence in the United
States, though as is apparent from Davis (1993), much of the debate prior to the ill-fated
Clinton reforms in the 1990s focused on the link or lack of it. The bipartisan Pepper
Commission, for example, in its Blueprintfor Health Care Reform [Rockefeller (1991 )]
urged that progressive taxation be used to raise new revenues to finance the Commis-
sion's proposed reforms. It is also evident that policy-makers in the European countries
are committed to the notion that all citizens should have access to health care.ll In
many countries this is taken further, it being made clear that access to and receipt of
health care should depend on need, rather than on ability to pay. The commitment to the
notion of universal and equal access is less evident in countries like the United States.
Despite this, it is apparent from Davis (1993) that much of the debate about reform in
the US in the early 1990s was motivated by a concern about growing inequalities in
access. Hence the Pepper Commission's desire to "guarantee all Americans, no mat-
ter what their income, employment status, or place of residence, access to affordable
insurance protection" [Rockefeller (1991, p. 2509)]. A number of countries have also
shown a concern to reduce inequalities in health status. In Britain, governments have
set up two working groups to examine health inequalities in the last twenty years. 12 In
the Scandinavian countries too, health inequalities have been widely discussed in policy
documents, as has also been the case in the Netherlands. 13
2.1.4. Ideology and the empirical literature on equity in health care
The empirical work to date on equity in health care reflects the apparently pro-
egalitarian bias amongst policy-makers.
Many studies of equity in the delivery of health care - in both Europe and the United
States - start from the premise that health care ought to be distributed according to need
rather than willingness and ability to pay. Andersen (1975) (an American) suggests
that an equitable distribution of health care is one in which the amount of health care
received correlates highly with indicators of need and is independent of variables such as
income, which are irrelevant to need. This definition is adopted by Benham and Benham
(1975) in their study of equity in the delivery of health care in the US before and after
9 Cf. e.g. Table 2.1 in Wagstaff and van Doorslaer (1993), which reproduces quotations from policy state-
ments in a number of OECD countries.
l0 The OECD (1992) concludes the same in its comparison of the health care systems of Belgium, France,
Germany, the Netherlands, Ireland, Spain and the United Kingdom.
II Again, the same conclusion has been reached by the OECD (1992) in its comparative study.
12 Cf. the groups chaired by Sir Douglas Black [Townsend and Davidson (1982)] and Sir Donald Acheson.
13 Wetenschappelijke Raad voor het Regeringsbeleid (1987).
the introduction of Medicare and Medicaid. Rosenzweig and Schultz (1991), in their
analysis of the distribution of medical care to pregnant women in the US, test various
hypotheses including the hypothesis that these services are allocated solely on the basis
of medical need. Le Grand (1978) (a Briton) also starts from the premise that receipt of
health care should depend on need and not on socioeconomic status. In addition to these
studies, a large number of studies have explored the issue of inequalities in health, many
of which come from non-economists. The concern here is how far health is distributed
unequally, especially across socioeconomic groups such as income groups.
Studies of equity in the finance of health care, by contrast, have tended to take as
their starting point the premise that health care ought to be financed according to abil-
ity to pay.14 An exception to this is a group of studies seeking to establish the extent
of net income redistribution associated with a particular mode of financing health care
- i.e. examining the combined redistributive effect of financing and utilization of pub-
lic health care. Such studies have been undertaken in Finland, France, Germany, Italy,
the Netherlands, Portugal, Switzerland and the United Kingdom. 15 It appears that these
studies start from a variant of the egalitarian viewpoint, where reducing inequality in "fi-
nal" incomes is regarded as the equity goal. The usefulness of these studies is somewhat
unclear, as there is little evidence that equity objectives in the health field are couched
in terms of net income redistribution. 16
2.1.5. Some interim conclusions
On the face of it, then, there appears to be a reasonably clear picture emerging on the
issue of what equity entails in the context of health care. There are the libertarians
whose concern, if they have one, is with ensuring that minimum standards are achieved.
And there are the egalitarians who are concerned to ensure that health care is financed
according to ability to pay and that the delivery of health care is organized in such a
way that everyone enjoys the same access to care and that the care is allocated on the
basis of need with a view to promoting equality of health.
2.2. What is equity?A closer look
Appearances can, however, be deceptive. The conclusions reached in the last subsection,
while useful, leave a number of key questions unanswered. Over the last few years,
economists have helped to answer them. One set of questions relates to definitions. What
14 Early studies include those of Hurst (1985), who compares the American, British and Canadian financing
systems, and Gottschalk et al. (1989), who compare the American, British and Dutch systems. A number of
recent studies reviewed below have taken this as their starting point.
15 For a thorough study, in English, see Leu and Frey (1985).
16 Insofar as the poor also tend to be the sick, such redistribution is, of course, implied by a commitment
to financing health care according to ability to pay. But it is not income redistribution that is the equity goal
being pursued.
exactly is meant by "access" to health care? Is it different from receipt of health care? If
so, how? What is meant by "need"? Another set of questions concerns the compatibility
of the various interpretations of equity. For example, is equalizing access consistent with
seeking to equalize health? A final set of questions concerns justification. What is the
justification for financing health care according to ability to pay? Or equalizing access?
Or allocating care according to need? Or equalizing health? If the last three objectives
are mutually incompatible, what cases can be made for pursuing one rather than the
others?
2.2.1. Sorting out definitions
When the term "access" is used in policy statements and in much of the academic litera-
ture (including much of the literature written by economists) it is clear that what is often
meant - indeed perhaps what is usually meant - is "receipt of treatment". This is well
illustrated by Tobin's (1970) remarks, where, after noting Americans' apparent concern
with equality of access, he suggests that equality in health care might be taken to mean
that "the treatment of an individual depends on his medical condition and symptoms, not
on his ability or willingness to pay" (emphasis added). 17 In a similar vein, although the
so-called RAWP formula used to allocate resources to NHS regions 18 claims to attempt
to equalize access, in practice the focus is firmly on expenditures, or - more precisely
- resources [cf. Mooney and McGuire (1987)]. Finally, several American and British
empirical studies of equity' 9 claim to examine the extent to which access to health care
is linked to need, but actually interpret access in terms of treatment received.
Le Grand (1982) and Mooney (1983, 1994) argue that access to treatment and receipt
of treatment are not the same thing. The former refers to the opportunities open to peo-
ple, while the latter concerns both whether these opportunities exist and if so whether
people have availed themselves of them. But how can access, so conceived, best be de-
fined? One possibility, suggested by Le Grand (1982) and strongly endorsed by Mooney
(1983), is that access be thought of in terms of the money and time costs that people
incur in obtaining health care. There is an implication of defining access in this way that
some find unsatisfactory, including Le Grand (1991), namely that if two people face the
same time and money costs, they are said to have the same access irrespective of their
income. It is, indeed, far from obvious whether it makes sense to say that someone with
virtually no income at all has the same access to health care as a millionaire simply
because they both face the same time and money prices. An alternative approach that
does not have this implication is suggested by Olson and Rodgers (1991). They sug-
gest defining access as the maximum attainable level of consumption of medical care,
given the individual's income, and the time and money prices associated with consum-
ing medical care. In this approach whether someone has access to a service, and if so
17 Cf. also Andersen (1975).

18 RAWP is the abbreviation of 'Resource Allocation Working Party'.
19 Cf. Aday et al. (1980), Collins and Klein (1980) and Puffer (1986).
how much access they have, depends not only on the time and money prices they face,
but also on their income.
Another term whose meaning is far from self-evident is "need". Often it is equated
with ill-health - people who are relatively ill are held to have a relatively high need
for medical care. This, as several writers have suggested [cf. e.g. Culyer and Wagstaff
(1993)], is over-simplistic. Someone can be said to need medical care when ill only if
there is medical care available that can improve their health. Need is an inherently in-
strumental concept and one that ought to permit the non-ill to be said to be in need of
medical care, in the sense that their health in the future could be better than it would
otherwise be if they received (preventive) care now. Defining need in terms of one's
current health state is thus unattractive. An alternative, suggested by Culyer (1976) and
Williams (1974, 1978), is to define need in terms of one's capacity to benefit from
health care. This tackles the instrumentality of need head on, but is unsatisfactory in
that it measures need in terms of the entity the care will affect (health) rather than in
terms of the entity that is needed (health care). A technological breakthrough, such as
keyhole surgery, that leaves a person's capacity to benefit (i.e. potential health improve-
ment) the same, but requires far fewer resources leaves need unchanged according to the
capacity-to-benefit definition. Culyer and Wagstaff (1993) therefore suggest an alterna-
tive definition of need as the minimum amount of resources requiredto exhaust capacity
to benefit. If capacity to benefit is, at the margin, zero, so too is need. Where marginal
capacity to benefit is positive, assessment of need requires an assessment of the amount
of expenditure required to reduce capacity to benefit to zero.
2.2.2. Conflicts between equity principles
Policy-makers in particular talk as if the three most frequently encountered interpreta-

tions of equity in health care delivery - equality of access, allocation according to need,
and equality of health - are all compatible with one another. But are they?
Consider first the compatibility between equality of access and the other two defini-
tions. As Mooney (1983) emphasizes, access is but one of the factors influencing receipt
of medical care. Other factors include the individual's perception of the benefits asso-
ciated with the treatment and the incentives facing the physician - in short, anything
affecting the demand for health care as opposed to its cost. Two people may thus enjoy
the same access to health care and be in the same degree of "need", and yet the treatment
they receive may differ. A poorly educated person may, for example, perceive the health
benefits to be lower than a well educated peer, with the result that the poorly educated
person does not contact a physician whilst the well educated person does. Conversely,
access may differ across individuals and yet the amount of medical care received may
be the same. It follows immediately [cf. Culyer and Wagstaff (1993)] that equality of
access will not necessarily result in health care being allocated according to need, or in
the attainment of - or even promotion of - equality of health.
The remaining potential conflict to be considered is that between allocation according
to need and equality of health. It is, in fact, often claimed by Marxists and some egal-
itarians that allocating medical care according to need will promote equality of health,
if not result in equality of health. Indeed, this is the principal justification offered for
adoption of the principle of allocation according to need [cf. e.g. Miller (1976)]. Culyer
and Wagstaff (1993) investigate the validity of this claim using the three definitions of
need discussed in Section 2.2.1. They show that allocating medical care according to
need will result in differing degrees of inequality in final health, depending on which
of the definitions one adopts and that, in general, it will not be the case that allocating
health care expenditures according to need will result in equality of health, or that such
an allocation rule will even promote equality of health.
2.2.3. Justifying equity principlesin the delivery of health care
Contrary to what appears to be believed in many policy-making circles, then, one cannot
logically espouse simultaneously equality of access, allocation according to need, and
equality of health. A choice has to be made. Given what we have just said, we can
already, it would seem, dispose of one of the contenders: allocation according to need.
The justification for this - that its adoption will promote equality of health - has already
found to be wanting. This leaves two contenders: equality of access and equality of
health.
Mooney et al. (1991, 1992) and Mooney (1994) argue strongly for the adoption of
equality of access. They argue that allocating care according to need or seeking to at-
tain equality of health would imply a willingness to override consumer preferences and
hence to depart from the premises underlying Paretian welfare economics. 2 0 It would
require, for example, that individuals' preferences concerning the consumption of med-
ical care be ignored. Such a departure from Paretian welfare economics would, in their
view, be ethically unjustified. According to Mooney et al., one should equalize access
and then accept whatever distribution of utilization and health the market throws up.
Let us put on one side for a moment the question of whether equalizing access is
compatible with an acceptance of the value judgements underlying Paretian welfare
economics. A difficulty with the argument of Mooney et al. is that it seems to take it for
granted that departing from the Paretian value judgements is ethically unacceptable -
something that economists and policy-makers would find anathema. This is surprising,
since there is widespread evidence that policy-makers in the health field and the public
at large do not accept the Paretian value judgements in the context of health and medical
care. 21 Thus even if it were the case that a concern with equalizing access to medical
care is consistent with the Paretian value judgements, this would not necessarily be a
point in its favour. 22
20 Cf. Mooney et al. (1991, 1992).

21 The writings of several economists reflect this viewpoint. Williams has been one of the staunchest critics
of the Paretian value judgements in the health context - see e.g. Williams (1976, 1981). These ideas have
also been developed by Culyer - see e.g. Culyer (1976), Williams (1981), Culyer (1989, 1990). See also
Rice (1997, 1998).
22 Cf. Culyer et al. (1992a, 1992b).
As it is, it is far from obvious why someone who is wedded to the value judgements
underlying Paretian welfare economics would want to ensure equal access to medical
care. If these value judgements appeal to anyone in this context, they are likely to appeal
to someone of a libertarian disposition. But, as we have noted in Section 2.1, part and
parcel of such a viewpoint is a view that access to medical care ought to be seen as part
of society's reward system with people being able, at the margin at least, to use their
income and wealth to get more or better health care than their fellow citizens if they
want. 23 A libertarian might, as we have noted, be willing to use taxation to ensure some
minimum standards are put in place, but it seems unlikely they would feel compelled to
equalize access. On the Le Grand-Mooney interpretation, that would mean equalizing
money and time prices; the latter would probably entail providing more facilities in
low-income areas to ensure that waiting times and accessibility are similar. Equalizing
access in the Olson-Rodgers sense would require even higher taxes and a substantial
narrowing of the income distribution - if not complete equality of income. This is likely
to appeal even less to a libertarian. In short, acceptance of the Paretian value judgements
seems to be inconsistent with a desire to equalize access to medical care. Or, to put it the
other way round, a desire to equalize access suggests a rejection of the Paretian value
judgements. 24
The argument in favour of equalizing access also begs the question: Why medical
care? Why not equalize access to skiing holidays, or swimming pools? To our minds
it is hard to defend according special ethical status to access and utilization without
acknowledging the role of medical care in promoting good health. 2 5 What makes access
to medical care special is that it influences the utilization of medical care, and this, in
turn, influences health. It is hard to justify being concerned about the distribution of
access to medical care, or about the distribution of medical care, without having a more
fundamental concern about the distribution of the ultimate upstream variable - health.
The distributions of access and utilization matter, but not in their own right; they are
merely instruments to achieving a desired distribution of health in whose distribution
our interest ultimately lies.
What is it, then, that makes health special? And why is it that the favoured distri-
bution of it is an equal distribution? Moral philosophers in the Aristotelian tradition 26
suggest that what makes entities such as health special is that they are necessary for
an individual to "flourish" as a human being. Insofar as medical care is necessary to
good health, this provides a strong ethical justification for being concerned with the dis-
tribution of medical care and not with the distribution of, say, skiing holidays. It also
provides a justification for using the word "need" in the context of medical care and not
in the context of, say, skiing holidays [Culyer and Wagstaff (1993)]. But this argument
also adds a new angle to "need": the extent to which medical care is needed is to be
23 Cf. Donabedian (1971), Maynard and Williams (1984), Wiliams (1988, 1993).
24 Cf. Rice (1997, pp. 396-397) on this point.
25 Cf. e.g. Daniels (1985).
26 Miller (1976), Daniels (1985), Gillon (1986), Braybrooke (1987), Wiggins (1987), Lockwood (1988).
1816 A. Wagstaff and E. van DoorsIaer
judged not so much in terms of its impact on health, as reflected in, say, freedom from
pain and mobility, but rather more generally in terms of its ability to enable individuals
to flourish. A hip replacement, for example, aids mobility and enables a person to flour-
ish. But some items of care, such as IVF, might do little to improve a person's health
narrowly defined and yet might make a big impact on their ability to flourish as a human
being.
Whatever its implications for the interpretation of "need", the flourishing argument
seems to point towards the pursuit of health equality, or at least as close to it as one can
get, since giving some a better chance to flourish as human beings than others would
seem hard to defend. The work of Sen (1992) lends support to this view. It also adds a
twist to the tale. Sen draws a distinction betweenfunctionings and capabilitiesto func-
tion. Functioningsconcern what people do or are and are seen as constitutive of person's
well-being. Together they determine the extent to which a person flourishes. Sen offers
being in good health - via, for example, avoiding premature mortality or avoiding mor-
bidity - as an example of a functioning. Other examples include being well nourished,
having self-respect and taking part in the life of the community. Capabilities to func-
tion are the various combinations of functionings from which a person can choose. For
example, a person may have the opportunity to be in good health or to participate in the
life of the community. Sen argues for equality of capabilities.
Sen's argument raises a couple of issues. His argument, like the flourishing argu-
ment, provides a rationale for talking of medical care "needs". But it is not obvious that
the set of medical care services that might be deemed necessary in order to enable a
person to flourish as a human being would always coincide with the set deemed neces-
sary to provide the person with whatever set of capabilities is considered appropriate. If
functionings are interpreted narrowly, the two former sets might well be considerably
larger than the latter, and the set of medical care services deemed to be "needed" from
a functionings perspective might well end up as a fairly narrow set of basic services.
Whether or not this is the case will depend crucially on how large the set of capabili-
ties is on which one wants to define equality. The other issue relates to the distinction
between functionings and capabilities. Two people may have the same capabilities -
i.e. the same set of functionings from which to choose - and yet end up with different
functionings. One of them may choose not to participate in the life of the community
at all even though there was the opportunity to do so. The other may choose to eat junk
food and be under-nourished even though there was the opportunity to be well nour-
ished. This has an important implication - one cannot infer that because the level of
one particular functioning is low, the person necessarily had a low level of capability
relative to that functioning. This prompts the question: Is it functionings which should
be distributed equally (the conclusion we were leaning towards above following the dis-
cussion of flourishing) or capabilities? Sen, in fact, comes down firmly in favour of the
latter, on the grounds that people may, quite legitimately, have different objectives and
they should be free to choose whichever they want.
This has an important implication in the present context - we may accept that being
in good health is an important element of a person's functioning or flourishing, but if
people have the opportunity to achieve this functioning and yet choose not to do so,
we cannot infer automatically that all inequalities in health are inequitable. What is
important is not that everyone achieves the same level of health but rather that everyone
has the opportunity to achieve the same level. This distinction is fine in theory, but
working out in practice whether people are in poor health or seem set to die at a young
age because they had the necessary capabilities but chose not to avail themselwes of
them, or because they didn't have them in the first place, will be hard work. It is no
surprise, then, that in the applied work that Sen and others have undertaken using the
capabilities approach, the focus has been firmly on what people achieve in terms of life
expectancy, literacy, and so on, rather than on what they might have achieved [cf. e.g.
UNDP (1993, p. 10 0 ff.)].
2.2.4. Justifying the ability to pay principle
The widespread commitment amongst policy-makers in the OECD countries to financ-

ing health care according to ability to pay raises the question: What underlies this com-
mitment? One motivation that is sometimes advanced for financing health care accord-
ing to ability to pay is a desire to promote equity in the delivery of medical care [Culyer
(1993)]. For example, linking payments for medical care to utilization in the spirit of the
benefit principle would go against the principle of equality of access, if we interpret ac-
cess along any of the lines indicated in Section 2.2.1 above. Or alternatively, one might
believe that linking payments to the utilization of medical care would deter people from
using medical care facilities and that this would therefore reduce one's chances of see-
ing medical care distributed equitably. Whichever of these justifications for decoupling
payments from utilization one offers, it is important to realize that one is only providing
a justification for the decoupling. These arguments provide a justification for rejecting
the benefit principle, not for accepting the ability-to-pay principle. One could, for exam-
ple, decouple payments from utilization by financing health care through a lump-sum
tax. This would ensure equal access to health care and would not generate any deter-
rence effects. But it would not result in health care being financed according to people's
ability to pay.27
A justification of the ability-to-pay principle - i.e. a justification for relating pay-
ments for protection against out-of-pocket payments to ability to pay - must, therefore,
come from elsewhere. Another possibility is that it may stem from a concern about the
distribution of income after health care payments have been netted out [Culyer (1993)].
For example, a reason for wanting to decouple payments from utilization might stem
from a desire to protect patients from health care payments that would threaten their
ability to purchase other goods and services that have a special ethical status, or that
27 Of course, if linking utilization to payments does not deter use, the case for the ability-to-pay principle
becomes even weaker. However, empirical evidence - e.g. from the RAND Health Insurance Experiment in
the US - suggests that cost-sharing does deter usage [cf. e.g. Newhouse et al. (1993, pp. 338-340)].
are used to produce household commodities that have a special ethical status. Hous-
ing and education are examples that spring to mind. Given that utilization of medical
care is frequently concentrated amongst the lowest income groups, a failure to decouple
payments from utilization would result in households at the bottom end of the income
distribution suffering the largest proportionate reductions in their disposable income as
a result of health care utilization. But this argument too provides a justification only for
decoupling payments from utilization. Financing health care via a lump-sum tax, or via
an insurance scheme with flat-rate premiums, would ensure that the users of health ser-
vices would not face a disproportionately high reduction in their incomes as a result of
falling ill. But neither would result in payments for health care being linked to ability to
pay. Rationalizing the widespread commitment to the ability-to-pay principle in health
care financing appears, therefore, to be harder than might at first be imagined. 28
2.2.5. Where does this leave equity?
The recent debate amongst economists on the nature of equity in the health field has
raised but not resolved a number of questions left unanswered by the earlier discussions
in the literature. Some useful things have been written about the problems involved in
defining terms such as "access" and "need", though it is not obvious that any of the
definitions proposed commands widespread - let alone universal - support amongst
economists. For example, as Le Grand (1991) notes, the general thrust of the argument
of Olsen and Rodgers leads one to wonder whether equalizing access to health care
should not logically lead to equalizing entire budget sets. Yet this makes a mockery of
the idea that one can equalize access to some commodities but not to others. The re-
cent debate has also explored the compatibility of the various interpretations of equity,
though here too it is most unlikely that the final word has been said. For example, even if
it is the case that allocating care according to need does not necessarily result in equal-
ity of health, it may well be that, given that patients present themselves for treatment
sequentially, allocating care on the basis of need (suitably defined) might be the best
that providers of health care can do to reduce health inequalities. The recent debate has
generated a somewhat heated discussion about which interpretation of equity in health
care delivery has the greatest appeal. The debate has helped to clarify the issues but has
certainly not led to any meeting of minds. There are those who firmly believe in equality
of access, and those who champion equality of health. The allocation-according-to-need
rule has rather fewer supporters but if it is true that allocating care according to need
might give providers the best chance of helping reducing health inequalities, then in
practice it might not be such a poor rule of thumb to adopt in the allocation of health
care. There has been less discussion of the ethics of the ability-to-pay principle, which
28 Another issue that merits discussion is: Why do policy-makers make such a point about wanting payments
for health care to be linked to ability to pay rather than looking more broadly at the impact on the income
distribution of the financing of a range of services?
is somewhat surprising: one can easily mount a case for divorcing payments from uti-
lization, but it does not logically follow that payments ought to be related to ability to
pay.
Overall, then, despite the recent literature, there are still a number of issues relating
to the meaning of "equity" that are unresolved. Moreover, although in some respects
the recent literature has helped to place the empirical work on firmer conceptual and
theoretical foundations, this is not true of all the recent literature, some of which has
left the conceptual foundations of some of the empirical work looking somewhat shaky.
It is to the empirical work that we now turn.
3. Equity in health care finance
The empirical literature to date on equity in health care financing has focused on the
issue of how far health care is financed according to ability to pay. This can be inter-
preted in terms of both vertical equity (in this case, persons or families of unequal ability
to pay making appropriately dissimilar payments for health care) and horizontalequity
(persons or families of the same ability to pay making the same contribution). 2 9 Most of
the empirical work to date has focused on the former; more specifically, it has focused
on the issue of progressivity. 3 0 This work, which aims to measure the progressivity of
the various financing sources in different countries, is surveyed in Section 3.2. Recently,
some work has been done on horizontal equity in health care finance in the context of
a broader study of the income redistribution associated with health care financing ar-
rangements. This work is reviewed in Section 3.3. This section starts, however, with a
discussion of health care financing arrangements and their variation across countries.
3.1. Health carefinancing typologies
Health care is typically financed from a mixture of four sources - taxes, social insur-
ance, private insurance and out-of-pocket payments. Social insurance is like income tax
29 The work reviewed here interprets this in terms of the link between payments in a specific year and pay-
ments in the same year. It might be argued that taking a lifetime perspective is more appropriate. For some
purposes this may be true, but it is not at all obvious that this is always true. One study that takes a lifetime
perspective is that of Propper (1995). To anticipate the conclusions of the empirical work reviewed later in
the paper, she concludes: "Results from LIFEMOD indicate that the shift from cross-sectional to a lifetime
analysis results in more equal distributions of income and morbidity, but the distributions of health care fi-
nance relative to income, and of health care receipt relative to need, are relatively unchanged by this move in
perspective. The reason is these latter distributions are functions of two distributions, both of which are flatter
across the lifetime than in the cross-section" (p. 202).
30 The progressivity of a health care financing system refers to the extent to which payments for health care
rise or fall as a proportion of a person's income as his or her income rises. A progressive system is one in
which health care payments rise as a proportion of income as income rises, whilst a regressive system is one
in which payments fall as a proportion of income as income rises. A proportional system is one in which
health care payments account for the same proportion of income for everyone, irrespective of their income.
in that it is compulsory; it is therefore unlike private insurance, which is usually vol-

untary. Social insurance is levied on earnings; this distinguishes it from income tax,
which is assessed on the basis of taxable income, and from private insurance, which in
some countries is sometimes assessed on the basis of risk factors, such as age, smoking
behaviour, etc.
The roles that each of the four financing sources plays varies across countries. In the
context of taxation, the principal source of variation is in the degree of earmarking in-
volved. In some countries, such as the UK, the taxes that go towards funding health care
are simply general tax revenues. In others, such as the Scandinavian countries, the taxes
that finance health care are largely local income taxes whose purpose is almost entirely
to raise revenues for health care. The degree of earmarking involved in social insurance
also varies from country to country. In some countries, the social insurance contribu-
tions that go to fund health care are general contributions. In many countries, however,
social insurance is earmarked. This is true not only of the countries with sickness funds,
such as Germany and the Netherlands, but also countries such as Italy, which used to
have a system of sickness funds, and countries like Ireland, which did not. It is also true
of the US in respect of Medicare Part A. The role of private insurance also varies from
one country to the next. In some cases (e.g. Ireland, Switzerland and the US) it provides
cover for sections of the population without any public cover. 3 1 In the case of Germany,
it provides cover to persons who have chosen not to have public cover - i.e. who have
opted out of the public sickness fund scheme. In countries, such as the Netherlands, it
provides cover to sections of the population with less-than-comprehensive public cover.
In some countries (e.g. Denmark, France, and again Ireland) it provides cover against
copayments levied by the public sector. In some countries, such as Ireland (yet again),
Italy, Portugal, Spain and the UK, private insurance provides double cover to those who
have comprehensive public cover. As is clear from the examples given, the private insur-
ance industry sometimes fulfills more than one of these functions at a time. Finally, the
role of out-of-pocket payments varies across countries, sometimes taking the form of
copayments (e.g. Denmark and the UK), there being large differences across countries
in the relative importance of copayments, extra billing and direct payments for services.
Table 1 shows the financing mixes in thirteen OECD countries for a recent year.
The clustering of countries is best seen via the "health care financing triangle" (Fig-
ure 1). At the bottom right corner are the social insurance countries: France, Germany
and the Netherlands. Countries such as Denmark, Finland and Sweden belong to the
tax-financed cluster of countries in the top left corner, while countries such as Switzer-
land and the US lie in the predominantly private cluster at the bottom left corner. Some
countries, such as Italy, are a half-way house between a social insurance system and
a tax-financed system. Figure 1 does not, of course, show the mix between private in-
surance and out-of-pocket payments. In most countries, the majority of private health
expenditures are out-of-pocket, this being especially true of the Nordic countries and
31 In the Swiss case, this is almost the entire population.

Ch. 34: Equity in Healtlh Care Finance and Delivery 1821
Table I
Financing mixes - thirteen countries
Direct Indirect General Social Total Private Direct Total

Country taxes taxes taxes insurance public insurance payments private
Denmark (1987) 72.5% 12.2% 84.7% 0.0% 84.7% 1.5% 13.8% 15.3%
Finland (1990) 51.0% 24.0% 75.0% 11.0% 86.0% 0.0% 14.0% 14.0%
France (1989) 0.0% 73.6% 73.6% 6.3% 20.1% 26.4%
Germany (1989) 10.5% 7.2% 17.7% 65.0% 82.7% 7.1% 10.2% 17.3%
Ireland (1987) 28.5% 39.3% 67.8% 7.3% 75.1% 10.0% 14.9% 24.9%
Italy (1991) 21.0% 17.2% 38.2% 39.2% 77.4% 1.8% 20.9% 22.6%
Netherlands (1992) 6.3% 5.0% 11.3% 64.6% 75.9% 16.3% 7.7% 24.1%
Portugal (1990) 20.7% 34.5% 55.2% 6.0% 61.2% 1.4% 37.4% 38.8%
Spain (1990) 30.8% 25.5% 56.3% 22.0% 78.3% 2.4% 19.3% 21.7%
Sweden (1990) 63.5% 8.4% 71.9% 17.8% 89.7% 0.0% 10.3% 10.3%
Switzerland (1992) 23.9% 4.8% 28.7% 6.9% 35.6% 40.5% 23.9% 64.4%
UK (1993) 29.0% 35.0% 64.0% 20.0% 84.0% 7.0% 9.0% 16.0%
US (1987) 28.1% 7.4% 35.5% 13.3% 48.7% 29.2% 22.1% 51.3%
100%
80%
60%
.o
t 40%
20%
0%
0% 20% 40% 60% 80% 100%
%raised through social insurance
Figure 1. The health care financing triangle [Wagstaff et al. (1999)1.
the southern European countries, where private insurance is still relatively uncommon.
The Netherlands, Switzerland and the US stand out as the three countries where the
majority of private expenditures are on private insurance premiums rather than on out-
of-pocket payments. Germany, Ireland and the UK come close behind in terms of the
relative importance of private insurance vis-A-vis out-of-pocket payments.
3.2. Vertical equity andprogressivity of health carefinance
Early work on progressivity in the finance of health care was based on tabulations of
health care payments by income group. In his comparison of Britain, Canada and the
United States, Hurst (1985), for example, presents tables indicating average payments
for health care by income group for each country. Payments are, however, presented in
absolute terms rather than as a proportion of income, so that it is impossible to assess
from the tables the degree of progressivity of each country's financing system. 32 Can-
tor's (1988) results for the United States - reported in Figure 18.13 of Davis (1993) -
are easier to interpret. They show that the proportion of income spent on health care in
the United States falls continuously as one moves up the income distribution, implying
that the American financing system is regressive. An implication of a progressive fi-
nancing system is that the share of the total financing burden borne by the lower income
groups is less than their share of society's income, whilst the share borne by the top
income groups exceeds their share of society's income. Comparing the share of income
received by each income decile with its share of health care payments thus provides an
alternative way of assessing progressivity. This is the approach adopted by Gottschalk
et al. (1989) in their comparison of the health care financing systems of the Netherlands,
the UK and the US. Their results for the US show that the American system is regres-
sive: thus, for example, the bottom income decile in 1981 received 1.4% of (post-tax)
income but made 3.9% of health care payments.
Tabulations of the proportion of income spent on health care and of the shares of
income and health care payments received and borne by different income groups do
not enable one to answer the question of how much more (or less) progressive one sys-
tem (or source of finance) is than another. At best they can indicate whether a system
is progressive, regressive or proportional. A more illuminating approach to assessing
the progressivity of health care financing systems is to employ progressivity indices
[Wagstaff et al. (1989)]. A variety of such indices have been proposed in the literature
on tax progressivity [Lambert (1993)1. The work to date seeking to measure the pro-
gressivity of health care financing has tended to employ just one of these, namely, that
of Kakwani (1977). 33
3.2.1. Kakwani'sprogressivity index
Kakwani's index is based on the extent to which a tax - or, more generally, source of
finance - departs from proportionality. It can be illustrated using Figure 2. The curve la-
32 In the text Hurst does remark that in Britain "household income rises about 41 times between the second
and ninth deciles whereas household tax contributions rose about seven-fold over this range" [Hurst (1985,
p. 117)].
33 Wagstaff et al. (1992) also use Suits' (1977) index.
100 %
cum
prop
of
income
and
payments
0% cum % of population p, ranked by income 100%
Figure 2. Lorenz curve for pre-payment income and concentration curve for payments.
belled Lpre (p) is the Lorenz curve for pre-payment income. The second curve - labelled
Lpay(p) - is the payment concentration curve, which plots the cumulative proportion
of the population [ranked according to pre-payment income as with Lpre(p)] against
the cumulative proportion of health care payments. If payments are levied strictly in
proportion to income, Lpay (p) and Lpre (p) coincide. If payments as a proportion of in-
come rise with income (so that the source of finance is progressive), Lpay (p) lies below
Lpre (p). The opposite is true if payments are regressive. The degree of progressivity can
therefore be assessed by looking at the size of the area between Lpre (p) and Lpay (p). If
Gpre is the Gini coefficient for pre-payment income, and Cpay is the concentration index
for payments, 34 Kakwani's index of progressivity, 7rK, is defined as
7rK = Cpay - Gpre, (1)
which is twice the area between Lpre(p) and Lpay(p). If the system is progressive, as
in Figure 2, rK is positive. If, by contrast, the system is regressive, so that Lpay(p) lies
above Lpre(p), 7rK is negative. The value of JTK ranges from -(1 + Gpre) (the entire
tax burden is concentrated in the hands on the poorest person and hence Cpay = -1) to
1 - Gpre (the tax burden is concentrated in the hands of the richest person and hence
Cpay = 1) [Lambert (1993, p. 178)].
34 The Gini coefficient is twice the area between the Lorenz curve and the line of equality (the diagonal),
whilst the concentration index is analogously defined but with reference to the concentration curve.
A useful property of Kakwani's index is that the overall index for a financing system
consisting of two or more sources of finance is a weighted average of the indices for the
individual sources, where the weights are the proportions of each source in total revenue
[cf. e.g. Suits (1977)]. Thus the progressivity characteristics of a health care financing
system depend on the proportion of total revenues raised from each source and the
degree of progressivity of each of these sources. Another feature of the index is worth
mentioning. It is perfectly possible for a source of finance (or a tax) to be progressive (or
regressive) at low income levels but regressive (or progressive) at high income levels.
Suppose, for example, that pensioners are exempt from social insurance contributions
and tend to be located in the lower income groups. Suppose too that contributions are
proportional (assume for simplicity to income) but only up to a ceiling. The exemption
of pensioners makes the system progressive at low income levels (the bottom income
groups will tend to pay a relatively small fraction of their income towards health care)
but regressive at high income levels (as a person's income rises above the ceiling, the
proportion of their income they pay towards health will fall). The result is that Lpay (p)
will cross Lpre(p) from below. Calculating the Kakwani index as the difference between
Cpay and Gpre in such cases implies that the regressivity at high incomes offsets - at least
partially - the progressivity at low incomes. The result could, of course, be a zero value
for the progressivity index.
3.2.2. Empirical work on progressivity and health care finance
Two recent papers [Wagstaff et al. (1992, 1999)] present estimates of progressivity by
source of finance for a number of OECD countries. 3 5 Both aim to apply a common
methodology and common definitions of income and health care payments to micro-
level data to attribute payments at the household level. The second paper is more suc-
cessful than the first in achieving comparable definitions and the summary here is of
these results. The progressivity indices for all financing sources are shown in Table 2
for all countries for selected years. Indices for total and subtotal payments were calcu-
lated using the revenue shares in Table 1 as weights. 3 6
The direct taxes used to finance health care are progressive in all countries. They
are especially progressive in the UK, Ireland and Germany, but far less progressive in
the two Scandinavian countries (Denmark and Sweden), reflecting the reliance in health
care financing there on the local income tax which is close to proportional. Indirect taxes
35 See the country reports in Van Doorslaer et al. (1993), for detailed results for specific countries.
36 Inevitably incidence assumptions have to be made in this analysis. In both of the aforementioned papers,
the same incidence assumptions have been employed in all countries. There are arguments for and against do-
ing this, of course, Wherever possible, personal income tax and property taxes have been assumed to be borne
by the taxpayers concerned, corporate income taxes by shareholders, sales and excise taxes by consumers, and
employee and employer social insurance contributions by employees. Private insurance premiums, whether
individual or group, have been assumed to be borne by the individual concerned. There are arguments for and
against these assumptions individually and collectively.
NCINm C0 00 Cm
-a, COCOOO '0~bN O-C .O 1
00
a O O CO
O CO CO CO CO
C
a
t Co CC CI N
oC N -C
o l Co
Co
- o C N C N mc C O- l
COO C C Coo l Co
a CCCO'C) CO0
1^ -7 00 'C
I l lJ[ I
NO -O_1O C$
o o'o
C
CO
o
, ln 0 0C 0c noC
C O N Co oooo
Cl
a
.)
a-
NI
O * - o Co -C o
a
NB C . -CO0coN to
c Oz C 6 6 6 6 6
C O- Cl=l| -Cl l l l3 l:
are regressive in all thirteen countries, but especially regressive in Spain and the UK.
General taxation, computed as a weighted average of direct and indirect, is progressive
in all countries. Interestingly, the general taxes used to finance health care appear to be
especially progressive in the two private financing countries, Switzerland and the US,
and also, albeit to a lesser extent, in Germany.
Social insurance emerges as progressive in all countries except the Netherlands and
Germany, two countries which exclude the higher income groups from the compulsory
sickness fund insurance. In contrast to Dutch and German social insurance schemes,
the French scheme (the Rdgime Generale) is almost universal and does not exclude
high earners; furthermore, pensioners and the unemployed, who are more likely to be
in the bottom income groups, pay much lower contribution rates. Also in countries such
as Ireland, Italy, Spain, and the UK, where it raises a not insignificant proportion of
revenues, social insurance emerges as a progressive source of revenue. This is probably
due to exemptions amongst the lower income groups and the fact that contributions are
assessed on the individual's own earnings rather than on their household's equivalent
income.
For the interpretation of the progressivity indices for private health insurance it is im-
portant to bear in mind the cover that private insurance buys in each country. Broadly-
speaking, countries belong to one of three groups, the exception being Ireland, and to a
lesser extent Switzerland and the US. The first comprises countries where private insur-
ance buys cover against public sector copayments and includes Denmark and France.
Private insurance is progressive in Denmark but regressive in France. This reflects the
fact that private insurance against public sector copayments is more widespread among
the lower income groups in France than it is in Denmark, which, in turn, probably re-
flects in part the higher copayments in France. The second group of countries comprises
those where private insurance is mostly taken out as supplementary cover (mostly "dou-
ble" cover) to that provided by the state and includes Italy, Portugal, Spain and the UK.
Private insurance of this type emerges as progressive, except in Spain, suggesting that
in Italy, Portugal and the UK, this insurance is a "luxury" good. The third group com-
prises countries where, for the individuals concerned, private insurance - albeit often
subsidized - is (or is nearly) the sole source of cover. This group includes Germany,
the Netherlands, Switzerland and the US. Switzerland is unusual in this group in that
private insurance is bought by almost everyone. In the other countries, only persons
with restricted or non-existent public cover generally take out private insurance. 3 7 In
the US, persons purchasing private insurance as their sole source of cover make up the
bulk of the population, whilst in the Netherlands they comprised (in 1992) 36% of the
population. Where it is relied upon by the majority of the population for cover, as in
Switzerland and the US, private insurance is highly regressive. The positive Dutch and
German indices stem from the fact that private insurance in these countries is almost
37 In the US a small proportion of expenditures on private insurance is accounted for by persons with public
cover purchasing supplementary insurance.
exclusively bought by the higher income groups. Ireland, and to some degree Switzer-
land and the US, span two or more of these groupings. In these three countries, private
insurance emerges as regressive. In Ireland this reflects the fact that private insurance
premium payments are computed net of tax relief, which benefits the better-off house-
holds most. The surprisingly small value - in absolute terms - of the Kakwani index
for private insurance for the US is attributable to coverage gaps and under-insurance
amongst the lower income groups [cf. Rasell and Tang (1994)].
Out-of-pocket payments are a regressive means of raising revenue. They are not par-
ticularly regressive in Ireland and the Netherlands, reflecting the incomplete cover of the
better-off privately insured in these countries. The very high regressiveness of out-of-
pocket payments in Switzerland and the US reflects the fact that persons on low incomes
in these countries are liable in full for any out-of-pocket payments, whereas their coun-
terparts in many European countries would be exempt from charges, either because of
their low income or because of other exempting factors (e.g. chronic ill-health, pen-
sioner status, etc.) that are often correlated with income.
The broad conclusions from this study concerning the overall progressivity of health
care financing systems confirm the earlier findings of Wagstaff et al. (1992). Health care
finance in two of the three social insurance countries (Germany and the Netherlands)
is regressive, whilst it is progressive in the third (France). In the tax-financed systems,
by contrast, health care finance typically emerges as proportional or mildly progressive.
The exception to this is Portugal, where the system overall emerges as regressive - this
reflects the high share of out-of-pocket payments in that country in 1990. Finally, in
the two predominantly privately financed systems (Switzerland and the US), health care
finance emerges as regressive.
In addition to the cross-country comparisons and cross-source comparisons reported
above, work has also been undertaken on the progressivity consequences of country-
specific trends or simulated proposed health reforms. Several studies 38 have analysed
the progressivity of the US health care financing system and some of these have con-
sidered the progressivity consequences of alternative reform packages. These studies
all confirm that the US health care financing system is highly regressive 3 9 and that out-
of-pocket payments are especially regressive. Rasell and Tang (1994) consider several
proposals for reform that aim at universal coverage. They find that all the proposals
would make the system less regressive, but that those relying more on taxes would do
so to a greater extent than those relying mainly on employer-paid insurance premiums.
Similar findings are reported by Holahan and Zedlewski (1992).
Janssen et al. (1994) assess the progressivity implications of the ill-fated Dekker
health insurance reform. The net effects of this scheme are unclear a priori. On the one
hand, the extension of the compulsory basic insurance with income-related payments
might be expected to increase the relative contribution of the higher income groups.
38 See, for example, Rasell et al. (1994), Rasell and Tang (1994) and Holahan and Zedlewski (1992).
39 Rasell and Tang (1994) report a Kakwani index of -0.114 for 1992.
On the other hand, the introduction of flat-rate premiums (these were to cover the gap
between the cost of insurance and the value of the voucher received from the State)
is clearly likely to work in the opposite direction. In the event, Janssen et al. find that
the overall regressiveness of the Dutch financing system would have been reduced by
the implementation of the Dekker Plan, but the system overall would still have been
regressive.
Lairson et al. (1995) analyse changes in the progressivity of the Australian health
insurance system. This provides an interesting mix of public and private finance. After
the inception of the universal public-coverage Medicare system in 1984, the overall
financing burden was found to be strongly progressive (K = 0.10), mainly as a result
of progressive taxation and Medicare levies. Due to the expansion of private insurance
and direct payments over the 1980s, progressivity was reduced and the overall financing
was roughly proportional in 1989.
3.3. Horizontal equity and income redistribution
The issue of horizontal equity in the finance of health care has received relatively little
attention. Horizontal equity can be defined in terms of the extent to which those of equal
ability to pay actually end up making equal payments, regardless of, for example, gen-
der, marital status, trade union membership, place of residence, etc. Horizontal inequity
might arise for a number of reasons. In private insurance, high-risk groups (e.g. the
elderly, those with pre-existing conditions, smokers) often pay higher premiums than
lower-risk persons of the same ability to pay. In the direct taxation part of the system,
horizontal inequity can arise through anomalies in the personal income tax system (e.g.
tax reliefs on mortgage interest payments, or on private health insurance premiums). In
a social insurance system, different groups may be eligible for different health insurance
schemes and hence may face different contribution schedules. In some cases the groups
may be defined in terms of earnings or income but it may be that this measure does not
properly reflect the individual's or household's ability to pay, in which case households
with similar abilities to pay may end up paying quite different amounts for similar levels
of cover. Rutten and Janssen (1987), for example, find that in the Netherlands, because
of the diversity of arrangements for different categories of persons, single persons on an
income of Dfl 17,000 in 1981 could have ended up paying as little as 2% of their income
towards health care if they were over 65 but as much as 13% if they were under 65 but
self-employed. 40 Horizontal differences of this kind, especially between the privately
and publicly insured, have been a major factor underlying the pressure to reform health
care financing arrangements in countries such as Germany and the Netherlands.
40 See Von der Schulenburg (1994) for variations in contribution rates between and within the sickness funds
in Germany.
3.3.1. Measuring horizontal equity
A number of measures of horizontal inequity have been developed in the literature on

public finance and income distribution, but with the exception of that proposed by Aron-
son et al. (1994), none is entirely satisfactory. The most popular is the measure proposed
by Atkinson (1980), Plotnick (1981) and King (1983), which involves assessing the
number of rerankings in the move from the pre-tax income distribution to the post-tax
distribution. As Aronson et al. emphasize, however, horizontal inequity refers to the
(unequal) treatment of equals, whilst reranking refers to the treatment of unequals.
The approach suggested by Aronson et al. (1994) enables - at least in principle -
a clear distinction to be made between reranking and horizontal inequity. The latter is
measured in terms of the variation in tax (or health care payments) amongst groups of
pre-tax (or pre-payment) equals. If there is no variation, there is no horizontal inequity.
The variation in each group of pre-tax equals is measured using the Gini coefficient
and then an overall index of horizontal inequity is constructed by taking a weighted
sum of the Gini coefficients for each group of pre-tax equals. The weights chosen are
the products of the square of the population share of the group of pre-tax equals and
the post-tax income share of these households. The reason for choosing these weights
is that the degree of horizontal inequity can then be measured in terms of its impact
on the distribution of income. This effect can then be compared to the impact on the
distribution of income attributable to the progressivity of and the reranking associated
with the tax in question.
More formally, the redistributive impact associated with a tax can be measured by the
reduction in the Gini coefficient caused by the tax. Thus
RE = Gx - GX-T, (2)
where Gx and GX-T are the pre-tax and post-tax Gini coefficients, respectively. In a
world where everyone faces the same tax schedule, irrespective of their non-income
characteristics (e.g. whether or not they are married, whether or not they own a home),
we have
RE= ( g)KT, (3)
where g is the share of income taken in tax and KT is the Kakwani (1977) index of tax
progressivity. Equation (3) clarifies the role of progressivity in the redistribution of in-
come: a progressive health care finance system will result in there being less inequality
in income after payments for health care have been made, and, in this sense, the pay-
ments will have a pro-poor redistributive effect. As is clear from Equation (3), however,
the extent of redistributive effect depends not only the degree of progressivity of the
tax but also on its importance, in terms of the average share of income taken up by the
tax. So, countries' health care systems may have similar degrees of progressivity and
yet be associated with quite different levels of income redistribution simply because in
one country health care payments absorb a larger share of income. The analysis in this
section, in addition to shedding light on the issue of horizontal equity, sheds light on
this important issue too.
Suppose, in contrast to what was assumed above, that people do not face the same
tax schedule and that the tax liability of household h is equal to
Th = T(x) +sh(x), (4)
where T(x) is the common amount of tax paid by all households with income x and
eh (x) is household h's deviation from this amount. It is assumed that these deviations
average to zero across all households - as Aronson et al. put it, "on average, at each x,
the tax system gets it right". The presence of h(X) in Equation (4) means that house-
holds with the same income x can end up paying different amounts of tax. This is the
classical notion of horizontal inequity. Only if eh(x) is zero for all h and x, is the tax
system horizontally equitable. Furthermore, the presence of e h(x) in Equation (4) may
result in households moving up or down the income distribution after they have paid
their taxes. There may, in other words, be reranking as one moves from the pre-tax
income distribution to the post-tax distribution.
These two possibilities are illustrated in Figure 3, which shows the relationship be-
tween post-tax income, x - Th, and pre-tax income, x, for a progressive tax. The "fans"
show the effect of differential tax treatment - for example, households starting off with
pre-tax income X3 will, on average, end up paying x3 - T (x3) in tax, but there will be
variation about this amount reflecting the presence of s h (x) in Equation (4). The exis-
tence of fans thus indicates the presence of horizontal inequity. If the fans overlap (as
they do in the case of households starting out with pre-tax incomes of xl and x2), then
reranking occurs - the shaded region of the fans indicate that the household that was
richer before tax has become the poorer after tax.
The presence of differential tax treatment means that Equation (3) is no longer valid.
Aronson et al. (1994) show that it can be replaced by
RE= (1g g)KT - EaXxGF(x) - [GX-T -CX-T], (5)
where KT is the Kakwani index computed on the assumption that everyone faces the
same tax schedule, ax is the product of the population share squared and the post-tax
income share of households with income x, GF(x) is the Gini coefficient for post-tax
income for households with pre-tax income x and CX-T is the post-tax concentration
index obtained by ranking households first according to their pre-tax income and then
within each group of pre-tax equals by their post-tax income. The first term, which
Aronson et al. call V, measures the inequality reduction that would have obtained if
there had been no differential tax treatment. The second term, which they call H, mea-
sures the extent of classic horizontal inequity - i.e. the unequal treatment of equals -
Z J
Figure 3. Horizontal inequity and reranking [Wagstaff and van Doorslaer (1994)].
by taking a weighted sum of the Gini coefficients GF(x) of the fans. These Gini coeffi-
cients are zero only if the eh (x) are zero for all x and h. The third term, which Aronson
et al. call R, measures the extent of reranking in the move from the pre-tax distribution
to the post-tax distribution by comparing the post-tax Gini coefficient with the post-tax
concentration coefficient. 4 1 If there is no reranking, R is zero.
The decomposition in Equation (5) helps - at least on the face of it - to clarify the
distinction between horizontal inequity and reranking. As Figure 3 makes clear, hori-
zontal inequity (the existence of fans) does not necessarily give rise to reranking (the
existence of fan overlap). Furthermore, if reranking is to be deemed inequitable, then it
must be on the basis of vertical equity considerations not horizontal equity considera-
tions. This much is clear. Things get less clear when one considers the possible sources
of reranking. In Figure 3 the only possible source of reranking is the existence of dif-
ferential treatment - i.e. the fans. Thus an occurrence which offends the principle of
vertical equity - if it offends any equity principle - can arise solely through the ex-
istence of horizontal inequity. As Aronson et al. point out, there is, however, another
possible source of reranking - a marginal tax rate in excess of 100%, which may oc-
cur over limited ranges of actual tax schedules. 4 2 This could cause reranking even if
everyone faces the same tax schedule.
The terms H and R are always non-negative, so differential treatment always reduces
the vertical effect. Empirically, Aronson et al. show that H increases and R decreases
when the income range used to define "equals" is expanded, but the total differential
41 This is similar to the measure of reranking proposed by Atkinson (1980) and Plotnick (1981).
42 This would give rise to a downward-sloping section of the relationship between x and x - T(x) and may
result in a household to the left of the peak swapping places with a household to the right of the peak in the
move from the pre-tax distribution to the post-tax distribution.
treatment (H + R) remains fairly constant. Thus, whilst conceptually distinct, H and R

seem likely to be difficult to disentangle in practice. Expressing V as a percentage of RE
facilitates a comparison of the relative importance of vertical effects versus horizontal
inequity and reranking.
3.3.2. Empiricalwork on horizontal equity
Wagstaff and van Doorslaer (1994) have illustrated the application of Aronson et al.'s
methods to health care financing by decomposing the pro-rich redistributive effect of the
Dutch health care financing system. They that most of this effect is due to the duality of
the system's insurance payments, with income-related payments mainly for the lower
two thirds of the income distribution and non-income-related premiums for the higher
income groups. They show, however, that some of the redistributive effect is due to
horizontal inequity and reranking, and that redistributive effect would have been 14%
lower than it was in 1987 if all households at each level of (equivalent) income had made
exactly the same (equivalent) health care payments. They also illustrate - by means of
a microsimulation exercise - the effects of changing the contribution rules of one of the
social insurance schemes on each of the components of the redistributive effect. The
abolition of exemptions and contribution ceilings is shown to reduce the gap between
the actual and potential redistributive effect.
Van Doorslaer et al. (1999) provide a comparison of the income redistribution conse-
quences of the health care financing mixes adopted in twelve OECD countries using the
methods described in Section 3.3.1. Figure 4 illustrates how the various components of
g=0.12
0.0200U
0.0150
g=0.068
0.0100
0.0050 g=0.020
0.0000
-0.0050
n A Inn
-0.10 -0.05 0.00 0.05 0.10 0.15 0.20

Kakwani index i I
I V x t
RE
Figure 4. Redistributive effect of social insurance as a function of KT and g.

redistributive effect are related for the six countries for just one source: social insurance
payments. It shows RE as well as V as a function of g and KT. At any given level
of funding (i.e. g), the vertical effect V is proportional to progressivity (or regressive-
ness). Similarly, at any given level of progressivity, V can be seen to increase with g.
For example, social insurance payments are only slightly more progressive in France
than in Finland, but generate a much larger redistributive effect in France because there
they represent about 12% of gross income whereas in Finland they are less than 2% of
gross income. The graph also shows that the presence of differential treatment of equals
(H + R) can substantially reduce the vertical redistributive effect. This is not visible for
the four countries where social insurance payments only represent a very small share
of income, but in the two countries with important shares of social insurance financing,
RE is well below V. In France, (H + R) lowers the positive redistributive effect, and
in Germany, it increases the negative redistribution. Horizontal inequity and reranking
in these two countries are mainly generated by the differences in sickness fund pre-
miums between households with similar gross incomes due to premium exemptions or
reductions (e.g. among elderly in France) or variation in contribution rates (e.g. among
Krankenkassenin Germany). In other words: in the absence of this differential treatment
of households with equal incomes, the Gini reduction would have been a lot higher in
France and the Gini increase somewhat lower in Germany.
Van Doorslaer et al. (1999) also find that unequal treatment of unequals (the ver-
tical effect V) is far more important in terms of redistributive effect than differential
treatment. There are, however, differences across sources, as can be seen from Table
3 which shows V100 , the vertical redistributive effect V expressed as a percentage of
the total RE. In general, and not surprisingly, large discrepancies between V and RE
occur in the voluntary private payments, where there is little or no relationship between
payment and ability to pay. Smaller discrepancies occur in public payments, where, at
least in the cases of direct taxes and social insurance, there is a link between payments
and ability to pay. Within public sources, there is, however, some variation. The discrep-
ancy between V and RE is very small in the case of taxation, despite the well-known
anomalies in personal income tax systems (such as tax relief on mortgage interest pay-
ments and health insurance) and the inevitable differences at a given income level in
household spending levels and patterns. By contrast, social insurance payments show a
non-negligible degree of differential treatment, mainly due to varying contribution rates
and exemptions on the basis of criteria other than income. There are also variations
across countries. Surprisingly, perhaps, the discrepancy between RE and V is fairly
small in the case of the direct taxes used to finance health care in the Nordic countries.
By contrast, and less surprisingly, the discrepancy is fairly high in the case of the social
insurance scheme operating in France and (to a lesser extent) in Germany.
4. Equity in health care utilization
The interest in the work reported in this section is whether, on average, persons in equal
need of treatment receive similar treatment, regardless of their income. The issue of ver-
1834 A. Wagstaff and E. van Doorslae
se sesgz 8 8
- C
N NN
cC cz l0
COC ) 00C n N 4 oo
_
CO
o sN *0s 0 - cN
N -N 00
Cf
re (C
CI
u>; o(:E rB cc
od:
o r
C)
o.
el
on se8s s > o o
t) NO
I NNN u 0 Om00
21) 0'>[ oCo
'0 0'CN0
QR8 e- e-
0) r- c7s D oh b
2
.S0 CT C1-m r CT V
o
0 q o no o7 o mo o
Cs S o O v,or 'm
tm on-ON Sm
m
Cs-
r.
Eo C1 NnN -Nc-C
m 8 Cs
rc' N 00 cC C N C- n c N
.;
0i
9
~~C
~~~~*C~~~~j
N O C~
C ) t- C
is
C.) C, n tO- o N o0
o ' No 0 o' ON c o No o
5Cs
dc CUo- ) 6n o io R> -m o
Cs
0
Cs
m6
o ~~ c9oooo
zeo Fo
C
'E
i - ON _ v, 0 _ ON ON
00' 0' 000 0 00

Cs
ce cec)>o
o o o oe~B
Cs
oON
t C
s CO >.C c sr
0- N 0
C _
Cs 0 ,, Cs °> ::C a _
U 8 · tN2=
c. < Orm
a, b
o33
XZvZ
Ch. 34: Equity inHealth Care Finance and Delivery 1835
tical equity - whether persons in different degrees of need are treated in appropriately
different ways - has hardly been the subject of any research by economists. 4 3
4.1. Regression-basedtests of inequity in health care utilization
4.1.1. Testing for inequity
Suppose there are two income groups, "rich" and "poor" and we measure "need" (set-
ting aside our earlier objections to this definition) by the presence or absence of illness,
so that we have two illness categories, "ill" and "not ill". Let mi be the medical expen-
diture received by person i, and hi be a dummy taking a value of 1 if person i is ill
(and therefore in need of medical care) and zero otherwise. Suppose we estimate the
following model:
cIr + Brhi + Uri if rich,

mi -= |ap+ phi + upi if poor, (6)
where the a's and P's are coefficients and the u's error terms. If, on average, people
who are not ill are treated the same, irrespective of whether they are rich or poor, we
would find ar = ap. If, on average, people who are ill are treated alike, whether they
be rich or poor, then we would find ar + -r = ap + ip. If equity obtains amongst both
need categories, then obviously we have both ar = p and r = fp. One could estimate
the parameters of interest either by running two separate regressions - one for the rich
group, one for the poor - or by running a single equation with the appropriate interac-
tions:
mi = ro + 7rlyi + 7r2hi + 7r3yihi + ui, (7)
where yi is a dummy taking a value of one if person i is rich and zero otherwise.
The relationships between the parameters in Equation (6) and those in Equation (7)
are simply: r0 = p, rl = r - ap, 7r2 = hp, and r3 = r - fp. Thus Wagstaff et al.
(1991b) propose testing for inequity by testing the joint hypothesis
Ho: ar = up and fr = p, (8)
or, alternatively, by testing the null hypothesis 4 4
Ho: ;rl =0 and Yr3 = 0. (9)
43 See, however, Cullis and West [(1979, pp. 237-239)] and Mooney (1996).
44 Essentially the same test is proposed by Birch et al. (1993) but they propose to use it not only for income-
related inequity but also for inequity arising with respect to region of residence, education and even the level
of community contact.
It is important to note that estimating Equation (7) without the interaction between yi
and hi - as is often done in the literature - is tantamount to assuming that the extra care
received by the ill is the same, irrespective of whether they are rich or poor. The general
approach outlined here can obviously be extended to the case where hi is a vector of
need indicators (not even necessarily dummy variables).4 5
There is a parallel here with the literature on discrimination in the labour market.
There the aim is to detect whether workers of equal productivity are paid the same,
irrespective of whether they are male or female, or Black or White. Empirical analysis
of this issue proceeds by estimating a model of the type in Equation (6), in which mi is
hourly wages (or the log thereof), hi is a productivity proxy (or vector of proxies) and
the sample is split by gender or by race. It is concluded that discrimination is absent if
the intercepts and slopes are the same across the two groups and to exist if they differ.
Discrimination would be measured either as (r - ap) hp(/r - p) or as (r - ap) +
hr(/Br - Pp), depending on which group was used as the reference group.46
One issue that merits discussion is the appropriate specification of Equation (6). As
specified above, the model includes only variables that are proxies for need. Ought one
to include factors other than need that may influence medical care utilization? And if
one does, what then is the appropriate test of inequity? There is a good deal of confusion
in the literature on both points.
The first of these points has been the subject of some debate in the discrimination
literature. 4 7 The view there is that one should include in one's wage equation only those
factors that are legitimate sources of variation in average hourly wages between men
and women. All agree that the core set of variables to be included are those capturing
human capital, because these capture productivity and differences in productivity are a
legitimate source of wage differences. This is not to say that there are not influences on
wages other than human capital, but rather that these are probably not legitimate rea-
sons for men and women having, on average, different hourly wages. By omitting such
factors from one's wage equation, one forces their effects to get channelled into differ-
ent intercepts or different slope coefficients on the productivity proxies. Including these
other factors would result in an unwarranted reduction in the estimate of discrimination,
since one would be taking out of one's estimate some differences that should not taken
out because they are not legitimate sources of wage differences. There are a couple of
twists to this story. One is that there may be variables other than human capital that
could, depending on one's view, be regarded as legitimate influences on wages but are
hard to measure. One could argue that women have a preference for jobs with flexible
hours to make raising children easy, even though such jobs are badly paid. Some argue
that this preference ought to be netted out of the discrimination measure but acknowl-
edge that in practice it cannot be done since this preference is hard to measure. The
45 It is worth noting that since we are using regression analysis simply to obtain group means, the approach
is not undermined by poor health reducing income.
46 Cf. e.g. Joshi and Paci (1998).
47 Cf. Joshi and Paci (1998) and Gill (1994).
other twist is that there may be variables that are neither wholly legitimate nor wholly
illegitimate sources of wage differences between men and women. Occupation is an
example. 4 8 It may be that women prefer certain occupations even though they are not
well paid because they are attractive for other reasons. This might be argued to be a
legitimate source of wage differences. On the other hand, it may instead be the case that
employers force women into these occupations by segregating workers (employment
discrimination).This would not be viewed as a legitimate source of wage differences.
Omitting occupation would result in discrimination being overestimated, but including
it would lower the estimate of discrimination by too much.
The logic of these arguments applies, it would seem, with equal force to the mea-
surement of inequity in the delivery of health care. In one's estimating equation one
ought to include only those variables that are regarded as legitimate sources of differ-
ences in medical care utilization across income groups. If one is interested in testing
for equal treatment for equal need, this means including only indicators of need: not
only measures of health but also demographic variables, since most health indicators
must be used in conjunction with demographic variables if they are to capture need
properly. Non-need variables, by contrast, ought to be excluded from the estimating
equations.
This leaves the question: How should inequity be assessed if non-need variables are
included in the estimating equations? Suppose, for the moment, that one estimates Equa-
tion (6). Then the difference between the mean medical care utilization of rich and poor
is equal to:
mr -mp = (r - p) + hr(r - p) + p(hr - hp). (10)
The first and second terms reflect inequity, while the last is the legitimate source of
difference between rich and poor due to the fact that the two groups differ in their need
for care. Now, suppose that instead of estimating Equation (6), one estimates the pair of
equations:
I r + brhi + CrXi + $ri if rich,

mi a + bphi + cpXi + £pi if poor,
where xi is a variable that does not reflect medical need. The individual's area of resi-
dence would be an example. The difference in expected values between the mean med-
ical care utilization of rich and poor is equal to:
mr- mp = (ar- ap) + hr(br- bp) + bp(hr - hp)

+xr(cr - cp) + cp(xr - Xp). (12)
48 Cf. Gill (1994).

The first and second terms reflect inequity, as before, while the third, as before, does
not. What of the fourth and fifth terms? The fourth clearly reflects inequity - it captures
differences between rich and poor in the effects on utilization of a variable that does
not reflect medical need. The variable xi ought not, of course, to influence utilization in
either group, so that the fifth term should also count as inequity. Thus the only legitimate
source of difference between rich and poor in their mean medical utilization is the term
capturing need differences - the third term. All the rest is inequity.
What does this imply about testing for inequity in a model containing both hi and
xi ? It indicates that, at a minimum, one ought to include in one's assessment of inequity
differences across income groups in the a's, b's and c's - not simply differences in
the a's and b's, or worse still only differences in the a's. But even if one does include
differences in c's in one's assessment, one would still get an inaccurate estimate the
degree of inequity, since one would have captured only the first, second and fourth terms
in Equation (12), not the last term. In a study where there are just two income groups,
including the last term is straightforward. But in studies where income is measured in
terms of quintiles, it is far from obvious how one ought to proceed. Furthermore, unless
hi and xi are uncorrelated, the estimates of fr and Bp obtained from Equation (6) will
differ from the estimates of br and bp obtained from Equation (11). So, if one were
to seek to assess inequity by focusing on the term capturing the legitimate source of
difference in utilization [the third term in Equations (10) and (12)], one would get a
different picture depending on whether one estimated Equation (6) or Equation (11). In
general, the best that can be hoped from studies that include x-variables is some tests of
differences in the a's, b's and c's.
This is not to say that including x -variables is totally counter-productive. Their inclu-
sion could shed light on the channels through which discrimination occurs, providing
one has estimated first a model with only the h-variables in it. Suppose that one esti-
mates Equation (6) and rejects the null hypothesis of no inequity in Equation (8). Sup-
pose one then estimates a model along the lines of Equation (11) with private insurance
coverage as the x-variable. Suppose this turns out to be significant and one cannot reject
the hypothesis:
HO: ar = ap, br = bp and Cr = Cp =O. (13)
Then a reasonable inference would be that the rejection of the hypothesis in Equation (8)
was due to differential private insurance coverage (xr xp) and that it was this that
was accounting for the difference in intercepts and slope coefficients in the first model.
Again, a single-equation version of Equation (12) could be devised. The analogue of
Equation (7) for Equation (12) is:
mi = o + nyr i r2hi + 7r3yihi + r4xi + r5xiyi ui. (14)
The null hypothesis of no inequity equivalent to that in Equation (13) then becomes
H0 : 7rl=0, 7 3 =0, 4 =0 and T5s=0.( (15)

4.1.2. Empiricaltests of inequity
Benham and Benham (1975) estimate an equation similar to Equation (7) with yi as a
continuous variable but 73 constrained to be zero. They find that the t-statistic on their
estimate of 51 is just below the critical value in their 1963 equation but is well below
the critical value in their 1970 equation. They conclude that "the US has moved in the
direction of greater equity" (p. 101). However, since n3 is set equal to zero, this is an
incomplete test.
Puffer (1986) estimates equations similar to Equation (7) for the UK and US, where
mi is a categorical variable defined over the number of primary care physician visits
and yi is one if the respondent is in the bottom income quartile. 4 9 In the case of the UK,
1ri and 513 are typically zero, but the coefficient in the 573 vector pertaining to gender
is significant: the probability of contact is lower amongst those in the bottom quartile
simply because being female has a smaller impact on contact probability amongst per-
sons in the lower quartile. In the case of the US, rtl and 513 are insignificant except in
the specification where excellent health is interacted with the bottom income quartile
dummy; in this case, ri turns out to be significantly negative, implying that the care
received by those not in excellent health is less if the person is in the bottom income
quartile than otherwise.
Van Doorslaer et al. (1992) report results for eight countries of regressions along the
lines of Equation (6). In their study, mi is imputed medical care expenditures, based
on the number of primary care visits, specialist/outpatient visits and hospital days, and
separate regressions along the lines of Equation (1) are estimated for the five income
quintiles. A two part model is run with the first part (zero versus positive imputed ex-
penditures) being estimated as a logit model and the second part (the amount of imputed
expenditures conditional on the figure being positive) by truncated OLS. The likelihood
ratio tests of inequity in the first part are rejected in all countries except Denmark. By
contrast, in all but two countries the hypothesis of no inequity is rejected in the second
part and in all but two countries the null hypothesis is rejected for the two-part model
overall. The rejection is especially decisive in the case of the US, but the intercepts
and the coefficients on the health and demographic variables do not increase or decline
monotonically across income quintiles. 50
Birch et al. (1993, p. 96) estimate a two-part model version of Equation (7) for the
use of family physician visits in Canada. They impose the constraint 7r5 = 0 and correct
for sample selectivity. Household income is entered as a continuous variable and need
is proxied by self-assessed health status dummies. They find that the coefficients rl and
49 Comparing the UK and US in this area is difficult, of course, because primary care physicians in the US
deliver care in some specialties that in the UK would be delivered by specialists.
50 The largest intercept, for example, is the bottom quintile's. The smallest is the 2nd bottom quintile's.
:r3 are not significantly different from zero and conclude that "household income is not
associated in any significant way with utilization" (p. 99).51
Hamilton et al. (1997) estimate ordered probability models for physician visits for
both Canada and the US along the lines of Equation (14). In their paper, x-variables
include variables such as years of education, marital status, race and doctors per 1000
population in the individual's local area. The income variable yi is specified as income,
income squared and income cubed. The coefficients 73 and rs are constrained to be
equal to zero, thereby ruling out some aspects of inequity - i.e. differences in the a's
are allowed for, but not differences in the b's or c's. In the US, but not in Canada, the 2r
coefficient on income is significantly positive in the positive visits equation for males,
and in the positive visits equation for females. The rl coefficients on all three income
variables are significant in the US but not in Canada. However, given the inclusion of the
x-variables in the models, and the zero restrictions placed on the 7r3 and 7r5 coefficients,
one has to be careful about reading too much into the results.
The same is true of the results reported by Grytten et al. (1995), who estimate logistic
regression models for patient- and physician-initiated visits in Norway for 1975 and
1985. Their models contain no interactions and their test of inequity is simply whether
or not rcl is significantly different from zero. They find that it is not and conclude from
this "that equality of utilization has been achieved within a publicly financed primary
health service" (p. 950). A fuller test would of course be to remove the zero restrictions
on the interactions and test the restrictions in Equation (15).
In his study of equity in Sweden, Gerdtham (1997) also estimates a restricted model
and tests only a subset of the restrictions implied by an equitable system. He estimates
a variant of the single-equation version of Equation (14), where the x-variables include
variables such as years of education, whether the individual lives in a large city and
marital status, the income variable yi is specified as a vector of quintile dummies, and
as in Hamilton et al.'s paper, the coefficients Jt3 and s5are constrained to be equal
to zero. The model is estimated separately for different types of care using count-data
methods. Some of the rl coefficients were significant in some of the physician contact
regressions but not all.
Rosenzweig and Schultz (1991) also use a model along the lines of Equation (14)
to test the hypothesis that prenatal medical services are provided to pregnant married
women on the basis of medical need in the United States in 1980. They constrain r3
and 7r5 to be zero and test the joint hypothesis 7rl = 7r4 = 0. They reject the hypothesis
except in the case of amniocentesis. Their analysis goes further than the other literature
in this area because (i) they attempt to control for unobserved initial health endow-
ment of the child, and (ii) they attempt to infer which of four implicit pricing regimes
prevails (including the allocation-according-to-need regime). With respect to (i), it has
51 The authors then estimate a variant of Equation (7) with 7r3 = 0. They find some significant r5 coefficients
on x variables such as education, social support and region of residence and interpret this as pointing towards
"non-income-related barriers to reasonable access" (p. 99).
been shown by Manning et al. (1982) that using subjective health measures ascertained
subsequent to the use of medical services may lead to significant biases in the estimates
of income or schooling effects. However, Rosenzweig and Schultz show that even the
use of objective health indicators prior to treatment as need indicators may lead to in-
consistent estimates if they are correlated with unmeasured initial health endowments.
They get round this by employing instrumental variables for the child's birth endow-
ment. With respect to (ii), they show that the difference between the income effect as
estimated in the unconditional equations and in those conditional on endogenous health
can distinguish which of four different regimes dominates prenatal care allocations:
(a) the normal market regime where all consumers face the same care prices, (b) one in
which implicit prices are lower for higher income groups because of tax deductibility
of medical care expenses under progressive taxation, (c) a regime in which the poor
face lower prices because of subsidy programmes such as Medicaid, and (d) a regime
in which nobody pays user fees and health care is allocated on the basis of need. For all
four prenatal services they examine (amniocentesis, Caesarian, ultrasound and X-ray)
their estimates are consistent with regime (b): the conditional income effect is lower
than the unconditional one, which means that the implicit price for these services is
lower for the rich than the poor. According to these results, then, not only is it the case
that prenatal care in the US is not allocated according to need; it is also the case that
access to these services, defined in terms of prices, is unequal, with the rich enjoying
better access than the poor. This, the authors argue, is because the distributional effects
of tax subsidies, which benefit the better-off most, offset the distributional effects of
Medicaid and other public programmes aimed at assisting the poor. 52
4.2. Measuring inequity
The discussion of the previous section was concerned with testing for inequity. It does
not enable inequity to be quantified - something that is essential if cross-country com-
parisons or comparisons over time are to be performed. The regression approach can,
however, be extended to allow an index of inequity to be derived, providing one is pre-
pared to accept a more general definition of horizontal equity. 5 3 So far, equity has been
taken to mean that the intercepts and slope coefficients in the medical utilization equa-
tions should be the same for all income groups, i.e. in the case of Equation (6), ar = ap
and fr = ip. In other words a health care delivery system cannot be said to be hori-
zontally equitable if the rich and poor are treated differently in any morbidity category.
But what if the rich are treated favourably in one morbidity category (e.g. the non-sick
category) but the poor are treated favourably in the other (e.g. the sick category)? A less
restrictive definition of equity would regard such a situation as horizontally equitable
52 The relative weakness of the latter effects may have something to do with the low fees paid by Medicaid
for the procedures analysed by Rozenzweig and Schultz (1991).
53 What follows draws heavily on Wagstaff et al. (1991b) and Wagstaff and Van Doorslaer (2000).
on balance, providing any favourable treatment afforded to the poor amongst the sick
was sufficiently large to offset the favourable treatment afforded to the rich amongst the
non-sick. But how large does "sufficiently large" have to be before one can say that, on
balance, no inequity exists?
4.2.1. A direct standardization-basedindex
One approach, suggested by Wagstaff, van Doorslaer and Paci (1991 b), is to divide one's
sample into income groups and then compute need-standardized medical care figures for
each income group using the direct standardization method [cf. e.g. Rothman (1986)].
These figures indicate how much medical care people in each income group would have
received if they had been in the same degree of need as the sample as a whole. The
figures are obtained by applying the need characteristics of the sample to the mean
medical care figures of the income group in question.
Let mi denote the amount of medical care received by individual i in a given period.
The distribution of medical care by income is captured by the medical care concen-
tration curve LM(p) in Figure 5, which graphs the cumulative proportion of medical
care against the cumulative proportion p of the sample, ranked by income. The con-
centration index, CM, corresponding to LM(p) indicates the degree of inequality in the
distribution of medical care but will tell us something about the degree of inequity only
in the unlikely event that need for medical care does not vary with income. The extent of
horizontal inequity can then be assessed by comparing the standardized concentration
I.0
cumulative
proportion of
medical care
0.0
0.0 1.0
cumulative proportion of sample p, ranked by income
Figure 5. Concentration curves of medical care and need. Source: Wagstaff and van Doorslaer (2000).
curve, labelled in L+(p) Figure 5, with the diagonal. If L+(p) lies below (above) the
diagonal, inequity exists and favours the better-off (worse-off). Two conditions can arise
under which there is no inequity: (i) if L+(p) coincides with the diagonal (and the stan-
dardized medical care use is the same for all income groups) or (ii) if L + (p) crosses
the diagonal (and inequity favouring the rich exactly compensates inequity favouring
the poor). The degree of inequity can be measured as twice the area between L+ (p)
and the diagonal, or equivalently as:
HIwv = - 2 L(P)dp = C (16)
where C + is the concentration index for directly standardized medical care, defined as
twice the area between L+(p) and the diagonal. A positive (negative) value of CM in-
dicates horizontal inequity favouring the better-off (worse-off). The standardized values
of medical care can be computed simply by running a regression of mi on xi for each
income group and then applying the population average value of xi to the gth group's
regression coefficients.
The discussion of statistical inference in Section 4.1 focused on testing the signifi-
cance of differences across income groups in the regression coefficients used to com-
pute the directly standardized medical care values underlying the Hlwvp index [cf. Van
Doorslaer et al. (1992)]. This provides only a partial answer to the statistical inference
question, since non-rejection of the null hypothesis is only a sufficient condition for
Hlwvp to be zero. Hlwvp can also be zero if the concentration curve crosses the di-
agonal. Furthermore, no such test is available in the indirect standardization method
discussed below. A complementary test of the necessary condition for no inequity re-
quires a test of the index itself to be zero. Wagstaff and van Doorslaer (2000) present
both convenient and more accurate estimators for the standard error of Hlwvp based on
the work of Kakwani et al. (1997) that allow for the statistical testing of the index itself
to be zero or to be different from other indices.
4.2.2. Empiricalwork using the direct-standardization-based

index
Van Doorslaer et al. (1992) report Hlwvp indices for eight countries using a measure
of imputed expenditures for GP visits, specialist visits and inpatient days.5 4 In the di-
rect standardization they include various indicators classified as either belonging to a
medical, functional or subjective model of ill-health [cf. Blaxter (1989)]. However, not
all indicators are available for each country and there is substantial variation in those
that were included. Their results indicate that, in general, when multiple indicators are
used in the standardization, the standardized medical expenditure distributions are less
54 More detailed results for each of these eight countries, and for France and Portugal, can be found in the
country-specific chapters of Van Doorslaer et al. (1993).
pro-poor, or more pro-rich, than when only one indicator is included at a time. This sug-
gests that inequalities in morbidity also exist within some of the morbidity categories
used. When the most general need standardization specification available per country
was used, four countries show a positive index (the Netherlands, Spain, UK and US)
and four countries (Denmark, Ireland, Italy, Switzerland) show a negative index. But
the authors warn against reading too much into these results in view of their sensitivity
to the need indicators used in the standardization. Given their suspicion that the in-
equity values reported are likely to understate any inequity favouring the well-off, they
conclude that pro-rich inequity almost certainly exists only in countries where inequity
favouring the well-off was detected and the null-hypothesis was rejected (i.e. Spain, the
UK and the US).
Propper and Upward (1992) report Hlwvp indices for the UK for the years 1974,
1982, 1985 and 1987. The measure of utilization is imputed expenditures as in the study
by van Doorslaer et al. (1992) and the standardizing variables are age and gender, and
the presence or absence of acute, limiting and non-limiting chronic illness. In each year
the authors find inequity favouring the poor, but with an apparent trend upwards in the
Hlwvp index from 1982 onwards (Hlwvp = -0.06 in 1987).
Lairson et al. (1995) analyse inequity in the delivery of health care in Australia in
1990 employing the same methods. They compute the Hlwvp index separately for the
imputed expenditures for physician consultations (both GP and specialist), outpatient
and inpatient care, and for the total of these three types of care. They too observe a
sensitivity of their findings to the indicators used in the standardization. When using
self-assessed health in the need standardization, they find substantial inequity favouring
the rich (Hlwvp = 0.06), mainly as a result of the rich using more physician consulta-
tions and inpatient care than the poor, given their need. Inequity favouring the poor is
found for outpatient care, and for all types of care if serious or chronic illness is included
in the standardization.
4.2.3. An indirect standardization-basedindex
The method based on the direct standardization has a major disadvantage: it requires
the use of grouped data and its usefulness is therefore limited by the fact that the value
of CM will depend on the number of income groups G. An alternative [Wagstaff and
van Doorslaer (2000)] is to employ the method of indirect standardization, which can
be employed on individual-level data, as well as on grouped data. The indirect standard-
ization generates a figure for each individual indicating the amount of medical care she
would have received if she had been treated as others with the same need characteris-
tics were, on average, treated. Wagstaff and van Doorslaer interpret this as her need for
medical care.
The extent of horizontal inequity can be assessed by comparing the concentration
curve of actual medical care utilization LM(p) with the need concentration curve, la-
belled LN(p) in Figure 5: if the latter lies above (below) the former, there is horizontal
inequity favouring the better-off (worse-off). They define an alternative measure of hor-
izontal inequity (HIwv) as twice the area between the need and medical care concen-
tration curves:
HIwv = 2 [LN((p) - LM(p)] dp = CM - CN, (17)
where CN is the concentration index for need (i.e. indirectly standardized medical care).
A positive (negative) value of HIwv indicates horizontal inequity favouring the better-
off (worse-off), whilst a zero value indicates that the factor of proportionality (between
medical care and need) is the same irrespective of income. The indirectly standardized
medical care figure, m*, can also easily be computed using regression methods. Only
one regression model needs to be estimated, rather than G. The standardized values are
computed as the predicted values saved from an equation where medical care use is
regressed on a vector of need indicators. Wagstaff and van Doorslaer (2000) show how
both convenient-regression based and more accurate standard errors for Hlwv can be
used for statistical inference.
4.2.4. Empiricalwork using the indirect standardization-basedindex
Wagstaff and van Doorslaer (2000) illustrate the use of the indirect standardization-
based index using data from the Netherlands for 1992. Using a two-part model, they
estimate standardized values for GP care, outpatient care and inpatient care utilization.
They find mild pro-poor and non-significant inequity indices for GP and inpatient care,
but high and significant pro-rich inequity for specialist outpatient care.
Van Doorslaer et al. (2000) have applied the indirect standardization approach to mea-
suring and testing for horizontal inequity in the health care delivery systems in eleven
countries. They compute HIwv inequity indices for the imputed expenditures for two
types of medical care utilization (physician visits and inpatient days) and for the total
of these two. For some countries, total physician visits are further subdivided into GP
and specialist visits. Some selected results for eight of the eleven countries - which are
presented in Table 4 - illustrate their main conclusions. 5 5 They find little or no signif-
icant inequity in the distribution of total medical care expenditures, though the disag-
gregation shows that this is the result of opposite utilization patterns for inpatient care
and physician consultations. Significant pro-rich inequity emerges for physician visits
in four countries: Denmark, Finland, Sweden and the US. Further disaggregation for
those countries for which it was feasible shows that this in turn results from even higher
pro-rich inequity in specialist visits: the higher income groups use a lot more specialist
services than is to be expected on the basis of need. Some pro-poor inequity is found for
GP visits, but this is much smaller and only significant in the case of Belgium. Inpatient
care is also distributed pro-poor but significantly so only in the cases of Belgium and
55 Results for Ireland and for East and West Germany have been excluded because the surveys used for these
countries did not have all need indicators used in the specification reported here.
Table 4
HIwv indices of horizontal inequity - selected countries
Country Type of medical care utilization

GP visits Specialist All physician Inpatient Total medical
visits visits days care
Belgium (1995) -0.033 0.039 -0.013 -0.090 -0.071

Denmark (1994) -0.003 0.0834 0.049 -0.065 -0.060
Finland (1996) 0.005 0.056 0.036 -0.070 -0.029
Netherlands (1992) -0.006 0.079 0.021 -0.070 -0.038
Sweden (1990) n.a. n.a. 0.030 -0.029 -0.014
Switzerland (1992) - - -0.004 0.056 0.040
UK (1989) 0.001 0.0275 0.014 -0.051 -0.016
US (1987) n.a. n.a. 0.043 -0.007 0.009
Notes:
(i) Hlwv indices are for need specification including age, sex, SAH vector and dummy
chronic illness;
(ii) significant indices in bold (p < 0.05).
Source: Van Doorslaer, Wagstaff et al. (1999).
the UK. Because of the importance of inpatient care utilization in overall expenditures,
the latter is also distributed pro-poor in all countries except the US and Switzerland. It
is, however, never significant except in the case of Belgium. Surprisingly, the finding of
significant pro-rich inequity in the utilization of physician visits, especially in specialist
visits, does not seem to be clearly associated with specific delivery system character-
istics. It seems to occur in countries with universal coverage (e.g. the Scandinavian
countries) as well as in those with incomplete coverage (e.g. the US), in countries with
(e.g. Denmark, Netherlands) and without (e.g. Belgium) a GP gatekeeper role, and in
countries with (e.g. Belgium) and without (e.g. Denmark) substantial cost sharing by
patients. On the other hand, it does not emerge in two countries (Switzerland and the
UK) which seem to have few system characteristics in common.
5. Equality of health
As indicated in Section 2, it can plausibly be argued that all concerns about the distribu-
tion of health care - or access to health care - stem ultimately from a more fundamental
concern about the distribution of health itself. The absence of any health inequality at
all may well be an unattainable goal but health care systems can influence the extent to
which health inequalities exist and the extent to which they are systematically related
to characteristics such as socioeconomic status, place of residence, race, etc. In any dis-
cussion of equity in health care delivery it is of some interest, therefore, to consider (if
not focus on) the extent to which existing delivery systems bring health distributions
closer to an equal distribution. Any attempt to measure such an impact encounters the
problems of how to measure health, how to measure inequalities in its distribution and
how to establish the marginal impact on this distribution of the various determinants of
health inequality. The problem is compounded by the well-known fact that variations
in health seem to be largely determined by factors outside of the health care system -
some known, some not known - which have to be adequately controlled for when trying
to single out the impact of a particular system's characteristics. 56
There is, in fact, a large literature on inequalities in health, some of which comes from
economists. It is this part of the literature that is the subject of this section. Some of this
work is concerned with pure inequalities in health - i.e. the variation in health within a
country at any particular time. The rest of the work, like most of the work in this area
by non-economists, concerns socioeconomic inequalities in health - i.e. the variation in
health which is systematically related to socioeconomic status. We survey both types of
work and report the results of efforts to see whether there is any systematic variation in
health inequalities across countries, and if so, whether there seems to be any relationship
between health inequalities and country characteristics, especially those relating to the
features of their health care systems.
5.1. Pure inequalitiesin health
The literature here is concerned with pure inequality, in much the same way as the
economics literature on income inequality is largely concerned with pure inequality.
The aim is simply to see how far there is inequality in measures of health or ill-health
across people, irrespective of where they happen to be on the socioeconomic ladder.
This approach does not, in contrast to the approach considered in Section 5.2, capture
whether persons in poor health are rich or poor, or professionals or unskilled manual
workers, or highly educated or educated only to school-leaving age. That, argue the
advocates of the pure inequality approach, is properly viewed as part of the process of
explaining health inequalities, not part of the process of measuringthem [cf. e.g. Illsley
and Le Grand (1987)].
5.1.1. Measurementof pure health inequality
The approach proceeds using standard measures of inequality developed in the income
inequality literature. For example, suppose that health is being measured in terms of
the number of years a person lives. One then lines people up according to their age at
death and plots on the horizontal axis the cumulative percentage of the population and
on the vertical axis the cumulative percentage of years of life. Since not everyone dies
at the same age, the resultant graph will be a Lorenz curve along the lines of Figure 6. 57
56 For a recent account of the wide spectrum of determinants of population health, see e.g. Evans et al. (1994).
57 The use of the Lorenz curve to measure inequalities in age at death was first proposed by Le Grand (1985).
100%
cum %
of years
of life
0% cum % of pop, ranked by age-at-death 100%
Figure 6. Lorenz curve of inequality in age-at-death.
This can be compared with the diagonal to assess the extent of inequality and the Gini
coefficient thus provides a measure of pure inequality in age at death. 5 8 It takes a value
of zero when everyone dies at the same age and a value of one when all but one person
die at birth. This is a measure of relative inequality - if everyone's age at death doubles,
the Gini coefficient doesn't change. By multiplying the Gini coefficient by the mean age
a death, one obtains a measure of absolute inequality - the absolute Gini coefficient or
the average mean deviation. This doubles if everyone's age at death doubles. Alterna-
tives to the Gini index could, of course, be used. Amongst those used in this literature
are the variance and Atkinson's (1970) index, but in principle one could use any of the
other measures of inequality used in the income inequality literature [cf. Cowell (1995),
Lambert (1993)].
5.1.2. Empirical work on pure health inequalities
Illsley and Le Grand (1987) report mean age-at-death and the Gini coefficient for mean
age-at-death for England and Wales. They find that mean age-at-death has risen almost
continuously over the period 1921-1983: from 60 to 70 in the case of males, and from
58 Or equivalently it is equal to the area between the Lorenz curve and the diagonal expressed as a proportion
of the area underneath the diagonal. The two are equivalent because the area under the diagonal is equal to
one half.
69 to 77 in the case of females. Over the same period, the Gini coefficient for age-
at-death has fallen almost continuously, from 0.24 to 0.12 for males and from 0.18 to
0.11 for females. They also find that inequality varies across causes of death (the causes
with the lowest variance in age-at-death in 1981 are cancer and circulatory diseases and
those with the highest variance are accidents and infectious diseases) and that for some
causes of death the level of inequality has changed considerably over time (the variance
in age-at-death for infectious diseases fell dramatically over the period in question).
Le Grand (1987, 1989) reports the results of an international comparison of inequal-
ities in age at death using the Gini coefficient, the absolute Gini and Atkinson's index.
Le Grand presents both unstandardized and standardized results. The former are based
on crude death rates. The latter are obtained using the indirect standardization: the re-
sultant rates thus give the number of deaths that would have occurred at each age in the
country in question, if it had had the same population distribution as the standard coun-
try (England and Wales). Le Grand (1989) finds low Ginis for the standardized values in
Finland, Germany, Ireland, Luxembourg, the Netherlands and the UK, and high Ginis
in France, Poland, Portugal, Romania, the US and Yugoslavia.
Given what is known about the importance of non-medical determinants of health,
one cannot conclude that all of the cross-country differences in inequality in age-at-
death are attributable to differences in the health care system. Le Grand (1987) reports
some rather interesting regression results that shed some light on the sources of differ-
ences in inequality in age-at-death. Surprisingly, he finds that the more a country spends
on medical care per capita, the higheris its inequality in age-at-death but the degree of
inequality does not depend on the share of medical care expenditure that is publicly fi-
nanced. What does seem to be important is a country's GNP per capita (the higher this
is, the less inequality there is in age-at-death) and the degree of inequality in income
(the lower this is, the less inequality there is).
5.2. Socioeconomic inequalitiesin health
The pure inequality approach does not pick up the socioeconomic dimension to health
inequalities: it does not pick up whether the people in bad health are rich or poor. This
aspect of the approach has been criticized by non-economists who feel the socioeco-
nomic dimension is an integral part of the measurement exercise and should not be
pushed back to the explanation exercise.
5.2.1. Measuring socioeconomic inequalitiesin health
An alternative to the Gini coefficient that takes into account each person's rank in the
socioeconomic distribution is the concentration curve approach. 59 Suppose, as before,
health is measured in terms of the number of years a person lives. This time people are
59 Wagstaff et al. (1989, 1991a), Kakwani et al. (1997).

100%
cum %
of years
of life
O% cum % of pop, ranked by soc-econ status 100%
Figure 7. Concentration curve of age-at-death.
lined up not according to their age at death but according to their socioeconomic status,
beginning with the most disadvantaged. 6 0 We might measure socioeconomic status in
terms of social (i.e. occupational) class, or educational attainment, or income, or what-
ever. We then plot on the horizontal axis the cumulative percentage of the population
(ranked by socioeconomic status) and on the vertical axis the cumulative percentage of
years of life. Insofar as those towards the bottom of the socioeconomic ladder die ear-
lier than those at the top, the resultant graph will look like Figure 7. This concentration
curve differs from the Lorenz curve in Figure 6 in that, unlike the latter, it remembers
people's socioeconomic status.
The concentration curve can be compared with the diagonal to assess the extent of
socioeconomic inequality in health. If country X has a concentration curve that lies
everywhere closer to the diagonal than country Y, we can reasonably say that X has
a lower level of socioeconomic inequality in health than Y. The concentration index
provides a measure of socioeconomic inequality in health, being defined as twice the
area between the concentration curve and the diagonal and taking a value of zero when
everyone dies at the same age and a value of one when all but the least disadvantaged
60 The procedure outlined in this paragraph and the next was proposed by Wagstaff et al. (1989). See Wagstaff
et al. (1991a) for further details.
person die at birth.61 Like the Gini coefficient, this is a measure of relative inequality - if
everyone's age at death doubles, the concentration index doesn't change. By multiplying
the concentration index by the mean age a death, one obtains a measure of absolute
inequality - the absolute concentration index. This, of course, doubles if everyone's
age at death doubles. The concentration index will give the same result as the Gini
coefficient only if people's order in the health parade is the same as their position in
their socioeconomic status parade.
It is possible, as we shall see below, to adapt the concentration index for any measure
of health or indeed ill-health. In the case of ill-health, the concentration curve will lie
above the diagonal if ill-health is concentrated amongst those at the bottom of the so-
cioeconomic ladder. In this case, the concentration index is negative. The concentration
curve approach can also be used with grouped data. For example, the data may refer
to differences by social class or by some other categorical measure of socioeconomic
status.
The approach can be used with unstandardized or standardized data [Kakwani et al.
(1997)]. In the case where the direct standardization is used, one has to work necessarily
with grouped data (e.g. income groups) and one obtains age-sex standardized values
of one's health or ill-health variable for each group from which a new concentration
curve can be constructed. To assess the degree of inequality, this is compared to the
diagonal. Twice the area between the concentration curve and the diagonal is the directly
standardized concentration index and is denoted in Kakwani et al. by C + . If the indirect
standardization is used, one obtains a concentration curve that indicates the distribution
of health or ill-health by socioeconomic status that would be observed simply through
the covariance between demographic factors and socio-economic status. Twice the area
between the actual concentration curve and this counterfactual curve gives the inequality
index appropriate for measuring inequalities using the indirect standardization. This is
denoted by I* in Kakwani et al.
The concentration index has a number of attractions as a measure of socioeconomic
inequalities in health. Unlike the range, used often by non-economists, it reflects the ex-
periences of the entire population and not just those of the two extreme groups. It is also
sensitive to the distribution of the population across socioeconomic groups - unlike the
range, for example, the index would change if the sizes of the various groups changed
even if their mean age at death did not. Furthermore, since the concentration curve
remembers people's socioeconomic status, the index ensures that the socioeconomic
dimension to inequalities is taken into account. This distinguishes the concentration in-
dex from the Gini coefficient as well as several other inequality measures that have been
61 Or equivalently it is equal to the area between the Lorenz curve and the diagonal expressed as a proportion
of the area underneath the diagonal. The two are equivalent because the area under the diagonal is equal to
one half.
used by non-economists in the mistaken belief that their index does indeed capture the
socioeconomic dimension. 62
One index used by non-economists that does capture the socioeconomic dimension
to health inequalities is the slope index of inequality (SII). This, it turns out, is closely
related to the concentration index. 6 3 As in that approach, people are ranked by their
socioeconomic status, beginning with the most disadvantaged. A bar is then drawn for
each socioeconomic group whose height is equal to the class's mean health (or age at
death or whatever) and whose width is equal to the proportion of the population in the
class in question. The midpoints of each bar measured on the horizontal axis indicate
the relative rank of each class. For example, if the bottom class contains 20% of the
population, its relative rank is 0.1, and so on. A regression line is then estimated with
the observations being the midpoints of the top of each bar. 64 The slope of this line is
the SII - it indicates the change in health associated with moving from the bottom of
the parade to the top. The SII is a measure of absolute inequality - if everyone's health
doubles, the SII doubles. A variation on the SII which is a measure of relative inequality
is the relative index of inequality (RII), which is simply the SII divided by mean health.
Clearly, if everyone's health doubles, the RII remains unaffected.
It turns out that the RII is equal to the concentration index, divided by twice the
variance of the relative rank variable, and the absolute concentration index is equal
to the RII, multiplied by the same amount.6 5 So, the concentration index and the RII
ought to produce identical rankings when comparisons are being made over time or
across countries, as will the absolute concentration index and the SII. The equivalence
between the concentration index and the SII also suggests a quick way of computing the
concentration index - line people up by their socioeconomic status and run a regression
of their health or ill-health on their relative rank. This in turn suggests a straightforward
way of obtaining standard errors for the concentration index. However, as Kakwani et
al. (1997) have shown, such standard errors are potentially unreliable due to the serial
correlation induced by the ranking variable. They develop alternative standard error
estimators.
5.2.2. Empiricalwork on socioeconomic inequalitiesin health
A large number of studies of socioeconomic inequalities in morbidity and mortality has

been undertaken by non-economists using the SII or RII. These are not surveyed here.
We focus instead on contributions to this field from economists; this has, for the most
part, been based on concentration indices.
62 This is true of the pseudo Lorenz curves used by Preston et al. (1981) and Leclerc et al. (1990), the index of
dissimilarity used by Preston et al. (1981), and Koskinen (1985), and the index of inequality used by Pappas
et al. (1993). See Wagstaff et al. (1991a) for further details.
63 Wagstaff et al. (1991a), Kakwani et al. (1997).
64 To ensure homoscedasticity the equation has to be estimated using weighted least squares rather than
ordinary least squares. See Wagstaff et al. (1991a) for details.
65 Wagstaff et al. (1991a), Kakwani et al. (1997).
Propper and Upward (1992) use the concentration index approach to analyse health
inequalities in the UK for the years 1974, 1982, 1985 and 1987. They employ four
different measures of health: the presence or absence of acute illness (illness or injury
restricting activity in the previous two weeks); the presence or absence of non-limiting
chronic illness; the presence or absence of limiting chronic illness; and whether or not
people rate their health as "not good" on a scale including "not good", "fairly good" and
"good". Individuals were ranked by equivalent household income and the analysis was
restricted to adults. With the exception of non-limiting chronic illness in 1985 and 1987,
pro-rich inequalities were found in each year for each indicator. Inequalities were most
pronounced for the self-assessed health variable, but were also fairly pronounced for
the limiting chronic illness variable. Except in the case of non-limiting chronic illness,
inequalities increased between 1974 and 1982, and then again between 1982 and 1985.
Between 1985 and 1987, however, they fell - in the cases of acute illness and limiting
chronic illness, back to their 1982 levels; in the case of self-assessed health, even further.
Van Doorslaer et al. (1997) present the results of an analysis of inequalities in self-
assessed health for nine countries. As in the study of Propper and Upward, individuals
were ranked by equivalent household income. Ill-health was measured by the multiple-
category responses from a question in which respondents were asked to rate their gen-
eral health status, with categories such as Excellent, Very Good, Good, Fair and Poor.
Responses to this question have been found to be good predictors of subsequent mor-
tality in a variety of industrialized countries [Idler and Benyamini (1997)]. A difficulty
in an comparative context is that the number of categories is not always the same in
all surveys. In any case, converting the variable into a dichotomous one by dividing the
sample into those whose health is, say, at least good, and the rest, by choosing some
arbitrary cut-off point, as is often done,6 6 is problematic, since it can lead to different
conclusions concerning trends in or differences in inequalities in health, depending on
where the cut-off point is chosen.6 7 Instead of dichotomizing the self-assessed health
question, it was assumed that underlying the responses to it is a latent variable with
a standard lognormal distribution. 6 8 In effect, the latent health scores for each of the
response categories are obtained by dividing up the area under the standard lognormal
distribution according to sample proportions falling into each of the response categories.
Each respondent is assigned the latent health score corresponding to their response. This
score is increasing in ill health, since the best response category is put at the left-hand
tail of the distribution. The lognormality assumption means that the difference between
Excellent and Good is smaller than the difference between Good and Fair, which is, in
turn, smaller than the difference between Fair and Poor, and so on. Individuals were
ranked by equivalent household income and the direct method of standardization was
used. The study found pro-rich inequalities in latent ill-health in all nine countries, with
66 Cf. e.g. Wagstaff et al. (1989).

67 Cf, Kunst (1992), Kunst et al. (1995) and Wagstaff and Van Doorslaer (1994).
68 This method was proposed by Wagstaff and Van Doorslaer (1994).
low levels of inequality in Sweden and East Germany and high levels in the UK and
US. In all countries inequality was found to be significant. The US was found to have a
significantly higher level of inequality than the UK, which, in turn, had a significantly
higher level of inequality than the remaining countries, amongst which no significant
differences in inequality were found. Dominance-checking was also undertaken. The
results showed that the US concentration curve lay everywhere outside the UK's, which
in turn lay everywhere outside all the remaining curves apart from that of the Nether-
lands.
Van Doorslaer et al. go on to explore the statistical association between health in-
equality indices and two measures of health spending and the level and distribution of
income for the nine countries in the study. 69 Neither total health care expenditure per
capita, nor the percentage of total expenditure spent publicly appear to have any sta-
tistical association with health inequality, suggesting that neither higher spending, nor
higher public sector shares are associated with lower health inequality. Of the two other
variables - the GDP per capita and the Gini coefficient of income inequality - only
the latter proved to bear a consistent and significant positive association with health
inequality. It appears, therefore, that income-related inequality in health is more associ-
ated - in these countries, at least - with the distribution of income in a society than to
its aggregate income level or its levels of health spending.
The results reported in the study by van Doorslaer et al. may be sensitive to the
choice of transformation of the self-assessed health responses. Two recent studies shed
light on this issue. Gerdtham et al. (1998) obtain, by means of a survey undertaken
in Uppsala County, information on income, self-assessed health and two measures of
quality of life (QoL) - a visual analogue rating scale (RS) measure and a time trade-off
(TTO) measure. They then compare the average QoL scores of persons reporting their
health as Excellent, Very Good, Good, Fair and Poor. They find that for both the RS
and TTO results, the difference between Excellent and Very Good is smaller than the
difference between Very Good and Good, which is, in turn, smaller than the difference
between Good and Fair, which is, in turn, smaller than the difference between Fair and
Poor. This lends some support to the lognormal transformation. Gerdtham et al. go on to
compute concentration indices for self-assessed health and the two QoL measures. They
find no statistically significant differences between the three inequality index values. A
similar finding is reported by Humphries and van Doorslaer (1998). They compute the
directly standardized concentration index for two alternative measures of health: the
self-assessed health question, with responses scored using the latent variable approach,
and the McMaster Health Utility Index, a generic health index that captures both quan-
titative and qualitative aspects of eight dimensions of health, namely vision, hearing,
69 These variables were chosen because they had been used before in a cross-country comparison of (non-
income-related) health inequality by Le Grand (1987). One rationale for including aggregate health spending
would be that its coefficient would capture any differences across the income distribution in the impact of
health spending on health [cf. Bidani and Ravallion (1997)].
speech, ambulation, dexterity, emotion, cognition, and pain. The authors find that in-
equalities in ill-health are slightly higher when measured using the self-assessed health
variable than when measured using the HUI, but not significantly so.
6. Conclusions
In all of the areas covered by the survey, useful progress has been made and important
conclusions have been reached. There are, of course, important issues that remain unre-
solved - both at the conceptual and empirical levels - but work in the area has advanced,
especially over the last 15 years or so, and is continuing to do so.
On the issue of the nature of equity, a number of useful lessons have been learnt, but
a number of questions remain unanswered. These have been outlined in Section 2.2.5
and do not need repeating here. Important contributions have been made by economists
in this area, with the issues now being more sharply focused than was the case 10 or
15 years ago. Nonetheless, there is still a divergence of views on key normative issues.
It seems quite probable that such differences will persist. Differences persist amongst
philosophers and others on these issues, and it seems unrealistic to expect economists
- who are, after all, not noted for their ability to agree with one another - to agree on
them. On the empirical work reviewed in Sections 3-5, a clearer picture is emerging.
This work has continued apace, with substantial but successful importation from other
areas of economics, notably the fields of public finance, income distribution and redis-
tribution, and labour economics. Like other areas on the microeconomic side of health
economics, the work has benefitted from the huge advances in personal computer tech-
nology over the last 15 years.
What, then, has been learnt from this empirical work? On the issue of health care
financing, there is now a body of evidence showing the regressiveness of out-of-pocket
payments, especially in countries such as the US. Clear pictures have emerged concern-
ing the progressivity of other financing sources. Taxation tends to be a progressive way
of raising revenues, but the degree of progressivity depends on the mix between direct
and indirect taxes, and whether the direct taxes are general taxes or semi-earmarked
local taxes of the type used in Scandinavia. Social insurance emerges as progressive
in countries where the higher income groups are included in the scheme, and regres-
sive where they are not; in such countries, this is partially offset by a more progressive
private insurance structure than would otherwise be the case. Private insurance is regres-
sive in countries such as the US where the bulk of the population relies on it for cover.
The literature has also produced useful insights into the progressivity consequences of
financing reforms and changes: the reforms proposed for the US in the early 1990s
would have reduced the system's overall regressiveness; the increase in private insur-
ance and out-of-pocket payments in Australia between 1984 and 1989 moved the system
from progressive to broadly proportional; and the ill-fated Dekker proposals would have
reduced but not eliminated the regressiveness of the Dutch health care financing system.
Useful results have also been produced on the issue of horizontal equity in health care
finance. These suggest that although horizontal inequity does have an impact on the
distribution of income, the impact is very small compared to the effect of progressivity:
in cases where the source in question is progressive, horizontal equity reduces the pro-
poor redistributive effect, but not by much. The results also point to cross-source and
cross-country variation: horizontal inequity in social insurance is more pronounced in
absolute and relative terms in France and Germany than in the UK and US; horizontal
differences are more pronounced in the case of private payments than public payments,
reflecting in part the greater degree of voluntariness (some choose to privately insure,
whilst others choose not to), but also the randomness associated with ill-health which is
part of the cause of the large horizontal differences observed in out-of-pocket payments.
Studies testing for inequity in the delivery of health care (i.e. unequal treatment
for equal need) have varied somewhat in their model specifications and methods, but
nonetheless some tentative conclusions can be drawn. There is growing evidence that
in the US the distribution of health care by income is not consistent with health care
being allocated according to need: this emerges in studies just of the US, as well as in
comparative studies in which other countries (Canada and various of the EU countries)
typically emerge in a more favourable light. Studies that have sought to measure the
degree of inequity have reached much the same conclusion, though there is some ev-
idence that in some countries (including the US) pro-poor inequities in inpatient care
are compensating for pro-rich inequity in specialist and outpatient care. Interestingly,
this does not seem to be true of Australia where, overall, there appears to be substantial
pro-rich inequity in the delivery of health care. Interestingly, too, there does not appear
to be any straightforward link within the EU countries between the degree of inequity
overall and the features of the system (e.g. whether GPs have a gatekeeper role, whether
copayments are high, etc.).
The research by economists on health inequalities has examined both pure inequali-
ties and socioeconomic inequalities in health. Unsurprisingly, given that the rank corre-
lation between health and socioeconomic status is not unity, the results produce rather
different findings. The UK, for example, appears to have a relatively low level of pure
inequality in age-at-death but a relatively high level of income-related inequality in self-
assessed health. The US, by contrast, has high levels of both types of inequality. Work
on both areas has also shed light on the causes of cross-country variations in health
inequality; interestingly, in both exercises, income inequality emerges as an important
predictor of health inequality.
In all three empirical areas, then, important lessons have been learnt. There is clearly
scope for more work in all areas, especially work aimed at unravelling the causes of
inequity. We know a fair amount about the factors responsible for the cross-country
and cross-source variation in the progressivity of health care finance. We know much
less about the relative contributions of factors that give rise to differences in horizontal
equity in this area. Nor do we know much about the ultimate causes of inequity in
health care delivery - what the principle factors are within countries that prevent equals
being treated equally by the health care system, or what role system differences have in
accounting for cross-country variations in inequity. We also need to know more about
the economic causes of health inequalities and the factors accounting for cross-country
differences. Given the progress that has been made over the last 15 years, it would not
seem forlorn to hope that these issues will be much better understood by the time the
second edition of the Handbook of Health Economics is published.
References
Aday, L.A., R. Andersen and G.V. Fleming (1980), Health Care in the U.S.: Equitable for Whom? (Sage
Publications, Beverly Hills, CA).
Andersen, R. (1975), "Health service distribution and equity", in: R. Andersen, J. Kravits and O. Anderson,
eds., Equity in Health Services (Ballinger, Cambridge, MA).
Aronson, J.R., P. Johnson and P.J. Lambert (1994), "Redistributive effect and unequal tax treatment", Eco-
nomic Journal 104:262-270.
Atkinson, A.B. (1970), "On the measurement of inequality", Journal of Economic Theory 2:244-263.
Atkinson, A.B. (1980), "Horizontal equity and the distribution of tax burden", in: H. Aaron and M. Boskin,
eds., The Economics of Taxation (Brookings Institution, Washington, DC).
Baker, J.L., and J. van der Gaag (1993), "Equity in health care and health care financing: Evidence from five
developing countries", in: E. van Doorslaer, A. Wagstaff and F. Rutten, eds., Equity in the Finance and
Delivery of Health Care (Oxford University Press, Oxford).
Barry, B.M. (1989), Theories of Justice (University of California Press, Berkeley, CA).
Benham, L., and A. Benham (1975), "Utilisation of physician services across income groups 1963-1970", in:
R. Andersen, J. Kravits and O. Anderson, eds., Equity in Health Services (Ballinger, Cambridge, MA).
Bidani, B., and M. Ravallion (1997), "Decomposing social indicators using distributional data", Journal of
Econometrics 77(1):125-140.
Birch, S., J. Eyles and K.B. Newbold (1993), "Equitable access to health care: Methodological extensions to
the analysis of physician utilization in Canada", Health Economics 2(2):87-101.
Blaxter, M. (1989), "A comparison of measures of inequality in morbidity", in: J. Fox, ed., Health Inequalities
in European Countries (Gower, Aldershot).
Braybrooke, D. (1987), Meeting Needs (Princeton University Press, Princeton, NJ).
Cantor, J. (1988), The Burden of Financing Health Care in the United States (John Hopkins School of Hygiene
and Public Health, Baltimore).
Collins, E., and K. Klein (1980), "Equity and the NHS: Self reported morbidity, access and primary care",
Cowell, F.A. (1995), Measuring Inequality (Prentice Hall/Harvester Wheatsheaf, London, New York).
Cullis, J.G., and P.A. West (1979), The Economics of Health: An Introduction (Martin Robertson, Oxford).
Culyer, A.J. (1971), "Medical care and the economics of giving", Economica 151:295-303.
Culyer, A.J. (1976), Need and the National Health Service: Economics and Social Choice (Martin Robertson,
Oxford).
Culyer, A.J. (1980), The Political Economy of Social Policy (Martin Robertson, Oxford).
Culyer, A.J. (1989), "The normative economics of health care finance and privision", Oxford Review of
Economic Policy 5:34-58.
Culyer, A.J. (1990), "Commodities, characteristics of commodities, characteristics of people, utilities and the
quality of life", in: Baldwin, S., et al., eds., The Quality of Life: Perspectives and Policies (Routledge,
London).
Culyer, A.J. (1993), "Health, health expenditures, and equity", in: E. van Doorslaer, A. Wagstaff and E Rutten,
ed., Equity in the Finance and Delivery of Health Care (Oxford University Press, Oxford).
Culyer, A.J., A. Maynard and A. Williams (1981), "Alternative systems of health care provision: An essay
on motes and beans", in: M. Olson, eds., A New Approach to the Economics of Health Care (American
Enterprise Institute, Washington, DC).
1858 A. Wagstaffand E. van Doorslaer
Culyer, A.J., E. van Doorslaer and A. Wagstaff (1992a), "Access, utilisation and equity: A further comment",
Culyer, A.J., E. van Doorslaer and A. Wagstaff (1992b), "Utilization as a measure of equity by Mooney, Hall,
Donaldson and Gerard: Comment", Journal of Health Economics 11(1):93-98.
Culyer, A.J., and A. Wagstaff (1993), "Equity and equality in health and health care", Journal of Health
Economics 12(4):431-457.
Daniels, N. (1985), Just Health Care (Cambridge University Press, Cambridge).
Davis, K. (1993), "Equity in the distribution of health care: The British debate", in: E. van Doorslaer,
A. Wagstaff and F. Rutten, eds., Equity in the Finance and Delivery of Health Care (Oxford University
Press, Oxford).
Deolalikar, A.B. (1995). "Government health spending in Indonesia: Impacts on children in different eco-
nomic groups", in: D. van de Walle and K. Nead, eds., Public Spending and the Poor (Johns Hopkins
University Press, Baltimore, MD).
Donabedian, A. (1971), "Social responsibility for personal health services: An examination of basic values",
Inquiry 8:3-19.
Evans, R.G., M.L. Barer and T.R. Marmor, eds. (1994), Why are Some People Healthy and Others Not?
(Aldine de Gruyter, New York).
Gerdtham, U.-G. (1997), "Equity in health care utilization: Further tests based on hurdle models and Swedish
micro data", Health Economics 6(3):303-319.
Gerdtham, U.-G., M. Johannesson, L. Lundberg, et al. (1998), "A note on validating Wagstaff and van
Doorslaer's health measure in the analysis of inequalities in health", Journal of Health Economics (forth-
coming).
Gill, A.M. (1994), "Incorporating the causes of occupational differences in studies of racial wage differen-
cials", Journal of Human Resources 29(1):20-41.
Gillon, R. (1986), Philosophical Medical Ethics (Wiley, Chichester, New York).
Goodin, R.E., and J. Le Grand (1987), Not Only the Poor: The Middle Classes and the Welfare State (Allen
&Unwin, London, Boston).
Gottschalk, P., B. Wolfe and R. Haveman (1989), "Health care financing in the US, UK and The Nether-
lands: Distributional consequences", in: A. Chiancone and K. Messere, eds., Changes in Revenue Struc-
tures (Wayne State University Press, Detroit).
Grytten, J, G. Rongen and R. Sorensen (1995), "Can a public health care system achieve equity? The Nor-
wegian experience", Med. Care 33(9):938-951.
Hamilton, B.H., V. Ho and H.J. Paarsch (1997), "The distribution of outpatient services in Canada and the
US: An empirical model of physician visits", Working Paper (John M. Olin School of Business, St. Louis,
Washington University).
Hammer, J.S., I. Nabi and J. Cercone (1995), "Distributional effects of social sector expenditures in Malaysia,
1974 to 1989", in: D. van de Walle and K. Nead, eds., Public Spending and the Poor (Johns Hopkins
University Press, Baltimore, MD).
Hochman, H.M., and J.D. Rodgers (1969), "Pareto optimal redistribution", American Economic Review
59:542-557.
Holahan, J., and S. Zedlewski (1992), "Who pays for health care in the United States? Implications for health
system reform", Inquiry 29:231-248.
Humphries, K.H., and E. van Doorslaer (1998), "Income-related health inequality in Canada", Working Paper
HPRU 98:10D (University of British Columbia, Center for Health Services Policy and Research, Vancou-
ver).
Hurst, J.W. (1985), Financing health services in the United States, Canada and Britain, Nuffield/Leverhulme
Fellowship Report (King Edward's Hospital Fund for London, London).
Idler, E.L., and Y. Benyamini (1997), "Self-rated health and mortality: A review of twenty-seven community
studies", Journal of Health and Social Behavior 38(1):21-37.
Illsley, R., and J. Le Grand (1987), "The measurement of inequality in health", in: A. Williams, ed., Health
and Economics (Macmillan, London).
Janssen, R., E. van Doorslaer and A. Wagstaff (1994), "Health-insurance reform in the Netherlands: Assessing
the progressivity consequences", Economic and Social Review 25(4):303-320.
Joshi, H., and P. Paci (1998), Unequal pay for men and women: Evidence from the British birth cohort studies
(MIT Press, Cambridge, MA).
Kakwani, N.C. (1977), "Measurement of tax progressivity: An international comparison", Economic Journal
87(345):71-80.
Kakwani, N.C., A. Wagstaff and E. van Doorlsaer (1997), "Socioeconomic inequalities in health: Measure-
ment, computation and statistical inference", Journal of Econometrics 77(1):87-104.
King, M.A. (1983), "An index of inequality: With implications to horizontal equity and social mobility",
Econometrica 51:99-115.
Koskinen, S. (1985), 'Time trends in cause-specific mortality by occupational class in England and Wales",
IUSSP 20th General Conference, Florence.
Kunst, A.E. (1992), International Variation in Socioeconomic Inequalities in Self-Reported Health (Dutch
Central Bureau of Statistics, Voorburg).
Kunst, A.E., J.J. Geurts and J. van den Berg (1995), "International variation in socioeconomic inequalities in
self reported health", Journal of Epidemiology and Community Health 49(2): 117-123.
Lairson, D.R., P. Hindson and A. Hauquitz (1995), "Equity of health care in Australia", Social Science and
Medicine 41(4):475-482.
Lambert, P.J. (1993), The Distribution and Redistribution of Income: A Mathematical Analysis (Manchester
University Press, Manchester).
Lambert, P.J., and J.R. Aronson (1993), "Inequality decomposition analysis and the Gini coefficient revisited",
Economic Journal 103(420):1221-1227.
Le Grand, J. (1978), 'The distribution of public expenditure: The case of health care", Economica 45:125-
142.
Le Grand, J. (1982), The Strategy of Equality: Redistribution and the Social Services (Allen &Unwin, Lon-
don, Boston).
Le Grand, J. (1985), "Inequalities in health: The human capital approach", Welfare State Programme Discus-
sion Paper.
Le Grand, J. (1987), "Inequalities in health: Some international comparisons", European Economic Review
31(1/2):182-191.
Le Grand, J. (1989), "An international comparison of distributions of ages-at-death", in: J. Fox, ed., Health
Inequality in European Countries (Gower, Aldershot).
Le Grand, J. (1991), 'The distribution of health care revisited: A commentary", Journal of Health Economics
10(2):239-245.
Leclerc, A., E Lert and C. Fabien (1990), "Differential mortality: Some comparisons between England
and Wales, Finland and France, based on inequality measures", International Journal of Epidemiology
19(4):1001-1010.
Lerman, R.I., and S. Yitzhaki (1995), "Changing ranks and the inequality impacts of taxes and transfers",
National Tax Journal 48(1):45-59.
Leu, R.E., and R.L. Frey (1985), "Budget incidence, demographic change and health policy in Switzerland",
in: A.J. Culyer and G. Terny, eds., Public Finance and Social Policy (Wayne State University Press, Detroit).
Lindsay, C.M. (1969), "Medical care and the economics of sharing", Economica 144:351-362.
Lockwood, M. (1988), "Quality of life and resource allocation", in: M. Bell and S. Mendus, eds., Philosophy
and Medical Welfare (Cambridge University Press, Cambridge).
MacLachlan, G., and A. Maynard (1982), "The public/private mix in health care: The emerging lessons", in:
G. MacLachlan and A. Maynard, eds., The Public/Private Mix in Health Care: The Relevance and Effects
of Change (Nuffield Provincial Hospital Trust, London).
Manning, W.G., J.P. Newhouse and J.E. Ware Jr. (1982), "The status of health in demand estimation; or
beyond excellent, good, fair and poor", in: V. Fuchs, ed., Economic Aspects of Health (Chicago University
Press, Chicago).
Maynard, A., and A. Williams (1984), "Privitisation and the national health service", in: J. Le Grand and
R. Robinson, eds., Privitisation and the Welfare State (Allen &Unwin, London).
Miller, D. (1976), Social Justice (Clarendon Press, Oxford).
Mooney, G. (1983), "Equity in health care: Confronting the confusion", Effective Health Care 1:179-185.
Mooney, G. (1986), Economics, Medicine, and Health Care (Wheatsheaf, Brighton).
Mooney, G. (1994), Key Issues in Health Economics (Harvester Wheatsheaf, New York).
Mooney, G. (1996), "And now for vertical equity? Some concerns arizing from aboriginal health in Australia",
Mooney, G., J. Hall, C. Donaldson, et al. (1991), "Utilisation as a measure of equity: Weighing heat?", Journal
Mooney, G., J. Hall, C. Donaldson, et al. (1992), "Reweighing heat: Response", Journal of Health Economics
11(2):199-205.
Mooney, G., and A. McGuire (1987), "Distributive justice with special reference to geographical inequality
in health care", in: A. Williams, ed., Health and Economics (Macmillan, London).
Newhouse, J.P., and the Health Insurance Experiment Group (1993), Free for all? Lessons from the RAND
Health Insurance Experiment (Harvard University Press, Cambridge, MA).
Nozick, R. (1974), Anarchy, State, and Utopia (Basic Books, New York).
OECD (1992), The Reform of Health Care: A Comparative Analysis of Seven OECD Countries (OECD,
Paris).
Olson, E.O., and D.L. Rodgers (1991), "The welfare economics of equal access", Journal of Public Economics
45:91-106.
Pannarunothai, S., and A. Mills (1997), "The poor pay more: Health-related inequality in Thailand", Social
Science and Medicine 44(12):1781-1790.
Pappas, G., S. Queen, W. Hadden, et al. (1993), "The increasing disparity in mortality between socioeconomic
groups in the United States, 1960 and 1986", New England Journal of Medicine 329(2):103-109.
Plotnick, R. (1981), "A measure of horizontal inequity", Review of Economics and Statistics 63(2):283-288.
Preston, S.H. (1975), "The changing relationship between mortality and level of economic development",
Population Studies 29:231-248.
Preston, S.H., M.R. Haines and E. Pamuk (1981), "Effects of industrialisation and urbanisation on mortality
in developed countries", IUSSP 19th International Population Conference, Manilla (IUSSP).
Propper, C., and R. Upward (1992), "Need, equity and the NHS: The distribution of health care expenditure
1974-87", Fiscal Studies 13(2):1-21.
Propper, C. (1995), "For richer, for poorer, in sickness and in health: The lifetime distribution of NHS health
care", in: J. Falkingham and J. Hills, eds., The Dynamics of Welfare (Prentice Hall Wheatsheaf, London).
Puffer, F. (1986), "Access to primary care: A comparison of the US and the UK", Journal of Social Policy
(15):293-313.
Rasell, E., J. Bernstein and K. Tang (1994), "The impact of health care financing on family budgets", Inter-
national Journal of Health Services 24(4):691-714.
Rasell, E., and K. Tang (1994), "Paying for health care: Affordability and equity in proposals of health care
reform", Working Paper (Economic Policy Institute, Washington, DC).
Rawls, J. (1971), A Theory of Justice (Harvard University Press, Cambridge, MA).
Rice, T.H. (1997), "Can markets give us the health system we want?", Journal of Health Politics, Policy and
Law 22(2):283-426.
Rice, T.H. (1998), The Economics of Health Reconsidered (Health Administration Press, Chicago).
Rockefeller, J.D. (1991), "A call for action: The Pepper commission's blueprint for health care reform",
Rosenzweig, M.R., and T. Schultz (991), "Who receives medical-care: income, implicit prices, and the dis-
tribution of medical-services among pregnant-women in the United States", Journal of Human Resources
26(3):4-73-508.
Rothman, K. (1986), Modem Epidemiology (Little, Brown and Co., Boston, MA).
Rutten, E, and R. Janssen (1987), "Een economische beschouwing over geijkheid in de gezondheidszorg",
in: W.R.v.h. Regeringsbeleid, ed., De ongelijke verdeling van gezondheid (Staatsuitgeverij, Den Haag).
Sen, A.K. (1992), Inequality Reexamined (Clarendon Press, New York).
Sugden, R. (1983), "Who cares?", IEA Occasional Paper No. 67 (Institute for Economic Affairs, London).
Suits, D. (1977), "Measurement of tax progressivity", American Economic Review 67:747-752.
Tobin, J. (1970), "On limiting the domain of inequality", Journal of Law and Economics 13:263-278.
Townsend, P., and N. Davidson (1982), Inequalities in Health: The Black Report (Penguin, Harmondsworth).
UNDP (1993), Human Development Report 1993 (Oxford University Press, New York).
van de Walle, D. (1995), "The distribution of subsidies through public health services in Indonesia, 1978-
87", in: D. van de Walle and K. Nead, eds., Public Spending and the Poor (Johns Hopkins University Press,
Baltimore, MD).
van Doorslaer, E., A. Wagstaff, H. Bleichrodt, et al. (1997), "Income-related inequalities in health: Some
international comparisons", Journal of Health Economics 16:93-112.
van Doorslaer, E., A. Wagstaff, S. Calonge, et al. (1992), "Equity in the Delivery of Health Care: Some
International Comparisons", Journal of Health Economics 11(4):389-411.
van Doorslaer, E., A. Wagstaff, H. van der Burg, et al. (2000), "Equity in the delivery of health care: Further
international comparisons", Journal of Health Economics, forthcoming.
van Doorslaer, E., A. Wagstaff, H. van der Burg, et al. (1999), "The Redistributive effect of health care finance
in twelve OECD countries", Journal of Health Economics 18(3):291-3 14.
van Doorslaer, E.K.A., A. Wagstaff and EEH. Rutten, eds. (1993), "Equity in the finance and delivery of
health care", An international perspective Commission of the European Communities health services re-
search series, No. 8 (Oxford University Press, Oxford).
von der Schulenburg, J.M. (1994), "Forming and reforming the market for third-party purchasing of health
care: a German perspective", Soc. Sci. Med. 39(10):1473-1481.
Wagstaff, A., P. Paci and E. van Doorslaer (1991a), "On the measurement of inequalities in health", Soc. Sci.
Med. 33(5):545-557.
Wagstaff, A. and E. van Doorslaer (1993), "Equity in the finance and delivery of health care: Concepts and
definitions", in: E. van Doorslaer, A. Wagstaff and F Rutten, eds., Equity in the Finance and Delivery of
Health Care: An International Perspective (OUP, Oxford).
Wagstaff, A., and E. van Doorslaer (1994), "Measuring inequalities in health in the presence of multiple-
category morbidity indicators", Health Economics 3:281-291.
Wagstaff, A., and E. van Doorslaer (1997), "Progressivity, horizontal equity and reranking in health care
finance: a decomposition analysis for The Netherlands", Journal of Health Economics 16:499-516.
Wagstaff, A., and E. van Doorslaer (2000), "Measuring and testing for inequity in the delivery of health care",
Journal of Human Resources, forthcoming.
Wagstaff, A., E. van Doorslaer, S. Calonge, et al. (1992), "Equity in the finance of health care: Some interna-
tional comparisons", Journal of Health Economics 11(4):361-387.
Wagstaff, A., E. van Doorslaer and P. Paci (1989), "Equity in the finance and delivery of health care: Some
tentative cross-country comparisons", Oxford Review of Economic Policy 5(1):89-112.
Wagstaff, A., E. van Doorslaer and P. Paci (199 l1b), "On the measurement of horizontal inequity in the delivery
of health care", Journal of Health Economics 10(2):169-205.
Wagstaff, A., E. van Doorslaer and H. van der Burg, et al. (1999), "Equity in the finance of health care: Some
further international comparisons", Journal of Health Economics 18(3):263-290.
Wetenschappelijke Raad voor het Regeringsbeleid (1987), De ongelijke verdeling van gezondheid (Staatsuit-
geverij, Den Haag).
Wiggins, D. (1987), Needs, Values, Truth: Essays in the Philosophy of Value (Basil Blackwell, Oxford, UK;
New York, NY, USA).
Williams, A.H. (1974), "'Need' as a demand concept (with special reference to health)", in: A.J. Culyer, ed.,
Economic Policies and Social Goals: Aspects of Public Choice (Martin Robertson, London).
Williams, A.H. (1976), "Cost-benefit analyses in public health and medical care: Comments on a thesis written
by Bengt Jonsson" (Department of Economics, University of Lund, Lund).
Williams, A.H. (1978), "'Need' - an economic exegesis". in: A.J. Culyer and K.G. Wright, eds., Economic
Aspects of Health Services (Martin Robertson, London).
Williams, A.H. (1981), "Welfare economics and health status measurement", in: J. van der Gaag and M. Perl-
man, eds., Health, Economics, and Health Economics (North-Holland, Amsterdam).
Williams, A.H. (1988), "Priority setting in public and private health care: A guide through the ideological
jungle", Journal of Health Economics 7(2):173-183.
Williams, A.H. (1993), "Equity in health care: The role of ideology", in: E. van Doorslaer, A. Wagstaff and
F. Rutten, eds., Equity in the Finance and Delivery of Health Care (Oxford University Press, Oxford).
Williams, A., and R. Cookson (2000), "Equity in health", in: A.J. Culyer and J.P. Newhouse, eds., Handbook
Chapter 35
EQUITY IN HEALTH*
ALAN WILLIAMS and RICHARD COOKSON

Centrefor Health Economics, University of York, York YOO15DD, England
Contents
Abstract 1864
Keywords 1864
1. Scene-setting 1865
2. Philosophy and economics - an analytical preamble 1866
3. Theories which leave the opportunity set ethically unconstrained 1871
3.1. Assumed general properties of the opportunity set 1871
3.2. Theories without side conditions 1871
3.2.1. No side conditions: no maximand 1871
3.2.2. No side conditions: linear maximand: weights equal 1871
3.2.3. No side conditions: linear maximand: unequal weights 1876
3.2.4. No side conditions: non-linear and smooth maximand: equal weights 1876
3.2.5. No side conditions: non-linear and smooth maximand: unequal weights 1878
3.2.6. No side conditions: non-linear and kinked maximand: equal weights 1880
3.2.7. No side conditions: non-linear and kinked maximand: unequal weights 1881
3.3. Theories with side conditions 1884
3.3.1. One side condition: no maximand 1884
3.3.2. One side condition: linear maximand: equal weights 1884
3.3.3. One side condition: other permutations 1885
3.4. The opportunity set once more 1885
3.5. Needs-based theories 1887
4. Theories which ethically constrain the opportunity set 1889
4.1. Introductory comments 1889
4.2. Theories with no maximand 1890
4.2.1. Libertarianism 1890
4.2.2. Participatory democracy 1891
*This chapter has benefited from the comments of various colleagues, to whom we are most grateful for
the time they have spent reading and thinking about what we had written at draft stage. For this help we
would like to thank the following: Gwyn Bevan, James Buchanan, Tony Culyer, Diane Dawson, Paul Dolan,
Lucy Gilson, Julian LeGrand, Susan Mendus, Richard Musgrave, Peter Singer, Michael Spackman, and Aki
Tsuchiya. A special additional word of thanks goes to Vanda Castle, who coped stoically with the frequent
changes we made to her painstakingly constructed diagrams.
Handbookof Health Economics, Volume 1, Edited by A.J. Culyer and J.P Newvhouse
© 2000Elsevier Science B. V All rights reserved
1864 A. Williams and R. Cookson
4.2.3. Contractarianism 1892

4.2.4. No-envy principles and equality of resources 1892
4.2.5. Equality of access to health care 1893
4.2.6. Medical ethics and the rule of rescue 1894
4.2.7. General comments on these six theories 1895
4.3. Theories with a maximand 1895
4.3.1. Equality of opportunity for health using financial means only 1895
4.3.2. Rawlsian theory 896
4.4. Fitting population change into the framework 1898
5. How is health to be measured? 1899
6. The economist's role 1899
7. Estimating equity-efficiency trade-offs in health 1901
8. Further considerations 1906
References 1907
Abstract
Equity in health has to be distinguished from equity in access to health care, or equity
in the distribution of health care resources. We take as a working definition of health
for our purposes the number of quality adjusted life years that a person may expect to
enjoy over his or her lifetime. Although we mostly follow the economists' custom of re-
garding equity as synonymous with reducing inequalities in health, we also consider the
much richer variety of concepts employed by philosophers when discussing distributive
justice. Here however we have distinguished notions of justice which are essentially
procedural from those which are substantive, concentrating mainly on the latter. What
we have sought to do is to identify the implications of various philosophical theories
of justice for the way in which a welfare economist might appraise a particular distri-
bution of health within a community. To do this we distinguish theories which place
constraints on admissible outcomes (the health opportunity set), from theories which
require the social welfare function (or maximand) to have particular properties. This
classification is summarised in the Table 1, which is the key exhibit around which the
analysis and exposition is organised.
Keywords
equity, justice, inequality, other normative criteria and measurement, relationship of

economics to social values, economic methodology, distribution, publicly provided
goods, government expenditures and health, government policy, regulation, public
health
JEL classification:A13, B41, D30, D63, H40, H51, 118

Ch. 35: Equity in Health 1865
1. Scene-setting
In this Chapter our concern is health, and not health care. We view health care as one
of many possible determinants of health and not necessarily the pre-eminent one. We
therefore regard equity in access to health care, or in the distribution of health care re-
sources, as an instrumental matter, flowing from a more fundamental concern with the
distribution of health itself. We shall assume for the time being that health can be mea-
sured unproblematically, and that the unit in which it is measured is the quality-adjusted
life year or QALY. 1 We shall restrict our attention to various ways in which an equitable
distribution of health within a population might be defined in an operational manner
that enables economists to carry out empirical work evaluating possible ways in which
it might be brought about. We will assume that in order to achieve a more equitable
distribution of health in a population it will generally be necessary to reduce the overall
level of health in that population (in other words, we assume that an efficiency sacrifice
will usually be required to achieve equity). 2 We will not be concerned with equity in the
financing of health care (or in the financing of any other public policy designed to bring
about a more equitable distribution of health). 3 Nor will we be concerned here with the
use of financial mechanisms explicitly to promote equity in health, important though
these may be as instruments as policy. Our purpose is rather to clarify the equity objec-
tive itself. For convenience we will assume that the resources the community wishes to
be used for the pursuit of health have been set aside in an equitable manner, and our task
is to ensure that there are clear criteria for determining how best to use them in pursuit
of an equitable distribution of health within that community. 4 The term "community" is
used to encompass all those to whom the social decision-makers are responsible.
In economics the term "equity" is usually taken to refer to fairness in the distribution
of a good (in this case "health"), and "fairness" is taken almost unthinkingly to mean
reducing inequalities. In philosophical writings equity concerns would more likely be
broader than this, and called concerns about "distributive justice". Concerns about dis-
tributivejustice often become intertwined with concerns about procedural justice, a mat-
ter which we shall examine more closely shortly. 5 Philosophical writings tend to focus
on what is fair as between individuals with known characteristics. Economists, on the
other hand, tend to focus on what is fair as between groups of individuals distinguished
See Dolan (2000) for a general discussion of measurement and valuation issues. We shall return to some
of them later in the specific context of equity weights.
2 If this is not the case, the simultaneous pursuit of efficiency and equity is made that much simpler.
3 This is taken up in Wagstaff and van Doorslaer (2000).
4 It is possible that the social decision-makers will need to take a broader view of equity when deciding
on how much of the community's resources to set aside for health than they need to take into account when
focusing on how best to use those resources for health itself (and only for health).
5 From the standpoint of procedural justice, a distribution of health within a community would be said to
be just if it were the result of processes that were just. It would be the processes that would be the object of
attention, and not the substantive outcome. The position taken here is that it is the outcome that is the focus
of attention, and that the processes are instrumental.
only by some common characteristics, accepting that the other characteristics of each
individual will be ignored even though they may differ widely. This calls for the exer-
cise of moral sensitivity about "statistical lives" rather than about the lives of named
individuals whom we can see and touch and talk to. For many people this notion of
"statistical compassion" seems to create both intellectual and psychological difficulties.
It is as if personal empathy with one or two individuals is possible but, paradoxically,
if many individuals are involved this capacity to empathise diminishes. This difference
between focusing on groups and focusing on individuals also distinguishes economists
(and managers) from clinicians and others dealing with people at an individual level.
The latter often claim that they are under an ethical duty to do everything possible for
the person in front of them no matter what the consequences might be for everybody
else. If this assertion is taken at its face value, it would imply that clinicians should ig-
nore their responsibilities for the welfare of their other patients except when that patient
is in front of them. It seems most unlikely that any clinician would actually behave in
that way, so perhaps the statement should not be taken at its face value, but regarded
instead as part of the rhetoric of medical practice, designed to bolster the doctor-patient
relationship. But whatever may be the role of such statements, it is clear that in a public
policy context, where distributive justice is an explicit objective, it is clearly not ethical
for a clinician to ignore the consequences of his or her actions concerning the treatment
of one patient for the health of other patients for whom the system is also responsible.
The exposition which follows is written primarily for economists with little or no
familiarity with the relevant philosophical literature. In order to make this literature
more readily accessible, we have selected the main points made by each of the cited
authors and, in the interests of clarity we have stripped away the many qualifications
and elaborations that are contained in the original works. These original sources need
to be consulted carefully before claiming a proper appreciation of an author's position.
We have also judged it not to be necessary to present detailed practical examples to
illustrate each case (which is a common mode of exposition in medical ethics), but have
contented ourselves with indicating the kinds of issue addressed by each philosophical
theory, leaving the reader to think through the practical implications for particular cases.
What we are attempting here is the brutal task of forcing high-minded philosophical
theories about distributive justice into the procrustean bed of welfare economics! That
painful process will commence with a brief explanation of how we propose to delineate
the economist's conventional conceptual framework for that purpose.
2. Philosophy and economics - an analytical preamble
In the last twenty years, social science journals have published upwards of one thou-
sand articles on the subject of equity in health (the exact number depending on how
widely or narrowly you define that elusive concept). In contrast, the humanities journals
contain almost nothing on that topic (or indeed on health more generally). On the rare
occasions when philosophers write anything about health, they tend to do so either in
social science journals or in books. The voluminous social science literature on equity
in health really started to take off in the 1990s. Although the emergence of the World
Bank work on DALY's [World Bank (1993), Murray and Lopez (1996)] was a signifi-
cant element in this trend, it was part of a more general increase in attention to equity
right across the social sciences. A notable feature was the increasing use of the word
"fairness", which failed to get a mention in connection with health during the 1980s,
but logged around 10 mentions a year during the 1990s. By 1997, the most commonly
cited philosophical work in articles published in social science journals about equity in
health was Rawls' (1971) which did not in fact include health as one of the primary
goods which fairness required to be distributed equally. To understand why economists
may have had difficulty incorporating philosophical ideas into their thinking, we must
first consider carefully what their characteristic mode of thinking is.
Economists typically approach optimisation problems by listing the options to be
considered (the "opportunity set") and then choosing between them by applying some
maximand (the "objective function"). The opportunity set may be presented as a pro-
duction possibility frontier, or as a utility possibility frontier. If it is presented as a
production possibilityfrontier, this is the particular set of options (defined as a pack-
age of goods and services) that are not "dominated" in the optimisation process by any
other option. On and within the frontier, the complete set of options comprises those
satisfying two conditions: (i) they must be technically feasible 6 and (ii) they must be
producible with the resources available. The resources available can either be repre-
sented by a budget constraint (leaving flexible the actual input combinations), or by a
fixed allocation of (unpriced) real resources. If, on the other hand, the opportunity set is
presented as a utility possibilityfrontier, the same conditions apply, but the analysis has
already proceeded one step further and distributed the goods and services to individuals
in different ways so that the options presented are the various interpersonal distributions
of utility that result. To analyse the case of equity in health, the analogy with utility pos-
sibilities is more appropriate than with production possibilities, since health, like utility,
can only subsist in individuals. We shall be presenting the opportunity set as a health
possibility frontier which focuses on the interpersonal distribution of health outcomes
and assumes it to be measured in terms that are analogous to a utility measure (QALYs).
We shall use this separation of the health possibility set from the social maximand
in order to classify and analyse the implications of different theories of justice for the
interpersonal distribution of health. 7 We shall argue that some of them (notably those
that stress procedural requirements or the primacy of goods other than health) are best
seen as restricting, on ethical grounds, eligibility for inclusion in the health possibility
set. Others impose special restrictions on the nature of the maximand. Some do both.
6 Technical feasibility is to be understood broadly as any constraint on what can be achieved with given
resources, including incentive constraints on individual behaviour (such as those arising from asymmetry of
information), as well as constraints relating to the state of technology, more narrowly understood.
7 Mishan (1977) uses a similar approach, and generates a diagram that is very similar to one of those that
we shall be using later.
As was briefly explained above, the conventional definition of the opportunity set is
that in principle it contains every option that is both technically feasible and producible
within the resource constraints. But the resource constraints themselves will typically
not simply be those set by nature but also be the result of some human decision or de-
cisions. As such they are likely to reflect the judgements of managerial, professional,
technical and political actors concerning what it is wise or reasonable to devote to the
objectives in question. Similarly, some things which are technically feasible may be
ruled out from further consideration by one or more of these actors because they are
not considered sensible or politically feasible within their own particular sphere of au-
thority and expertise. Consequently, in practical terms the opportunity set presented to
economists engaged in the evaluation of health policy options may already have been
severely truncated by such additional restrictions and, during the process of problem for-
mulation it is important for such investigators to explore how these particular options
(and not others) come to be the ones to be investigated.
Here it will be assumed that however the health possibility set was defined, ethical
constraints concerning the requirements for equity in health were not part of the process.
We shall regard any ethical constraints emerging from theories of justice as additional
factors to be taken into account in defining an opportunity set which had previously
been innocent of any such considerations. It will furthermore be assumed that such
ethical constraints never enlarge the opportunity set.8 They may leave it untouched, if
all the previously considered options happened to fulfill the ethical constraints anyway.
They may delete options which were not on the frontier and which consequently would
not have been considered in any case. They may also delete options on the frontier that
would not have been chosen, in which case again they are irrelevant. In general we shall
assume that these ethical constraints do "bite", in that they remove from consideration
options that would otherwise have been chosen. Otherwise we would be treating them
as a prioriirrelevant.
Theories which restrict the nature of the maximand fall into two groups. The first
group establishes side conditions upon outcomes which have to be fulfilled before any
maximisation process comes into play. For instance, a side condition might be that there
must be some minimum level of health provided for some specified group and no trade-
off is permitted between this objective and any other. Once the minimum has been
provided, however, the maximand has unrestricted applicability (unless there is a whole
sequence of such side conditions, each of which has to be satisfied in the prescribed
order before we get to the residual set to which the maximand is applied). The second
8 It might be argued that, paradoxically, ethical constraints can actually enlarge the health possibility set by
facilitating trust and co-operation (e.g. between doctors and patients). For instance, an ethical constraint that
removed the fear that a certain category of hospitalised patient might be allowed to die, when they would
prefer to go on living, simply in order that organs which they had offered to donate might be "harvested"
sooner, might make people more willing to donate organs, with consequent additional benefits to others.
However, we shall regard such instances as exceptional, and generally regard procedural rules as restricting
rather than enlarging the opportunity set.
group of restrictions upon the nature of the maximand concerns its actual content and
the weights to be attached to its various elements. These differ from the previous case
in that trade-offs are permitted between all of the various ethical desiderata.
In order to deal systematically with these complexities, it will be useful to refer to the
accompanying Table 1, the upper half of which covers those cases where the opportunity
set is unaffected by the particular notion of equity under consideration and the lower
half of which covers those cases where it is (including those which affect both the
opportunity set and the maximand, such as the Rawlsian equity notions). In each half
of Table 1 there is a further subdivision according to whether the theory imposes side
conditions upon outcomes. Where equity is held to require that more than one such
condition has to be satisfied, there are two possibilities: one is that they are to be satisfied
in a prescribed order; the other is that they are all to be satisfied simultaneously. In the
former case, we first have to ensure (say) that Group A achieves some minimum level
of health and, when that has been done, we move on to Group B and so on. In the
other case, where all side conditions have to be satisfied simultaneously, there is no
such priority ranking between groups and all side conditions have equal salience. In
the remainder of the Table it is the particular characteristics of the optimisation criteria
are the focus of interest. Under "Nature of Maximand" the first column covers the case
where there is no maximand. The remaining columns focus on the shape of the social
welfare contours for health. Essentially these may be linear or non-linear and, if non-
linear they may be smooth or kinked. Within each of these three subgroups, equal weight
may be given to each class of person or unequal weights may be assigned to them.
These are the distinctions that we have found to be important discriminators between
the more popular notions of distributive justice. 9 In the analysis which follows, we shall
go through the cells in the Table one by one, starting in the top left-hand corner, and
proceeding from left to right, one row at a time. Where a cell is empty, we shall com-
ment on what it might have contained had such a notion of justice been propounded
by anybody. The main purpose, however, is to link each notion of equity to a particular
analytical device. This will be done by first outlining the theory of justice from which
the notion derives and then showing how it could be represented in welfare economics.
For ease of reference each subsection in the text has a summary heading identifying the
particular cell whose contents are there being discussed. Further reading on particular
theories of justice is suggested in the relevant sections; for a review of economic think-
ing about equity in health see Culyer and Wagstaff (1993); for a review of economic
thinking about equity in general see Hausman and McPherson (1996); for a review of
philosophical thinking about equity in health see Daniels (1985); for a review of philo-
sophical thinking about equity in general see Plant (1991).
9 Although we have considered each theory as a separate entity, it is quite likely that an individual citizen
will subscribe to more than one of them simultaneously. In such a case there will arise the need to establish
equity-equity trade-offs as well as equity-efficiency trade-offs, but this is a complication which we have left
to the imagination of the reader.
oa
t ;i
0
X7 a
0H tO=c : = ffi S
o o
z s<E ~WdE
a a
~~~O ~~~~a
aa
0 u
aE
aa
-aj a
a a
a ~a
~0) .
a.-
a a r.
aE
z
-a"-
~ a1 ao
5
a C, M a-0i
a
1,
E-
~a
au-
-a W a
.iXM
a
a
a
3s
a -
'a
n~aau
.- -i
a-
uC
a- - CL
a a a
oa
3a a
V00aa Z O Z O
"a
0 a
an a aa
'-- GO-1
I z t
3. Theories which leave the opportunity set ethically unconstrained
3.1. Assumed generalproperties of the opportunity set
Before embarking upon that task, however, something needs to be said about the gen-
eral properties of the health possibility set. Generally we shall assume this to have the
properties normally assumed for an opportunity set in welfare economics, namely that
it generates a frontier which is continuous, concave to the origin and monotonically de-
creasing from left to right. This means that there is a densely packed set of possibilities
for generating health for both A and B (in a two person world) but that on the frontier
the health of the one can be improved only by sacrificing some of the health of the
other. Later we shall consider a situation in which this is not the case, and point out its
consequences.
In each of the various figures to be used below, on the horizontal axis is measured the
Health of A, and on the vertical axis the Health of B.1 0 For the moment we assume that
this is measured in QALYs over some appropriate time horizon, using some appropriate
measure of central tendency, and can be treated as cardinally measurable and interper-
sonally comparable. Whatever unit is chosen, it is used uniformly on both axes. The
line FF represents the frontier of the health possibility set, representing those feasible
combinations of health for A and B which are under consideration at some point in time.
The purpose is to decide which is the optimal outcome according to each philosophical
position. In each case the optimum will be denoted by X.
3.2. Theories without side conditions
3.2.1. No side conditions: no maximand
For completeness, this cell accommodates radical nihilism or moral scepticism, which
claims that it is not possible to make sensible judgements about health outcomes being
fair or just. Whatever happens happens and we cannot judge whether it is good or bad
or right or wrong.
3.2.2. No side conditions:linearmaximand: weights equal
From a utilitarian standpoint, justice is ultimately a matter of maximising the sum total
of human happiness.1 1 This principle embodies a kind of equality since it gives equal
10 A and B may be any subgroups within the community whose relative health status is an object of policy
concern. Thus, depending on the context, it may be rich and poor, or men and women, or black and white, or
smokers and non-smokers, or northerners and southerners, or urban and rural, or those alive today and unborn
generations. At the level of principle the issues concerning the social welfare function will be the same. But
the health possibility frontier is likely to be very different according to which sub-groups are on the respective
axes.
11 Perhaps the best introduction to utilitarianism is Smart and Williams (1973). An excellent set of critical
readings is Sen and Williams (1982).
weight to each individual's happiness, in line with the Bentham's famous slogan: "each
to count for one, none for more than one". Although there are many different brands of
utilitarianism, we can identify three common features. First, consequentialism: things
must be evaluated in terms of their consequences. Second, welfarism: consequences
must be evaluated in terms of the welfare or utility of individual human beings. Third,
sum-ranking: the overall evaluation must be based on the sum total of individual utili-
ties in the relevant population. This latter principle means that the distribution of utili-
ties between people is of no intrinsic concern; it also means that utilities must be car-
dinally measurable and interpersonally comparable. 12 Different brands of utilitarianism
can then be distinguished, according, among other things, to how "utility" is understood
(e.g. pleasures and pains versus desires or preferences) and how the unit of evaluation
is defined (e.g. particular acts versus general rules or motives for acting).
Despite their lack of intrinsic concern with the interpersonal distribution of utilities,
the eighteenth- and nineteenth-century fathers of utilitarianism, Bentham and Mill, were
in fact passionate advocates of radical social reforms which would redistribute income,
health care and other utility-yielding goods from rich to poor. This is because utilitarian-
ism yields a clear case for redistribution of a good if one assumes diminishing marginal
utility of that good, and if utility declines in the same way for everybody. Redistribution
is then bound to be a good thing on a utilitarian calculus, since the gain in happiness to
the poor from one more unit of the good will be greater than the loss in happiness to the
rich from one less unit of the good. 13
It is important to distinguish this classical utilitarian argument for equality of health
(i.e. that greater equality of health generally increases the sum total of utility) from more
fundamental egalitarian principles which assume that inequality in health is intrinsically
bad. The classical utilitarian argument concerns the relationship between health and
utility: it assumes that utility is diminishing in health.1 4 By contrast, more fundamental
egalitarian principles focus attention on how individual utilities are aggregated to yield
social welfare. Rather than simply adding utilities up, as classical utilitarianism does,
these more fundamental egalitarian principles adopt a social welfare function which
gives greater weight to individuals with lower utility.
Despite its noble historical pedigree as a philosophy of social reform, utilitarianism
has been subjected to a great deal of criticism in the philosophical literature in recent
12 Conventional Paretian welfare economics, because it works with ordinal utilities, rejects sum-ranking.
Kenneth Arrow (1973) has called this position "ordinal utilitarianism".
13 Utilitarians also have a second general argument for equality, based on what Hare has called the "disutility
of envy" [Hare (1993)]. This is the idea that individual utility may depend on the degree of inequality in
society, both due to purely altruistic concerns for one's fellow human beings, and to concerns about the crime
and social disruption that may accompany gross inequalities.
14 It is more conventional, and perhaps more plausible, to assume diminishing marginal utility of commodi-
ties (such as consumer goods, or health care) rather than diminishing marginal utility of health. Furthermore,
it could be argued that our QALY measure should be an all-inclusive utility measure, which already takes ac-
count of any diminishing marginal utility of health [see Dolan (2000) for further discussion of this "welfarist"
argument].
Health of EI
F Health of A
Figure 1. Simple maximisation; linear contours; equal weight for A and B.
decades [Hampshire (1978)]. Each of the three key features of utilitarianism mentioned
above has been criticised, with perhaps the fiercest criticisms being directed against con-
sequentialism, on the grounds that the ends (i.e. the consequences) do not always justify
the means. 15 However, much of this criticism has been directed against utilitarianism
as a system of personal morality, rather than a way of evaluating public policies and
institutions. It has been argued, for instance, that utilitarianism recommends that one
should become a cold and calculating person, who lacks integrity and dignity, breaks
ties of personal affection to family and friends and violates other common-sense ethical
principles, in the pursuit of impersonal humanitarian ends [see, in particular, the essay
by Williams in Smart and Williams (1973)]. 16 Yet while a close attachment to family,
friends and personal projects may be virtues of personal morality, it seems reasonable
to argue that matters of public policy properly require a more detached and impartial
perspective which eschews personal favouritism of this kind [Goodin (1995)].
15 One of the most influential critics of consequentialism was Isaiah Berlin. His famous essay "two concepts
of liberty" [Berlin (1958)] contains, among other things, a powerful critique of the idea that there must be a
single true answer to moral questions about how good or bad particular consequences are for people and for
society.
16 Utilitarians have responded to this charge by claiming that the utilitarian calculus can operate at the level
of general rules for behaviour rather than particular acts. It may then turn out that the utility-maximising rules
for behaviour are ones which respect common-sense ethical codes (to keep promises, to honour thy father and
mother, and so on).
Health of B) (Pareto)
\. N
t \Y
\( AX
F Health of A
Figure A. Simple maximisation; linear contours; equal weight for A and B but with a Paretian restriction.
The conventional welfare economics approach is very close to the utilitarian position,
being a simple maximising one as exemplified in Figure 1. Here a unit of health is re-
garded of equal social value no matter who gets it, which is denoted by welfare contours
which are at 450 to each axis, implying that only the total amount of health is important
and its distribution between A and B is of no public policy interest. If the additional
assumption is made that there is diminishing marginal utility from health, and that it
diminishes in the same way for everybody, then we move on to Figure 3, which will be
discussed later when we come to the relevant cell in the Table 1.
This initial depiction actually departs from the currently conventional economic ap-
proach in two respects. Normally, what would be on the axes would be the utility en-
joyed by each individual (bringing it even closer to the utilitarian position), not the
health enjoyed by each individual. For the time being, however, we are restricting our
attention to health and health alone as a utility-generator for each individual, treating
a unit of health as if it were a unit of utility. The second respect in which Figure 1
differs from what modern economists normally do is that, in order to fulfill the Pare-
tian restriction on the admissibility of statements about improvements in social welfare,
the starting position becomes relevant. The reason for this is that a Paretian would not
accept that in Figure 1 the Health of A and the Health of B could be cardinally measur-
able and interpersonally comparable. To avoid these rather strong requirements, Pareto
suggested that, when making judgements about the relative desirability of different so-
cial states, we should restrict ourselves to situations in which all that would be required
would be for each person to say whether they were better off or worse off (according to
whatever criteria they each separately chose to apply). Only those situations in which
Health of B
(
F Health of A
Figure 2. Simple maximisation; linear contours; greater weight for A than B.
no-one was worse off and at least one person was better off should then be regarded as
clear social improvements. The effect of this restriction is illustrated in Figure 1A, in
which Z is the starting point, assumed to lie within the frontier. Pareto restricts admis-
sible judgements about improvements in welfare to the quadrant lying to the north east
of Z, in which Y would be the best attainable situation. In utilitarian terms, X is better
still but a Paretian can say nothing about a move from Z to X because B is worse off
at X than at Z. For a Paretian to break out of this impasse, some way would have to
be found for A to compensate B so that B would no longer be worse off than at Z and
A would still have some net gain left. The application of this "compensation principle"
is not explored further here, because direct transfers of health between individuals are
problematic, and the matter is better considered in a broader institutional context than
that which we are currently supposing.
3.2.3. No side conditions: linearmaximand: unequal weights
This is a variant of the utilitarian position in which a unit of health has a different social
value according to who gets it, while the weights are invariant with the amount of health
a person has. It represents a situation in which it might be argued that a particularly
deserving class of people (e.g. war heroes) should always have priority over others
when there are ways of improving people's health, no matter how much health they
already have. The analytical representation of this notion simply requires the (straight-
line) contours in Figure 1 to be at an angle greater than 45 ° to the axis of the more
favoured group. This is the situation shown in Figure 2, where A is held to be the more
deserving group.
3.2.4. No side conditions: non-linearand smooth maximand: equal weights
The egalitarian goal of reducing inequalities in health has powerful appeal, and has
motivated a large empirical literature on health inequalities (especially between social
classes in developed countries and between men and women in developing countries).
As indicated above, it may derive from classical utilitarianism combined with an as-
sumption of diminishing marginal utility of health. Alternatively it may derive from
more fundamental objections to inequalities in health, irrespective of any effect on the
sum total of health or utility. It seems unreasonable to insist on strict equality in health,
however, since this might require a levelling-down in everyone's health towards that of
the most unhealthy. So theories in this cell all accept a trade-off between equality and
efficiency - the degree of curvature of the maximand reflecting the degree of aversion
to inequality. 7
One such theory is the so-called "fair innings" argument, according to which there
is some length of life (e.g. "three-score-years-and-ten") which can be regarded as an
ethical entitlement. Those who get less than this are entitled to feel unfairly treated,
whereas those who get more than this have no cause to complain on equity grounds
when they eventually die. In the present context this argument needs reformulating in
terms of quality-adjusted life expectancy rather than just life expectancy, since a life
spent disabled and in pain cannot be held to be equal to one of the same length spent in
good health [Williams (1997)]. This fair innings argument would thus lead to priority
being given to the young over the old on equity grounds. This goes beyond the implicit
form of "utilitarian ageism" which follows from the fact that the elderly will, generally
speaking, have lower capacity to benefit from a given health intervention than the young.
The main criticism that has been levelled against the fair innings argument is that
this strong form of "ageism" is incompatible with the duty of care that a civilised soci-
ety owes to its elderly population. This objection is somewhat weakened, however, by
17 At one polar extreme, we can think of zero aversion to inequality as reflecting the straight line maximand
employed by classical utilitarianism. At the other polar extreme, maximum aversion to inequality would give
rise to a kinked L-shaped maximand, of a kind to be discussed shortly.
the fact that the principle operates alongside a continuing desire to improve everyone's
health, and allows for a variable degree of aversion to inequality according to the mores
of the society. The uncomfortable conclusion, however, is that the more strongly one
wishes to reduce inequalities (e.g. between social classes) in people's lifetime experi-
ence of health, the more one will have to discriminate against the elderly and in favour
of the young. This presents a further moral dilemma for all concerned.
A second theory that can be interpreted as falling within this cell is Sen's theory of
capability [Sen (1980, 1993)]. On this theory, justice involves taking note of individual
advantage as reflected by an objectively measurable index of people's capability to do
valuable acts (e.g. to work, to play sport) and to reach valuable states of being (e.g. self-
respect, good health). Aggregative considerations are assessed in the space of capability,
as well as distributive concerns and, insofar as equality is involved as one of the compet-
ing objectives of justice, the concern is with equalising - or moving towards equalising
- people's advantages as measured by their respective capabilities. What goes on to the
list of valuable functionings, and how the various weights are to be derived, are matters
which Sen deliberately leaves open, to be determined by "reasoned agreement" accord-
ing to the task in hand. We can interpret this theory as falling into this cell so long as (i)
we focus on health and set aside other capabilities, and (ii) we allow trade-offs between
equality and efficiency. 8 This interpretation is not unreasonable, since Sen himself has
recently argued that health is one of the most important indicators of well-being [Sen
(1998)].
The main criticism that has been levelled at Sen's theory is that it permits a high
degree of subjective judgement on the part of the analyst in selecting the list of func-
tionings and then determining how much weight to give each one. This strikes many
liberal political theorists and economists as giving the analyst too much scope for heavy-
handed paternalism [Sugden (1993)]. Sen's reply is that the capability approach is suf-
ficiently general to be compatible with a wide range of specific conceptions of the good
life and that "the need for selection and discrimination is neither an embarrassment,
nor a unique difficulty, for the conceptualisation of functioning and capability" [Sen
(1993)].
Compared with the simple quasi-utilitarian situation depicted in Figure 1, these par-
ticular concerns about distributional justice can be represented by making two special
assumptions: firstly that there is diminishing marginal social welfare from increasing
the health of each individual, and secondly that this schedule is identical for both indi-
viduals. This would have the effect, shown in Figure 3, of making the welfare contours
convex to the origin and symmetrical about the line OE (which is at 450 to each axis). In
this case there is some trade-off between maximisation and equalisation and the greater
18 One further difficulty with this interpretation is that it runs roughshod over the key distinction Sen makes
between "functionings" (achieved outcomes) and "capabilities" (opportunity to achieve good outcomes). Sen
argues that individual advantage should be assessed not simply as the bundle of valuable functionings (e.g.
good health) the individual actually achieves, but as (an index of) the whole set of bundles he is capable of
choosing from - his "capability set". See Wagstaff and van Doorslaer (2000).
1878 A. Williams andR. Cookson
(Equality of Health) E
Health of B
F Health of A
Figure 3. Simple maximisation; non-linear smooth contours; equal weights for A and B.
the weight given to equalisation relatively to maximisation the greater will be the cur-
vature of the social welfare contours.
3.2.5. No side conditions: non-linearand smooth maximand: unequal weights
According to theories of substantive equality of opportunity [LeGrand (1982, 1991),

Arneson (1989), Cohen (1989), Roemer (1998)], justice involves compensating people
for any disadvantages they suffer through no fault of their own but not for disadvantages
they suffer as a result of their own free choice. That is, everyone should have the same
opportunity to obtain good things in life (e.g. health, wealth, positions of authority)
but it should then be up to them to choose how they exercise that opportunity. It is
important to distinguish this principle of substantive equality of opportunity from the
much weaker principle of formal equality of opportunity, according to which people
should have equal basic liberties and there should be no formal legal barriers against
particular groups of people. Under formal equality of opportunity, however, people may
suffer disadvantage through accidents of birth over which they have no choice - such
as being born into a harmful social environment, or with disability, or with less natural
talent than others.
(Equality of Opportunity for Health)

E
I
Health of B II
II
II
I
F Health of A
Figure 4. Simple maximisation; non-linear smooth contours; greater weights for A than B.
Any theory of substantive equality of opportunity requires a theory of "free choice"

to determine what counts as being freely chosen and what counts as being beyond one's
own voluntary control. Such a theory might tell us, for example, that people in manage-
rial occupations generally have a higher degree of free choice about whether or not to
adopt a healthy lifestyle than those in manual occupations. If so, equality of opportu-
nity would then recommend that people in managerial occupations should receive less
health care for any condition affected by their behaviour (and/or face greater taxes on
unhealthy behaviours such as smoking) than those in manual occupations.
As before, a strict insistence on equality without any trade-off between equality and
efficiency is unattractive, because it might require a levelling-down of opportunities
for health (e.g. by withdrawing health care from those who currently enjoy higher-than-
average opportunity for health). So equal opportunity theorists generally accept the need
for balancing equality goals with efficiency goals.
If on such ethical grounds public policy takes a different view of the health of A
relatively to the health of B, the welfare contours between the two are no longer sym-
metrical about the locus of equality (OE) in Figure 3, but skewed in favour of one of
them (say A). This is the situation depicted in Figure 4.
One criticism of the principle of equality of opportunity is that it is rather harsh and
unforgiving to those who, for whatever reason, make decisions which they regret later
in life. It amounts to justice without mercy. At a more practical level, another difficulty
is that the available data may be too crude to measure adequately a concept as subtle
and context-dependent as "free choice". So, in practice, the application of an equality of
opportunity principle to public policy decisions will inevitably end up penalising some
individuals who appear to have a high degree of free choice according to the available
data, but who actually have suffered disadvantage through no fault of their own.
A final difficulty lies in restricting the principle of equality of opportunity to health
alone. Arguably, one cannot have genuine equality of opportunity for health without
also having equality of opportunity for all goods [LeGrand (1991), Culyer and Wagstaff
(1993, p. 445)]. This is because, if there is diminishing marginal utility of non-health
goods (e.g. income), then those who are disadvantaged in non-health terms will have to
make a greater sacrifice of utility in order to attain the same level of health.
3.2.6. No side conditions: non-linearand kinked maximand: equal weights
A strict egalitarian position would permit no concessions to health maximisation except

as between alternatives that gave people equal levels of health. Strict equality at a higher
level of health for both parties would be better than strict equality at a lower level of
health for both parties. This is not to be confused with the argument that if we want
to insure ourselves against being left in poor health when others enjoy better health,
it would be prudent for us to insist on equality of health outcomes as the criterion to
be employed in choosing between policy options. This justification for adopting the
"maximin" principle would be the outcome of strong risk aversion rather than a strong
desire for equity, where equity is held to be the moral imperative to maximise the level
of health attained by the worst-off group.
The L-shaped indifference curves in Figure 5 represent this situation, in which im-
proving the health of whichever group already has the greater amount of health does
not register as a social improvement but improving the health of the worse off person
does (up to the point where equality is achieved). Thus moves towards greater equality
improve welfare, as does maintaining equality at a higher level of health for both parties.
A variant of this maximin position may also be of some importance because it ap-
pears, from some empirical survey work designed to elicit popularly held views about
equity in health, that for some people public policy interest in reducing inequalities in
health is nil until such inequalities reach a certain magnitude, at which point they dom-
inate the situation and are regarded as intolerable. This threshold effect is depicted in
Figure 5A, where close to the locus of equal health the welfare contours are as they were
in Figure 1 but once inequality reaches a certain level the contours become parallel to
the axes,19 as in Figure 5.
It is also conceivable that people might hold the opposite view, namely that equality
is the primary objective and its pursuit should be moderated only when the cost in terms
19 It is also possible that these contours are curved as in Figure 3, though this case is not shown here.
F HEALTH OF A
Figure 5. Simple maximisation; non-linear kinked contours; equal weight for A and B.
of the overall health of the community becomes intolerable. This generates the situation
depicted in Figure 5B, which shows a case where the chosen option is one where there
is marked inequality, because the level of health that would have been enjoyed at the
best feasible level of equal health (point Z) is too low a level of health to be acceptable.
3.2.7. No side conditions: non-linearand kinked maximand: unequal weights
In the preceding case the characteristics of the two parties played no role in determining
the optimal outcome. It might, however, be the case that considerations such as those
discussed earlier in relation to Figure 4 might come into play. This is a situation in which
it might be held that one party is more deserving of special consideration by society be-
cause they were not in any way responsible for their own ill health. A similar argument
might be put forward when it is the fact of inequality in health that is the focus of ethi-
cal concern and not its cause, or when the welfare contours brook no compromise with
health maximisation except when choosing between options which satisfy the equity
objective.
One possible instance of such a position is the notion that health policies should
favour those whose current situation is worst, even though they are not the people
for whom the greatest improvements in health are possible. In this scenario, "need for
health" is the driving force and it is argued that the objective should be to choose those
policy options that would most quickly bring those people into equality with the others,
-- l- -r
riealt oi B
F Health of A
Figure 5A. Simple maximisation: non-linear kinked contours; equal weight for A and B.
U
F Health of A
Figure 5B. Simple maximisation: non-linear kinked contours; equal weight for A and B.
but without reducing the health of the better off. This means that the starting point be-
comes relevant, as it was in the Paretian case depicted in Figure 1A, but now the welfare
contours have become L shaped, indicating that one particular inequality-reducing path
r
vj F YA F1
Figure 6. Simple maximisation: non-linear kinked contours; greater weight to A than B.
is preferred to all others. This is the situation depicted in Figure 6, where Z is the start-
ing point, at which the health of A is worse than the health of B. The L-shaped welfare
contours are no longer symmetrical about the locus of points of complete equality, OE,
but about some locus such as ZY, where Y is the "target" level of health distribution
to which policy aspires. This represents a policy which seeks to reduce inequalities by
sharing health improvements between A and B in a way that systematically favours A.
In the case shown, where ZY is a straight line, this is in a constant ratio. It implies ac-
cording a greater weight to health improvements for A compared with the weight given
to the same improvement for B.20
20 This is not unlike the ethical position advocated by Murray and Lopez (1996) who argue that everyone
should be assumed to have a life expectancy at birth of about 80 years so as to prevent those with short
life expectancy being discriminated against when calculating the benefits of health care interventions. This
is equivalent to assigning a greater weight to each additional year of life gained by those who actually get
only a few, than is assigned to each year gained by those who actually get a lot. If the weights are inversely
proportional to the relative distance of the two parties to the common target, it would generate a straight line
such as ZY.
3.3. Theories with side conditions
3.3.1. One side condition:no maximnand
We have found no notion of distributive justice which fits into this cell, although such
a notion is quite plausible. It would simply assert that no situation could be considered
equitable which did not achieve some decent minimum level of health for people. 21
Lacking a maximand, there would be no means of discriminating between situations
which met this condition, or of deciding which is the better situation amongst those
which fail to meet the side condition. It would be a strictly dichotomous judgement,
equitable or not equitable, with no degrees of differentiation within either set. It is the
situation represented in Figure 7, where there may exist many health possibilities which
satisfy the side condition, but in the absence of a maximand there is no way of choosing
between them. This is equivalent to shifting the origin from O to 01, and ruling out
the shaded area to the southwest of CO 1C. It is also conceivable, however, that the
minimum requirement may be set so high that it is completely unattainable or, in a
particular policy context, that it has to be adjusted somehow so that it just touches FF at
a single point (its intersection with OE). In this latter case policy would be being driven
by practicality and, in the absence of any explicit optimising criterion, the adjustment of
the "decent minimum" to whatever is actually attainable would presumably be guided
by political expediency rather than by any ethic concerning distributive justice.
3.3.2. One side condition: linear maximand: equal weights
A utilitarian position, modified by imposing a side constraint that people must have
a decent minimum level of health, would fall into this cell of our classification. It is
a plausible position, which might generate three rather different outcomes. The first
would arise if the side condition were not satisfied. The second situation would arise if
the side condition were fulfilled, but it prevented the health-maximising situation from
being attained (i.e. no matter how big the sacrifice imposed on B, A's needs must be
satisfied). The third situation would be where the side condition was easily satisfied
and the outcome is no different from what it would have been had a simple utilitarian
position been adopted.
The first situation is depicted in Figure 8A, the second in Figure 8B, and the third
in Figure 8C. From a welfare economics viewpoint, Figure 8B is perhaps the most
interesting. The side constraint is satisfied anywhere along the segment of the feasibility
frontier that is marked XY but the best place to be in that segment is X. Z would be better
in terms of the maximand, but is ruled out by the side condition (because at Z, A would
21 The notion of a decent minimum of health is to be carefully distinguished from the notion of a decent
minimum of health care, as proposed, for example, by Fried (1976). The latter principle would act as a
constraint on health outcomes, requiring a minimum package of health care services irrespective of the health
gains that might be enjoyed by redeploying resources elsewhere.
Ch. 35. Equity in Health 1885
(Decent Minimum of Health)
IlICialtlL1+1 ofD
VI D
. I-.,
E
1r
U
F Health of A
Figure 7. Side condition equally applicable to both A and B; no maximand.
not be enjoying the postulated decent minimum level of health). So the "cost" of the
side condition is the difference in the level of overall health between X and Z, which is
the amount Q as indicated on the horizontal quasi axis O1C. Thus in this Figure we have
the underlying conceptual basis for attaching a "shadow price" to the binding constraint,
which quantifies the sacrifice in aggregate health that has been made in order to satisfy
that constraint.
3.3.3. One side condition: other permutations
In principle, each of the other maximands can be combined with a side condition. We
could also have more than one side condition. In this case, the side conditions can either
be satisfied simultaneously or sequentially. As before, the side conditions can in prin-
ciple be combined with any of the maximands. However, we have found no published
philosophical work which examines these complex permutations.
3.4. The opportunity set once more
So far the argument has been conducted with a feasibility set that has the normal prop-
erties assumed in economics. In the health field, however, there is an important set of
circumstances which require the feasibility frontier to take on a different shape. This
occurs when the health of one person is directly affected by the health of another person
Health of B
C'
U U
Health of A
Figure 8A. Side condition equally applicable to both A and B; linear contours; case A: no acceptable
possibility.
To
C
t'
Health of B
O I
I" Health of A
Figure 8B. Side condition equally applicable to both A and B; linear contours; case B: side condition binding.
Health of B
'
1P Health o A
Figure 8C. Side condition equally applicable to both A and B: Linear contours; case C: side condition not
binding.
(e.g. with infectious or contagious diseases, or in mental illness where a person may
become a threat to other people's safety). In such a case the health of (say) B can be
improved only if the health of A is improved.
This is the situation represented in Figure 9.22 This modification of the feasibility
set will not make any difference to cases where the maximand is of the orthodox kind
(namely where social welfare is improved, to some extent at least, if either party's health
improves). In cases such as were depicted in Figures 5 and 6, however, it might do so,
as shown in Figure 9, which is the same as Figure 5 but with the modified feasibility
frontier replacing the more orthodox one depicted earlier. It will be seen that the chosen
outcome X is not one of strict equality, because the best attainable level of equal health
(Y) would be inferior to the chosen point and point Z, which is the level of equal health
on the same social welfare contour as X, is simply not attainable.
3.5. Needs-based theories
Several philosophers have argued that health care should be distributed according to
need [including Williams (1962), Harris (1988), Lockwood (1988), Waltzer (1983)].
Typically, these philosophers do not argue that all goods and services should be dis-
tributed according to need; rather, they argue that health care is a special kind of good
which requires a special principle of distribution. Tobin (1970) calls this idea "specific
22 A similar diagram is used in Mishan (1977) and Atkinson and Stiglitz (1980).
Health of B
F Health of A
Figure 9. Health possibility frontier showing that, when the health of A is below X, an improvement in A's
health is necessary in order to improve B's health (contours non-linear and kinked).
egalitarianism". Waltzer (1983) develops this idea into a theory of "spheres of justice",
according to which different goods require different principles of distribution according
to the nature of the good (consumer goods, for instance, should be distributed according
to market forces; higher degrees according to academic merit; health care according to
need).
One difficulty with needs-based theories lies in drawing the line between goods which
should be distributed according to need and those which should be distributed according
to market forces. Should goods other than health care also be distributed according to
need (e.g. nutrition, shelter, transport, employment)? For that matter, should some forms
of health care be distributed according to market forces (e.g. cosmetic surgery)? A sec-
ond difficulty lies in defining what "need" means. Different definitions can recommend
very different distributions of health care. We can identify at least three possibilities: (i)
the person's initial level of ill-health, (ii) the person's capacity to benefit from health
care, and (iii) the expenditure required to equalise health. Definition (i) leads to the po-
sition shown in Figure 6. Definition (ii) leads to the simple utilitarian principle of health
maximisation shown in Figure 1 (since care is given to those who benefit most). Defini-
tion (iii) leads to a side constraint requiring strict equality of health (since care is given
so as to equalise health). For a more detailed analysis of these and other definitions of
need, the reader is referred to Culyer and Wagstaff (1993).
4. Theories which ethically constrain the opportunity set
4.1. Introductory comments
There is a strong philosophical tradition which approaches issues of distributive justice

in a totally different manner. It is argued that there are certain desiderata about the
way in which public policy decisions are made which, if observed, will ensure that the
outcomes will be just and no further criteria concerning fairness need to be applied
to them. To incorporate this mode of thinking into conventional economic analysis we
need to think of it as restricting the feasible opportunity set to a narrower "ethically
acceptable" set. Thus, in Figure 10 the original frontier FF has been shrunk illustratively
to F 1Fl.
This "illustration" requires some further commentary. First of all, there is no guaran-
tee that the specified desiderata will leave any feasible outcome "ethically acceptable" at
all. Conflicts between ethical principles may arise which require one or other principle
to give way. Collectively the desiderata may indeed be vulnerable to an "impossibility
theorem" which says that something will have to be relaxed for a feasible acceptable
outcome to emerge. To choose which desideratum to relax requires some higher level
optimisation criterion to be adduced, so that each constraint can be "shadow priced"
(Various Process Theories)

Health o BI
1) DEGENERATE CASE I:
F ,\ FFIF is a single noint
.......... x..
I~--
2) DEGENERATE CASE II:

F I F 1is null set
F1 -
\X
(-I S~~~I -lb
t
F F Health of A
Figure 10. Health possibilities reduced by ethical constraints: no side conditions on outcomes; no maximand.
in relation to it.23 Such theories do not typically have any such "maximand", however,
but merely assert the intrinsic value of each constraint. Nevertheless, suppose that as a
result of some further consideration, the key constraint were adjusted (downwards) to
make just one feasible outcome ethically acceptable, then Figure 10 would need to be
redrawn such that F 1F 1 became a single point somewhere within the original frontier
FF.
Not all theories which restrict the opportunity set are cast in these strict terms. Some
are hybrids which, although initially working their way through a series of conditions
which have to be satisfied lexicographically, allow whatever is left to be subjected to a
maximising rule. In Rawls' theory, for instance, the first task is to maximise the liberty
enjoyed by each individual, subject to that same amount of liberty being available to
all. This maximisation problem takes lexical priority over the second one, which is to
maximise the welfare of the worst off individual. Thus, once the maximal "equality of
liberty" condition has been satisfied, outcomes are to be ordered according to the extent
to which they improve the lot of the most disadvantaged in terms of an index of "primary
goods" such as income, authority and respect [Rawls (1982)].
In what follows we shall first of all discuss theories which can be interpreted as im-
posing ethical constraints upon the health possibility set but offer no maximand. The
bulk of philosophical thinking in this field seems to fall into this category. Then we
shall consider theories which do have such a maximand. As will be evident from the
table this appears to be sparsely populated territory. This may well be because we have
not come across any such theories in our search of the philosophical literature. It is an
interesting (and perhaps even a productive) intellectual exercise (which we leave to the
reader) to formulate theories of distributive justice which would fit into each of these
empty cells.
4.2. Theories with no maximand
4.2.1. Libertarianism
According to libertarians [e.g. Nozick (1974), Locke (1967)], justice is entirely a matter
of the enforcement of private property rights, including the right to the fruits of your
own labour. The role of the state should be to provide law and order but no more than
that. The "minimal" or "night-watchman" state should safeguard property, deter force or
fraud in private contracts and force criminals to pay adequate financial compensation to
their victims. The state should not raise taxes or interfere with individual liberty for any
other purpose - not to regulate markets or to provide public goods, nor to re-distribute
income and health from the fortunate to the less fortunate. So libertarians reject the
notion that there should be governmental agencies with direct responsibility for effi-
ciency and equity in the provision of health. Instead, they advise that state involvement
23 In a similar vein, ethical constraints may not take absolute and unconditional priority over efficiency con-
siderations: the constraint may be relaxed once the "shadow price" of upholding it becomes too high.
in health care should cease and be replaced by private health care insurance and private
charity for the poor and the disabled.
For libertarians, it makes no sense to enquire about the justice of a particular end-state
distribution of benefits and burdens (such as health outcomes). Instead, what matters is
the historical pattern of events. So long as the distribution of health outcomes results
from a process of free private contracts between consenting adults, then it is a just
distribution. Gray (1977) puts it thus: "Attempts to impose any other principle on the
free exchanges of free men involve imposing upon them a hierarchy of needs and merits,
about which no consensus exists in our society and which there is no reason to suppose
can be achieved".
An important feature of libertarianism is that rights are seen as absolute ethical con-
straints and that no trade-offs are permitted under any conditions. In particular, the state
should not violate minor rights in order to prevent major rights from being violated
(e.g. for the police to steal private documents in order to prevent suspected terrorists
from blowing people up). According to libertarianism, the state has no business violat-
ing one person's rights for the sake of someone else's rights, which would amount to
what Nozick (1974) has called a "utilitarianism of rights".
4.2.2. Participatorydemocracy
Another important strand of thinking about justice that fits into this cell is what might
be called the "participatory democracy" school of thought. This is the idea that justice
is entirely a matter of having a suitably free and fair process of democratic dialogue, in
which all citizens participate on an equal footing. So long as this process of dialogue is
in place, the resulting health outcomes must be just. A notable exponent of this dialogue
theory of justice in the health care literature is Rudolf Klein (1996); an exponent in
the philosophy literature is Ackerman (1980). Instead of inviolable rights to private
property, individuals are held to have inviolable rights to participate in public dialogue.
It is important to distinguish the principle of (actual) dialogue as a substantive theory
of justice from the advocacy of (hypothetical) dialogue as a means for choosing be-
tween rival theories of justice. As a substantive theory of justice, the dialogue principle
can be criticised as being too vague to give guidance to policy-makers and health care
managers. Without further specification, it gives no guidance as to what procedure of
dialogue is desirable; or how and why one procedure of dialogue is to be chosen over
another. Nor, once all the talking is done, does it help the person who is charged with
making the final decision.
Dialogue as a means of choosing between rival theories of justice has been advocated
most notably by Habermas (1984) in his theory of communicative action. Habermas
argues that the correct methodology for finding a theory of justice is to ask what rational
individuals would agree to after a process of dialogue in an "ideal speech situation"
- i.e. under conditions of free and uncoerced discussion. In principle, however, any
substantive theory of justice might be agreed to in the ideal speech situation (including
utilitarianism, libertarianism, and so on), so the task of analysing them all, in order to
inform such a dialogue, remains.
4.2.3. Contractarianism
Contractarianism, or social contract theory, is a broad tradition of political thought based

on the idea of a hypothetical agreement between rational individuals about what princi-
ples of justice should govern their social interactions. The notion of a hypothetical so-
cial contract is amenable to economic analysis, although this approach departs markedly
from the conventional welfare economics approach as described in Section 2. Contrac-
tarian analysis often leads to theories of justice which specify ethical constraints rather
than maximands [e.g. Buchanan (1975), Hayek (1976), Sugden (1986)]. In principle,
however, depending on the precise bargaining situation or "original position" the par-
ties to the social contract find themselves in, the contracting parties might come to agree
upon any substantive theory of justice, including ones with maximands. Rawls' (1971)
theory of justice, which specifies that the contracting parties are operating from behind
a "veil of ignorance" about their personal characteristics and position in society, is a
contractarian analysis which leads to a theory with both ethical constraints and a maxi-
mand. Harsanyi's (1955) theory of utilitarianism, which argues that rational individuals
from behind a "veil of ignorance" would choose to maximise lifetime expected util-
ity, is a contractarian analysis which leads to a theory with a maximand but no ethical
constraints. Contractarianism is thus perhaps best seen as a general methodology for
choosing between theories of justice, rather than as any one particular theory of justice
in its own right.
4.2.4. No-envy principles and equality of resources
Some economists have explored the idea that a fair situation is one in which no-one
envies anyone else [Varian (1974), Baumol (1986)]. The idea is that no-one should pre-
fer to be in any one else's shoes, taking into account not only their economic situation
(e.g. what goods they consume and what work they do) but all relevant aspects of their
personal circumstances (e.g. what disabilities they suffer from). The trouble with all
"no-envy" principles, however, is that they require a compensation principle to say any-
thing useful about real-world situations in which the ideal situation of no-envy cannot
be realised. Without a compensation principle, we can say nothing about how reductions
in envy for some can be traded-off against increases in envy for others.
One philosopher who has tried to develop a compensation principle within the no-
envy framework is Dworkin (1981a, 1981b). He uses the philosophical device of imag-
ining a hypothetical Walrasian insurance market for individuals who have not yet been
born. Compensation is then defined as the amount of insurance which (unborn) people
with equal cash endowments would take out against the possibility of being born dis-
abled or lacking in talent. 2 4 The motivation behind this hypothetical device is that it
allows us to set a market value not only on transferable "external" resources (goods and
24 Roemer (1996) contains a critical summary of the details of this compensation proposal.
services) but also on non-transferable "internal" resources (talent, intelligence, health).

Equal cash endowments in this hypothetical situation thus embodies a no-envy principle
which takes into account both external and internal resources.
Dworkin's central thesis is that justice requires that people should be compensated if
they have disabilities and lack of talent but not if they have expensive tastes or foolish
ambitions. He argues that people have a degree of responsibility for cultivating their
own preferences and so should not be compensated for having expensive tastes and
over-weaning ambitions. In the context of health care, for example, it can be argued that
preferences for frivolous beauty treatments, held by vain and spoilt patients, should not
attract resources away from less demanding patients who are in genuine need of health
care. More generally, Dworkin's idea is that people should be compensated for disabil-
ities, which are an involuntary accident of birth, but they should not be compensated
for expensive tastes, which are voluntarily cultivated. However, this way of drawing the
line between compensatable and non-compensatable factors has attracted criticism from
Cohen (1989) and Arneson (1989). These theorists point out that some expensive pref-
erences may be relatively involuntary (e.g. a taste for angling which suddenly becomes
expensive due to pollution) and that some disabilities may be relatively voluntary (e.g.
becoming disabled due to a sports injury). They argue that the degree of free choice
should determine the degree of compensation - which then leads back to an equality of
opportunity theory such as we considered earlier.
A second criticism of Dworkin's theory is that it makes no allowance for efficiency
goals and insists upon full equality of resources. This is unattractive, since it might
mean a levelling-down of everyone's resources. Dworkin replies that this consequence
only appears unattractive if one adopts an overly materialistic conception of individual
well-being. Dworkin adopts what he calls a "challenge" conception of well-being, ac-
cording to which individual well-being is entirely a matter of reacting appropriately to
challenges in appropriate circumstances - i.e. living well in a good society [Dworkin
(1988)]. He contrasts this with an "impact" conception of well-being, according to
which well-being is entirely a matter of achieving appropriate final ends (e.g. plea-
sure, health). On Dworkin's "challenge" conception, well-being is largely a matter of
individuals living their own lives well or badly. All the state can do is to get the circum-
stances right - which, in Dworkin's view, means equality of resources for all and not
necessarily more resources for all.
4.2.5. Equality of access to health care
A number of economists have proposed the principle that access to health care should be
the same for everyone - or for everyone with equal need [e.g. Mooney (1983), Mooney
et al. (1991), Olsen and Rodgers (1991)]. This principle focuses exclusively on health
care, ignoring all other possible ways of improving health and so differs markedly from
the principle of equality of opportunity for health outcomes. The principle of equal
access to health care places ethical constraints on the health possibility set: it rules out
all attainable health outcomes that require unequal access to health care. By contrast, the
principle of equal opportunity for health leaves the health opportunity set unconstrained,
but generates an asymmetrical maximand (as we saw earlier, in Figure 4).
The main criticism of the equal access principle is precisely that it focuses too nar-
rowly on the commodity of health care, rather than looking at what health care can
actually do for people. A second difficulty is that it is not clear how "access" should be
defined. We can identify at least four operationally different definitions: (i) access as the
utilisation of health care; (ii) access as the money and time costs incurred in receiving
health care [Mooney (1983)], (iii) access as the maximum attainable consumption of
health care [Olsen and Rodgers (1991)], and (iv) access as the forgone utility cost of
obtaining health care [LeGrand (1982)]. The reader is referred to Culyer and Wagstaff
(1993) for a discussion of the pros and cons of each definition. See also Wagstaff and
van Doorslaer (2000).
4.2.6. Medical ethics and the rule of rescue
So far, we have discussed ethical constraints that have been proposed in the context of
decisions about groups, rather than particular individuals, since that is the focus of this
chapter. However, ethical rules intended to govern decisions about particular individuals
and patients (e.g. do no harm, respect autonomy, tell the truth) will often have conse-
quences for resource allocation at the group level and to this extent can be thought of as
further ethical constraints on the health possibility set. There is a large medical ethics lit-
erature which proposes ethical principles for individual-level clinical decisions.2 5 One
such principle has been termed the "rule of rescue" [Hadorn (1991)]. This is the idea
that each individual, and society as a whole, has an ethical duty to do everything possi-
ble to help those in immediate life-threatening distress. In some cases, the rule of rescue
is merely a common-sense rule of thumb which moves us closer to the health possibility
frontier - for instance, it may sometimes be efficient to treat urgent cases at once, with-
out immediate regard to cost, because a stitch in time saves nine. Or, there may be more
convoluted justifications for this rule in terms of indirect or long-term consequences for
health - for instance, people may gain greater peace of mind if they know that those in
distress will always be rescued. However, it is hard to give a rational justification for the
rule of rescue insofar as it conflicts with other principles of justice. It seems irrational
to devote resources to people who happen to be in immediate distress if it is true that
other people with a greater claim on those resources (e.g. those who need them more, or
who are more disadvantaged) will lose out as a result. In defence of the rule of rescue in
such cases, however, it can be replied that some ethical rules are such deeply embedded
social conventions that they should be followed unless the consequences of doing so are
clearly disastrous.
25 Good introductions to the medical ethics literature are Glover (1977) and Gillon (1985).
4.2.7. Generalcomments on these six theories
In each of these six cases the implications for the analytical apparatus are the same and
are shown in Figure 10 in Section 4.1. Of the options that were in the health possibility
set when it was unconstrained by notions of equity or justice, some will have to be
removed. Since the principles for deciding which ones are inadmissible do not refer to
anything in the health outcome space, these deletions will follow no systematic pattern
in that space. The options that are ruled out may only be those which lie well inside
the frontier but we have assumed that some will be on the frontier, so that the frontier
shrinks (otherwise these theories would leave actual policy choices unaffected). Their
main problem from an economist's viewpoint is that they offer no maximand, so no
guidance is offered on which of the admissible options should be preferred, or on which
of the "unjust" options would be preferred in a second-best situation where something
has to be done (since maintaining the status quo in policy terms should be amongst the
options). Instead of optimising subject to constraints, what we have here are simply the
constraints.
4.3. Theories with a maximand
4.3.1. Equality of opportunityfor health usingfinancialmeans only
It can be argued that a more equal distribution of health is a desirable maximand but that
certain means of achieving this (e.g. via forcible redistribution of kidneys or other body
parts) are ethically unacceptable. A hybrid theory of this kind was proposed in LeGrand
(1987), which argues for equality of opportunity for health as a maximand but then
imposes an ethical constraint against using the health care delivery system to achieve
this goal. Equality of opportunity for health requires discrimination against those who
do not fully exercise their opportunity for health (e.g. against smokers from managerial
classes). LeGrand (1987) argues, however, that this discrimination should be done only
via health care financing (e.g. consumption taxes, insurance premiums, user chargers)
and not via the delivery of health care. This theory leaves in place an equal opportunity
maximand but imposes the ethical constraint against health possibilities which require
discrimination in the delivery of health care.
The difficulty with this kind of theory is that it is not clear how, or on what basis,
the line should be drawn between acceptable and unacceptable means of redistributing
health. In particular, the argument cannot be that financial redistribution is acceptable
because it involves redistributing only non-health outcomes (e.g. consumer pleasures)
which are less morally significant than health outcomes. This argument is false to the
extent that (a) income and health are linked, and (b) non-health outcomes are morally
significant. Furthermore, it is an argument for restricting the principle of equality of op-
portunity to non-health outcomes only and employing some other maximand for health.
1896 A. Williams andR. Cookson
(Rawlsian Maximin Health)

Health of B
Ft F Health of A
Figure 11. Health possibilities reduced by ethical constraints: no side conditions on outcomes: non-linear
kinked contours.
4.3.2. Rawlsian theory
According to Rawls' theory of justice [Rawls (1971)], there are two principles of jus-
tice. First, basic liberties are to be distributed equally and at the maximum level that is
compatible with everyone else enjoying the same level. Second, social and economic
inequalities are to be arranged so that they are to the greatest benefit of the least advan-
taged members of society. Basic liberties include the right to vote and to be eligible for
public office, the right to hold property, freedom of speech and assembly, and freedom
from arbitrary arrest. Social and economic inequalities are to be measured by an index
of "primary goods" that (i) are all-purpose means that every rational person needs to
pursue their own ends and that (ii) are distributed by society rather than nature. This
includes basic liberties, income and wealth, positions of responsibility, and self-respect,
but does not include "natural goods" such as health, intelligence and imagination. These
principles are to be satisfied sequentially, the first taking absolute lexicographic priority
over the second. This means, in particular, that people are not allowed to trade off basic
liberties for greater income (or, indeed, greater health).
Figure 11 illustrates our (re-)interpretation of the Rawlsian position as applied to
health. We have interpreted Rawls first principle - the "liberty principle" - as an ethical
constraint on the health possibility set. This is because it will rule out any attainable
distributions of health which require inequality in basic liberties. We have interpreted
the second principle - the "difference principle" - as an L-shaped maximand in the
(Extended Rawlsian Maximin Health)
Health of B
F1
O
F1 F Health of A
Figure 12. Health possibilities reduced by ethical constraints: side condition equally applicable to both A
and B; maximand; non-linear kinked contours. Case: no ethically acceptable possibility satisfied the side
condition.
space of health. 26 However, we would like to emphasise that this interpretation of the
difference principle is explicitly rejected by Rawls (1982), who insists that the principle
should operate in the space of primary goods rather than health. One reason is that health
is distributed by nature as much as by society. A second reason is that good health is
an end in itself as much as an all-purpose means that every rational person needs for
pursuing their own ends. 2 7 The third reason is that, as Arrow (1973) has pointed out, a
strict application of the principle of maximin health would require excessive resources
going to those with extreme health care needs, perhaps to the extent at which others
are reduced to subsistence levels of income. For these reasons, Rawls has explicitly
stated that health is not a primary good and that he restricts his basic theory of justice to
26 Other authors have re-interpreted Rawls in this way, notably Green (1976) who argues that all basic needs
(including housing, education, and so on) should count as primary goods.
27 This second reason is particularly important if our measure of health incorporates quality of life as well
as length of life, and hence requires investigation into the subjective utility values that people set on differing
levels of morbidity. One of the key motivations for Rawls' use of primary goods as the index of individual
advantage, rather than utility or welfare, was to avoid his theory having "to examine citizen's psychological
attitudes or their comparative levels of well-being" [Rawls (1982)]. Rawls argues that it is not possible to
find a cardinal and interpersonally comparable measure of the quality of life that is universally acceptable, on
the grounds that in a pluralist society people can and should have irreconcilably differing views about what
constitutes a good life.
"normal, active and fully co-operating members of society" [Rawls (1982)]. As Daniels
(1985) puts it, "In effect, there is no distributive theory for health care because no one
is sick!" [italics in original].
In his more recent work Rawls (1993) mentions one way in which his basic theory of
justice might be extended to incorporate the possibility of ill-health. Following Daniels
(1985) he suggests that, before the difference principle should come into effect, all
individuals should be brought up to the minimum level of health required for them to
be "normally functioning" members of society. This moves us to a different cell in our
Table 1, involving a side condition on health outcomes (i.e. a minimum level of health).
If we keep the same maximand as before (i.e. maximin health), we move to Figure 12
which shows the case where this decent minimum side condition (the shaded area) ...
any ethically acceptable situation (those in F1 F l ).
4.4. Fittingpopulation change into the framework
The issue of population change raises special difficulties which are not easy to handle
within the framework we have outlined in this Chapter. If population size is treated
as exogenously given, then health inequalities between born and unborn generations
can be dealt with in the same way as those between any other population subgroups A
and B. The difficulties arise when policy decisions influence population size in ways
that we judge to be ethically significant. In particular, the use of an average measure of
population health by our framework already presupposes a particular (and controversial)
theory of optimal population. It implies that the state should encourage the birth of
people who would have above-average health and discourage the birth of people who
would have below-average health.
Parfit (1984) and Broome (1996) have investigated the issue of optimal population
size and have shown that it leads to problems even when our framework is modified in
various ways. One obvious modification would be to take the sum total of health for the
population subgroup, rather than average health, as the measure of health on each axis.
However, this leads to what Parfit (1984) calls the "repugnant conclusion", that a suffi-
ciently large population of people living in (equal) pain and misery will turn out to be
better than a smaller, healthier population. This is because adding people whose health
is on average the same as that of the existing population will always increase the sum
total of health without changing the degree of inequality. To avoid this conclusion, a fur-
ther modification would be to subtract a "critical level" of health from each individual's
health score before taking the sum total [Blackorby, Bossert and Donaldson (1995)].
However, this critical level must be set fairly high to avoid the "repugnant conclusion"
that it is optimal for there to be a large population living just above the critical level
of health. Yet this has potentially controversial implications, since critical-level theory
recommends that the state should discourage the birth of those who would be below the
critical level.
So it remains to be seen whether our framework can satisfactorily accommodate the
issue of population change with further modifications, or whether this issue requires a
rather different ethical framework.
5. How is health to be measured?
Up until now we have assumed that a person's health is measured in QALYs over some
appropriate time period. In the context of any discussion of equity in health other pos-
sibilities have been advocated, so this now becomes a matter of direct concern, and of
sharp difference of opinion. In some contexts it is a person's current health state, in
terms of an immediate threat to life (life expectancy), that is taken to be the dominant
ethical concern ["the rule of rescue", see Hadorn (1991)]. In others it is the person's
current level of suffering or disability that dominates [Nord (1993)]. In yet others it is
the size of the health gain that is relevant rather than the resulting level of health (e.g.
the idea that it is better to give a small gain to many than large gains to a few). For
some [e.g. Harris (1988)] the rest of a person's life is taken as the unit and regarded as
equal between people irrespective of its expected length or quality, conditional only on
whether the person wishes to go on living. For the "fair innings" argument it is a per-
son's whole lifetime experience of health that is taken to be the valid unit of comparison
(measured in terms of life years [in Murray and Lopez's (1996) DALY approach] or in
terms of quality-adjusted life years [in Williams's (1997) approach]).
Thus each of the philosophical approaches outlined above could be further subdivided
according to which concept of health it employs when comparing the plight of A and
B.28 Each such concept will generate different conclusions for policy. This is true even
with the simple maximisation rule depicted in Figure 1. Maximising life expectancy
is not the same as minimising suffering or disability. The implications of maximising
health gains will be affected by how these are measured and if "the rest of a person's
life" is the maximand, only immediate survival (i.e. "saving a life") will count as a
health gain. The "fair innings" approach differs from all the others in taking people's
past health history into account as well as their actual present health state and their
expected future health prospects.
6. The economist's role
To monitor states of the world, in order to determine the extent to which they satisfy the
postulated criteria for equity in health, requires observation. In the case of "pure" pro-
cedural theories which restrict the set of options on ethical grounds, the empirical task
becomes that of determining whether or not the postulated conditions are met by each
potential candidate (including the status quo). In general they will not be met, which in-
evitably leads to the conclusion that the distribution of health is generally not equitable.
28 We can even imagine theories which employ different concepts of health for valuing equality than for
valuing efficiency. For example, a theory might restrict attention to current health for valuing equality (so that
inequalities in people's immediate level of pain and disability matter, but not inequalities in length of life), but
widens the scope of attention to lifetime health for valuing efficiency (so that both length of life and quality
of life matter).
The implied solution to this problem is some radical reorganisation of society so that the
conditions are in fact met, which typically will be such a vastly complex and socially
costly enterprise that it cannot possibly be undertaken in one fell swoop. Moreover, it
is usually difficult to decide where to start, because trade-offs are not permitted, so no
guidance is offered by which to judge, on a piecemeal basis, where the greatest progress
might be made at least cost. Generally speaking, therefore, economists have fought shy
of becoming engaged in the evaluation of such reform processes, since until their ad-
vocates accept the possibility of trade-offs we have little or nothing to contribute from
within our discipline, except to remind them of the ubiquitous existence of opportunity
costs.
Moving on to those theories which have side conditions on acceptable outcomes but
no maximand (as in Figure 7), here again economic analysis has no role to play, because
the purpose of policy analysis here is simply to establish whether there are any options
that would generate outcomes that lie in the acceptable zone to the northeast of 01.
Since there are no (explicit) criteria for choice apart from satisfying the side conditions,
there is nothing further to be done. This is equally true whether there are any acceptable
options or not. As before, we are offered no guidance as to what best to do to move from
wherever we are to some point in the acceptable zone (it does not matter whose health
we sacrifice, or by how much, in order to get there). Again, the economist's principal
role seems to be to draw attention to this hiatus in the formulation of the policy problem.
This brings us finally to those theories which have a role for a maximand, with or
without prior side conditions in the outcome space. Here there will be a role for eco-
nomic analysis to help enlighten the public policy process. In the simplest case, where
the satisfaction of an equity norm concerning the distribution of health may entail some
sacrifice in the total amount of health available to the community as a whole, in princi-
ple the matter is a relatively simple one of estimating an equity-efficiency trade-off for
health. As will soon become evident, however, it is not such a simple matter in prac-
tice, since estimating the properties of a social welfare function for health is almost as
difficult as estimating one for social welfare generally. Where the equity norm is cast
in terms of side conditions upon acceptable outcomes, then for those side conditions
that are binding the role of economic analysis lies in calculating the shadow price of
the constraint at the point where it binds. This tells us the cost, in terms of the max-
imand, of being where the constraint holds us, as compared with being wherever we
would otherwise have chosen to be in the outcome space (e.g., in Figure 8B above, the
difference between the aggregate health of the community at point Y and the aggregate
health of the community at point X). The community can then decide whether observ-
ing this ethical constraint is worth this sacrifice, or whether it wishes to moderate it
somewhat, or make it even more stringent. Thus the "shadow price" becomes an input
into an equity-efficiency trade-off consideration. In these cases an analyst needs to note
which health concept is the relevant one for the theory of justice under consideration,
and how groups A and B are defined. Each theory needs to be scrutinised carefully for
its chosen "target" in the equity stakes.
7. Estimating equity-efficiency trade-offs in health
If the nature and implications of these various equity principles are to be clarified in
a policy-relevant way, it is necessary to quantify what might otherwise merely remain
vaguely appealing but ambiguous slogans. Only with some quantification will it be pos-
sible to convert them into criteria that can be applied in a consistent manner, and with a
reasonable chance of checking on performance (i.e. holding people accountable). Quan-
tification thus has the potential for clarification, for performance measurement, for ac-
countability, and for policy formulation, analysis and reappraisal. The quest for more
quantification of equity issues is worth pursuing on these grounds alone, despite the
hostility that that it is likely to engender from those who mistakenly equate greater pre-
cision with lack of humanity. This seems to be another manifestation of the difficulties
some people have in feeling sympathy for large numbers of people whose fate is rep-
resented by the relevant statistics, when they would respond quite readily if these same
facts were presented to them as a series of individual case studies.
Generally speaking, the work of economists in seeking greater quantification in this
field falls into two classes: that which addresses equity-efficiency trade-offs in the dis-
tribution of health explicitly in a quantitative manner, but currently lack the empirical
data with which to support the assumed numerical relationship; and that which attempts
to estimate such trade-offs empirically using questionnaire methods, but without having
an explicit theory of justice into which to insert and interpret the findings. It is a rather
unsatisfactory state of affairs.
Williams' (1997) exploration of the "fair innings" argument is an example of the
former type of enterprise. It starts from the commonplace observation that people who
die young are widely regarded as not having had a "fair innings" compared with those
who live to a ripe old age. Thus if one adopted a person's whole lifetime as the ap-
propriate period of observation and took something approaching quality-adjusted-life-
expectancy-at-birth in that community as the "norm", then someone enjoying a poorer
prospect than this would be entitled to feel unjustly treated and deserving of more effort
on the part of the community to improve his prospects than should be devoted to some-
one whose prospects are better than this norm. The exact magnitude of such "equity
weights" will depend on how averse that community is to inequalities in people's life-
time experience of health. Going back to Figure 3 (and interpreting health in terms of
whole lifetime health prospects) the greater the community's aversion to such inequali-
ties between A and B, the stronger will be the curvature of each social welfare contour.
The greater the curvature, the greater the sacrifice in total health that will be acceptable
in order to improve the distribution of health. Where no such sacrifice is acceptable, we
are back with Figure 1. Where having an equal distribution is all that counts, anywhere
on OE is preferable to anywhere off it. Williams assumes that the general situation will
be as in Figure 3 and uses hypothetical data about inequality aversion to generate equity
weights between social classes, and between old and young within social classes, in the
UK, that would be consistent with the "fair innings" approach to equity in health. He
also speculates that different weights might be relevant for different subgroups within
the community, since people's aversion to inequality might depend on who the inequal-
ities were between, and what had caused them. This last consideration would make
Figure 4 the appropriate analytical framework.
Murray and Lopez (1996) are also concerned to reduce inequalities in lifetime health,
but they concentrate on life expectancy rather than on quality-adjusted life expectancy.
Their objective is to reduce inequalities in the burden of disease between different re-
gions of the world. They generate implied equity weights in a rather more indirect
way. 29 In essence, they require that when calculating the distribution of the global bur-
den of a disease, it should be assumed that in the absence of that disease everyone
would enjoy the age-specific life expectancy of a hypothetical population whose life
expectancy at birth is 80 years for males and 82.5 years for females. In the simplest
terms, this means that although in a poor country someone's actual life expectancy at
some age might be 30 years, it will be assumed to be much greater, say twice as great,
thereby implicitly weighting each year at twice its actual amount (60/30). In a richer
country, where the person's actual life expectancy at that same age would have been
(say) 40 years, the implied weight is 60/40 or 1.5, and so on. There has been no attempt
to test empirically whether this view of equity in health has any widespread support, or
whether the choice of 80 or 82.5 years of life expectancy at birth as the norm for men
and women respectively, is the one that best reflects in quantitative terms the desired
equity-efficiency trade-off. It is the situation represented in Figure 6 earlier.
Turning next to the published studies that have used questionnaire methods to obtain
empirical estimates of equity-efficiency trade-offs, we find disappointingly little evi-
dence that can be usefully inserted into theoretical frameworks of the kind discussed
above. Although a good deal of progress has been made in identifying key factors and
key regularities in people's preferences for equity in health, very few studies have ac-
tually attempted to infer numerical parameters from their results that could be slotted
into theoretical models of equity-efficiency trade-offs. In reviewing this work, we first
describe the questionnaire methods that have been used and then turn to the main find-
ings.
In principle, standard health valuation methods such as "standard gamble" or
"time-trade-off" [Dolan (2000)] could be converted into methods for eliciting equity-
efficiency trade-offs. In effect, people could be asked how much of their own health
they would be willing to sacrifice in order to achieve equity in the distribution of health.
To our knowledge, however, no study has used this personalised approach to eliciting
equity trade-offs. Instead, most of the studies reviewed below have asked respondents
to adopt the impersonal perspective of a social decision-maker who has to choose be-
tween alternative distributions of other people's health. A third approach, which lies
somewhere between the personal and impersonal approaches, asks people to base their
29 These equity weights are not to be confused with the further age-weights that Murray and Lopez apply to
account for the value of people's differing social roles (e.g. being a parent) at different stages in their lives,
which is a utilitarian efficiency consideration rather than an equity consideration.
answers on what they themselves would prefer were they to face the (known or un-
known) probability of being the individuals they are asked to choose between. In other
words, respondents would be required to make personal choices - yet from behind a
"veil of ignorance" about their own personal situation [Rawls (1971)].
The most popular variant of the impersonal social decision-maker approach is known
as the "Person-Trade-Off" (PTO) approach [Nord (1995)]. Respondents are asked to
choose between two groups of people who differ in certain specified respects (e.g. age,
lifestyle, health) and who stand to benefit from some pair of alternative health interven-
tions. The trade-off is established by asking them to vary the number of people in one
group while holding constant the number in the other group, so as to find a point of
indifference. Valuations from this technique can be seen as representing the trade-offs
that people are prepared to make between health gains for different kinds of people and
numbers of persons treated. A less popular variant of this approach is to vary the health
gain (or distribution of gains) for one group while holding everything else constant (in-
cluding the number of people in each group).
Whichever variant is used, the great difficulty lies in defining the value elicitation
questions tightly enough to elicit precisely targeted information on equity-efficiency
trade-offs of a kind which can be used in economic analysis. 30 As we shall see, a num-
ber of studies have obtained results which cannot be interpreted as equity-efficiency
trade-offs, since they fail to distinguish between equity and efficiency considerations
sufficiently clearly. Furthermore, they may also fail to distinguish different equity con-
siderations. 3 1 Responses to PTO questions, for instance, are likely to contain the relative
weights a respondent attaches to a number of potentially relevant equity factors (in par-
ticular, the severity of the pre- and post-intervention health states, the health gain as
a result of intervening, and the number of persons treated) and it may be impossible
to disentangle what are the relative weights attached to each of these considerations.
The weights attributed to these different equity considerations may have quite different
implications for resource allocation decisions.
This problem of confounding factors is compounded by the psychological difficulties
faced by respondents who are not used to thinking about the highly focused and sim-
plified questions about equity and fairness that economists tend to ask. Consequently,
they may well interpret questions in unanticipated ways and bring to bear wider con-
siderations that the economist wishes to set aside [Dolan and Cookson (1998)]. In such
30 This requirement means that questionnaires used by economists tend to be far more abstract and tightly
defined than those used by opinion pollsters and sociologists.
31 Dolan (1998) suggests an alternative approach which uses a particular class of health-related social welfare
functions (HRSWF) which allows efficiency and equity to be considered independently. The function has two
parameters: one which determines the general degree of aversion to inequality (the curvature), and one which
determines the specific equity weight given to one class of recipients over another (the slope). Unlike the PTO
methodology, it allows us to address two key issues separately: (1) how individuals value one health state
compared with another, and (2) how society values those same health states. Dolan argues that it is important
that these two issues are kept separate since each may need to be answered in a different way in different
contexts.
circumstances, it is often hard to know whether or not respondents would accept the
real-world policy implications of their hypothetical answers, once those answers have
been fed into the economist's theoretical model and translated into practical policy guid-
ance. Thus questionnaire studies need to steer a somewhat heroic course between the
Scylla of confounding factors and the Charybdis of befuddled respondents. Given these
twin perils, the handful of published numerical trade-off results summarised below must
be viewed with much caution.
The issue of age is the only area in which economists have come close to estimating a
practically useful equity-efficiency trade-off parameter. Studies have consistently found
that people are, broadly speaking, willing to discriminate in favour of the young and
against the elderly in distributing health care treatments [Charny, Lewis and Farrow
(1989), Nord et al. 1996].32 Two studies have estimated the size of this trade-off, using
large-sample telephone surveys, by asking people to choose between two life saving
programmes which save different numbers of people of different ages. Cropper et al.
(1994) find that their median respondent was willing to let seven 60-year olds die to save
one 20-year old; Johannesson and Johansson (1997) find that their median respondent
was willing to let forty-one 70-year olds die to save one 30-year old.
Unfortunately, these estimates cannot be interpreted as equity-efficiency trade-offs
as they stand, because they fail to distinguish between efficiency reasons for ageism
(e.g. the utilitarian consideration that young people typically gain more health from
treatment than elderly people) and equity reasons (e.g. the "fair innings" argument). So
Johanneson and Gerdtham attempt to infer the relative strength of equity and efficiency
considerations through empirical modelling techniques. By making the heroic assump-
tion that respondents based their answers upon average QALY expectancy for people of
that age, they infer that nine QALYs gained for 70-year olds are judged equivalent to
one QALY gained for 30-year olds.
The only other equity trade-off estimate we are aware of was made by Johannesson
and Gerdtham (1996), who piloted a version of the "veil of ignorance" approach. Using
a telephone survey, they asked respondents to choose between two societies that differ
with respect to the number of QALYs received by two groups. They found that, on aver-
age, respondents were willing to give up 1 QALY in the group with more QALYs to gain
0.45 QALYs in the group with fewer QALYs (irrespective of how many fewer QALYs).
This result is hard to interpret, however, in the absence of any supporting qualitative evi-
dence about what background assumptions respondents were making about other equity
considerations (especially the age of people in the different groups).
32 More specifically, Chamy et al. (1989) find an "inverted-U" shaped valuation of life with age: the value
rises from a 2-year old up to an 8-year old (on the grounds that the 8-year old is "more of a person"), but
then falls back down again from a 35-year old to a 60-year old (on the grounds that the 35-year old is more
likely to have parental responsibilities). Since this study used scenarios involving clinical choices between
two patients, however, it is not clear how far this finding is transferable to the settings we are interested in,
involving choices between groups made by managers and policy makers.
Although no trade-off estimates have been made, economists have also begun to
investigate the issue of how far respondents are willing to discriminate against those
who are partly responsible for their own ill-health. 33 So far, it seems that the majority
of people do wish to discriminate against those with purportedly "self-inflicted" ill-
nesses, both in group choices and in clinical decisions [Kneeshaw (1997), Dolan and
Cookson (1998)]. However, (i) a substantial minority dispute this principle, (ii) there
is more agreement on the general principle than on its specific application to groups
such as smokers, heavy drinkers, the overweight and illegal drug users, and (iii) follow-
ing group discussion, people are less likely to support this principle when they become
more aware of the conceptual and practical difficulties associated with it [Dolan and
Cookson (1998)].
In conclusion, it is fair to say that this empirical work is still at an embryonic stage,
and that no-one has yet devised a fully satisfactory questionnaire method for eliciting
equity-efficiency trade-offs in health. It is worth noting that this is by no means a unique
failing of health economists. Economists have examined the distribution of a number of
things other than health - from income right through to grapefruits - without manag-
ing to elicit reliable equity-efficiency trade-offs [reviewed in Miller (1992)]. The basic
finding from this work is that people do not subscribe to any one "pure" theory of dis-
tributive justice, applicable in all circumstances. Rather, people's responses appear to
mix different concepts of equity together (e.g. some weight on procedural justice, some
on attaining better outcomes, some on meeting minimum constraints), and to vary the
chosen equity concept from one context to another. 34
The great challenge is to bridge the gap between the economic requirement to esti-
mate precisely targeted equity-efficiency trade-offs, and the psychological capabilities
of respondents to think about equity and efficiency in such a tightly defined manner. In
part, this will require greater sensitivity of economic theory to the philosophical sub-
tleties and context-sensitivities embedded in people's preferences. It will also require
more sophisticated methods for eliciting preferences which are capable of teasing out
key considerations from confounding factors. Researchers have begun to respond to this
latter challenge by combining questionnaires with in-depth group discussions, which (a)
allows respondents more time to digest and to reflect upon the questions, and (b) gathers
qualitative evidence about how respondents have interpreted the questions [Nord et al.
(1996), Dolan and Cookson (1998)].
33 Two issues of procedural equity (which are outside our remit) have also received particular attention: (i)
equality of access to health care and (ii) urgency of treatment. It has been found that people give substantial
weight to equality of access and are not willing straightforwardly to give higher priority to some patient groups
purely on the basis that they stand to gain more health from treatment [Nord (1993), Nord and Richardson
(1995), Dolan and Cookson (1998)]. It has also been found that people wish to give higher priority to those in
life-threatening emergency situations, independently of capacity to benefit, both in rankings of specific health
care services [Kneeshaw (1997)] and in clinical choices [Dolan and Cookson (1998)].
34 One particularly important aspect of the context is that preferences depend upon the status quo distribution
of outcomes [Kahneman, Knetsch and Thaler (1986), Bukszar and Knetsch (1997)]. That is, outcomes which
impose losses of income on others relative to a status quo situation are perceived as being unfair, quite apart
from egalitarian considerations relative to absolute levels of income.
8. Further considerations
At the level of principle the main weakness of the work reported here is that it regards
health as the sole element in the social welfare function, and ignores other objectives
of public policy. This is defensible to the extent that public policy responsibilities do
in reality tend to get fragmented, and in this case a concern for health is at least much
wider than a concern for health care. It might also be defended on the ground that if it
were only possible to work with one indicator of people's well-being, people's lifetime
experience of health, measured in QALYs, might be a better measure than any other
[see Sen 1998]. A person's quality-adjusted-life expectancy is more individualised than
(household) income or wealth, and is known to reflect differences in people's income,
education, occupation, location and general economic and social environment. Health
is an essential ingredient for every individual's well-being, no matter what values and
aspirations they have. So treating it as the primary concern of public policy may be quite
sensible and revealing, even though somewhat unorthodox.
The weakness of so doing, of course, is that it denies us the opportunity to explore
the extent to which people will trade off health against other desirable things - such
as wealth, leisure, pleasure, social status, self-respect, friendship, and so on. If, for in-
stance, individual well-being is seen as holding a balanced portfolio of health, wealth
and wisdom [see Williams (1988)] then all three need to be in both individual and social
welfare functions. Indeed, the United Nations [UN Development Programme (1992)]
has in fact devised an index for the comparison of levels of well-being between nations
that combines life expectancy, real income and literacy levels, with that specific purpose
in mind. From an equity viewpoint, however, it may well be that people have different,
rather specific, views about equity in each of these realms.3 5 For instance, they may be
very egalitarian concerning health but not at all so with respect to income and wealth. So
it cannot be assumed that an equity notion which is strongly supported (and well docu-
mented quantitatively) in the health field should also be applied to these other fields ...
and vice-versa!
At a more practical level, economists need to widen their repertoire of equity notions
in the health field and be rather more careful about specifying why a particular aspect
of health (e.g. someone's current health state) should be of policy interest from the
viewpoint of distributive justice, when presenting data (usually cross-sectional) about
inequalities in health. It would be helpful if it became the norm to link any statement
about equity to some specified philosophical position, so that the reader could see more
clearly how the data should be interpreted, and what are the strengths and weaknesses
of the author's position at a conceptual level.
In a day to day setting, there is likely to emerge some division of labour between
economists according to their immediate responsibilities. Many health economists are
35 A view elaborated by the political philosopher Walzer (1983).

engaged in economic evaluations of health care activities and employ methods of anal-
ysis that are designed to elucidate the relative efficiency of different options (in cost-
effectiveness, cost-utility, or cost-benefit terms). They can hardly be expected, in the
context of each such study, also to embark upon the generation of all sorts of different
equity weights for the diverse clientele of public and private agencies who might one
day want to use their results. The most that can reasonably be expected of them is that
they make it possible to identify the distributions of health and health benefits across the
subjects in their study, by the main personal characteristics that are likely to be relevant
to judgements about distributive justice. It is likely that this will need to go beyond the
usual background data on age and sex and area of residence to include factors such as
lifestyle characteristics, occupation, education, income, ethnic group etc.
Meanwhile a second group of economists needs to be working on the development
of the equity concepts themselves and on estimating the shadow prices and equity-
efficiency trade-offs they imply or require. The objective should be to derive context-
specific equity weights that could then be applied by policy analysts to the data gener-
ated by the efficiency-based evaluative studies, to ensure that considerations of equity
in health do not get lost but get treated in a systematic (rather than an arbitrary and
capricious) way when the results of an efficiency calculus are up for consideration. By
harnessing methods from psychology, sociology and philosophy to this process, health
economists may in the future succeed where others have failed in measuring equity-
efficiency trade-offs. It will be a major step forward to find a way of translating people's
considered preferences about equity in health into measurements that can give clear and
concrete guidance to social decision-makers in this field.
References
Ackerman, B. (1980), Social Justice in the Liberal State (Yale University Press, New Haven, CT).
Arneson, R. (1989), "Equality and equality of opportunity for welfare", Philosophical Studies 56.
Arrow, K.J. (1973), "Some ordinalist-utilitarian notes on Rawls' theory of justice", Journal of Philosophy
70(9):251.
Atkinson, A.B., and J.E. Stiglitz (1980), Lectures on Public Economics (McGraw-Hill International, Singa-
pore).
Baumol, W.J. (1986), Superfairness (MIT Press, Cambridge, MA).
Blackorby, C., W.Bossert and D. Donaldson (1995), "Intertemporal population ethics: Critical-level utilitarian
principles", Econometrica 65:1303-1320.
Broome, J. (1996), "The Welfare Economics of Population", Oxford Economic Papers 48:177-193.
Buchanan, J.M. (1975), The Limits of Liberty: Between Anarchy and Leviathan (University of Chicago Press,
Chicago).
Bukszar, E., and J.L. Knetsch (1997), "Fragile redistributive choices behind a veil of ignorance", Journal of
Risk and Uncertainty 14:63-74.
Charny, M.C., P.A. Lewis, et al. (1989), "Choosing who shall not be treated in the NHS", Social Science and
Medicine 28(12):1331-1338.
Cohen, G.A. (1989), "On the currency of egalitarian justice", Ethics 99:906-944.
Cohen, G.A. (1993), "Equality of what: Resources, capabilities and midfare", in: M. Nussbaum and A. Sen,
eds., The Quality of Life (Clarendon Press, Oxford).
Cropper, M.L., S.K. Aydede, et al. (1994), "Preferences for life saving programmes: How the public discounts
time and age", Journal of Risk and Uncertainty 8:243-265.
Culyer, A.J., and A. Wagstaff (1993), "Equity and equality in health and health care", Journal of Health
Daniels, N. (1985), Just Health Care (Cambridge University Press, Cambridge).
Dolan, P. (1998), "The measurement of individual utility and social welfare", Journal of Health Economics
17:39-52.
Dolan, P. (2000), "The measurement of health-related quality of life for use in resource allocation decisions",
in: A.J. Culyer and J.P. Newhouse, eds., Handbook of Health Economics (Elsevier, Amsterdam) Chapter 32.
Dolan, P., and R. Cookson (1998), "Fairness in health care: What the public thinks", Discussion Paper (Centre
for Health Economics, University of York).
Dworkin, R. (1981a), "Equality of welfare", Philosophy and Public Affairs 10.
Dworkin, R. (1981b), "Equality of resources", Philosophy and Public Affairs 10.
Dworkin, R. (1988), "Foundations of liberal equality", Tanner Lectures on Human Values (Stanford Univer-
sity).
Fried, C. (1976), "Equality and rights in medical care", Hastings Center Report 6, pp. 29-34.
Frolich, N., and J.A. Oppenheimer (1992), Choosing Justice: An Experimental Approach to Ethical Theory
(University of California Press, Berkeley, CA).
Gillon, R. (1985), Philosophical Medical Ethics (John Wiley and Sons, Chichester).
Glover, J. (1977), Causing Death and Saving Lives (Pelican Books, London).
Goodin, R.E. (1995), Utilitarianism as a Public Philosophy (Cambridge University Press, Cambridge).
Gray, J. (1977), "On the contestability of social and political concepts", Political Theory 5:335.
Green, R.M. (1976), "Health care and justice in contract theory perspective", in: R.M. Veatch and R. Branson,
eds., Ethics and Health Policy (Ballinger Publishing, Cambridge) 111-126.
Habermas, J. (1984), The Theory of Communicative Action, 2 Volumes, translated by T. McCarthy (Beacon
Press, Boston).
Hadorn, D. (1991), "Setting health priorities in Oregon: Cost-effectiveness meets the rule of rescue", Journal
of the American Medical Association 265:2218-2225.
Hampshire, S. (1978), "Morality and pessimism", in: S. Hampshire, ed., Public and Private Morality (Cam-
bridge University Press).
Hare, R. (1993), "Health care policy: Some options", in: R. Hare, Essays on Bioethics (Calendon Press,
Oxford) 210.
Harris, J. (1988). "More and better justice", in: M. Bell and S. Mendus, Philosophy and Medical Welfare
(Cambridge University Press, Cambridge) 75-96.
Harsanyi, J.C. (1955), "Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility",
Hausman, D., and J. McPherson (1996), Economic Analysis and Moral Philosophy.
Hayek, FA. (1976), Law, Legislation and Liberty (University of Chicago Press, Chicago).
Johannesson, M., and U.G. Gerdtham (1996), "A note on the estimation of the equity-efficiency trade-off for
QALYs", Journal of Health Economics 15:359-368.
Johannesson, M., and P.O. Johansson (1997), "Is the valuation of a QALY gained independent of age? Some
empirical evidence", Journal of Health Economics 16:589-599.
Kahneman, D., and J. Knetsch, et al. (1986), "Fairness as a constraint on profit-seeking: Entitlements in the
market", American Economic Review 76:728-741.
Klein, R., P. Day and S. Redmayne (1996), Managing Scarcity: Priority-Setting and Rationing in the National
Health Service (Open University Press, Buckingham).
Kneeshaw, J. (1997), "What does the public think about rationing? A review of the evidence", in: B. New,
ed., Rationing: Talk and Action in Health Care (BMJ Publishing Group).
LeGrand, J. (1982), The Strategy of Equality: Redistribution and the Social Services (Allen and Unwin,
London).
LeGrand, J. (1987), "Equity, health and health care", Social Justice Research 1:257-274.
LeGrand, J. (1991), Equity and Choice (Harper Collins, London).

Locke, J. (1967), Two Treatises of Government, P. Laslett, ed. (Cambridge University Press, Cambridge).
Lockwood, M. (1988), "Quality of life and resource allocation", in: M. Bell and S. Mendus, Philosophy and
Medical Welfare (Cambridge University Press, Cambridge) 33-55.
Miller, D. (1992), "Distributive justice: What the people think", Ethics 102:555-593.
Mishan, E.J. (1977), "Economic criteria for intergenerational comparisons", Futures: The Journal of Forcast-
ing and Planning 9(5):383-404.
Mooney, G. (1983), "Equity in health care: Confronting the confusion", Effective Health Care 1:179-185.
Mooney, G., J. Hall, C. Donaldson and K. Gerard (1991), "Utilisation as a measure of equity: Weighting
heat?" Journal of Health Economics 10:475-480.
Murray, C.J.L., and A.D. Lopez, eds. (1996), The Global Burden of Disease (Harvard University Press).
Nord, E. (1993), "The trade-off between severity of illness ad treatment effect in cost-value analysis of health
care", Health Policy 24:227-238.
Nord, E. (1995), "The person-trade-off approach to valuing health care programs", Medical Decision Making
15:201-208.
Nord, E., J. Richardson, et al. (1995), "Maximizing health benefits vs. egalitarianism - an Australian survey
of health issues", Social Science and Medicine 41(10):1429-1437.
Nord, E., A. Street, et al. (1996), "The significance of age and duration of effect in social evaluation of health
care", Health Care Analysis 4:103-111.
Nozick, R. (1974), Anarchy, State and Utopia (Basic Books, New York).
Olsen, E.O., and D.L. Rodgers (1991), "The welfare economics of equal access", Journal of Public Economics
45:91-106.
Parfit, D. (1984), Reasons and Persons (Oxford University Press, Oxford).
Plant, R. (1991), Modem Political Thought (Blackwell, Oxford).
Rawls, J. (1971), A Theory of Justice (Harvard University Press, Cambridge, MA).
Rawls, J. (1982), "Social unity and primary goods", in: A. Sen and B. Williams, eds., Utilitarianism and
Beyond (Cambridge University Press) 159-187.
Rawls, J. (1993), Political Liberalism (Columbia University Press, New York).
Roemer, J. (1996), Theories of Distributive Justice (Harvard University Press, Cambridge, MA).
Roemer, J. (1998), Equality of Opportunity (Harvard University Press).
Sen, A. (1980), "Equality of what", in: S. McMurrin, ed., Tanner Lectures on Human Values (Cambridge
University Press, Cambridge).
Sen, A. (1993), "Capability and well-being", in: M. Nussbaum and A. Sen, eds., The Quality of Life (Claren-
don Press, Oxford).
Sen, A. (1998), "Mortality as an indicator of economic success and failure", Economic Journal 108(446): 1-
26.
Sen, A., and B. Williams, eds. (1982), Utilitarianism and Beyond (Cambridge University Press).
Smart, J.J.C., and B. Williams (1973), Utilitarianism: For and Against (Cambridge University Press).
Sugden, R. (1986), The Economics of Rights, Cooperation and Welfare (Basil Blackwell, Oxford).
Sugden, R. (1993), "Welfare, resources, and capabilities - a review of inequality reexamined by Amartya
Sen", Journal of Economic Literature 31(4):1947-1962.
Tobin, J. (1970), "On limiting the domain of inequality", Journal of Law and Economics 13(2):263-277.
United Nations Development Programme (1992), Human Development Report 1992 (Oxford University
Press, New York and Oxford) 12-25.
Varian, H. (1974), "Equity, envy and efficiency", Journal of Economic Theory 9:63-91.
Wagstaff, A., and E.K.A. van Doorslaer (2000), "Equity in health care finance and delivery", in: A.J. Culyer
Walzer, M. (1983), Spheres of Justice (Basil Blackwell, Oxford).
Williams, A. (1988), ""...Makes a man healthy wealthy and wise" or from folklore to system science",
in: G. Duru et al., eds., System Science in Health Care, Vol. 2 - Health Care System and Actors, Fourth
International Conference of the Society for System Science in Health Care, Lyon, July 4-8, 1988 (Masson,
Paris) 57-60.
Williams, A. (1997), "Intergenerational equity: An exploration of the 'fair innings' argument", Health Eco-
nomics 6(2):117-132.
Williams, B. (1962), "The idea of equality", in: P. Laslett and W.G. Runciman, eds., Philosophy, Politics and
Society (Basil Blackwell, Oxford).
World Bank (1993), World Development Report 1993 (Oxford University Press, Oxford).
AUTHOR INDEX
n indicates citation in footnote.
Aaron, H.J. 1240 Allen, D., see Rosenberg, S. 717

Aarts, L. 1002, 1003, 1021n, 1037, 1038, 1038n, 1039, Allen, R. 857, 867, 871, 1119
1039n Allen, R.C. 1608
Aarts, L., see Lindeboom, M. 325, 330 Alston, J.M., see Sumner, D.A. 1603, 1604n
Abel-Smith, B. 22 Altemeier, W.A., see Hickson, G.B. 1189
Abernethy, A. 1587, 1594 Altman, D. 618, 797n
Abraham, J. 1456 Altman, S.H. 1498
Abraham, K.G. 123n, 130n Altman, S.H., see Guterman, S. 601
Abraham, K.S. 1377, 1381 Altmeier, W.A., see Hickson, G.B. 255
Abt Associates and Policy Analysis, Inc. 1508 Ament, R., see DesHarnais, S.I. 863-865
Abt, C. 1576 American Association of Health Plans 710
Acemoglu, D. 1033 American Dental Association 1255
Ackerman, B. 1891 American Economics Group 1606
Ackoff, R.L. 1587 American Enterprise Institute 1498n
ACS 1394 American Hospital Association 1144
Acton, J. I00n, 445, 447, 1222, 1225 American Medical Association 143n, 238n
Adamache, K.W. 719,1463 American Psychiatric Association 898
Adamache, K.W., see Pope, G.C. 798, 800, 805, 81 in, American Public Health Association 1696
812, 814 Andersen, R. 23, 1257, 1261. 1810, 1812n
Adams Business Media 1634 Andersen, R.,seeAday,L.A. 1812n
Adams, E.K. 1128 Anderson, G. 520
Adams, E.K., see Ellwood, D. 682, 683 Anderson, G., see Weiner, J.P. 143n
Adams, E.K., see Porell, .EW. 412n, 1131 Anderson, G.F 800
Adams-Watson, J., see Lichtenstein, R. 829n Anderson, G.F., see Davis, K. 1498
Aday,L.A. 1070n, 1812n Anderson, G.E, see Fowler. E.J. 814
Adler, G.S., see Lazenby, H.C. 174n Anderson, G.E, see Gaynor, M. 1453
Advisory Commission on Intergovernmental Relations Anderson, G.F., see Weiner, J.P. 932
1548, 1570 Anderson, H.A., see Peterson, D.E. 1548
Ager, B. 1322n Anderson, K.H., see Butler, J.S. 325, 330
Aghion, P. 1475 Anderson, OW., see Sager, M.A. 600
Agro-Economic Services, Ltd. 1599 Anderson, P.M. 660n, 669, 670, 693n
Ahituv, A. 1790, 1793 Anderson, R. 1763n
Ahmed, T., see Khandker, R. 1123 Anderson, R., see Grossman, M. 404
Ahn, H. 293, 294 Anderson, S.P. 1411, 1436
Ahomrn, J., see Soumerai, S.B. 1327 Anderson, T.F. 70n
Aigner, D. 333, 1277 Andersson, F., see Gerdtham, U.-G. 11, 18, 22, 26, 27,
Ainsworth, M. 1778n 37, 46
Ainsworth, M., see Mwabu, G. 282 Andresen, E.M. 1745
Akehurst, R. 1289 Andrews, H., see Link, B.G. 898, 899
Akerlof, G. 607, 1065, 1562, 1786 Andrews, R.L. 1586
Alanen, P., see Pietila, T. 1283 Angrist, J. 293, 336, 398
Alchian, A. 1144, 1190 Angrist, J., see Acemoglu, D. 1033
Alchian, A., see Klein, B. 1473 Angrist, J., see Imbens, G.W. 293
Alderson, C. 274 Angst, J., see Swendsen, J.D. 898
Alegria, M. 897 Angus, D.C. 733
Aletras, V. 331,332, 335 Anis, A. 1330n
Alexander, G., see Kogan, M. 1064 Anis, A., see Cockburn, I. 133n, 135, 163
Alexander, J.A. 1454 Antczak-Bouckoms, A.A. 1287, 1288, 1290
Alexander, K., see Mooney, G. 1181 Antczak-Bouckoms, A.A., see White, B.A. 1285, 1286
Allanson, J., see Hoyme, E. 1066 1288
Allaz, A.F, see Pemeger, TV. 927 Antebi, S., see Anderson. G.F 800
I-1
I-2 Author !ndeA
AntelJ.J. 1515, 1525, 1527, 1528 Baifrie, R.S., see Naylor, C.D. 1240
Appelbaum, PS., see Steadman, H.J. 898-899 Bailey, C., see Myerberg, D.Z. 1077
Appropriation Authorization for the Department of Bailey, J. 1186
Veterans Affairs 928 Bailey, J., see King, N. 1186
Aquilina, D., see Johnson, A.N. 741 Bailey, N. 1763n
Archibald. R.B. 139n Bailit, H. 442, 1274
2
Arinen, S. 315, 319, 1261-1263, 1265, 1267, 1270, 1291 Bailit, H., see Hay, J.W. 1258, 1 58n, 1261, 1262, 1268,
Arinen, S., see Rosenqvist, G. 1261, 1262, 1267 1271
Arluck, G.M., see Grossman, M. 1641 Bailit, H., see Manning, W.G. 440, 1255, 1261, 1263,
Armknecht, P.A. 156n. 158n. 160n 1264, 1267. 1271, 1273
Armstrong, D. 1184 Baker, C., see Lee, A.J. 506n
Armstrong, D., see Smith, C.H. 1181 Baker, C., see Lubitz, J. 984, 985
Armstrong, M.A.. see Klatsky, A.L. 1651 Baker, J.B. 1423. 1433, 1436, 1437, 1447, 1476
Armstrong, P., see Armstrong, D. 1184 Baker, J.L. 1806n
Armstrong, P.W., see Mark, D.B. 215 Baker, L. 624, 741, 742, 1073
Arneson, R. 1878, 1893 Baker,R. 1181
Amett III, R.H., see Cohen, J.W. 134n Baker, R., see Bound, J. 270, 300
Arnott, R. 1706 Baker, R.M.,see Bound, J. 381
Arnould, R.J. 733 Bakewell, D., see Ysseldyke, J.E. 1036n
Aro, S., see Hakkinen, U. 315, 319. 320 Bakke, O.M. 1315n
Aronson, J.R. 1829-1830 Bal, D.G., see Flewelling, R.L. 1548
Arrighetti, A. 1179 Baldwin, C.Y. 964
Arrow, K. 2, 3, 58, 67, 79-81, 225n, 256, 421, 463,466, Baldwin.M.L. 1009n, 1033, 1033n
467, 496, 497n, 498, 499, 499n, 520, 576, 590, 607, Ballance, R. 1299n, 1300, 1304, 1318, 1321, 1329
721, 723, 726, 759, 908, 917, 958, 1097-1099, 1148, Ballard, C. 662
1254,1409, 1413,1491,1732, 1872n, 1897 Baloff, N., see Griffith, M.J. 617
Arthur Andersen Economic Consulting 1600n Baltagi, B.H. 30, 1548, 1559, 1570 1587, 1588, 1594
Asch, D., see Ubel, P. 88 Baluch, W.M., see Clark, D.O. 798n, 802, 803n
Ash, A. 784, 786, 797, 798, 800, 810, 811, 81 In, Bamezai, A., see Mann, J. 1160
812-814, 822 Bamezai, A., see Melnick, G.A. 1429, 1448, 1449, 1463,
Ash, A., seeEllis, R.P. 143n, 789, 791, 798. 800, 81 In, 1468
814, 822 Banerjee, A. 41, 237
Ash, A., see Pope, G.C. 798, 800, 81 In. 812 Bangdiwala, S., see Criqui, M.H. 1651
Ashby, C., see Morrisey, M. 718, 727 Bangemann, M. 1330n
Ashcraft, M., see Berki, S.E. 728, 729 Banks,D.A. 1167
Ashenfelter, 0. 1568 Bar-Hillel, M.. see Yaari, M. 88, 95
Barbash, G., see Mark, D.B. 215
Ashley, J. 768n
Bardsley, P. 1561, 1594-1596
Ashmore, M. 2, 3
Barendregt, J.J. 1577
Assistant Secretary for Planning and Evaluation 900
Barenthin, I., see Widstr6m, E. 1255
Atkinson, A.B. 298, 1575, 1583, 1589, 1593, 1663, 1829,
Barer, M. 172, 173n
1831n, 1848, 1887n1
Barer, M., see Evans, R.G. 1847n
Atlantic Information Services, Inc. 710, 713
Barer, M., see Stoddart. G.L. 85
Attrill, M., see O'Rourke, C.A. 1288
Barker, D.J. 1062
Auerbach, R., see Zwanziger. J. 730, 738
Barlow, D., see Daly, E. 1748
Auld, M.C. 1769, 1771
Barnett, P.G. 1548, 1549, 1568. 1569
Auster, R. 299, 353,381,396, 510
4 Barnett, P.G., see Hu. T.-W. 1549, 1594, 1595
Ayanian, J., see Ellis, R.P. 1 3n, 798, 800. 81 In, 814. Barnett, P.G., see Keeler, T.E. 1548, 1549, 1561, 1569,
822, 932
1596, 1604n
Ayanian, J., see Pope, G.C. 798, 800, 811n, 812 Barnow, B. 1018n, 1019n
Aydede, S.K., see Cropper, M.L. 1904 Baron, A., see Miller, L.A. 1064
Azeni, H., see Dickey, B. 923 Baron,D.P. 869
Azzone, V., see Ellis, R.P. 788, 808, 809 Baron, J., see Ubel, P. 88
Barr,N. 20,26
Babcock, B.A. 1603 Barr, R.D. 1745
Babigian, H.M., see Cole, R.E. 734 Barr, R.D.,seeFeeny, D. 1745
Babigian, H.M., see Reed, S.K. 737 Barrand, N.L., see Gauthier, A.K. 835
Babor, T.E 1642, 1643 Barrett, M., see Gamick, D.W. 734
Baca, P., see Llodra, J.C. 1288, 1289 Barros, P.P. 11, 18,38, 39,46,48
Bachman, D., see Hombrook, M.C. 789 Barry, B.M. 1808
Bachmann, R., see Arrighetti, A. 1179 Barry, M.J. 216
Badger, G., seeBickel, W.K. 1564 Barry, M.J., see Fowler Jr, EJ. 216
Bae, J.P., see Frank, R.G. 778, 815 Barsky, R.B. 1682, 1692
Bai, J., see Hu, T.-W. 1549, 1594, 1595 Bartel, A. 665n, 673, 898, 899
Author Index I-3
Bartlett, H.P. 964 Benham, L., see Scitovsky, A.A. 617

Bartlett, J. 1575n, 1580n Benjamin, B., see Manning, W.G. 1255, 1261, 1263,
Bartlett, J., see Hu, T.-W. 1555 1264, 1267, 1271, 1273, 1285, 1286, 1288, 1289
Bartlett, W. 1242 Bennet, K. 1735
Bartlett, W., see Wilson, D. 1190 Bennett, B., see Newhouse, J. 1113n
Bartman, B., see Moy, E. 477 Bennett, M.D., see Hornbrook, M.C. 798n, 802
Bartosch, W.J., see Cromwell, J. 1544 Bennett, T., see Braveman, P.A. 733
Barzel,Y. 1212, 1568 Bentham, J. 349, 350
Basinski, A., see Naylor, C.D. 1240 Bentley, J.M. 1280
Baskin, R., see Leaver, S.G. 137n Benyamini, Y., see Idler, E.L. 1853
Batavia, A.I., see DeJong, G. 103 In Benzeval, M., see Ham, C. 20
Bates, D., see Ellis, R.P. 143n, 798, 800, 81 In, 814, 822 Beran, R., see Ganem, J. 471
Bates, D., see Pope, G.C. 798, 800, 81 In, 812 Berenson, R., see Gold, M.R. 714, 715
Bates, I. 1709 Berenson-Howard, J., see Black, M. 1081
Bates, J. 1225 Bergen, M.R., see Lenert, L.A. 1746, 1753
Bator, FEM. 62n Berger, M.C. 399, 1655
Battese, G.E. 335 Bergmann, B. 1077, 1082
Battista, R., see Kottke, T. 1079 Berk,M. 13,571
Battista, R.N., see Kottke, T.E. 1711 Berk, M., see Monheit, A. 690, 691
Bauer, E.J. 975, 976, 978 Berki, S. 728, 729
Bauer, K.G. 1502 Berki, S., see Thomas, J.W. 780
Bauer, M., see Cooper, B. 939, 940 Berkowitz, E. 1044n
Baugh, D.K., see Ray, W.A. 602, 630 Berkowitz, M. 999n, 1030
Bauman, R.A., see Hursh, S.R. 1563 Berkowitz, M., see Mashaw, J. 1002n, 1008
Baumann. K.E. 1643 Berman, J.J., see Oppenheim, G.L. 484n
Baumgardner, J. 413, 596, 605, 710, 724, 913 Berman, P. 171
Baumohl, J., see Segal, S. 898 Bemdt, E.R. 119, 133, 133n, 150n, 151 153, 154n, 158n,
Baumol, W. 89n, 107, 1892 159n, 163,898, 903,904, 1301, 1303n
Bazzoli, G. 676n Bernheim, B.D. 978, 979, 1022
Beales, J.J. 1328 Bemstein, J., see Rasell, E. 1827n
Beard, T.R., see Saba, R.P. 1570 Berrick, D.M., see Read, J.L. 1746
Beaulieu, N., see Madrian, B.C. 678, 680 Berry, C. 526
Beauregard, K.M., see Cohen, J.W. 134n Berry, R.E. 1664
Beck, K. 776n, 793, 804, 822n Berry, S. 1132, 1412n, 1430n, 1431, 1432, 1434, 1447
Beck, R.G. 430,582 Bertko, J., see Shewry, S. 831
Becker, E., see Sloan, E. 1436 Besanko, D., see Baron, D.P. 869
Becker, E.R. 1147, 1156 Besley, T. 20, 849, 885, 1230, 1238, 1693
Becker, E.R., see Antel, J.J. 1515, 1525, 1527, 1528 Bevan, G. 833
Becker, G. 311, 349, 350, 353, 392n, 398, 401, 403, 404, Bevan, G., see Mays, N. 83n, 92
448, 876, 1551, 1554, 1559, 1560, 1560n, 1561-1563, Bezembinder, T.G.G., see Stalmeier, PEM. 1741, 1742
1569, 1570, 1572n, 1636, 1682, 1685, 1713, 1713n, Bhatia, V., see Giacomini, M. 1190
1773 Bhattacharya, J. 325, 326, 583
Becker, G., see Chaloupka, EJ. 1646n Bice, T.W. 617
Becker, G., see Ehrlich, I. 256, 413, 541, 1364, 1683, Bice, T.W., see Salkever, D.S. 1494, 1506n, 1508, 1517,
1686, 1688 1519, 1526
Becker, G., see Ghez, G.R. 357n, 362n, 369n Bickel, W.K. 1563, 1564, 1564n
Becker, G., see Michael, R.T. 350, 352, 372 Bickel, W.K., see DeGrandpre, R.J. 1564
Becker, G., see Stigler, G. 1562 Bidani, B. 1854n
Bedney, B., see Harrington, C.A. 958, 965, 966 Bieler, G.S., see Norton, E.C, 306
Beebe, J. 777n, 798, 819n, 823, 835 Bien, T. 1711
Beebe, J., see Lubitz, J. 806, 984, 985 Biesecker, G.E., see Needleman, H.L. 1066
Beeuwkes Buntin, M., see Newhouse, J.P. 928,929 Bijl, R., see Alegria, M. 897
Begun, J., see Feldman, R. 226, 476 Bikhchandani, S. 237
BehNman, J. 300, 325-327, 329, 330, 1055 Billi, J.E. 729
Behrman, R., see Gomby, D. 1064 Billows, W.D., see Jason, L.A 1597
Beinecke, R.H. 906 Birch, S. 83n, 92, 198, 511, 1260, 1272, 1273, 1286,
Beinecke, R.H., see Callahan, J.J. 904, 906, 912, 924 1288, 1289, 1835n, 1839
Beland,FE 1183 Birch, S., see Donaldson, C. 101n
Bell, R.M., see Carter, G.M. 798n Birch, S., see Gafni, A. 103, 105, 212, 1735, 1737
Ben-Porath, Y. 349 Birch, S., see Hutchinson, B. 1189
Benedict, M.E., see Montgomery, E. 690n Birch, S., see Hutchison, B. 83n
Benham, A., see Benham, L. 1810, 1839 Bird, R.M. 13
Benham, L. 74, 226, 469, 474, 1810, 1839 Bishai,D.M. 310
Benham, L., see Andersen, R. 23, 1257, 1261 Bishop, C.E. 964
1-4 Author Index
Bishop, C.E., see Baldwin, C.Y. 964 Borrowitz, M., see Sheldon, T.A. 830, 833
Bishop,J.A. 1549, 1568, 1587, 1588, 1593 Borsch-Supan, A. 282, 977
Bishop,N.,seeFuchs,L.S. 1036n Boskin, M.J. 130n
Bjorklund, A. 311 Boss, L., see Thompson. G.B. 1707
Black, A., see Bailit, H. 1274 Bossert, W., see Blackorby, C. 1898
Black, M. 1081 Boston Consulting Group 1326
Blackorby, C. 96n. 1898 Bothwell, J.L. 743
Blackstone, E.A. 1477 Botvin, G. 1079
Blades, C.A. 1 Boulding, K. 70
Blair, R.D. 1457, 1459, 1468 Bound, J. 270, 300, 381. 676n, 1024n, 1025, 1026n, 1027
Blank, R. 683, 684, 1072n Bovbjerg, R. 818n, 821n. 1343, 1349. 1355n, 1358, 1373
Blank, R.H. 1702 Bovbjerg, R., see Marsteller, J.A. 721
Blau, D. 676, 678. 679, 1056n Bovbjerg, R., see Zuckerman, S. 1356, 1372
Blaug,M. 1,3,506 Bowen, B. 835
Blaxter, M. 1843 Bowles, R.A. 1239
Blaylock, J.R. 298, 299 Box, G.E.P. 29
Bleichrodt, H. 103, 105, 106, 1730, 1741, 1742, 1746, Boyd, N.E, see Llewellyn-Thomas, H. 1734, 1748
1747, 1751 Boyd, N.F, see Sutherland, H.J. 1741
Bleichrodt, H., see van Doorslaer, E. 1853 Boyd, R. 1586, 1662
Blendon, R.J. 516 Boyd, R., see Seldon, B.J. 1548, 1587, 1589
Blincoe, L.J., see Miller. T.R. 1661n Boyden, J.S., see Mills, D.H. 1351
Blind, K.V.K. 1688 Boyer, M. 1559
Blisard, W.N., see Blaylock, J.R. 298, 299 Boyle, H.H., see Torrance, G.W. 1739
Blomqvist, A. 23, 39, 43, 47, 77n, 499, 586-588, 715, Boyle, M., see Feeny, D. 1732
724,744,918. 1185 Boyle, R.G., see Ohsfeldt, R.L. 1550, 1564, 1565, 1596,
Blomquist, G.C. 1679n 1597
Blondal, S. 1039 Bradbury, R.C. 733
Bloom. B. 1067 Braden, B.R., see Levit, K.R. 125n. 159, 959, 960
Bloom, D. 1763n, 1767n, 1786n Bradford, W.D. 1191
Bloor, K., see Maynard, A. 96 Bradford, W.D., see DeFelice, L.C. 494
Blumberg, L., see Manning, W.G. 271, 1643 Bradshaw, J. 768n
Blumenthal, D., see Rizzo, J.A. 525, 1417 Branch, L.G. 961
Blumenthal, D., see Schlesinger, M. 743 Branch, L.G., see Experton, B. 734
Blundell, R.W. 301,302, 308, 313,324 Brandeau, M. 1763n
Blundell, R.W., see Smith. R.J. 302 Brattberg, G. 1241
Blustein, J., see Weiss, L.J. 1179 Braun, R.J., see Maryniuk, G.A. 1264, 1290
Boadway, R. 62n, 65, 97n, 999n, 1022n Braveman, P.A. 733
Boardman, A.E. 1184, 1185 Braverman, A. 228
Boardman, A.E., see Vining, A.R. 1169 Bravo, M., see Llodra, J.C. 1288. 1289
Boardman, A.R. 1169 Bray, JW., see Farrelly, M.C. 1553, 1573
Boaz. R.E 974 Bray, R.M. 1657n
Boddy, F.A., see Donaldson, C. 1285, 1288, 1290 Braybrooke, D. 1815n
Boenna. W.G.W. 1177 Brazier, J. 1744, 1745
Boland, B. 1640n Brazier, J.E., see Harper, R. 1745
Boland, J.P., see Berry, R.E. 1664 Brekke. M.. see Kottke, T. 1079
Bolduc, D. 282 Brekke, M.L., see Kottke, T.E. 1711
Bollen, K.A. 303 Bresnahan,T. 123n, 1117, 1437n, 1444, 1446, 1447,
Bolton, P. 1474 1455, 1461
Bolton, P., see Aghion, P. 1475 Bresnahan, T., see Baker, J.B. 1436, 1437, 1447
Bombardier, C. 1745 Breusch, T.S. 30
Bombardier, C., see Wolfson, A.D. 1746 Breyer, E 1157
Bond, R.S. 1318n, 1324n Breyer, E, see Zweifel, P. 1, 353, 375, 381, 383, 385,
Bondy, S., see Room, R. 1657 387, 415,464n, 498n, 499, 499n, 507,507n,51 In,
Bongaarts, J. 1763n 1153, 1154, 1686
Bonnano, J.B. 719 Bridgers, W.F., see Pickett, G. 1712
Bonneux, L., see Barendregt, J.J. 1577 Brierley, H., see Jachuk, S.J. 1726. 1732
Bonsel, G.J., see Krabbe, P.F.M. 1746 Briggs, A. 197
Boozer, M. 1786 Brisson, A.S. 912, 936
Boqi, L., see Peto, R. 1546 British American Tobacco 1570
Borcherding, T.E. 1169 Brito, D. 71, 1694, 1706, 1778n
Borchgrevink, C.E, see Hjortdahl, P. 1179 Britton, C., see Myerberg, D.Z. 1077
Borjas, G., see Bartel, A. 665n, 673 Broadbent, E., see Hendricks, W. 1010n
Bork, R. 1473, 1474 Broders, A.C. 1545
Born, P, see Simon, C. 469n, 474 Brodie, M., see Keeler, E. 512n
Author Index I-5
Brodie, R.J., see Chetwynd, J. 1587 Buntin, M.B., see Newhouse, J.P. 786, 787, 791n, 792,
Broman, S. 1062 793, 822, 826
Bronstein, J. 1132 Burchell, B. 1179
Brook, R. 442 Bureau of Labor Statistics 907
Brook, R., see Bailit, H. 1274 Bureau of the Census 724
Brook, R., see Keeler, E.B. 1162, 1163 Burge, R.T., see Pope, G.C. 495
Brook, R., see Valdez, R.B. 443 Burgess, S.M. 325
Brook, R., see Ware Jr., J.E. 739 Burke, R., see Fries, B.E. 968
Brook, R., see Wells. K.B. 1712 Burke, W. 1178
Brooks, J. 1130, 1160, 1449, 1467n, 1469 Burkhauser, R. 1000, 1002n, 1008, 1009, 1009n, 1010,
Brooks, R. 1732 1017, 1028, 1029n
Brooks, S.D., see Harford, T.C. 1657n Burkhauser, R., see Aarts, L. 1003, 1037-1039, 1039n
Brooks-Gunn, J. 1062 Burkhauser, R., see Butler, J.S. 325, 330
Broome, J. 1242, 1733, 1738, 1749, 1898 Burkhauser, R., see Haveman. R. 998n, 1038n
Brown, A.B. 1605 Burkhauser, R., see Mashaw, J. 1002n, 1008
Brown, B., see Bunker, J. 513, 1259n Bums, B.J., see Taube, C.A. 909, 911
Brown, B.W., see Lubeck, D.P. 735 Bums, B.V., see Taube, C.A. 937
Brown, C. 690n Burns,L. 1132, 1132n, 1431, 1433
Brown, D.M. 495 Burns, L., see Walston, S.L. 1478
Brown, J.P. 1343, 1345, 1346, 1364 Buron,L.,seeHaveman, R. 1000, 1001, 1002n, 1011,
Brown, L. 709, 714, 718, 720 1011nOl
Brown, L., see Glied, S. 728 Burstin, H., see Ellis, R.P. 143n, 798, 800, 81 In, 814, 822
Brown,L.L., seeCassileth, B.R. 1728 Burstin, H., see Pope, G.C. 798, 800, 811 n, 812
Brown, M.L., see Baker, L.C. 742 Burtless, G. 426n
Brown, R. 616, 618, 623, 730 Busch, S.M., see Bemdt, E.R. 133, 163
Bush, J., seeFanshel, S. 101
Brown, R., see Grannemann, T.W. 1157, 1452
Bush, J., see Patrick, D.L. 1732
Brown, R., see Hill, J. 728, 730
Butler, J.S. 325, 330
Brown, R.E. 1685n
Butler, J.S., see Burkhauser, R. 1028
Brown, S. 1060
Butler, R. 1017n, 1028, 1356, 1374
Brown, S.W., see Cole, R.E. 734
Butler, R., see Worrall, J.D. 1017n, 1027n, 1028, 1029n,
Browner, W., see Rice, D.P. 1575
1045n
Browning, E. 664
Butler, S.M. 831
Brownson, R.C. 1595
Buttery, R.B. 1228
Broyles, R.W. 429n Butz, D.A. 1476
Broyles, R.W., see Rosko, M.D. 1507n Buxton, M., seeBriggs, A. 197
Bruce, M.L. 899 Byrne, D.J., see Newhouse, J.P. 557, 599, 864, 1529n
Bruce, N., see Boadway, R. 62n, 65, 97n Byrne-Logan, S., see Ash, A.S. 810, 812, 813
Bruggerman, K., seeMooney, G. 1181
Brundin, I. 492n Cady, J. 226
Bruun, K. 1649 Cafferata, G.L., see Stone, R. 960
Bryant, R.R. 1655 Cahal, M.E 229
Bubolzt, T.A., see Wennberg, J.E. 240 Cain, K.C., see Diehr, P. 242
Buchanan, J. 71 Cairns, J. 307, 1209, 1731, 1742, 1751
Buchanan, J.L. 438, 587, 728. 729, 733 Calem, P.S. 1165
Buchanan, J.L., see Keeler, E.B. 256 Calfee, J., see Craswell, R. 1349
Buchanan, J.L., see Manning, W.G. 911, 922 Califf, R.M., see Mark, D.B. 215
Buchanan, J.L., see Mauldon, T. 736 Callahan, J.J. 904, 906, 912, 924
Buchanan, J.L., see Rosenthal, M. 474 Calnan, M. 1181
Buchanan,J.M. 13, 19, 1233, 1892 Calnan, M., see Williams, S.J. 1181
Buchberger, T., see Sheingold, S.H. 597 Calonge, S., see Wagstaff, A. 1822n, 1824, 1827
Buchinsky, M. 271 Calonge, S., see van Doorslaer, E. 1839, 1843, 1844
Buchmueller, T.C. 275, 278, 621, 668, 670, 672, 688, Camargo Jr., C.A. 1651, 1652
692, 694, 767, 831 Cambois, E. 963, 987
Buck, D. 1608 Camerer, C. 1734
Buck, J.A. 908, 926 Cameron, A.C. 315, 318, 319, 322
Buclingham, K. 103, 1735, 1736 Cameron, S. 1561
Buehler, J.W. 1068 Camp, P., see Ware Jr, J.E. 739
Buescher, P.A. 1077 Campbell, J.L. 1182
Bui, L., see Berndt, E.R. 1301, 1303n Campbell, J.Y 41
Bukszar, E. 1905n Canino, G.J., see Swendsen, J.D. 898
BulowJ.I. 1568 Cantor, J. 1822
2
Bunker, J. 513, 1 59n Cantor, J., see Short, P. 681n, 1061
Bunker, J., see Shoven, J.B. 1577, 1582 Capehart, T. 1601n, 1602
I-6 Author Index
Capilouto, E.I., see Antczak-Bouckoms, A.A. 1288 Chaloupka, E, see Warner. K.E. 1542n, 1574, 1583,
Capilouto, E.I., see Ohsfeldt, R.L. 1550. 1564, 1565, 1584, 1595
1596, 1597 Chamberlain,G. 310, 311, 1640
Card, D. 661n, 1032 Chamberlin, E. 1102n
Card, D., see Blank, R. 1072n Chambers, S., see Birch, S. 83n, 92
Cardenas,E.M. 155n, 156, 156n, 157, 157n Chang, C.E 960, 972, 974
Cardon, J. 619 Chang, F.E-R. 383, 393, 394
Carey, T.S. 733 Chang, Y.C., see Rigotti, N.A. 1597
Carliner, G., see Bloom, D.E. 1763n Chapman, J.D. 786
Carlsen, F. 18,507, 511 Chapman, J.D., see Newhouse, J.P. 786, 787, 791n, 792,
Carlsen, F., see Grytten, J. 1181. 1184, 1185 793,822, 826, 928,929
Carlson, B.L., see Murtaugh, C.M. 959, 975, 977 Chapman, S. 1548, 1588. 1599, 1608
Carlton, D.W. 1473 Charles, C. 75
Carneyand, M.E, see Greenfield, S. 734 Charnes, A. 1278
Carpenter, R., see Myerberg, D.Z. 1077 Charney, M.C. 94
Carr-Hill, R. 1182 Charny, M.C. 1904, 1904n
Carr-Hill. R., see Dixon, P. 1181 Chase Econometrics 1606, 1606n
Carr-Hill, R., see Sheldon, T.A. 830, 833 Chassin, M. 484
Carrillo, H., see Harrington, C.A. 958, 965, 966 Chanlk,P. 1077, 1081, 1082
Carroll, C.E., see Warner, K.E. 1575-1577 Chayet 1494
Carroll, N.V. 597 Cheadle. A., see Haas-Wilson, D. 279
Carson, R.T., see Mitchell. R.C. 1752 Chelluri, L., see Angus, D.C. 733
Carter, G. 782,798n Chen, M.M., see Patrick, D.L. 1732
Carter, G., see Farley, D.O. 814 Chen, W., see Escarce, J.J. 714
Carter, G., see Ginsburg, P.B. 1530 Chenet, L., see Leon, D.A. 1646
Carter, G., see Keeler, E.B. 595, 772n, 773, 777n, 817, Cheng, S.H. 433
819, 825, 826, 913 Cher, D.J., see Lenert, L.A. 1746, 1753
Carter, R.A.L., see Blomqvist, A. 23, 39, 43, 47 Cherkin, D. 420, 432, 583, 713, 1708
Carter, W.B., see Andresen, E.M. 1745 Chern, W., see Just, R. 1461n
Cartland, J., see Yudkowsky, B. 1060 Chemew, M. 714,742, 860, 1133, 1456
Cartwright, A., see Gray, P.G. 229 Cherichovsky, D. 830
Casanova, C. 1066 Chesney, J., see DesHarnais, S. 598, 630, 863 865
Cassileth. B.R. 1728 Chetwynd, J. 1587
Casson, M. 1179 Chiang, T.L., see Cheng, S.H. 433
Cassou, S.. see Dowd, B. 279, 280 Chiang, Y., see Riley, G. 832
Cassou, S., see Feldman, R. 279-281,620 Child Trends, Inc. 1080
Castillo-Chavez, C. 1763n Chinitz, D. 761
4
Catron, B. 140n, 149, 1 9n Chinitz, D., see Chernichovsky, D. 830
Cauley, S.D. 315. 1263 Chiriboga, D.A., see Hay, J.W. 1258, 1258n, 1261, 1262,
Cavanaugh, D., see Callahan, J.J. 904, 906, 912, 924 1268, 1271
2
Cave, J. 609,1415 Chirikos, T. 1001, 100 n, 101 ln
Caves RE. 1322 Choi. J.P. 1474, 1474n
Cebu Study Team 1681 Choi. W.S., see Pierce, J.P. 1592
Celebucki, C., see King, C. 1592 Chollet, D., see Needleman, J. 1166
Center for Mental Health Servicies 903 Chollet, D., see White-Means, S.I. 974
Center for Science in the Public Interests 1659, 1662n Choodnovskiy, I., see Soumerai, S.B. 1327
Center for Studying Health System Change 719, 725 Chorba, T., see Rosenberg, M. 1056
Centers for Disease Control and Prevention 1578, 1646, Chou, S.Y., see Sloan, EA. 1162, 1165
1685, 1707 Chow, G. 1558
Cercone, J., see Hammer, J.S. 1806n Christenson, S.L., see Ysseldyke, J.E. 1036n
Chaffin, M., see Kelleher, K. 898, 899 Christianson, J. 742, 744, 912, 924, 975
Chaikind, S. 1062, 1066 Christianson, J., see Dowd, B. 557
Chait, E., see Rowland, D. 1061 Christianson, J., see Lurie, N. 735, 925
Chalkley, M. 62n, 76, 501n, 522, 550, 553, 856, 857, Christianson, J., see Wholey, D. 624, 744
859-861,866, 874, 881n, 1098, 1407n Chu, D.K., see Saywell, R.M. 1072n
Chalmers, A., see Bombardier, C. 1745 Chu, K., see Gold, M. 1529
Chaloupka, F 276, 1543, 1548, 1549n, 1550, 1552, 1553, Chuma, H., see Ehrlich, 1. 351, 361. 363, 375n, 390,
1560,1560n, 1561, 1565,1570,1573, 1578, 1680, 1682
1595-1597, 1641,1642,1646n, 1648. 1679, 1699, Ciampi, A., see Llewellyn-Thomas, H. 1734, 1748
1685, 1707 Cistemas. M., see Yelin, E. 1007
Chaloupka, F. see Grossman. M. 1551, 1595, 1641, Citrin, T., see Warner, K.E. 1611
1642, 1685 Clancy, C.,see Franks, P. 1177
Chaloupka, F., see Hoyt, G.M. 1639n Clancy, C., see Moy, E. 477
Chaloupka, F., see Tauras, J.A. 1553, 1555 Clark, D.O. 798n, 802, 803n
Author Index I-7
Clark, J.D., see Paul-Shaheen, P. 70n Connolly, G.N., see Harris, J.E. 1549, 1595
Clark, R. 690n Connolly, G.N., see King, C. 1592
Clark,R.C. 1152n Connor, R.A. 1454
67
Clark, R.L., see Headen, A.E. 676, 677, 9n Conover, C. 961
Clark, S., see Freed, G. 1065 Conover, C., see Sloan, FA. 961, 1166
Clark, W.B., see Hilton, M.E. 1657 Conrad, D. 619, 1165, 1261-1263, 1270, 1272, 1478
Clarke, A.E., see Garber, A.M. 209n Conrad, D., see Whitney, C. 1116n
Clary, M. 1079 Conrad, K.L., see Hughes, S.L. 961
Claxton, K., see Lerner. C. 1184, 1185 Conroy, P 1795
Clayton, E., see Hickson, G. 1360 Contandriopoulos, A., see Dionne, G. 76, 1184, 1185
Clements, K.W. 1638 Cook, E.F., see Pearson, S.D. 737
Clifford, P., see McPherson, K. 243 Cook, J. 1729
Coakley, J., see Malin, H. 1660 Cook, P.J. 543, 1372, 1575, 1579n, 1639, 1641, 1646n,
Coalition on Smoking or Health 1577, 1578 1647, 1648, 1650, 1652, 1655, 1657, 1659, 1662n,
Coase, R. 1374, 1473 1663n, 1679,1688, 1707
Coast, J. 96, 1242 Cook, P.J., see Moore, M.J. 1641, 1642
Coate, D. 1641 Cook, P.J., see Warner, K.E. 1542n, 1574, 1583, 1584,
Coate, D., see Grossman, M. 1551, 1552, 1641 1595
Coate, D., see Lewit, E.M. 299, 1549n, 1550-1552, 1559, Cookson, R., see Dolan, P. 1903, 1905, 1905n
1578, 1588, 1594, 1699 Cookson, R., see Williams, A. 87n, 89n, 568, 1728, 1733,
Cobb-Clark, D.A,, see Wellington, A.J. 688 1806
Cochrane, J. 627, 722, 759 Cooley, T.F, see Bothwell, J.L. 743
Cockburn, I. 133n, 135, 163, 1308 Coope, P., see Chetwynd, J. 1587
Cockburn, I., see Berndt, E.R. 133n, 152, 153, 154n, Cooper, B. 939, 940, 1010n
158n, 159n Cooper, J., see Townsend, J.L. 1547, 1549, 1553, 1573
Cockburn, I., see Ellison, S.F 162n, 1301 Cooper, J.D. 1305
Cockburn, I., see Griliches, Z. 135, 152, 1322n Cooper. M.H., see Culyer, A.J. 2
Cockbum, I., see Henderson, R. 1305 Cooper, P. 665, 668
Cocks, D.L., see Berndt, E.R. 154n, 159n Cooper, R.B. 1210
Coelen, C. 1510, 1517, 1520, 1522 Cooper, T. 1476
Coelen, C., see Gaumer, G.L. 599 Cooper, W.W., see Charnes, A. 1278
Coelli, T.J., see Battese, G.E. 335 Cooter, R. 63n
Coffey, R. 1222 Corder, L., see Manton, K.G. 986
Coffey, R., see Friedman, B. 1514, 1525 Cordes, J.J. 1583
Cohen, A.J., see Hodgson, T.A. 173 Corea, J.,see Brown, R.E. 1685n
Cohen, D., see Edgar, A. 1739 Corlett, W.J. 1663
Cohen, G.A. 1878, 1893 Corman, H. 381, 1084
Cohen. H.A. 1493 Corman, H., see Chaikind, S. 1062, 1066
Cohen, J.W. 134n, 557 Corman, H., see Kaestner, R. 1062
Cohen, M.A. 972, 978 Cornelius, E., see Fries, B.E. 968
Cohen, S.B., see Cohen, J.W. 134n Cornelius, L., see Moy, E. 477
Cohen, W.M. 1318, 1318n Corrigan, J. 1100n
Cohn, R., see Criqui, M.H. 1651 Corts, K. 1476n
Colditz, G.A., see Oster, G. 1575 Cons, K., see Baker, L. 624, 741
Colditz, G.A., see Stampfer, M.J. 1651 Cosgrove, J., see Montgomery, M. 692, 695
Cole, N. 1069, 1071 Cottrell, K. 1234
Cole, R., see Olds, D. 1081 Coughlin, T., see Liu, K. 975. 976
Cole, R.E. 734 Coughlin, T.A. 960
Coleman, K., see Weiner, J.P. 143n, 932 Coulam R., J. Smith 923
Colle, A.D. 582 Coulam, R.F 1529
Collins, E. 1812n Coulam, R.E, see Branch, L.G. 961
Collins, J.F, see Elkin, 1. 904 Coulson, N.E. 315
Collins, J.J. 1647 Coulter, A. 1185
Collins, J.J.,see Harwood, H.J. 1664 Council on Wage and Price Stability 139n
Collins, J.L., see Lowry, R. 900 Cowan, C.A., see Lazenby, H.C. 174n
Collishaw, N.E. 1575 Cowan, C.A., see Levit, K.R. 959, 960
Comanor, W.S. 51 In, 1302, 1328, 1475 Cowan, L.D., see Criqui, M.H. 1651
Comanor, W.S., see Lu, J.L. 1320 Cowell, FA. 1848
Comstock, L., see Hooper, E. 1065 Cowing, T.G. 1452
Cone, K. 1503 Cowing, T.G., see Graham, G.G. 1524, 1526
Cone, K., see Dranove, D. 1512 Cowling, K., see McGuinness, T. 1587, 1593
Conner,M.,seeGrogan, S. 1181 Cox, D. 326
Conner,R. 1117 Cox,D.R. 1210
Conniffe, D. 1561 Cox, D.R., see Box, G.E.P. 29
I-8 Author Index
Cox, H. 1587-1589 Danger, K.L., see Frech 111,H.E. 495, 1471, 1477
Coyle. J., see Calnan, M. 1181 Daniels, N. 1739, 1815n, 1869, 1898
Coyte, P., see Dewees, D. 1344, 1357 Danzon, P. 257, 516, 526, 549, 662, 744, 877, 1020n,
Craswell, R. 1349 1023n, 1025n, 1224n. 1329, 1330n, 1344, 1348, 1350,
Crawford, R.G., see Klein, B. 1473 1354-1362, 1364-1366, 1369,1371-1373, 1373n,
Crawford, V. 1416n 1374, 1375, 1380, 1383-1385, 1387, 1392
Cready, C.M., see Weissert, W.G. 967 Danzon, P.,see Cummins, J.D. 1362, 1363
Creten, D., see Goldstein, A.O. 1592 Danzon, P., see Halrington, S. 1362, 1372
Cretin, S., see Buchanan, J. 729 Danzon, P., see Towse, A. 1344, 1387
Crew, M. 413. 1413, 1413n Darby, M.R. 1416n, 1417
Criqui, M.H. 1651 Dardanoni, V. 383, 393. 394, 1680
Criswell, L.A., see Yelin, E.H. 738 Daugherty, J.C. 155n, 156n
Crivelli, L., see Zweifel, P. 426 Daum, R., see Goldstein, K.P. 1794n
Crocker, K.J. 1473, 1476 Davidson, G., see Moscovice, 1. 975
Cromwell, J. 17, 495n, 510, 919, 1520, 1521, 1526, 1544 Davidson, L.L. 1082
Cromwell, J., see Mitchell, J.B. 485 Davidson, N., see Townsend, P. 1075, 1810n
Cromwell, J., see Sloan, FA. 552, 557, 1075 Davies, A.R., see Ware Jr., J.E. 739
Cromwell, J.. see Wedig, G. 525 Davies, G.N. 1286-1288
Cropper, M. 365, 383, 393, 1679n, 1680, 1682, 1904 Davies, K., see McCarthy, T. 830, 831, 833, 834n
Cropper, M., see Krupnick, A. 1225 Davis, B., see Harris, J.E. 1549, 1595
Cross, PA., see Cassileth, B.R. 1728 Davis, K. 548. 581, 1498, 1810, 1822
Croxson, B. 2, 1187 Dawson, S., see Barr, R.D. 1745
Csaba, 1. 875 Day, N., see Larkby, C. 1646
Cuddington, J.T. 1767n Day, P., see Klein, R. 1891
Cuenca, E., see Manau, C. 1285, 1286, 1288, 1289 deAbajo, F. seeBakke, O.M. 1315n
Cuffel, B. 901 de Graaff, J.V. 57, 62n
Cuffel, B., see Scheffler, R.M. 912 de Haan, A.F.J., see Verhoef, L.C.G. 1741
Cullen, ET., see Link, B.G. 898, 899 de Haes, J.C.J.M., see Stiggelbout, A.M. 1741, 1746,
Cullis, J.G. 1, 1223, 1232, 1233n, 1235, 1835n 1747
Cullis, J.G., see Culyer, A.J. 1214, 1240. 1242n de Jong, P., see Aarts, L. 1002, 1003, 1021n, 1037, 1038,
1038n, 1039, 1039n
Cullis, J.G., see Jones, P.R. 1242
de Jong, P., see Haveman, R. 1024
Culp, W.J., see Wennberg, J.E. 240
de Lange, B., see Schokkaert, E. 806, 830, 834n
Culyer, A. 2, 13, 18-20, 22, 23, 26, 58, 63, 64, 64n,
de Lissovoy, G., see Gabel, 1. 719
65-67, 70, 71-73, 87n, 91, 92, 92n, 93, 95, 103, 505,
2 de Lissovoy, G., see Weiner, J.P. 711.712, 718
1214, 1240,124 n, 1727.1731,1736, 1807, 1808.
De Meza, D. 577,1147
1809n, 1813,1814, 1814n. 1815, 1817, 1869, 1880,
de Serpa, A.C. 1226
1889, 1894
Deakin, S., see Arrighetti, A. 1179
Culyer, A., see Blades, C.A. I
Dean, D. 1018n
Cumella, E.J., see Epstein, A.M. 780, 782, 783, 785. 806 44
Dean, D., see Berkowitz, E. 10 n
Cummings, K.M., see Lewit, E.M. 1553
Deardorff, A.V. 1319
Cummins, J.D. 1362. 1363
Deb, P. 315,321,927
Currie, J. 367, 676n, 684, 1060, 1062, 1068-1073, 1075,
DeBakey, S.F 1649, 1649n
1077n, 1080, 1708n
Debrock, L.W., see Arnould. R.J. 733
Currim, LS. 1748 DeCicca, P. 1554, 1555
Curtis, M., see Harrington, C.A. 958, 965, 966
Decker, S. 683, 684, 1060, 1075
Curtis, P., see Rogers, J. 1178 Dee, T.S. 1554, 1555
Custer, W.S., see Willke, R.J. 602 DeFelice, L.C. 494
Cutler, D. 135, 159, 163, 173, 237n, 602, 603, 614, 616, DeForge, B.R., see Sobal, J. 1682
616n, 621,622, 624, 627, 628,630, 630n, 650, 651, DeFriese, G., see Kottke, T. 1079, 1711
656, 661n, 686, 687, 687n, 692, 696, 719, 722, 728, DeGrandpre, R.J. 1564
729, 741,743,759,773,865,932,978-980, 1074, DeGrandpre, R.J., see Bickel, W.K. 1563, 1564
1121 Delong, G. 1031n
Cutler, D., see Altman, D. 618, 797n Dekay, M., see Ubel, P. 88
Cutler, D., see Berndt, E.R. 119 Delahoz, K.E, see Regidor, E. 1222n
Delea, T.E., see Oster, G. 1544
Dalen, D.M., see Grytten, J. 1277, 1283 Delgado-Rodrigues, M., see Llodra, J.C. 1288, 1289
Daly, E. 1748 DeMaio, S., see Charles. C. 75
Daly, M. 1010 Demateo, S., see Regidor, E. 1222n
Daly. M., see Burkhauser, R. 1000, 1002n Demsetz, H. 1446
Damberg, C., see Buchanan, J.L. 728, 729, 733 Demsetz, H., see Alchian, A. 1144, 1190
Damberg, C., see Mauldon, T. 736 Deneckere, R. 1412n
Damsgaard, M.T., see Krasnik, A. 1189 Deolalikar, A. 1806n
Danger, K.L. 1477 Deolalikar, A., seeBehrman, J. 1055
Author Index I-9
DePalma, A., see Anderson, S.P 1411, 1436 Donaldson, C. 1, 57n, 101n, 278, 1187, 1188, 1285,
Department of Health 1242, 1243n 1288, 1290, 1734
Department of Health and Human Services 238n Donaldson, C., see Mooney, G. 64n, 89, 92, 1814, 1814n,
Department of Health and Social Security 1241n 1893
Department of Health. Education and Welfare 523 Donaldson, D., see Blackorby, C. 96n, 1898
Derrick, FW., see Nguyen, N.X. 504n, 515 Donaldson, L.J. 1241n
Derzon, R.A., see Lewin, L.S. 1156 Donat, P.,see Riportella-Muller, R. 1060
Desai, K., see Young, G.J. 1160 Donham, C.S., see Levit, K.R. 959, 960
DesHarnais, S.I. 598, 601, 630, 863-865 Donovan, J., see Coast, J. 96, 1242
DeSimone, J., see Philipson, T. 1796n Doppmann, R.J., see Leu, R.E. 448
Desvousges, W.H., see Smith, V.K. 1701 Dor, A. 280, 334, 1452
Detsky, A.S. 261 Dor, A., see Brooks, J. 1130, 1160, 1449, 1467n, 1469
Deverill, M., see Brazier, J. 1744 Dorfman, R. 1109n, 1303
Devlin, H.B. 1243 Doroodian, K., see Seldon, B.J. 1587, 1589
Dewa, C.S., see Frank, R.G. 1709 Dorrington, C., see Zambrana, R. 1060
Dewees,D. 1344, 1357 Dorwart, R., see Schlesinger, M. 873, 1160, 1161, 1163
Dhaene, G.. see Schokkaert, E. 806, 830, 834n, 835 Doty, PA. 962, 963,984
Dhir, R., see Leigh, J.P. 400, 401 Douglas, G. 1412
Diamond, P. 586n, 759, 933, 1021n, 1022n, 1662 Douglas, R.M., see Steven, I.D. 1181
Dick, A. 959 Douglas, R.M., see Veale, B.M. 1183
Dickens, W. 666n Douglas, S. 317, 325, 328, 1553, 1554, 1561
Dickert, J., see Scott, E. 1577 Douglass, C.W. 1288, 1290
Dickey, B. 923 Douglass, C.W., see Niessen, L.C. 1285-1289
Dickey, B.,seeFisher,W.H. 873 Dow, W. 1055, 1774, 1775
Dickey, T.S., see Hirsh, H.L. 516 Dowd, B. 279, 280, 557, 619, 1689, 1690
Dowd, B., see Boardman, A.E. 1184, 1185
Diehr, P. 242, 922
Dowd, B.,seeConner,R. 1117
Diehr, P., see Martin, D. 736, 1177
Dowd, B., see Connor, R.A. 1454
Dietrich, A.J. 1179
Dowd, B., see Feldman, R. 79, 279-281, 475, 614, 620,
Diewert, W.E. 139n
624, 711, 713, 719, 727-729, 740, 741
DiFranza. J.R. 1592
Dowd, B., see Johnson, A.N. 735
DiFranza, J.R., see Rigotti, N.A. 1597
Dowd, B., see Zellner, B.B. 821n
Dill, A.P. 937
Dowie,R. 1186
Diller, M. 1012n
Dowling, W. 1508
DiMasi,J.A. 1309n, 1310
Downey, L.A. 1598
Dingwall, R., see Ham, C. 1387
Dranove, D. 73n, 155n, 274, 468, 473, 475, 476, 477n,
Dionne, G. 76, 227, 235, 1184, 1185, 1413 484n, 490, 493,499,501,507, 507n, 508n, 510, 511,
Dionne, G., see Doherty, N.A. 1364 518,652, 726, 743, 854, 858, 867, 871,887, 1064,
Director, A. 1474 1097, 1107, 1109n, 1112, 1117, 1118, 1120, 1123,
Disney, R. 1001, 1002n 1124,1126, 1128-1131, 1131n, 1132,1133,1158,
Dittus, R.S., see Fitzgerald, J.E 598, 630, 734 4
1168n. 1184, 1327, 1 07n, 1413n, 1414. 1416n, 1429,
Dixit, A. 1103n, 1411, 1558 1445, 1448-1450, 1450n, 1452, 1512, 1702, 1711
Dixon, J. 1241 Dranove, D., see Cone, K. 1503
Dixon, P. 1181 Draper, B. 1653
Dixon, S., see Hoyme, E. 1066 Draper, D., see Keeler, E.B. 1162, 1163
Dobson, A., see Gutennan, S. 597 Dreyfus, T., see Kronick, R. 790, 791, 798, 814, 834
Dobson. A., see Weiner, J.P. 143n, 798, 799, 811n, 814, Drummond, M. 100, 102n, 103, 105, 188, 205, 1284,
932 1285, 1291
Dobson,E 1386 Drummond, M., see Buckingham, K. 1735
Docherty, J.P., see Elkin, I. 904 Duan, N. 286, 287, 290, 291, 788, 1265, 1267, 1273,
Doering, C.R., see Lombard, H.L. 1545 1638n
Doessel, D.P. 1286-1288 Duan, N., see Bailit, H. 1274
Doherty, N. 712, 1280, 1288, 1364 Duan, N., see Manning, W.G. 23, 286, 288, 438, 439,
Dolado, J., see Banerjee, A. 41 1263-1265, 1267, 1285, 1286, 1288, 1289
Dolan, P 87r, 96, 101,212, 215, 1184, 1728, 1735, 1738, Duan, N., see Wells, K.B. 911
1740-1742, 1745-1748, 1752, 1753, 1865n, 1872n, Dubay, L. 687, 687n, 1074
1902, 1903,1903n, 1905, 1905n Dubay, L., see Cohen, J.W. 557
Dolan, P., see Gudex, C. 1740 Dubois, R.W., see Carter, G.M. 798n
Dolan, R.C., see Dean, D.H. 1018n Dubois. V., see van de Kar, A. 1181, 1182
Doll, R. 1545, 1593 Dubowitz, H., see Black, M. 1081
Domino, M., see Salkever, D. 1033n Duckett, S. 1234, 1235
Don, B., see Goldacre, M.J. 1228, 1234 Duesenberry, J.S. 1558
Don, B., see Lee, A. 1232n Duffy, M. 1561, 1586-1588
DonabedianA. 1112n, 1441n, 1809, 1815n Dufour. M.C. 1652, 1710
1-10 Author Index
Dufour, M.C., see DeBakey, S.F 1649, 1649n Eklund. G., see Wasserman, D. 1653
Dufour, M.C., see Stinson, FS. 1632 El-Safty, A.E. 1557
Dufour, M.C., see Williams, G.D. 1632 Eldenburg, L. 864, 1427
Dufresne, E., see Hamilton, V.H. 302 Elder, J.P., see Flewelling, R.L. 1548
Duke University Center for Health Policy 1166 Eldridge, L.P. 138n
Dulberger, E.R., see Boskin, M.J. 130n Eleazer, G.P., see Eng, C. 961, 962
DuMouchel, W.H. 1646n, 1648 Elekes, Z., see Skog, O.-J. 1653
DuMouchel, W.H., see Sisk, J.E. 737 Elixhauser, A., see Brown, R.E. 1685n
Duncan, A.S. 272, 273, 298 Elkhatib, M.B., see Garg, M.L. 516
Dunet, D.O., see Haddix, A.C. 1686, 1705 Elkin, E.P., see Kupperman, M. 1743
Dunkelberg, W., see Staten, M. 1126, 1130, 1449, 1463, Elkin, I. 904
1464n Ell, K., see Zambrana, R. 1060
Dunmeyer, S., see Rice, D.P 1660, 1664 Ellis, R.P. 143n, 449, 487. 491n, 522, 522n. 551, 555,
Dunn, D. 786, 808, 810, 829, 831,834, 934n 596, 603,606, 607, 616, 619, 620, 715,723, 724, 773,
Dunn, W.L. 1498n 780, 784, 786,788, 789, 791,798, 800, 803, 806-810,
Dunne, J.P. 48 81 n, 814, 819, 822, 827, 835, 856, 856n, 857. 863,
Duran-Arenas, L., see Billi, J.E. 729 864, 866, 867, 871,872, 911,917,919,920,927,928,
Dusansky, R. 968 932, 1119, 1120, 1167, 1350, 1354, 1361
Dusenbury, K., see Maibach, E. 623 Ellis, R.P., see Ash, A.S. 784, 786, 797, 798, 800, 810,
Dusenbury, K.. see Neuman, P. 83 811.811n, 812, 814
Dustmann, C. 313, 323 Ellis, R.P., see Cromwell, J. 919
Duston. T.E. 712 Ellis, R.P.. see Freiman, M.P 863, 864, 920
Dworkin, R. 89n, 1892, 1893 Ellis, R.P., see Harrow, B. 904, 920n
Dwyer, D. 1025n Ellis, R.P, see Jennison, K. 492, 522
Dyckman. Z. 465n, 504 Ellis, R.P, see Pope, G.C. 798, 800. 811n, 812
Dyer, J.S. 1735, 1737 Ellis, R.P., see Shen, Y. 816, 933
Dykacz, J. 1026 Ellis, R.P., see van de Ven, W.PM.M. 83, 625, 744, 929n.
Dyson, R. 1280 1180, 1188
Ellison, S.F 162n, 1301
Ellwood, D. 682. 683
Eakin,B.K. 1513, 1514, 1526
Elsten, H., see Weinstein, M. 478n
Earls, F. 1082
Elster, J. 95, 1546, 1556
Easley, D. 1149 Elston, J.M., see Weissert. W.G. 960
Eastaugh, S. 464n Elzinga, K.G. 1423, 1428
Easterling, D.V., see Sager, M.A. 600 Emanuel, E.J. 521
Eaton.C. 1102n Emons, D.W. 474, 488,493
Eaton, K., see Widstrom, E. 1255, 1257 Emons, W. 500, 1416n
Eberts, R. 690, 691, 694 Employee Benefit Research Institute 647-651, 663, 676
Echeverria, M.U., see Picone, G. 395n Encinosa, W. 817, 933, 1191, 1477
Eddy, D. 1350, 1683 Eng, C. 961, 962
Edelman, P.L., see Hughes, S.L. 961 Engberg, J., see Wholey, D. 744
Edgar, A. 1739 Enggard, K. 19n
Edlin, A.S. 859 Engle, R. 40, 43
Edlund, M., see Kessler, R.C. 900n, 901.90 n English, E.C., see Oppenheim, G.L. 484n
Edwards, G. 1633. 1634, 1638, 1638n, 1649. 1658, 1660 Ennett, S.T., see Baumann, K.E. 1643
Edwards, G., see Bruun, K. 1649 Ennett, S.T., see Norton. E.C. 306. 1644
Edwards, L. 1062 Ennew, C. 1193
Edwards, L., see Shakotko, R. 1062 Ensor. T., see Gold, M.R. 714, 715
Edwards, R.T. 1235, 1240, 1241, 1241n Enterline,P. 430n
Efron, B. 790 Enterline, P., see Ricci, E.M. 430n
Egerter, S., see Braveman, PA. 733 Enthoven, A. 719, 724, 725, 760. 778, 822, 827, 1115,
Eggers, P. 617, 730 1125
Eggers, P., see Beebe, J.C. 777n, 798 Entman, S. 1360
Ehrenberg, R. 662n, 690n, 692, 695 Environmental Protection Agency 1546, 1581, 1581n,
Ehrenberg, R., see Smith, R. 690n 1582, 1596
Ehrlich, I. 256, 351, 361, 363. 375n, 390, 413. 541, 1364, Epstein, A. 492, 780, 782, 783. 785, 806
1680, 1682, 1683, 1686, 1688 Epstein, A., see Berndt, E.R. 154n, 159n
Eichenholz, J., see Altman, S.H. 1498 Epstein, A., see Gatsonis, C.A. 306
Eichner, M. 572. 583, 1059n, 1415 Epstein, A., see McPherson, K. 243
Eijkemans, M.J.C., see Stiggelbout, A.M. 1747 Epstein, A., see Rosenthal, M. 474
Eisenberg, J.M. 475, 482, 505, 521 Epstein, A., see Schneider, E.C. 75
Eisenberg, J.M., see Boardman, A.E. 1184, 1185 Epstein, A., see Stern, R.S. 737
Eisenberg, T., see Henderson, J.A. 1356 Epstein, A., see Udvarhelyi, I.S. 737
Ekelund, R.B., see Saba, R.PE 1570 Epstein, A., see Weissman, J. 570, 629
Author Index I-11
Epstein, R. 1379, 1381 Feenberg, D., see Feldstein, M. 689

Epstein, S., see Schlesinger, M. 1160, 1161, 1163 Feeny, D. 102n, 105, 1732, 1745
Epstein, W.V, see Yelin, E.H. 738 Feeny, D., see Barr, R.D. 1745
Eraker, S.A.,see Miyamoto, J.M. 1730, 1740 Feeny, D., see Kupperman, M. 1743
Erbsland, M. 300, 383,387 Feeny, D., see Torrance, G.W. 1734
Erickson, P., see Patrick, D.L. 216 Feigenbaum, B., see Lindsay, C.M. 1214
Ermann, D. 717,720 Feigenbaum, P.G., see Yelin, E.H. 738
Ermann, D., see Gabel, J. 719 Feighan, T., see Ennew, C. 1193
Erner, B.K., see Palmer, R.M. 600, 630 Fein, R. 3, 98
Emshoff, J.R., see Ackoff, R.L. 1587 Feinstein, J.S. 961
Emster, V.L., see Warner, K.E. 1587, 1588 Felder, S., see Zweifel, P. 451, 984, 985
Erwin, W.G., see Carroll, N.V. 597 Feldman, B. 1223
Escarce, J. 325, 327, 474, 476, 478n, 495, 514, 714 Feldman, H.A., see McKinlay, J.B. 506
Eshleman, S., see Kessler, R.C. 897 Feldman, R. 58, 68, 75, 78, 78n, 79, 109, 226, 279-281,
Esposito, A., see Greenwald, L.M. 829 463n, 475, 476, 489n, 510,511,514n, 525, 614, 620,
Essink-Bot, M.L., see Krabbe, P.F.EM. 1746 624, 659n, 711,713,719, 727-729, 740, 741,744,
Etter, J.E, see Perneger, T.V. 927 1463
Ettner, S.L. 778, 810, 815, 898, 899, 933, 934, 967, 974 Feldman, R., see Christianson, J.B. 742, 744
EuroQol Group 102 Feldman, R., see Conner, R. 1117
Euwals, R., see Soest,. A. van 1005, 1005n Feldman, R., see Connor, R.A. 1454
Evans, A., see Holahan, J. 720 Feldman, R., see Dowd, B. 279, 280, 557, 619
Evans,N. 1592 Feldman, R., see Kralewski, J.E. 727
Evans, R. 465, 503, 507, 520, 526, 1113 Feldman, R., see Wholey, D. 624, 744
Evans, R.G. 1, 17, 27n, 57, 63, 67n, 68, 68n, 72. 74-76. Feldstein, M. 3, 63, 134, 134n, 163n, 166n, 332, 414,
79,84,86,107-109,414, 1184,1185, 1260, 1847n 436, 465n, 520, 522n, 523n, 524, 580, 581, 587, 661n,
Evans, R.G.,see Barer, M.L. 172, 173n 689, 1022n, 1113, 1152n, 1185, 1218, 1492, 1505n
Evans, R.G., see Culyer, A.J. 63 Feldstein, P. , 163, 464n, 581, 717, 741, 1123, 1254
Evans, R.G., see Stoddart, G.L. 85 Feldstein, P., see Buchmueller, T.C. 275, 621,767
Evans, W. 1446, 1553, 1555, 1578, 1596, 1597 Feldstein, P., see Wickizer, T.M. 717, 1123
Evans, W., see Dee, T.S. 1554, 1555 Felt, S., see Gold, M. 1529
Evans, W., see Viscusi, W.K. 1734 Fendrick, A.. see Chemew, M. 1456
Fendrick, M.A., see Chernew, M. 742
Experton, B. 734
Fenn, P. 874, 875, 1344
Eyles, J., see Birch, S. 83n, 92, 1835n, 1839
Fenn, P., see Csaba, I. 875
Eyles, J., see Hutchison, B. 83n
Fernandez, M.P., see Gervas, J. 1177, 1188
Eze, P. 602
Ferris, J., see Room. R. 1657
Festinger, L. 1738
Fabian, R., see Tolley, G.S. 1687, 1706 Fiester, S.J., see Elkin, I. 904
Fabien, C., see Leclerc, A. 1852n Files, A. 830
Fagan, J. 1652 Finch, M. 904
Fagan, L.E, see Fitzgerald, J.E 598. 630 Finch, M., see Dowd, B. 279, 280
Fairbank, A. 904 Finch, M., see Feldman, R. 279-281, 620, 711, 728, 729
Fanara, P. 1504 Finch, M., see Lurie, N. 735,925
Fanshel, S. 101 Finch, M., see Nyman, J.A. 960
Farber, H.S. 1357, 1358 Fineberg, H.V., see Read, J.L. 1746
Fare, R. 1279, 1279n Fineberg, H.V., see Weinstein. M. 478n
Faresjb. T., see Jnsson, B. 1281 Fink, B., see Myerberg, D.Z. 1077
Farkas, A. 1563 Finkler, S. 207, 1115
Farkas, A., see Evans, N. 1592 Fiore, M.C., see Cromwell, J. 1544
Farley, D.E. 1427n Fireman, B.H., see Selby, J.V. 433
Farley, D.E., see Dor, A. 1452 Fischer, E., see Kelleher, K. 898, 899
Farley, D.O. 814 Fischer, M., see Currie, J. 1060
Farley, P. 468, 477, 523, 619, 1185, 1364 Fischoff, B. 1753
Farley, P., see Monheit, A. 690, 691 Fishback. P.V. 693n
Farrell, J. 494 Fishe, R.P.H., see Conroy, P. 1795
Farrell, M.J. 1558 Fisher, A., see Smith, V.K. 1701
Farrell, P. 399, 400, 1682 Fisher, C.R. 602
Farrelly, M.C. 1553, 1573 Fisher, F. 129n, 131,135, 139n, 1444
Farrelly, M.C., see Evans, W.N. 1555, 1596, 1597 Fisher, I. 1654
Fattore, G. 1330 Fisher, P.M., see Goldstein, A.O. 1592
Faustman, W.O., see Lenert, L.A. 1753 Fisher, W.H. 873
Feder, J. 597, 965 Fitzgerald, J.E. 598, 630, 734
Feder, J., see Hadley, J. 600, 1124 Fitzpatrick, R. 1751
Federal Trade Commission 1451, 1568, 1584 Fixler, D. 147n, 148n, 149n, 155n, 168
1-12 Author Index
Flacco, P.R., see Baldwin, M.L. 1009n Frankel, S.J. 1218, 1233, 1234, 1242, 1243
Fleiss, J. 259n Franks, P. 1177
Fleming, D. 1184 Fray, J., see Cole, R.E. 734
7
Fleming, G.V.,seeAday,L.A. 1812n Frazier, H., see Weinstein, M. 4 8n
Fleming, S., see DesHarnais, S.I. 598, 630, 863 865 Frech III. H.E. 470, 471n, 474, 475,495, 505, 511 n, 546,
Flewelling. R.L. 1548 1407n, 1413,1416, 1445, 1463,1471, 1475n, 1477
Flierman, H.A., see Krasnik, A. 1189 Frech III, H.E., see Comanor, W.S. 1475
Flint, S., see Yudkowsky, B. 1060 Frech III, H.E., see Danger; K.L. 1477
Foch, C. 1286 Frech III, H.E., see Mobley, L.R. 1424n, 1435
Foege, W.H., see McGinnis, J.M. 1677, 1697 Freed, A.J., see McLaughlin, C.G. 740, 741
Fogel, R.W. 1678, 1712 Freed, G. 1065
Folberg, HJ., see Rosenberg, J.D. 1376 Freeman,G.K. 1179, 1183
Foley, W.J., see Fries, B.E. 968 Freeman, J.L., see Wennberg, J.E. 240
Folland, S. 1,67, 70n, 78, 464n, 475, 477n, 601 Freiberg, L. 433
4
Folland, S., see Stano, M. 2 4n Freiman, M.P. 863, 864, 920, 934
Fontein, P., see Soest, A. van 1005, 1005n French, M.T. 1655
Food and Drug Administration 1592 Freund, D.A., see Palmer, R.M. 600, 630
Forbes, J., see Donaldson, C. 1285, 1288, 1290 Frey, B.S. 1190
Forbes, W.F. 1575 Frey, R.L.,see Leu,R.E. 1811n
Ford, A.B., see Katz, S. 975 Fried, C. 1884n
Ford, I.K. 137n, 154n, 155n, 156n, 158n, 161, 161n Friedman, B. 1125n, 1453, 1514, 1525
Forder, J. 964, 972 Friedman, B., see Feldstein, M.S. 587
Foreman, S.E. 1463 Friedman, E.S. 718
Forster, J.L. 1597 Friedman, G.D., see Klatsky, A.L. 1651
Forster, M. 325-327 Friedman, M. 472
Forstner, H., see Ballance, R. 1299n, 1300, 1304, 1318, Fries, B.E. 968
1321, 1329 Fries, B.E., see Ikegami, N. 968
Forsyth, G. 1240n Fries, J.F. 1678
Fortus, R., see Berki, S.E. 728, 729 Fries, T., see Cooper, T. 1476
Fossett, J. 1075 Frisch, R. 384, 392n
Foster,R. 1123n Frisman. L.K. 941
Foster, W.E., see Babcock, B.A. 1603 Fritschler, A.L. 1545n, 1603
Fountain, D., see Harwood. H. 1664
Froberg, D.G. 1745, 1747, 1748
Fournier, G.M. 1452
Froeb. L., see Evans, W. 1446
Fowler, E., seeFowles, J.B. 805
Froeb, L.M. 1427, 1430
Fowler, E.J. 814
Froeb, L.M., see Werden, G.J. 1433
Fowler, F.J. 216
Frost, C.E.B. 1219, 1228, 1236, 1236n
Fowler, F.J., see Barry, M.J. 216
Fry, J. 1177
Fowles, J.B. 805
Fry, J., see Armstrong, D. 1184
Fox, C.H., see Douglass. C.W. 1288, 1290
Fryer, J., see Carey, T.S. 733
Fox, D. 96
Fuchs,D.,seeFuchs, L.S. 1036n
Fox, N., see Eng, C. 961, 962
Foye, H.R., see Szilagyi, PG. 737 Fuchs, L.S. 1036n
Francis, B.J.,see Frost, C.E.B. 1219, 1228, 1236 Fuchs, V. 1 27n, 107, 127n, 349. 397, 398, 401 436,
Francis, J., see Klarman, H. 101 463, 465, 469,476, 504. 507, 510,518-520, 581,985,
Francis, P.J. 1695, 1778n 1095, 1113,1492, 1682, 1777n
Frank, R. 157n, 429n, 469, 474, 477n, 510, 522n, 595, Fuchs, V., see Farrell, P. 399, 410, 1682
597, 599, 609n, 711n, 712, 722, 728, 740, 778, 815, Fudenberg, D. 883
816, 854, 864, 899, 901,904, 906, 909, 913, 915n, Fuhr, J.P., see Blackstone, E.A. 1477
916, 920, 924-926, 928, 930n. 931,938-941, 1084, Fujii,E.T. 1547, 1587, 1593. 1594
1160, 1322, 1323, 1445n, 1709 Fulton, G.A.,see Warner. K.E. 1600n, 1607-1609
Frank, R., see Berndt, E.R. 119, 133, 163, 898, 903-904 Furlong, W., see Barr, R.D. 1745
Frank, R., see Brisson, A.S. 912, 936 Furlong, W., see Feeny, D. 1732, 1745
Frank, R., see Ettner, S.L. 778, 810, 815. 898, 899, 933,
934 Gabel, J.R. 526, 713, 719, 725
Frank, R., see Kessler, R.C. 900n, 901, 901n Gaffney, S., see Jensen, G.A. 712
Frank, R., see Rosenthal, M. 474 Gafni, A. 100, 103, 105, 212, 1734. 1735, 1737
Frank, R., see Lave, J.R. 598,920 Gafni, A., see Birch, S. 198
Frank, R., see Salkever, D. 872 Gafni, A., see Bleichrodt, H. 1751
Frank, R.H. 108 Gafni, A., see Donaldson, C. 101n
Franke, G.R. 1587, 1589 Gafni, A., see Levine, M. 75
Franke, G.R., see Andrews, R.L. 1586 Gafni, A., see Mehrez, A. 102, 103, 212, 1729,
Frankel, J., see Coast, J. 1242 1735-1737, 1742
Frankel, S., see Coast, J. 96, 1242 Gafni, A., see O'Brien, B. 100
Author Index I-13
Gaisford, J., see McCarthy. T. 830, 831, 833, 834n Gertler, P., see Dor, A. 280
Gal-Or, E. 1120, 1477 Gertler, P.,see Dow. W. 1055
Galbraith, J.K. 1558 Gertler, P., see Frank, R.G. 899
Galbraith, J., see Banerjee, A. 41 Gertler, P., see Gaynor, M. 465, 476, 487n, 494, 715,
Gale, F. 1599, 1600, 1600n, 1601, 1609, 1610 1191
Gallup, C.L., see Ellis, R.P 1350, 1354, 1361 Gertler, P., see Stern, R.S. 737
Galves, R., see Llodra, J.C. 1288, 1289 Gertler, P., see Thorpe, K.E. 557, 968
Gambardella, A. 1307n Gertman, P.M. 440n
Ganem, J. 471 Gervas,J. 1177, 1188
Garber, A.M. 96, 104, 193, 195, 199, 209n, 236, 256n, Getzen, T.E. 1, 20, 154, 154n, 155n, 451, 464n, 470, 475,
261,395n, 791, 975, 977, 978, 1284, 1285, 1705 517
Garber, A.M., see Dick, A. 959 Getzen, T.E., see Kendix, M. 20
Garber, A.M., see Lenert, L.A. 1746, 1753 Geurts, J., see Kunst, A.E. 1853n
Garber, J.E., see Palmer, R.M. 600, 630 Geurts, J., see van Doorslaer, E. 508n
Garber, S. 1320 Ghert, L.S., see Headen, A.E. 676, 677, 679n
Garcia, J. 298 Ghez, G.R. 357n, 362n, 369n
Garcia dos Santos, J. 1558 Giacomini, M. 83, 1190
Gardiner, J.A., see Downey, L.A. 1598 Gifford, G., see Feldman, R. 740, 741
Gardner, L.B., see Gamick, D.W. 734 Gilbert, M. 127n
Gardner, L.B., see Robinson, J.C. 728 730, 829n Gill, A.M. 1836n, 1837n
Garfinkel, S.A., see Zarkin, G.A. 1690 Gill, K.C., see Marmor, T.R. 940, 941
Garg, M.L. 516 Gilleskie, D.B., see Blau, D.M. 676, 678, 679
Gamick, D. 734, 1132n, 1431, 1433 Gillon, R. 1807, 1808, 1815n, 1894n
Garnick, D., see Luft, H.S. 1434, 1450 Gilman, B., see Pope, G.C. 798,81 In, 812
Gamick, D., see Robinson, J. 1116, 1450, 1522 Gilpin, E., see Evans, N. 1592
Garrett, J., see Carey, T.S. 733 Gilpin, E.A., see Pierce, J.P. 1592
Gaskin, D. 742 Ginsburg, D.H. 156n
Gatsonis, C.A. 306 Ginsburg, D.H., see Armknecht, P.A. 156n, 160n
Gaumer, G.L. 599 Ginsburg, D.H., see Ford, I.K. 137n, 154n, 156n, 158n
Gaumer, G.L., see Coulam, R.F 1529 Ginsburg, M., see Fixler, D. 147n, 148n, 149n
Gaumer, G.L., see Staiger, D. 603, 630 Ginsburg, P.B. 1158, 1497, 1498, 1509, 1510, 1516,
Gauthier, A.K. 835 1527, 1530
Gauthier, A.K., see Rogal, D.L. 836 Ginsburg, P.B., see Frech HI, H.E. 475, 546
Gavazzi, M., see Fries, B.E. 968 Giovino, G., see Zhang, P. 1604
Gaynor, M. 73n, 304, 333, 463n, 465, 469, 473, 475-477, Githens, P.B., see Sloan, FA. 1373n, 1663
487n, 494, 499, 499n, 502, 714, 715, 743, 1103n, Gittelsohn, A., see Wennberg, J. 240
1130, 1191, 1407n, 1409, 1413, 1413n, 1416, 1453, Giuffrida, A. 484n
1465, 1470n, 1471, 1477 Gius, M.P. 1645
Gaynor, M., see Abraham, J. 1456 Given, R. 743
Gaynor, M., see Encinosa, WE. 1191 Glantz, S.A. 1546, 1581
Gaynor, M., see Haas-Wilson, D. 493 Glantz, S.A., see Tye, J.B. 1592
Gaynor, M., see Frank, R.G. 941 Glaser, W.A. 1145
Gaziano, J.M., see Camargo Jr., C.A. 1651, 1652 Glass, C., see Entman, S. 1360
Gbsemete, K. 18,22 Glass, D. 864
Gehlbach, S., see Lee, A.J. 506n Glass, D.R., see Elkin, I. 904
Gehlbach, S., see Rapoport, J. 737 Glazer, J. 482n, 485, 486, 486n, 487, 773, 815, 856, 913,
Geil, P. 315, 319 933, 1120, 1416n
Genuardi, J., see Wrightson, W. 619 Glazer, J., see Frank, R.G. 595, 712, 722, 728, 816, 913,
Geoffard, P. 1695, 1696, 1763n, 1769-1771. 1771n, 915n, 930n
1778n, 1779n, 1783n Glennerster, H., see Matsaganis, M. 829n, 830
Gerard, K., see Donaldson, C. 1, 57n, 1187, 1188 Glesnes-Anderson, V., see Anderson, G. 520
Gerard, K., see McGuire, A. 19, 23, 48 Gliebe, W.A.. see Garg, M.L. 516
Gerard, K., see Mooney, G. 64n, 89, 92, 1893 Glied, S. 123, 412n, 449, 591, 604, 650, 712, 728, 1191n,
Gerdtham, U,-G. 11, 18 20, 22, 24, 24n, 26, 27, 30, 32, 1325
37-39, 45, 46, 48, 49, 268, 316, 319, 320, 412, 451, Glied, S., see Bloom, D.E. 1786n
849n, 1840, 1854 Glied, S., see Sisk, J.E. 737
Gerdtham, U.-G., see Gbsemete, K. 18, 22 Globerman, S. 1223, 1233n, 1234
Gerdtbam, U-G., see Johannesson, M. 1904 Glover, A.E 239
Gerety, M.B. 599, 630 Glover, J. 1894n
German, P., see Shapiro, S. 936 Gluck, M.E., see Garber, A.M. 209n
Geroski, P. 1427n Glynn, R.J., see Camargo Jr., C.A. 1651, 1652
Gerstein, D.R., see Moore, M.H. 1660 Godber, E. 1192
Gertler, P. 280, 283, 556, 968, 1121, 1222 Goddard, J. 1240, 1243
Gertler, P., see Allen, R. 857, 867, 871, 1119 Goddard, M. 1179
I-14 Author Index
Goddeeris, J. 413,723, 1167 Graham, E.A., see Wynder, E.L. 1545, 1593
Goddeeris, J., see Ballard, C. 662 Graham,G.G. 1524, 1526
Godfrey, C. 1663 Graham, J.D.. see Tengs, T.O. 1705
Godfrey, C., see Buck, D. 1608 Granger, C., see Engle, R. 40, 43
Godfrey, C., see Sutton, M. 278 Granger, C.W. 42
Goel, R.K. 1586, 1587, 1589, 1594 Grannemann, T.W. 1157, 1452
Goel. R.K., see Baltagi, B.H. 1548 Grant, B.F., see DeBakey, S.F. 1649, 1649n
Gouveia, M., see Besley, T. 20 Gravelle, H. 1687, 1706
Gold, M. 183, 188, 212, 213, 714, 715, 1284, 1529, 1705, Gravelle, H., see Dixon, P. 1181
1739, 1752 Gravelle, H., see Giuffrida, A. 484n
Goldacre, M.J. 1228, 1234 Gravelle, J.G. 1580, 1582
Goldacre, M.J., see Lee, A. 1232n Gray, A., see Daly, E. 1748
Goldacre, M.J., see Newton, J. 1228 Gray, A.M. 1281
Goldberg, G., see Bailit, H. 1274 Gray, B.H. 1144, 1147, 1155n, 1168
Goldberg, G., see Lillard, L. 1708 Gray, J. 1891
Goldberg, G., see Manning, W.G. 728, 729, 732, 735 Gray, P.G. 229
Goldberg, G., see Ware Jr., J.E. 739 Gray,W. 1017n
Goldberg, L.G. 718, 742 Green, J. 511, 1416n
Goldberger, A.S. 383, 398 Green, P., see Bentley, J.M. 1280
Goldenhar, L.M., see Warner, K.E. 1585 Green, R., see Phillips, C.D. 961
Goldfarb, R.S., see Suranovic, S.M. 1562, 1584 Green, R.M. 1897n
Goldman, B.S., see Naylor, C.D. 1240 Greenberg, I., see Babor, T.F. 1642. 1643
Goldman. D.P. 729 Greenberg, P.E., see Berndt, E.R. 151
Goldman, D.P., see Garber, A.M. 209n Greenberg, W. 493
Goldman,E 583 Greenberg, W., see Fanara, P. 1504
Goldman, H., see Lave, J.R. 598
Greenberg, W., see Feldman, R. 1463
Goldman, H.H. 903
Greenberg, W., see Goldberg, L.G. 718, 742
Goldman, H.H., see Jencks, S.F 933.934
Greene, W. 29, 30, 1262n, 1264
Goldman, L., see Emanuel, E.J. 521
Greenfield, S. 492, 734
Goldman, L., see Pearson, S.D. 737
Greenfield, S., see Kaplan, S.H. 1179
Goldman, P., see Thorpe, K.E. 557, 968
Greenhouse, J.B., see Needleman, H.L. 1066
Goldman, W. 906, 912, 924
Greenlees, J.S., see Abraham, K.G. 123n, 130n
Goldman, W., see Cuffel, B. 901
Greenwald, H.B. 435
Goldsmith, L., see Giacomini. M. 1190
Goldstein, A.O. 1592 Greenwald, L.M. 829
Goldstein. K.P. 1794n Grembowski, D., see Conrad, D.A. 619, 1261-1263,
1270, 1272
Goldstein, M.K., see Lenert, L.A. 1746, 1753
Golec, J.H., see Bradbury. R.C. 733 Gribben, B. 1222n
Gomby, D. 1064 Griffin, M.R., see Ray, W.A. 602, 630
Gomulka, J., see Atkinson, A.B. 298 Griffith, M.J. 617
Goodin, R.E. 1807, 1873 Griffiths, D.A.T.
4 Griliches, Z. 122n, 127n, 131, 133n, 135, 152, 398, 1322n
Goodman, A., see Folland, S. 1, 67, 78, 64n, 475, 477n
Goodman, J. 1545 Griliches, Z., see Bemdt, E.R. 119, 133n, 150n, 151-153,
Goodman, M., see Beinecke, R.H. 906 154n, 158n, 159n
Goodman, M.J., see Hornbrook, M.C. 798n., 802-804, Grliches, Z., see Boskin, M.J. 130n
806 Griliches, Z., see Ellison, S.F. 162n, 1301
Goodwin. J., see Hooper, E. 1065 Griliches, Z., see Fisher, F.M. 129n, 131, 135
Gordon, N. 729 Grimes, D.R., see Warner, K.E. 1600n, 1607-1609
Gordon, R.J., see Boskin, M.J. 130n Grise, V.N. 1600, 1601n, 1602. 16(04
Gordon, R.J., see Bresnahan, T.F. 123n Griss, R., see DeJong, G. 1031n
Gorecki, PK. 1325 Grisso, T., see Steadman. H.J. 898, 899
Gorman, S.A., see Sisk, J.E. 737 Grob, G.N. 903, 938, 941, 942, 943n
Gorman, W.M. 1557 Groenewegen, P.P., see Boerma, W.G.W. 1177
Gosden,T. 1187, 1188 Groenewegen, P.P. see Krasnik, A. 1189
Gottschalk, P. 1811n, 1822 Grogan, C., see Scheffler, R.M. 912
Gottschau, A., see Krasnik, A. 1189 GroganS. 1181
Gourieroux, C.A. 271, 317 Grol, R., see Wensing, M. 1181
Gouveia, M., see Besley, T. 849, 885, 1238. 1693 Gronfein, W. 939
Gowrisankaran, G., see Chernew, M. 1456 Grootendorst. P.V. 314, 315, 319-321
2
Grabowski, H.G. 1310, 1316, 1317, 1322, 1323n, 1324, Grosskopf, S., see Fare, R. 1279, 1 79n
1587, 1588, 1702, 1703 Grossman, M. 23, 68, 68n, 126n, 129, 238n, 349, 351,
Grabowski, H.G, see DiMasi, J.A. 1309n, 1310 351n, 352n, 353, 356, 371n, 374, 377, 377n, 380n,
Graham, D., see Cook, P.J. 543. 1372, 1688 381,382, 387, 388, 393,395, 397, 398n, 404, 413,
Graham, D., see Vernon, J.M. 1474 417, 421,428,439, 448, 498, 1062, 1084, 1182, 1253,
Author Index I-15
1258, 1551, 1552, 1595, 1636, 1636n, 1641, 1642, Hadley, J., see Zuckerman, S. 333, 1157
1680, 1680n, 1681, 1682, 1684, 1685 Hadlock, C.J. 1147
Grossman, M., see Becker, G.S. 392n, 401, 1551, 1560, Hadom, D. 1894,1899
1560n, 1561, 1569, 1570, 1572n, 1636. 1685 Haes, J.C., see Ong, L.M. 1179
Grossman, M., see Chaloupka, EJ. 1549n, 1550, 1552, Hafner-Eaton, C. 1708
1565, 1578, 1595-1597, 1642, 1646n, 1648 Hagan, M., see Monheit, A. 690, 691
Grossman, M., seeCoate,D. 1641 Haglund, B., see Makuc, D. 1429
Grossman, M., see Colle, A.D. 582 Hague, D.C., see Corlett, W.J. 1663
Grossman, M., see Corman, H. 381, 1084 Haines, M.R., see Preston, S.H. 1852n
Grossman, M., see Edwards, L. 1062 Hajivassiliou, V. 272, 305
Grossman, M., see Goldman, F 583 Hajivassiliou, V., see Borsch-Supan, A. 282, 977
Grossman, M., see Lewit, E.M. 299, 1549n, 1550-1552, Hakkinen, U. 300, 315, 319-320
1578, 1588, 1594, 1699 Halberstadt, V., see Haveman, R. 998n, 1038n
Grossman, M., see Markowitz, S. 1653 Halfon, N. 1066
Grossman, M., see Saffer, H. 1648 Halfon, N., see St. Peter, R. 1060
Grossman, M., see Shakotko, R. 1062 Hall, A. 271
Grossman,S. 1101, 1149 Hall, J., see Besley, T. 1230, 1238
Grothaus, L., see Cherkin, D. 420, 432, 583, 713, 1708 Hall, J., see Mooney, G. 64n, 89, 92, 1814, 1814n, 1893
Gruber, J. 125n, 134n, 160n, 164n, 507, 508, 510, 511, Hall, J., see Richardson, J. 1743
518, 525, 651, 651n, 652n. 654, 659, 660n, 661n, 662, Hall, J., see Scott, A. 1188
668, 669, 671,671n, 672. 673, 674n, 675,675n, 677, Halpern, J. 1026
679, 680n, 684, 691,693,693n, 695-697, 697n, 698, Halper, M.T., see Alexander, J.A. 1454
765,982, 1024, 1026, 1027, 1129,1161, 1167, 1428, Ham, C. 20, 830, 1387, 1389
1429. 1451 Hamermesh, D. 693n
Gruber, J., see Currie, J. 684, 1060, 1068-1073, 1080 Hamermesh, D., see Woodbury, S. 696
Gruber, J., see Cutler. D. 686, 687, 687n, 1074 Hamilton, B. 305, 325, 330, 331, 1451, 1840
Gruber. L.R. 718-720 Hamilton, G.J. 761
Gruenberg, L. 803, 806, 817, 819, 967 Hamilton, J.D. 41
Gruenberg, L., see Ash, A.S. 798, 800, 810, 822 Hamilton, J.L. 1587, 1589, 1593, 1594
Grunewald, P.J. 1640 Hamilton, V. 302, 960
Grytten, J. 511, 1181, 1184, 1185, 1259, 1261-1263, Hamilton, V., see Hamilton, B. 325, 330, 331
1265, 1267, 1271. 1273, 1277, 1283, 1840 Hamilton, V., see PrimoffVistnes, J. 315
Grytten, J., see Carlsen, F. 18, 507, 511 Hammer, J.S. 1806n
Gudex,C. 1241, 1740 Hammerton, M., see Jones-Lee, M.W. 1225
Gudex, C., see Dolan, P. 1742, 1745-1747, 1752, 1753 Hammond, E.D. 1545
Guerra, N., see Tolan, P. 1080 Hammond, P.J. 1557
Guilkey, D.K. 303 Hampshire, S. 1873
Guilkey. D.K., see Bollen, K.A. 303 Han, J., see Phillips, C.D. 961
Gullickson, W. 170, 170n Handte, J., see Rosenberg, S. 717
Gunderson, M. 690n Hanemann, W.M. 1226
Gurmu, S. 316, 322 Hanks, C., see Olds, D. 1081
Gustman, A. 665n, 676, 677, 679 Hanley, D., see Fowler Jr., EJ. 216
Guterman, S. 597, 601 Hanley, J., see Bailit, H. 1274
Gutierrezfisac, J.L., see Regidor, E. 1222n Hanley, R.J., see Rivlin, A.M. 984
Gutman, G., see Lane, D. 964 Hanning, M. 1235, 1243
Guyer, B., see Frank, R.G. 1709 Hanratty, M. 1069
Gyourko, J. 690n Hanratty, M., see Gruber, J. 660n, 698
Hansen, P. 43, 47, 48
Haas, P.J., see Perkoff, G.T. 728 Hansen, R.W., see DiMasi, J.A. 1309n, 1310
Haas-Wilson, D. 279, 280, 476, 493, 860, 1112n Hansmann,H.B. 1144, 1146-1148, 1151
Haas-Wilson, D., see Gaynor, M. 1413, 1413n, 1465, Hanson, J., see Hoyme, E. 1066
1470n, 1471 Harberger, A.C. 93, 252n
Haavelmo, T. 1558 Harbold, J.R., see Bray, R.M. 1657n
Habeck, R.V., see Hunt. H.A. 1044n Hare, R. 1872n
Habermas, J. 1891 Harford, T.C. 1657n
Habicht, J.P., see Martorell, R. 1062n Hargraves, M.A., see Hogue, C.J. 1083
Hadden, W., see Pappas, G. 1852n Hariharan, G., see Douglas, S. 317, 325, 328, 1553, 1554
Haddix, A.C. 1686, 1705 Harper, M.J., see Gullickson, W. 170, 170n
Hadley, J. 514, 600, 1124, 1513, 1518, 1526, 1527 Harper, R. 1745
Hadley, J., see Feder, J. 597 Harrington, C.A. 958, 965, 966
Hadley, J., see Gaskin, D. 742 Harrington, M., see Gold, M. 1529
Hadley, J., see Langwell, K.M. 618 Harrington, S. 1362, 1372
Hadley, J., see Lee, R. 552 Harris, B.L. 432
I-16 Author Index
Harris, J. 73n, 495n, 1100. 1100n, 1549, 1556, 1568, Heinonen, M., see Sintonen, H. 1277
1577,1578, 1595,1604n, 1887, 1899 Heins, H. 1081
Harrison, E., see Long, S. 1067 Heiss, G., see Criqui, M.H. 1651
Harrison, E.R., see Keeler, E.B. 1162, 1163 Held, P.J. 556
Harrison, J.L., see Blair, R.D. 1457, 1459, 1468 Held, PJ., see Berry, C. 526
Harrison, L., see Godfrey, C. 1663 Held, P.J., see Pauly, M.V. 414
Harrow. B. 904, 920n Heller, D.N., see Redelmeier, D.A. 1742
Harrow, B., see Cromwell, J. 919 Hellinger, F. 488, 492n, 721, 728, 829n, 1415, 1508n,
Harsanyi, J.C. 1892 1519
Hart. L.G. 714 Hellwig, M. 1415n
Hart, O. 851, 852, 1101, 1149, 1150, 1474 Helms, L.J. 431
Hart, O., see Grossman, S. 1101, 1149 Hemenway, D. 492, 1692
Hartunian,N. 101On Hemingway, H. 1232n
Hartz, A.J. 1162 Hendel, .,see Cardon, J. 619
Harvey. B., see Garnick, D.W. 734 Henderson, C., see Olds, D. 1066, 1081
Harwood, H. 1664 Henderson, J., see McGuire, A. I, 17
Harwood, H.J. 1664 Henderson, J., see Newton, J. 1228
Hashimoto, M. 661n Henderson, J.A. 1356
Hassan, M., see Mon'isey, M.A. 1160, 1165 Henderson, R. 1305
Hassan, M., see Sloan, FA. 1164, 1165 Henderson, R., see Cockburn, I. 1308
Hassan,M.,seeWedig, G.. 1163-1165 Henderson, T., see McGuire, A. 64n
Haugen, D.K., see Zellner, B.B. 821n Henderson, V.. see Ricci, E.M. 430n
Hauquitz, A., see Lairson, D.R. 1828, 1844 Hendricks, W. 1010n
Hause, J.C. 398 Hendry, D.F., see Banerjee, A. 41
Hausman, D. 66, 1869 Hennekens, C.H., see Camargo Jr., C.A. 1651, 1652
Hausman, J. 30. 400, 427n, 449, 1226n. 1436 Hennekens, C.H., see Stampfer, M.J. 1651
Hausman, J., see Burtless, G. 426n Hennessey, J., see Dykacz, J. 1026
Hausman, J., see Ellison, S.F. 162n, 1301 Hennessey, K.D., see Reed, S.K. 737
Hausman, J., see Halpern, J. 1026 Hensher, D.A., see Truong, P. 1226
Haveman, R. 998n, 1000, 1001, 1002n, 1009n, 1011, Henteleff, P.D., see Roos, N.P. 240, 244
11O1n, 1015n, 1024, 1025n, 1029,1038n Heppell, S. 1314
Haveman, R., see Burkhauser, R. 1002n. 1008, 1009, Herbers, J.E., see Reed, W.W. 1745
1009n, 1010, 1029n Herbert, R., see Anderson, G.F 800
Haveman, R., see Gottschalk, P. 181 In, 1822 Herd, A., see Lairson, D. 728, 729
Haveman, R., see Wolfe, B.L. 1003-1005 Herrero, C. 1727
Havighurst, C.C. 474, 474n, 1381, 1383, 1385 Hersch, P.L. 1450, 1451
Hawe, P., see Shiell, A. 64n Hershey, J.C. 1747
Hawes, C., see Phillips, C.D. 961 Hertzman, C. 1062
Hawkings, S.T., see Pearson, S.D. 737 Herzlinger, R.E. 1156, 1162, 1165
Hay, J. 286, 304, 513, 740. 741, 788n, 1258, 1258n, Hey, J.D. 228, 234
1259n, 1261,1262, 1268. 1271, 1575, 1577, 1579, Hiatt, H.H., see Weiler, PC. 1344, 1351, 1352n, 1353n,
1581, 1767n 1354, 1358, 1361n, 1368, 1370,1371,1373,
Hayek, F.A. 1892 1378-1380, 1380n
Hayes, K., see Stearns, S.C. 985 Hibbard, J. 75,860, 1133, 1414
Hayes, V., see Newton, J. 1186 Hicks, D. 1240n
Hayter, P., see Robinson, R. 1193 Hicks, J. 61n, 187
Hayward, L. 1653 Hickson, G 255,484,491, 1189. 1360
Hayward, R., see Chemew, M. 714 Higgings, J. 1240, 1240n
HCIA, Inc. 744 Higgins, S.T., see Bickel, W.K. 1563, 1564
Headen, A.E. 676. 677, 679n, 975-977 Higgins, S.T., see DeGrandpre, R.J. 1564
Heal, G.M.,see Ryder, H.E. 1559 Higgins, V., see Rowlands, O0. 1178
Healey,A. 1186 Hill, A.B., see Doll, R. 1545, 1593
Healey, A., see Shackley, P. 77, 77n Hill, J. 728, 730
4
Health Care Financing Administration 169n, 17 n Hill, S., see Haveman, R. 1000, 1001, 1002n, 1011,
Health Insurance Association of America 650 1011n
Heaney, C.T. 429 Hill, S., see Wolfe, B. 1008n
Heaney, D.J., see Howie, J.G.R. 1180 Hiller, M.D. 520
Heath Jr., C., see Steenland, K. 1596 Hillman, A. 504, 517,521,548, 557, 715,716
Heaton, J.M., see Mills, R.P 1234 Hilton, M.E. 1657
Heckman, J. 269, 285, 313, 322, 329, 336, 1641, 1773n Himmelstein, D. 1060
Hedges, L., see Philipson T. 1796n Hinmmelstein, D., see Woolhandler, S. 82
Heidenreich, P. 173 Hindson, P., see Lairson, D.R. 1828, 1844
Heien, D.M. 1661 Hirschman, A.O. 1238
Heifitz, S.B.. see Horowitz, H.S. 1285, 1286, 1288, 1289 Hirsh, H.L. 516
Author Index I-17
Hirshleifer, D., see Bikhchandani, S. 237 Hoover, E.D. 136n

Hirshleifer, D.. see Png, I.P.L. 1476 Hopkins, C.E., see Roemer, M.I. 431
Hirshleifer, J. 256, 1700 Hopton, J.L., see Howie, J.G.R. 1180
Hirth, R. 958 Horbar, J. 1068
Hirth, R., see Chernew, M. 742 Horder, J., see Fry, J. 1177
Hirth, R., see Pauly, M. 627, 628n Horgan, C.M. 909,911,912,933
Hitiris, T. 11, 18-20, 31, 49 Horn, D., see Hammond, E.D. 1545
Hjortdahl,P. 1179, 1181 Horn, S., see Anderson, G.F 800
Hjortdahl. P., see Kristiansen, I.S. 1189 Hornberger, J. 1702, 1746
Hlatky, M.A., see Mark, D.B. 215 Hombrook, M.C. 789, 798n, 802-804, 806
Ho, T., see Scheffler, R. 1123 Hombrook, M.C., see Gruenberg, L. 803
Ho, V., see Hamilton, B. 1451, 1840 Horowitz, H.S. 1285, 1286, 1288, 1289
Hochman, H.M. 1807 Horowitz, L.A., see Rice, D.P. 151n
Hocutt, A. 1035n, 1036n, 1037 Horwood, S.P., see Torrance, GW. 1739
Hodge, D., see Zambrana, R. 1060 Hosek, S.D. 729, 735, 739
Hodgkin, D. 556, 603, 857, 865, 922, 1132, 1528n Hosek, S.D., see Wells, K.B. 738
Hodgson, T.A. 173, 1575-1577, 1664 Hosmer, D., see Lee, A.J. 506n
Hodgson, T.A., see Rice, D. 1010n, 1575 Hotz, V.J., see Ahituv, A. 1790, 1793
Hodgson, T.A., see Warner, K.E. 1575 1577 Hoult, J. 904
Hoechst. M.R. 147n Houthakker, H.S. 1558
Hoefler, J.M., see Fritschler, A.L. 1545n, 1603 Hovind, O.B., see McPherson, K. 243
Hoerger, T. 282, 502, 975, 977, 1158, 1165, 1444 Howard, L., see Mennemeyer, S. 1134
Hoerger, T., see Lindrooth, R.C. 981 Howard, L.Z., see Hoerger, T.J. 502
Hoerger, T., see Sloan, EA. 960, 972, 981, 1164, 1165, Howard, L.Z., see Mennemeyer, S.T. 860
1357. 1358, 1360 Howard, P., see Harper, R. 1745
Hofflander, A.E., see Nye, B.E 1354, 1360 Howie,J.G.R. 1180
Hoffman, R., see Miller, L.A. 1064 Hoyme, E. 1066
Hoffmeyer, U. 1145 Hoynes, H. 1043n
Hoffmeyer, U., see McCarthy, T. 830, 831, 833, 834n Hoyt, G.M. 1639n
Hogan, C., see Farley, D.E. 1427n Hsiao, C. 30
Hogarty, T.E, see Elzinga, K.G. 1423, 1428 Hsiao, W.C. 482, 513
Hogue, C. 1083 Hsieh, C.-R. 1548, 1594
Hogue, C., see Buehler, J.W. 1068 Hsu, H.-P. 1778n
Hogue, C., see Schoendorf, K. 1083 Hu, T.-W. 1079, 1261, 1262, 1266, 1544, 1549, 1555,
Holahan, J. 514n, 720, 1827, 1827n 1594, 1595
Holahan, J., see Kenney, G. 967 Hu, T.-W., see Barnett, PG. 1548, 1549, 1568, 1569
Holahan, J.. see Zuckerman, S. 475, 485 Hu, T.-W., see Hsieh, C.-R. 1548
Holak, S.L. 1588 Hu, T.-W., see Keeler, T.E. 1548, 1549, 1561, 1569,
Holbrook, J.H, see Warnmer, K.E. 1587, 1588 1596, 1604n
Holden, K.C., see Haveman, R. 1015n, 1029 Hu, T.-W., see Sung, H.-Y. 1548,1549, 1561, 1569
Holland, W., see Malin, H. 1660 Hu,T.-W.,seeXu,X. 1548
Hollenberg, J., see Kelleher, K. 898, 899 Hu, T.Y. 242, 1658
Hollingsworth, E.J. 936 Huang, L., see Rosett, R.N. 433,580, 582
Holloway, PJ., see O'Rourke, C.A. 1288 Huang, L.X., see Evans, W.N. 1553, 1555
Holman, H.R., see Lubeck, D.P. 735 Hubbard, R.G. 981
Holmer, M.R., see Buchanan, J.L. 587 Hubbard, T. 1415n
Holmes, A., see Deb, P. 927 Hudes, J. 728, 729
Holmlund, B. 693n Hughart, N., see Frank, R.G. 1709
Holmstr6m, B. 414, 851, 1101n, 1122n, 1190 Hughes, D. 434, 1188
Holmstr6m, B., see Hart, 0. 851 Hughes, D., see McGuire, A. 19, 23, 48
Holst, D.. see Grytten, J. 511, 1259, 1261 1263, 1265, Hughes, D., see Newacheck, P. 1065
1267, 1271, 1273 Hughes, E.FX., see Shortell, S. 1162, 1451
Holt, E. 1077 Hughes, J.R., see Bickel, W.K. 1564
Holt, E., see Frank, R.G. 1709 Hughes, J.R., see DeGrandpre, R.J. 1564
Holtedahl, K., see Kristiansen, I.S, 1189 Hughes, M., see Kessler, R.C. 897
Holtman, A.G. 1258, 1261-1263, 1265, 1268, 1270, 1271 Hughes, S.L. 961
Holtman, A.G., see Cowing, T.G. 1452 Humphries, K.H. 1854
Holtz-Eakin, D. 647, 667, 668, 672 Hunt, H.A. 1044n
Holtzman, D., see Lowry. R. 900 Hunt, K.A., see Gabel, J.A. 725
Honor, B.E. 314 Hunt, S., see Luft, H. 1115n
Hooijmans, E.M., see van de Ven, W.P.M.M. 300 Hunt, S., see Shewry, S. 831
Hooper, E. 1065 Hunt-McCool, J. 289
Hoos,A.M.,seeOng,L.M. 1179 Hurd, M. 676, 678, 977, 978, 1682, 1698
Hoover, C., see Schlesinger, M. 1160, 1161, 1163 Hurdle, S. 525, 1417
1-18 Author Index
Hurley, J. 64n, 66. 107, 108, 184, 514, 568 Jackson, C., see MelTrrill,J. 618
Hurley, J., see Birch, S. 83n 92 Jackson, C., see Sturm, R. 737
Hurley, J., see Giacomini, M. 1190 Jackson, T., see Rosenberg, S. 717
Hurley, J., see Hutchinson, B. 1189 Jackson-Beeck, M. 617, 730
Hurley, J., see Hutchison. B. 83n Jackson-Thompson, J., see Brownson, R.C. 1595
Hurley, R.. see Gold, M.R. 714,715 Jacobs, P. 1
Hursh, S.R. 1563 Jacobsen, K. 940
Hurst, J. 2, 181n, 1822, 1822n Jacobson, B., see Hemingway, H. 1232n
Hurst, K.M., see Gabel, J.A. 725 Jaditz, T., see Fixler, D. 168
Hurwitz, M., see Caves, R.E. 1322 Jadlow, J.M., see Wilson, G.W. 1157
Huse, D.M., see Oster, G. 1544 Jaeger, D., see Bound, J. 270, 300, 381
Huskamp, H.A. 906, 924 Jaffe, M.W., see Katz, S. 975
Huskamp, H.A., see Frank, R.G. 924, 926 James, M. 1241
Hussain, I., see Doherty, N. 1280 Jan, S., see Mooney, G. 1181
Hustead, E. 909 Janssen, R. 1827
Husten, C., see Zhang, P. 1601n 1603, 1604 Janssen, R., see Rutten, F. 1828
Huston, J., see Weisbrod, B. 1767n Jason, L.A 1597
Hutcheson, J., see Black, M. 1081 Jaynes, G. 1415n
Hutchinson, A., see Newton, J. 1186 Jefferson, T.O., see Drummond, M.F. 1291
Hutchinson, B. 1189 Jencks, C. 898, 900
Hutchison, B. 83n Jencks, S. 933, 934, 936
Hutchison, B., see Birch, S. 83n, 92 Jencks, S., see Horgan, C.M. 933
Hyatt, D., see Gunderson, M. 690n Jennison, K. 492, 522
Hyland, A., see Lewit, E.M. 1553 Jennison, K., see Udvarhelyi, I.S. 737
Hynes, M.M., see Sisk, J.E. 737 Jensen, G. 133, 712, 780
Jensen,M.C. 1144
Ichimura, H. 294 Jeong, J. 272
Idelson, R.K., see Wechsler, H. 1712 Jewell, N.P., see Watts, C.A. 909
Idler, E.L. 1853 Jewett, J., see Hibbard, J. 75, 860, 1133
Jewett, T., see Hoyme, E. 1066
lezzoni, L. 758
Jin, S.G. 1575
lezzoni, L., see Ellis, R.P 143, 784, 786, 798, 800, 806.
807, 810, 811n, 814, 822, 835,932 Johannesson, M. 98, 100, 102n, 103, 194, 212, 1227.
lezzoni, L., see Pope, G. 798, 800, 811n, 812 1729, 1735, 1904
Johannesson, M.. see Bleichrodt, H. 1730, 1741, 1742,
lezzoni, L., see Zuckerman, S. 333, 1157
1746, 1747
IIT Research Institute 1308
Johannesson, M., see Gerdtham, U.-G. 1854
Ikeda, S., see Ikegami, N. 968, 1330
Johannesson, M., see Karlsson, G. 192, 194
Ikegami, N. 962, 963, 968, 1330
Johansen, S. 43, 44, 1119n
Illsley. R. 1847, 1848
Johansson, P.-O. 100, 1679n
Im, K.S. 41, 44
Johansson, P-O., see Cameron, A.C. 315, 319, 322
Im, K.S., see Pesaran. M.H. 39
Johansson. P.-O., see Johannesson, M. 1227. 1904
Imbens, G.W. 293
Johnson, A., see Feldman, R. 727, 741
Imbens, G.W, see Angrist, J.D. 336
Johnson, A.N. 735, 741
Imber, S.D.. see Elkin, I. 904
Johnson, F.R., see Smith, V.K. 1701
/ngber, M.J. 780. 783.810
Johnson, K., see Mitry, D.J. 1280
Ingber, M.J., see Greenwald, L.M. 829 Johnson,L.W. 1587, 1589
Ingram, D., see Makuc, D. 1429 Johnson, M., see Flewelling, R.L. 1548
Ingram, R.M. 1234
Johnson, P., see Aronson, J.R. 1829. 1830
Institute of Medicine 717, 718, 720, 913n, 919, 968, Johnson, P.R. 1603
1083, 1156, 1162, 1164 Johnson, T.R. 1568
Inter-Authority Comparisons and Consultancy 1235 Johnson, V., see Ohsfeldt, R. 493
Intrator, O., see Mor, V. 982 Johnson, W.G. 1031n
Intriligator, M.. see Brito, D.L. 71, 1694, 1706, 1778n Johnson, W.G., see Baldwin, M.L. 1009n, 1033n
Ippolito, P. 1700-1702 Johnson, W.G.,see Berkowitz, M. 999n
Ippolito, R.A. 1547, 1588, 1593, 1594 Johnson, W.H., see Leaver, S.G. 137n
Ireland, M., see Fowles, J.B. 805 Johnston, K., see Cairns, J. 1731
Iversen, 1. 1188 Jolly, D. I
Iverson, T. 1215, 1218, 1235, 1238. 1238n Jones, A. 273,283,298,788, 1170, 1183, 1265, 1267,
1277, 1559, 1572n, 1711
Jachuk, S., see Jachuk, S.J. 1726, 1732 Jones, A., see Donaldson, C. 278
Jachuk, S.J. 1726, 1732 Jones, A., see Duncan, A.S. 272, 273,298
Jackman, A., see Carey, T.S. 733 Jones, A., see Forster, M. 325-327
Jackson, B.A., see Katz, S. 975 Jones, A., see Rice, N. 306, 307
Jackson. C., see Frank, R. 920, 1084 Jones, A., see Yen, S.T. 297 299
Author Index 1-19
Jones, C.A., see Gray, W. 1017n Kaplan, E., see Brandeau, M. 1763n
Jones, K., see Hoyme, E. 1066 Kaplan, G., see Gordon, N. 729
Jones, L., see McPherson, K. 243 Kaplan, R.M. 917
Jones, N., see Brazier, J. 1745 Kaplan, S.H. 1179
Jones. N.M.B., see Harper, R. 1745 Kapur, K. 669, 670, 672
Jones, P.R. 1242 Karlsson, G. 192, 194
Jones, P.R., see Cullis, J.G. 1223, 1232, 1235 Kami, E., see Darby, M.R. 1416n, 1417
Jones, R. 1186 Karoly, L. 676, 677
Jones-Lee, M. 100, 1225 Kasper, J. 730, 1067
Jones-Lee, M., see Dolan, P. 1735 Kasper, J., see McCombs, J.S. 736
Jnsson, B. 16,48, 1281 Kasteler, J., see Olsen. D.M. 229
Jonsson, B., see Gerdtham, U.-G. 11, 18, 19, 22, 24, 24n, Katsoyiannopolous, V., see Calnan. M. 1181
26, 27, 32, 37-39, 45, 46, 48, 49, 268, 412, 451,849n Katz, H.M., see Pattison, R.V. 1156
Joo, H., see Goldstein, K.P 1794n Katz, L. 661n
Joossens, L. 1548n, 1570, 1599 Katz, L., see Dickens, W. 666n
Joreskog, K.G. 383 Katz, M.L. 1475
Jorgenson, D.W., see Boskin, M.J. 130n Katz, S. 975
Joshi, H. 1836n Kauer, R.T., see Silvers, J.B. 1165
Joskow, P. 874, 1144, 1450, 1523-1525 Kawai, H., see Ikegami, N. 1330
Jourdans, M., see Gudex, C. 1241 Kay, J. 1659, 1663
Joyce, T. 381 Kee, D.W., see Strumwasser, I. 730, 737
Joyce, T., see Corman, H. 381, 1084 Keeler, E. 256, 316, 325, 419, 422-425, 441. 442, 512,
Joyce, T., see Grossman, M. 381, 1084 579n, 595, 772n, 773,777n, 817, 819, 825, 826, 911,
Juba, D.A. 617 913,934, 1130, 1159, 1162, 1163, 1269, 1271, 1448,
Juhn, P.I., see Stem, R.S. 737 1449
Just,R. 1461n Keeler, E., see Buchanan, J.L. 587
Juster, ET., see Barsky, R.B. 1682, 1692 Keeler, E., see Manning, W.G. 23, 286, 288, 911, 922,
1010n, 1263-1265,1267, 1285,1286,1288, 1289,
1575, 1577, 1579, 1580, 1580n, 1581, 1581n, 1582,
Kaelber, C., see Malin, H. 1660 1650, 1651n, 1660, 1687, 1707
Kaestner, R. 1062 Keeler, E., see Newhouse, J.P. 437n, 438, 441, 442, 790,
Kaestner, R., see Grossman, M. 395, 397, 398n, 1681 791n, 792, 792n, 793, 793n, 797, 805, 814, 932, 933
Kagan, A., see Yano, K. 1651 Keeler, E., see Ware Jr., J.E. 739
Kaganova, M,H., see Gruenberg, L. 803 Keeler, T.E. 1548, 1549, 1561, 1569, 1596, 1604n
Kahn. J.G. 814 Keeler, T.E., see Barnett, P.G. 1548, 1549, 1568, 1569
Kahn, K.L. 601,630, 630n Keeler, T.E., see Hu, T.-W. 1079, 1544, 1549, 1555,
Kahn, K.L., see Keeler, E.B. 1162, 1163 1594, 1595
Kahn, L., see Perkoff, G.T. 728 Keeler, T.E., see Sung, H.-Y. 1548, 1549, 1561, 1569
Kahneman, D. 88, 100, 1905n Keeler, T.E., see Xu, X. 1548
Kahneman, D., see Tversky, A. 1750 Keen, M., see Kay, J. 1659, 1663
Kaitin, K.I. 1310 Keeney, R.L. 31, 194, 1732
Kaitin, K.I., see Bakke, O.M. 1315n Keesey, J., see Buchanan, J.L. 728, 729, 733
Kaitin, K.I., see Shulman, S.R. 1314 Kehrer, B., see BetTy, C. 526
Kakwani, N.C. 1822, 1829, 1843, 1849n, 1851, 1852, Keith, A., see Hu, T.-W. 1544
1852n Kelleher, K. 898, 899
Kaldor, N. 61n, 187 Keller, A., see Greenfield, S. 492, 734
Kallapur, S., see Eldenburg, L. 864, 1427 Keller, M. 1632
Kallich, J.D., see Farley, D.O. 814 Kelly, G.G. 151, 151n, 152
Kalyanaram, G., see Robinson, W.T. 1324n Kelly, N.L., see Oster, G. 1575
Kamberg, C., see Bailit, H. 1274 Kelman, S., see Rice, D. 1010n, 1660, 1664
Kamlet, M.S., see Garber, A.M. 104 Kemper, K. 1064
Kanak, J., see Cromwell. J. 1521 Kemper, P. 959, 962
Kandori, M. 237 Kemper, P., see Murtaugh, C.M. 959, 975-979
Kane, R.A. 972 Kemper, P., see Pezzin, L.E. 962, 975
Kane, R.A., see Nyman, J.A. 960 Kemper, P., see Spillman, B.C. 982
Kane, R.L. 972 Kendix, M. 20
Kane, R.L., see Froberg, D.G. 1745, 1747, 1748 Kendler. K., see Kessler, R.C. 897
Kane, R.L., see Kane, R.A. 972 Kendrick, J.S. 1079
Kane, R.L., see Olsen, D.M. 229 Kendrick, J.W. 122n
Kann, L., see Lowry, R. 900 Kenkel, D. 185, 276, 283, 284, 302, 303, 316, 396, 404,
Kanoza, D. 141n, 151n, 152 1595, 1647, 1648, 1655, 1656, 1658, 1663, 1681,
Kantor, S.E., see Fishback, P.V. 693n 1683,1684, 1690,1692n, 1694.1697,1699, 1708,
Kao, C., see McCoskey. S. 43 1709, 1711
Kao, K. 1587, 1588, 1593, 1594 Kenkel, D.. see DeCicca, P. 1554, 1555
1-20 Author Index
Kenkel, D., see Tolley, G.S. 1687, 1706 Klein, S., see Szilagyi, P.G. 737
Kennedy, A. 830, 834n Kleinman, J., see Buehler, J.W. 1068
Kennedy, F. 1222 Kleinman. J.. see Makuc, D. 1429
Kennedy, G., see Dubay, L. 687, 687n Kleinman, J., see Schoendorf, K. 1083
Kennedy, W., see Broman, S. 1062 Kleinman, J.H., see Jackson-Beeck, M. 617, 730
Kenney, E., see Flewelling, R.L. 1548 Kletke, P., see Escarce, J. 474
Kenney, G. 967, 1427 Klevorick, A.K. 475, 476
Kenney, G., see Dubay, L. 1074 Klevorick, A.K., see Levin, R.C. 1317
Kentala, J., see Utriainen, P. 1282 Kliegman, R. 1083
Kerkhofs, M. 277, 309 Kluger, R. 1601
Kerkhofs, M., see Lindeboom, M. 325, 330 Knapp, M. 918,939
Kerrebrock, N., see Lewit, E.M. 1553 Kneeshaw, J. 1905, 1905n
Kerstein, J., see Hillman, A.L. 557, 716 Kneisner, T.J., seeEakin, B.K. 1514
Kesenne, J. 830, 834n Knesper, D.J. 904
Kessel, R. 76n, 464, 473. 474, 520, 1491 Knetsch, J., see Kahneman, D. 100, 1905n
Kessler, D. 516, 1356. 1368, 1451 Knetsch, J.L., see Bukszar, E. 1905n
Kessler, L.G., see Taube, C.A. 909, 911 Knottnerus, A., see van de Kar, A. 1181, 182
Kessler, L.G., see Thompson, G.B. 1707 Knutson, D., see Fowles, J.B. 805
Kessler, R. 897, 900n, 901, 901n Kobrinski, E., see DesHarnais, S.I. 863-865
Kessler, R., see Alegria. M. 897 Koch, G.G., see Steams, S.C. 985
Kessler, R., see Ettner, S.L. 898, 899 Koch, M. 1010n
Kessler, R., see Swendsen, J.D. 898 Koepsell, T.D., see Magid, D.J. 435
Kestens, P., see Schokkaert, E. 806, 830, 834n Kogan, M. 1064
Khandker, R. 717, 1123 Kohler, B., see Brazier, J. 1744, 1745
Khandker, R., see Pope, G.C. 798, 800, 805, 811n, 814 Koitz, D. 1043n
Khayat,K. 1181 Kok, G., see van de Kar, A. 1181, 1182
Kiebert, G.M., see Stiggelbout, A.M. 1741, 1746, 1747 Kolbe, L.J., see Lowry, R. 900
Kiefer, N. 328 Koller, C.E, see Bovbjerg, R.R. 821n
Kievit, J., see Stiggelbout, A.M. 1741, 1746, 1747 Koop,G. 334,1157
Kiker, B.F. see Hunt-McCool, J. 289 Koopmans, T. 62n
Kim, B. 1132 Kopit, W.G. 1440, 1448
Kim, Y.W., see Burkhauser, R. 1028 Kopp, R.J. 333
Kimball, M.S., see Barsky, R.B. 1682, 1692 Kopp, S.W. 1324, 1328
Kimberley, J.R., see Walston, S.L. 1478 Kopstein, A.N., see Rice, D. 101On, 1575
Kind, P., see Brazier, J. 1745 Korenman, S. 1063
Kind, P., see Dolan, P. 1742, 1745-1747, 1752, 1753 Koskinen, S. 1852n
Kind, P., see Gudex, C. 1740 Kotch, J., see Williams, B. 1080
Kind, P., see Rosser, R. 101, 102, 1748 Kotelchuck, M., see Kogan, M. 1064
Kindig, D., see Sager, M.A. 600 Kotlikoff, L. 690n
Kindig, D, see Stearns, S. 492, 716, 1189 Kotlikoff, L., see Borsch-Supan, A. 282, 977
King, A., see Hansen, P. 43, 47, 48 Kotowitz, Y. 576
King, C. 1592 Kottke, T. 1079,1711
King, C., see Jason, L.A 1597 Kovar, M.G. 984
King, E., see McKusick, D. 902 Kovar, M.G., see Steams, S.C. 985
King, M.A. 1829 Koyanagi, C., see Frank, R.G. 916
King, N. 1186 Krabbe, REM. 1746
King, N., see Bailey, J. 1186 Krakauer, H., see Hartz, A.J. 1162
Kington, R., see Danzon, P.M. 1366 Krakower, J., see Ganem, J. 471
Kirby, D. 1079 Kralewski, J. 727
Kitch, E.W. 1318n Kralewski, J., see Feldman, R. 711, 713, 719, 728.729
Kitchman, M., see Neuman, P. 83 Kramer, M., see von Korff, M. 909
Kitzman, H., see Olds, D. 1081, 1082 Kramer, M.J., see Fuchs, VR. 436, 581
Klaassen, M.P., see Wakker, P. 197 Krasker, W.S., see Herzlinger, R.E. 1156, 1162, 1165
Klarman, H. 3, 67, 71,98, 100, l00n, 101, 1491 Krasnik, A. 1189
Klass, M.W. 1473n Krattenmaker. T.G. 1475
Klassen, T.H., see Kralewski, J.E. 727 Krauss, N., see Rhoades, J. 958
Klatsky, A.L. 1651 Kravitz, R., see Greenfield, S. 492, 734
Klebanos, P.K., see Brooks-Gunn, J. 1062 Kravitz, R., see Rolph, J.E. 1361
Kleiman, E. 18, 22, 25 Kreider. B. 1023n, 1026
Kleiman, R., see Folland, S. 601 Kreider, B., see Riphahn, R. 1026n
Klein, B. 725, 1473 Kremer, M. 1777n, 1778n
Klein, B., see Schneider, L. 1586-1589, 1593, 1594, 1699 Kreuter, W., see Diehr, P. 242
Klein, K., see Collins, E. 1812n Krishnan, R. 1455
Klein, R. 1891 Kristiansen, I.S. 1177, 1185, 1188, 1189
Author Index 1-21
Kristiansen, PL., see Harwood, H.J. 1664 Lamphere, J., see Needleman, J. 1166
Kronick, R. 790, 791,798, 814, 834, 927 Lancaster, K.J. 350
Krueger, A. 741, 1011 Lands, W.E.M. 1651n
Krueger, A., see Angrist, J.D. 398 Landsburg, S.E. 1692
Krueger, A., see Card, D. 661n Lane, D. 964
Krueger, A., see Katz, L.E 661n Lane, J.D., see Stinson, ES. 1632
Kruger, H.B., see Gruber, J. 659, 662, 691, 693, 695, Lane, J.D., see Williams, G.D. 1632
697n Lane, J.R., see Angus, D.C. 733
Krupnick, A. 1225 Lane, W. 137n, 141n
Kruse, D., see Krueger. A. 1011 Langa,K. 1120
Kubik, J., see Gruber, J. 1026 Lange, M., see Mooney, G. 69n
Kuehnle, J., see Babor, T.F. 1642, 1643 Langford, E.A. 154n
Kuhn, E.M., see Hartz, A.J. 1162 Langwell, K. 618
Kuhn, T. 468 Langwell, K., see Lamphere, J.A. 766
Kuhse, H., see Nord, E. 88, 89, 94 Lanjouw, J.O. 1318, 1330
Kumar, A.K.N., see Cohen, M.A. 978 Lankford, H. 1036
Kumar, V., see Norton, E.C. 980 Lanning,J.A. 1504, 1514, 1527
Kunreuther, H., see Pauly, M. 627, 628n Lanoie, P. 1017n
Kunst, A.E. 1853n Larkby, C. 1646
Kuols, L.K., see Woodward, R.S. 498 Larry, L.O., see Manning, W.G. 1265, 1267
Kupfer, D.J. 904 Larson, C. 1081
Kupperman, M. 1743 Larson, C., see Gomby, D. 1064
Kushman, J.E., see Scheffler, R.M. 1276, 1277, 1280 Larson, E., see Welch, H. 1128
Kuttner, R. 1160 Larson, M.G., see Bombardier, C. 1745
Kuznets, S., see Friedman, M. 472 Larson, M.H., see Callahan, J.J. 904, 906, 912, 924
Kwoka, J.E. 74,226 Lasagna, L., see Bakke, O.M. 1315n
Lasagna, L., see DiMasi, J.A. 1309n, 1310
Lasagna, L., see Shulman, S.R. 1329
Laake, P.,seeGrytten,J. 511, 1259, 1261-1263, 1265, Lasagna, L., see Wardell, W. 1311
1267, 1271, 1273 Lau, L.J. 1276n, 1277n
Laasman, J.M., see Schokkaert, E. 806, 830, 834n Laugesen, M. 1589, 1590
Labeaga, J.M. 311,312 Lave, J.R. 598, 920
Labeaga, J.M., see Garcia, J. 298 Lave, J.R., see Frank, R.G. 157n, 597, 599, 854, 864, 920
Labelle, R. 63, 78, 1259, 1259n Lave, J.R., see Juba, D.A. 617
Labelle, R., see Hurley, J. 514 Lawson, A.M. 140n
Labelle, R.J., see Rice, T. 78, 525 Lawthers, A. 516
Labelle, R.J., see Stoddart, G.L. 26 Lawthers, A., see Localio, A.R. 1368
LaBov, A.D., see Palmer, R.M. 600, 630 Lazear. E. 367
Lacroix, G., see Bolduc, D. 282 Lazenby, H.C. 174n
Laditka. S.B. 987 Lazenby, H.C., see Levit, K.R. 125n, 159, 959, 960
Laerum, E., see Hjortdahl, P. 1179, 1181 Le Grand, J. 90, 1235, 1811, 1812, 1818, 1847n, 1849,
Laffont, J.-J. 499n, 852, 867-870, 872, 874, 876, 878, 1854n, 1878, 1880, 1894, 1895
879 Le Grand, J., see Goodin, R.E. 1807
LaFontaine, F. 1473 Le Grand, J., see Illsley, R. 1847, 1848
Lairson, D. 728, 729, 1828, 1844 Leaf, P.J., see Bruce, M.L. 899
Laixuthai, A., see Chaloupka, FJ. 1642 Leahy, M.J., see Hay, J. 513, 740, 741, 1259n
Laixuthai, A., see Grossman, M. 1641 Leake, B., see Wells, K.B. 1712
Lakatos, I. 6 Lean, D.F, see Bond, R.S. 1318n, 1324n
Lakdawalla, D. 984, 986, 1152, 1166, 1436, 1438 Leape, L.L. 244
Lake, T., see Gold, M. 714, 715, 1529 Leaver, S.G. 137n
Laliberte, L., see Mor. V. 982 Leber, W.R., see Elkin, I. 904
Lally, C., see Steenland, K. 1596 Leclerc, A. 1852n
Lambert, P.J. 1822, 1823, 1848 Lee,A. 1232n
Lambert, P.J., see Aronson, J.R. 1829, 1830 Lee, A., see Goldacre, M.J. 1228, 1234
Lambin, J.J. 1587 Lee, A.J. 506n
Lambrinos, J. 943 Lee, C. 831,834
Lamers, L.M. 793, 798, 801,802, 805, 827n Lee, K.L., see Mark, D.B. 215
Lamers, L.M., see van Barneveld, E.M. 771, 821, 821n, Lee, L., see Kronick, R. 798, 814
824, 824n, 825, 826, 828 Lee, L.-F 274, 280, 295, 296
Lamers, L.M., see van Vliet, R.C.J.A. 793n, 804 Lee, L.F., see Ichimura, H. 294
Lamers, L.M., see van de Ven, W.P.M.M. 793n, 794, 820n Lee, M.L. 1436
Lammes, EB., see Ong, L.M. 1179 Lee, R. 552, 984, 1417
Lamphere, J. 766 Lee, R., see Hadley, J. 514
Lamphere, J., see Gauthier, A.K. 835 Lee,R.,seeRomeo,A.A. 1161, 1429, 1522
1-22 Author Index
Lee, S.D., see Alexander, J.A. 1454 Lewis, C.E., see Wells, K.B. 1712
Lee, T.H.. see Pearson, S.D. 737 Lewis, J.M. 1288, 1290
Leeflang, P.S.H. 1587 Lewis, M. 16
Leer, J.W.H., see Stiggelbout. A.M. 1741, 1746, 1747 Lewis, P.A., see Charny, M.C. 1904, 1904n
Lees, D.S. 3,71 Lewis, T.R. 867,869
Leffler, KB. 472 Lewit, E. 299, 1549n, 1550-1553, 1559, 1578, 1588,
Leffler, K.B., see Klein, B. 725 1594. 1699
Lefgren, L.J., see Madrian, B.C. 669, 672 Lewit, E., see Grossman, M. 1551, 1552
Lefkowitz, B., seeDunn, W.L. 1498n Lewit, E., see Gomby, D. 1064
Lefkowitz, D., see Short, P. 1067 Li, J., see Ehrenberg, R. 692, 695
Lelkowitz, D., see Cohen, J.W. 134n Li, Z., see Experton, B. 734
Leger, S.St., see James, M. 1241 Liang, J. 959
Leibowitz, A. 286. 439, 440, 690, 691, 1301 Liaw, .ER.,see Brooks-Gunn, J. 1062
Leibowitz, A., see Buchanan, J.L. 728, 729, 733 Licari, M.J.. see Meier, K.J. 1593, 1594
Leibowitz, A., see Manning, W.G. 23, 286, 288, 728, 729, Lichtenberg, E., see Gamick, D.W. 1431, 1433
732, 735, 1263-1265, 1267, 1285, 1286, 1288, 1289 Lichtenberg, E., see Luft. H.S. 1434
Leibowitz, A., see Mauldon, T. 736 Lichtenberg, F.R. 1302
Leibowitz, A., see Valdez, R.B. 443 Lichtenstein, R. 829n
Leigh, J.P 400, 401, 1681 Lichtenstein, R., see Thomas, J.W. 780, 803, 806
Leigh, J.P., see Berger, M.C. 399, 1655 Lichter, H.M., see Fox, D. 96
Leighton-Read, J., see Bombardier, C. 1745 Lieberman, M.B. 1473, 1475
Leimer, D. 1022n Light, D. 496, 821, 1243
Leiper, A., see Feeny, D. 1745 Liljas, B. 383, 395. 1680
Leitzel, J., see Cook, P.J. 1663n Lillard, L. 419, 1708
Lelbach, K. 1649 Lillard, L., see Danzon, P.M. 1357-1359, 1362, 1373n
Lemeshow, S., see Rapoport, J. 737 Lin, C.-F., see Levin, A. 43, 44
Lenert, L.A. 1746, 1753 Lin, C.J., see Hsieh, C.-R. 1594
Lenhard-Reisinger, A., see Sisk, J.E. 737 Lin, E., see Alegria, M. 897
Leon, D.A. 1646 Linde-Zwible, W.T., see Angs, D.C. 733
Leonard, G., see Hausman. J. 1436 Lindeboom, M. 325, 330
Leonard, J.S. 1024, 1024n Lindeboom, M., see Kerkhofs, M. 277, 309
Leonard, T.C., see Suranovic, S.M. 1562, 1584 Lindrooth, R.C. 981
Lepkowski, J., see Lichtenstein, R. 829n Lindrooth, R.C., see Fisher, W.H. 873
Lerner, C. 1184, 1185 Lindrooth, R.C., see Norton, E.C. 1644
Lert, F., see Leclerc, A. 1852n Lindsay, C.M. 72, 72n, 1214, 1219, 1808
Lesnoy, S., see Leimer, D. 1022n Lindsey, E., see Pearson, S.D. 737
Lessler, D., see Wickizer, T. 1123 Lindsey, P.A. 717
Lestina, D.C., see Miller. T.R. 1664 Lindsey, P.A.,see Newhouse, J.P. 717
Leto, L., see Rosenberg, S. 717 Link, B.G. 896n, 898, 899
Letsch. S.W., see Lazenby, H.C. 174n Linna, M. 335
Lettau, M., see Buchmueller, T. 692. 694 Linna, M., see Nordblad, A. 1278, 1283
Leu, R.E. 11. 13, 18, 19, 22, 25, 33, 37, 39, 46, 448, Linnerooth, J. 543
1565, 1577,1580n. 1593, 1811n Linnosmaa, I.L,see Sintonen, H. 429n
Leung, S. 288, 289, 1638n Lippman, S.A. 228
Leveson, I., see Auster, R. 299, 353, 381, 396 Lipscomb, J. 1732
Levi, E.H., see Director, A. 1474 Lipscomb, J., see Luce, B.R. 189
Leviatan, L.I. 22 Lipsey, R., see Eaton, C. 1102n
Levin, A. 43,44 Liston, D.K., see Jensen, G.A. 712
Levin, D., see Baltagi, B.H. 1548. 1559. 1570, 1587, Litan, R., see Harrington, S. 1362
1588, 1594 Little, I.M.D. 107
Levin, R.C. 1317 Liu, B.M., see Liu, K. 960
Levine, D., see Fudenberg, D. 883 Liu, CF., see Pope, G.C. 798, 800, 811n, 812
Levine, H.J., see Wallack, S.S. 937 Liu, J.-T., see Hsieh, C.-R. 1594
Levine, M/I. 75 Liu, K. 960, 975, 976
Levine, P., see Zimmerman. D. 686 Liu, K., see Coughlin, T.A. 960
Levine, S., see Wechsler, H. 1712 Liu, K., see Manton, K.G. 987
Levinsohn, J., see Berry, S. 1431, 1432, 1434 Liu, X., see Liang, J. 959
Levit, K.R. 125n, 159, 959, 960 Livennore, G., see Harwood, H. 1664
Levit, K.R., see Lazenby, H.C. 174n Llewellyn-Thomas, H. 1734, 1748
Levy, H. 651 Llewellyn-Thomas, H., see Sutherland, H.J. 1741
Levy. H., see Krueger, A.B. 741 Llodra, J.C. 1288, 1289
Levy, J.M. 515 Lluch, C. 1559
Levy, J.M., see Greenwald, L.M. 829 Localio, A.R. 516. 1368
Lewin, L.S. 1156 Locay, L., see Gertler, P. 280, 283, 1222
Author Index I-23
Locke, J. 1890 Lynk, W. 1129, 1130, 1159, 1161,1426,1429. 1440,

Lockwood 1755 1448,1449, 1453, 1477
Lockwood, B., see De Meza, D. 1147 Lyon, A.B. 1573, 1660n
Lockwood, M. 1815n, 1887 Lyon, T.P, see Crocker, K.J. 1476
Loewenstein, G. 1742, 1743, 1751
Loewenstein, G., see Ubel, P. 88 Ma, C.A. 466n, 468, 477n, 485, 489n, 490, 491, 491n,
Logan, R., see Forsyth, G. 1240n 492, 494, 495, 518, 521,522, 542, 544, 549, 604, 857,
Lohr, J., see Freed, G. 1065 859, 861,867, 871,871n, 906, 912, 917, 924, 925,
Lohr, K. 85 1120, 1416n
Lohr, K., see Bailit, H. 1274 Ma, C.A., see Brundin, I. 492n
Lomas, J., see Giacomini, M. 1190 Ma, C.A., see Gaynor, M. 1470n, 1477
Lomas, J., see Hutchinson, B. 1189 MacArthur, C. 1056
Lomas, J., see Hutchison, B. 83n Macbeth, C.A., see Saywell, R.M. 1072n
Lombard, H.L. 1545 MacFarlan, M., see Gerdtham, U.-G. 11, 18, 32, 38, 39,
Lona, M., see Schokkaert, E. 806, 830, 834n 45,46,48,49
Londono, J.L. 830 MacFarlane, D., see Levine, M. 75
Lonergan, J., see Veale, B.M. 1183 Machina, M. 102n, 416n
Long, A.M., see Levit, K.R. 959, 960 MacLachlan, G. 1805n
Long, M., see DesHamais, SI. 863-865 MaCurdy, T. 362n
Long, S. 619, 654, 1067 MaCurdy, T., see Dick, A. 959
Loomes, G. 102n, 1737, 1743, 1747, 1752-1753 MaCurdy, T., see Garber, A.M. 791, 975, 977, 978
Lopez, A. 312 MacWhinney, J., see Szilagyi, P.G. 737
Lopez, A., see Murray, C. 94, 102, 1007, 1542, 1867, Maddala, G.S. 287, 288, 302
1883n, 1899, 1902 Maddala, G.S., see Jeong, J. 272
Lopez, A., see Peto, R. 1546 Madden, G.J., see Bickel, W.K. 1564, 1564n
Lovell, K., see Aigner, D. 333, 1277 Madrian, B. 666, 666n, 667-669, 670n, 672, 673,
Lowry, R. 900 675-678, 679n, 680, 1415
Lu, B.Y., see Jin, S.G. 1575 Madrian, B., see Currie, J. 676n, 1062
Madrian, B., see Cutler, D. 661n, 692, 696
Lu, J.L. 1320
Madrian, B., see Gruber, J. 652n, 654, 668, 669, 671,
Lu, M. 498n
671n, 672, 673,674n, 675, 675n, 677, 679, 680n
Lubeck, D.P. 735
Maerki, S., see Luft, H. 620, 730, 735, 741, 1115n, 1450
Lubitz, J. 774, 798, 804, 806, 984, 985
Magid, D.J. 435
Lubitz, J., see Beebe, J.C. 777n, 798
Magotich, M., see Greenfield, S. 734
Lucas, C., see Young, G.J. 1160
Maher, J., see Rowlands, O. 1178
Luce, B.R. 189
Maibach, E. 623
Luce, B.R.,see Brown, R.E. 1685n
Mailath, G.J., see Kandori, M. 237
Luchok, K., see Riportella-Muller, R. 1060 Makela, K., see Bruun, K. 1649
Ludwick, R., see Dranove, D. 1130, 1449 Maki, D.R. 1043n
Luft, H. 620, 713, 714, 716, 720, 723, 725, 728-732, 735, Makuc, D. 1429
739, 741, 742, 779, 782, 822, 829n, 1115n, 1116, Malcomlson, J. 866
1132n, 1434, 1450, 1709 Malcomlson, J., see Chalkley, M. 62n, 76, 550, 501n,
Luft, H., see Garnick, D.W. 734, 1431, 1433 522, 553, 856, 857, 859 861, 874, 881n, 1098, 1407n
Luft, H., see Giacomini, M. 83 Males,M.A. 1647,1648
Luft, H., see Kahn, J.G. 814 Malin, H. 1660
Luft, H., see Miller, R.H. 604, 630, 711, 725, 731, 732, Malitz, FE., see Wartman, S.A. 1179
738, 739, 742, 925,928 Maljanen, T., see Sintonen, H. 1260-1263, 1265, 1267,
Luft, H., see Robinson, J. 728, 729, 829n, 1114, 1116, 1270, 1271. 1273, 1277
1168n, 1412, 1428, 1450, 1451 Malmgren, J.A., see Andresen, E.M. 1745
Luik,J.C. 1588, 1591 Manau, C. 1285, 1286, 1288 1289
Lum, S.K.S. 169n Manchester, P.B. 1570
Lumio, M., see Bruun, K. 1649 Manderscheid, T., see Frank, R.G. 901
Lumsdaine, R. 677, 679, 680 Manelli, A.M. 873
Lundback, M. 499 Mangen, S.P. 936, 940
Lundberg, G.D., see Werman, B.G. 516 Manheim, L.M., see Hughes, S.L. 961
Lundberg, L., see Gerdtham, U.-G. 1854 Manion, R., see Goddard, M. 1179
Luoma, K., see Nordblad, A. 1278, 1283 Mann, C.C. 1306
Luras, H., see Iversen, I. 1188 Mann,J. 1160, 1763n
Lurie, N. 431, 735, 925 Mann, J., see Buchanan, J.L. 728, 729, 733
Lurie, N., see Johnson, A.N. 735 Manning, C.L., see Russell, L.B. 600
Lurie, N., see Lillard, L. 1708 Manning, R., see Hu, T.-W. 1544
Lusk, E.J., see Cassileth, B.R. 1728 Manning, R.L. 1320
Lusted, L. 516 Manning, W.G. 79, 249, 271, 286, 288, 290, 291, 426,
Lustig, H., see Schokkaert, E. 806, 830, 834n 437n, 438-440,475, 587,728, 729, 732, 735, 775n,
1-24 Author Index
788,808,827n, 910-9i2, 922, 100llOn, 1255, 1259, Marquis, M.S., see Newhouse, J.P. 286, 429n, 433, 449,
1261,1261n. 1262-1267,1269,1269n, 1271-1273, 1262
1285, 1286, 1288, 1289, 1575, 1577, 1579, 1580. Marquis, M.S., see Wells, K.B. 738
1580n, 1581,1581n, 1582, 1643, 1650, 1651n, 1660, Marsden, M.E.,see Bray, R.M. 1657n
1687, 1707, 1841 Marshall, A. 61, 1556
Manning, W.G., see Danzon, P. 526 Marsteller, J.A. 721
Manning, W.G., see Duan, N. 286, 287, 290, 291, 1265, Martial, L.R., see Sweanor, D.T. 1567, 1571
1267. 1273 Martie, C.W., see Doherty, N.J.G. 1288
Manning, W.G., see Frank, R.G. 909 Martin, AIB., see Levit, K.R. 959, 960
Manning, W.G., see Greenfield, S. 492, 734 Martin, D. 736, 1177
Manning, W.G., see Keeler, E.B. 316, 325, 911,934 Martin, S. 1229, 1234
Manning, W.G., see Keeler, T.E. 1548, 1549, 1561, 1569, Martin, S., see Sheldon, T.A. 830, 833
1596, 1604n Martinez-Carretero, J., see Manau, C. 1285, 1286, 1288,
Manning, W.G., see Khandker, R. 717, 1123 1289
Manning, W.G., see Leibowitz, A. 286, 439, 440, 1301 Marton, K.I., see Dietrich, A.J. 1179
Manning, W.G., see Lillard, L.A. 419, 1708 Martorell, R. 1062n
Manning, W.G., see Luce, B.R. 189 Martyn, C.. see Barker, D.J. 1062
Manning, W.G., see Mullahy, J. 197, 268, 269 Maryniuk, G.A. 1264, 1290
Manning, W.G., see Newhouse, J.P 436n, 437n, 438, Mashaw, J. 1002n, 1008
441,442,790, 791n, 792, 792n, 793, 793n, 797,805, Maskin, E., see Fudenberg, D. 883
814, 932, 933 Mason, B., see Gudex, C. 1241
Manning, W.G., see O'Grady, K. 440 Mason, W.M., see Griliches, Z. 398
Manning, W.G., see Sturm, R. 737 Masse, L., see Tremblay, R. 1080
Manning, W.G., see Wasserman, J. 299, 1547, 1550, Masson, A. 1321, 1324
1552, 1596, 1597 Masten, S.E. 1473
Manning, W.G., see Wells, K.B. 911 Mathers, C. 1725
Manning, W.G., see Zweifel, P. 382n, 580n, 584, 759, Mathewson, F.J. 727
1182, 1264 Mathewson, G.F 1475
Manning, W.J. 23 Mathios, A., see DeCicca, P. 1554, 1555
Mannix, E. 95 Mathios, A., see Ippolito, P. 1700-1702
Manochia, M., see Bakke, O.M. 1315n Matsaganis, M. 829n, 830
Mansfield, E. 1308, 1310 Matus, M. 896
Manski, C.E 292, 1644 Mauldon, T. 736
Manson, J.E., see Camargo Jr., C.A. 1651, 1652 Maurizi, A. 1257
Manton, K.G. 864, 986, 987 Maurizi, A., see Benham, L. 469, 474
Manton, K.G., see Liu, K. 960 Max, W.. see Bartlett, J.C. 1575n, 1580n
Manton, K.G., see Tolley, H.D. 804 Max, W.,see Miller, L.S. 1575n, 1580n
Maxwell, R.J. 18
Mao, Z.Z. 1548
Maxwell, S., seeWeiner, J.P 143n,798, 799, 81ln, 814,
Mapp, T.J., see Donaldson, C. 278
932
Maranvanyika, E. 1548
May, R., see Anderson, R. 1763n
Marback, M.E., see Neuman, P. 83
Mayer, J. 1081
Marceau, N., see Boadway, R. 999n, 1022n Maynard, A. 96, 1187, 1809n, 1815n
Marciniak, M., see Hu, T.-W. 1544 Maynard, A., see Culyer, A.J. 1809n
Marder. W.D. 472 Maynard, A., see Gudex, C. 1241
Margulies, R., see Lewin, L.S. 1156 Maynard, A., see MacLachlan, G. 1805n
Marine, W., see Miller, L.A. 1064 Mayo, J.W. 1524, 1526
Marinker, M., see Maynard, A. 1187 Mayo, N.E., see Hamilton, B.H. 325
Mark, D.B. 215 Mays, J.W., see Price, J.R. 616, 620, 773
Mark, T. 735, 1163 Mays, N. 83n, 92
Mark, T., see McKusick, D. 902 McAfee, R.P. 872
Markham, B., see Levine, M. 75 MeAlpine, D., see Mechanic, D. 912
Markheim, L., see Berry, C. 526 McAuliffe, R. 1587, 1588
Markowitz, S. 1653 McAvinchey, I.D. 412n, 583, 1220, 1230
Marmor. T.R. 940,941 McBride, T., see Liu, K. 975, 976
47
Marmor, T.R., see Evans, R.G. 18 n McBride, T., see Coughlin, T.A. 960
Marquis, K.H.,seeManning, W.G. 1265, 1267 McCaffrey, D., see Moscovice, I. 975
Marquis, M.S. 236, 437n, 439, 620, 621, 773 McCall, B., see Card, D. 1032
Marquis, M.S., see Danzon, P.M. 526 McCall, D., see Donaldson, C. 1285, 1288, 1290
Marquis, M.S., see Hosek, S.D. 729, 735.739 McCall, J.J., see Lippman, S.A. 228
Marquis, M.S., see Long, S. 654, 1067 McCall, N. 792, 984
Marquis, M.S., see Manning, W.G. 23, 286, 288, 426, McCall, N., see Rice, T. 515n
437n, 587, 775n, 827n, 912,1263,1264,1265,1267, McCall, N., see Scitovsky, A.A. 430, 582, 617
1285, 1286, 1288, 1289 McCall, T.B. 487
Author Index I-25
McCallum, J., see Veale, B.M. 1183 McGuire, T., see Klevorick. A.K. 475, 476
McCann, D., see Feldman, R. 727, 741 McGuire, T., see Ma, C.A. 466n, 468, 485, 489n, 491n,
McCann, R., see Eng, C. 961, 962 492, 494, 518, 521,542, 549, 604, 906, 912, 917, 924,
McCann, R.W., see Kopit, W.G. 1440, 1448 925, 1416n
McCarthy, T. 476, 512, 830, 831, 833, 834n Mclnerny, T., see Szilagyi, P.G. 737
McCarthy. T., see Hoffmeyer, U. 1145 Mclntosh, E., see Ryan, M. 1181
McCashin, B., see Lane, D. 964 McKee, M., see Leon, D.A. 1646
McClellan, M. 270, 557, 863, 866, 1120n, 1499n, 1530 McKenna, C.J. 228
McClellan, M., see Cutler, D. 237n, 630n, 135, 159, 163, McKenna, C.J., see Hey, J.D. 228, 234
173 McKenzie, G.W. 252n
McClellan, M., see Eichner, M. 572, 1415 McKenzie, L., see Cairns, J. 1731
McClellan, M., see Garber, A.M. 791 McKenzie, L., see Loomes, G. 102n, 1743, 1752
McClellan, M., see Heidenreich, P. 173 McKenzie, R.B. 1556
McClellan, M., see Kessler, D. 516, 1356, 1368, 1451 McKeown, T. 1725
McCloskey, D.N. 1753 McKinlay, J.B. 506
McClure, W. 782, 798 McKusick, D. 902
McClure, W., see Christianson, J.B. 742 McLachlan, W., see Ashley, J. 768n
McCombs, J.S. 736 McLaughlin, C., see Carey, T.S. 733
McCombs, J.S., see Kasper, J.D. 730 McLaughlin, C., see Merrill, J. 741
McCormick, M. 1062,1069 McLaughlin, C.G. 740, 741
McCoskey, S. 43, 44, 48 McLaughlin, C.G., see Warner, K.E. 1585
McCulloch, J., see Goldman, W. 906, 912, 924 McLaughlin, T.J., see Soumerai, S.B. 1327
McCusker, J. 736 McLean. R.A. 476
McDermed. A.A., see Clark, R. 690n McLeod, I., see Thompson, M.E. 1565, 1593
McDonnell, P.A., see Levit, K.R. 959, 960 McLeod, P.B. 1589
McDowell, I. 1743 McMillan, J., see McAfee, R.P 872
McFadden, D. 1431 McNamee, H.B., see Mello, N.K. 1642
McFarland, D.A., see Mayo, J.W. 1524, 1526 McNeil, B. 1741
McFetridge, D.G. 1319 McNeil, B., see Gatsonis, C.A. 306
McGarry, K., see Hurd, M. 676, 678, 1682, 1698 McNeil, B., see Weinstein, M. 478n
McGeer, A., see Wolfson, A.D. 1746 McNeil, J.M. i000n, 1001
McGinnis, J. 1677, 1697 McPhee, S., see Gamick, D.W. 1431, 1433
McGinnis, J., see Strumwasser, I. 730, 737 McPhee, S., see Luft, H.S. 1434, 1450
McGinty, M.J., see Keeler, E.B. 1162, 1163 McPhee, S., see Robinson, J. 1116, 1450, 1522
McGlynn, E. 725, 1133 McPherson, J., see Hausman, D. 66, 1869
McGonagle, K., see Kessler, R.C. 897 McPherson, K. 243, 1228
McGowan, J., see Fisher, F. 1444 McPherson, K., see Daly, E. 1748
McGuigan, K.A., see Mauldon, T. 736 McRae, J.J. 1325
McGuinness,T. 1587, 1593 McVicar, D., see Chalkley, M. 866
McGuire, A. 1,17, 19, 23, 48, 64n Meade, T.W., see Atkinson, A.B. 1663
McGuire, A., see Fenn, P. 874, 875 Meads, C., see Laugesen, M. 1589, 1590
McGuire, A., see Hughes, D. 434 Mechanic, D. 463n, 520, 521, 521n, 904, 912, 918, 925,
McGuire, A., see Mooney, G. 1812 937
McGuire, A., see Parkin, D. 18, 19, 22-24 Mechoulan, S. 1786n, 1787
McGuire, T. 79n, 242n, 252, 414, 475, 502, 507, 507n, Meckling, W.H., see Jensen, M.C. 1144
508,509n, 518,524, 589, 617, 725, 826n, 908, 909, Meenan, R., see Hombrook, M.C. 789
910n, 911,913n, 918, 923, 926, 928, 941, 1182, 1188, Meertens, R.,seevandeKar,A. 1181, 1182
1368, 1407n, 1416, 1417 Meghir, C., see Blundell, R.W. 313
McGuire, T., see Bemdt, E.R. 898, 903, 904 Mehrez, A. 102, 103, 212, 1729, 1735-1737, 1742
McGuire, T., see Cromwell, J. 919 Mehrez, A., see Gafni, A. 1735, 1737
McGuire, T., see Ellis, R.P. 449, 487, 491n, 522, 522n, Meier, K.J. 1593, 1594
551,596, 603, 715, 723, 724, 786, 798, 819, 827, 856, Meier, M., see Zweifel, P. 451,984, 985
856n. 863,864, 866, 911,917, 919,920, 1119, 1120, Meiners, M., see Hodgson, T.A. 1575, 1664
1167 Melinkovich, P., see Miller, L.A. 1064
McGuire, T., see Etmer, S.L. 778, 810, 815, 933, 934 Mello, N.K. 1642
McGuire, T., see Frank, R.G. 429n, 595, 609n, 711n, 712, Melnick, G.A. 742, 1126, 1130, 1429, 1448, 1449, 1451,
722, 728, 778, 815, 816, 901,904, 906, 909, 913, 1463, 1468
915n, 916, 924-926, 930n, 931,940 Melnick, G.A., see Keeler, E. 1130, 1159, 1448, 1449
McGuire, T., see Freiman, M.P 863, 864, 920 Melnick, G.A., see Mann, J. 1160
McGuire, T., see Frisman, L.K. 941 Melnick, G.A., see Zwanziger, J. 742, 1116, 1451, 1527,
McGuire, T., see Glazer, J. 482n, 485, 486, 486n, 487, 1529
773,815,856,913,933, 1120, 1416n Meltzer, D. 129, 203, 395n, 1705
McGuire, T., see Hodgkin, D. 556, 603, 857, 865, 1528n Meltzer, D., see Johannesson, M. 98
McGuire, T., see Jacobsen, K. 940 Mendelson, D.N., see Rubin, R.J. 1366, 1369
I-26 Author Index
Mendelson, D.N., see Schwartz, W.B. 1361 Mitchell, J.B., see Cromwell, J. 17,510
Mendelson, J.H., see Babor, T.F 1642, 1643 Mitchell, J.B., see Frank, R.G. 1160
Mendelson., J.H., see Mello, N.K. 1642 Mitchell, J.B., see Freiman, M.P. 934
Mendelson, M.A. 968 Mitchell, J.B., see Sloan, F. 552, 557, 1075
Menke. T. 602 Mitchell, J.B., see Wedig, G. 525
Mennemeyer, S. 860, 1134 Mitchell, J.M. 517, 1193
Meredith, L.S., see Sturm, R. 737 Mitchell, J.M., see Fournier, G.M. 1452
Mergenhagen, P.M., see Sloan, F.A. 1350, 1354, 1360, Mitchell, L. 1288, 1289
1362 Mitchell, 0. 665, 665n, 668, 676
Merikangas, K.R., see Swendsen, J.D. 898 Mitchell, O., see Dwyer, D. 1025n
Meriam-Webster 89 Mitchell, O.S., see Reed, S.K. 737
Merrick, E.L. 906, 925 Mitchell, R.C. 1752
Merrigan, P., see Hamnilton, V.H. 302 Mitchell, S.A., see Morrisey, M.A. 1512, 1525
Merrill, J. 618, 741 Mitry, D.J. 1280
Merrill, J., see McLaughlin, C.G. 740, 741 Mitry, N.W., see Mitry, D.J. 1280
Merrill, J., see Somers, S.A. 978 Miyamoto, J.M. 1730, 1740
Metra Consulting Group 1587, 1594 Miyazaki, H. 1415
Meyer, B.D., see Anderson, P.M. 660n, 693n Mobley. L.R. 1424n, 1435
Meyerowitz, B.E. 1728 Modigliani, F. 1558
Meyers, S.M., see Roddy, P.C. 418 Moffit, R.E., see Butler, S.M. 831
Michael, R.J. 938, 939 Moffitt, R. 449, 682, 683, 684n, 982
Michael, R.T. 350, 352, 372, 376 Moffitt, R., seeHoynes, H. 1043n
Midanik, L. 1639 Mokyr, J. 1056
Miles,J.J. 1408n, 1409, 1420, 1458n, 1471n Molana, H., see Milne, R. 24, 24n
Milgrom, P. 1122n, 1416n Monfort, A., see Gourieroux, C.A. 271, 317
Milgrom, P., see Conrad, D.A. 619, 1261-1263, 1270, Monheit, A. 690,691
1272 Monheit, A., see Berk, M.L. 571
Milgrom, P., see Holmstrm, B. 851, 1 10n, 1190 Monheit, A., see Cohen, J.W. 134n
Milgrom, P., see Whitney, C. 1116n Monheit, A., see Cooper, P. 665, 668
Milgrom, P.R. 872 Monheit, A., see Farley, P.J. 619
Mill, J.S. 61 Monheit, A., see Mueller, C.D. 1255, 1259, 1261-1265,
Millar, AB., see Grunewald, PJ. 1640 1267, 1269-1272
Miller, A., see Williams, B. 1056, 1076, 1077 Monheit, A., see Short, P. 681 n. 1061
Miller. D. 1814, 1815n, 1905 Monohan, J., see Steadman, H.J. 898, 899
Miller, D.S., see Cassileth, B.R. 1728 Montgomery, E. 683, 684, 690n
Miller, J., see Douglas, G. 1412 Montgomery, E.,see Evans, W.N. 1596, 1597
Miller, J., see Korenman, S. 1063 Montgomery, J., see McGuire, T.G. 926, 928
Miller, L.A. 1064 Montgomery, M. 692,695
Miller, L.S. 1575n, 1580n Moon, M. 937
Miller, L.S., see Bartlett, J.C. 1575n, 1580n Mooney, C., see Phelps, C.E. 240, 240n, 244, 245, 252,
Miller. L.S., see Rice, D. 1010n, 1660, 1664 254, 256, 904
Miller, M.E. 864 Mooney, G. 1, 64, 64n, 65, 69n. 75, 89, 92, 92n, 107,
Miller, P. 95, 1187 108,414,499,1181,1725, 1805n, 1807, 1812-1814,
Miller, P., see Rao, R. 1587 1814n, 1835n. 1893, 1894
Miller, R., see Szilagyi, P.G. 737 Mooney, G., see Anderson, T.F. 70n
Miller, R.D. 692, 694 Mooney, G., see Krasnik, A. 1189
Miller. R.H. 604, 630, 711, 725, 731,732, 738, 739, 742, Mooney, G., see Kristiansen, I.S. 1177, 1188, 1189
925, 928 Mooney, G., see McGuire, A. 1, 17, 64n
Miller, R.H., see Luft, H.S. 829n Mooney, G., see Ryan, M. 1259
Miller, T.R. 1661n, 1664 Mooney, G., see Wiseman, V. 1703
Miller, W., see Bien, T. 1711 Moore, J., see Hart, O. 1149
Million, A., see Geil, P. 315, 319 Moore, J.D. 1494, 1495
Mills, A., see Pannarunothai, S. 1806n Moore, M. 1578, 1641, 1642, 1650n
Mills, D.H. 1351 Moore, M., see Cook, P.J. 1579n, 1639, 1641, 1652,
Mills, R.P 1234 1657, 1662n, 1679, 1707
Millward, R. 1209 Moore, M.H. 1660
Milne, E. see Cameron, A.C. 315, 319 Moore, P.S., see Fitzgerald. J.F. 598, 630, 734
Milne, R. 24, 24n Moore, W.J. 1327
Mincer, J. 349, 397 Mor, V. 982
Miron, J.A. 1649n Moran, J.R., see Nelson, J.P. 1645
Mirrlees. J.A. 586n Morey, M.J., see Goel, R.K. 1586, 1587, 1589. 1594
Misamore, G.W., see Palmer. R.M. 600, 630 Morgan, M.V., see Lewis, J.M. 1288, 1290
Mishan, E. 99, 109, 1867n, 1887n Morgenstern, O., see von Neumann, J. 1734
Mitchell, J.B. 485 Morlock, L., see Taube, C.A. 937
Author Index I-27
Morlock, L.L. 936 Murphy, E.H., see Berkowitz, M. 999n

Morlock, L.L., see Wartman, S.A. 1179 Murphy, K., see Schneider, L. 1586-1589, 1593, 1594
Morris, C.N. 325, 327, 437n, 975, 978 Murphy, K.M., see Becker, G.S. 311,392n, 398, 401,
Morris, C.N., see Duan, N. 286, 287, 290. 291, 1265, 403, 1551, 1554, 1559, 1560, 1560n, 1561-1563,
1267, 1273 1569,1570, 1572n, 1636, 1682, 1685, 1713
Morris. C.N., see Manning, W.G. 1265, 1267 Murphy, K.M., see Chaloupka, EJ. 1646n
Morris, J.N., see Borsch-Supan, A. 282, 977 Murphy, K.M., see Schneider, L. 1699
Morris, N.E., see Johnson, A.N. 735 Murphy, R.D., see Ippolito, R.A. 1547, 1588, 1593, 1594
Morris, R., see Roland, M. 1228 Murray, C. 94, 102, 686, 1007, 1542, 1726. 1867, 1883n,
Morris, S., seeLenert, L.A. 1753 1899, 1902
Morrisey, M. 596, 598, 718, 727, 778, 780, 1123n, 1124, Murray, D.M.,see Forster, J.L. 1597
1158-1160,1165, 1423,1428n, 1512, 1525 Murray, J., see Mitchell, L. 1288, 1289
Morrisey, M., see Bronstein, J. 1132 Murray, M., see Files, A. 830
Morrisey, M., see Feldman, R. 58, 68, 78n Murray, R. 122n, 170, 172, 172n
Morrisey, M., see Jensen, G. 133, 712, 780 Murtaugh, C.M. 959, 975-979
Morrisey, M., see Lanning, J.A. 1504, 1514, 1527 Murtaugh, C.M., see Kemper.P. 959
Morrisey, M., see Lynk, W.J. 1477 Murthy, V.N.R. 24
Morrisey, M., see Mennemeyer, S. 860, 1134 Musacchio, R.A., see Willke, R.J. 602
Morrisey, M., see Ohsfeldt, R. 493 Musgrove,P. 71, 74n
Morrisey, M.,see Sloan, EA. 599, 1164, 1165, 1528 Mushkin, S. 3, 67, 349, 1491, 1680
Morrisey, M., see Wedig, G.J. 1163, 1164 Mushlin, A.I., see Phelps, C.E. 103, 185, 252
Morrison, E.M., see Garnick, D.W. 734 Musliner, M.C., see Weissert, W.G. 960
Mornison.,G.C. 1737 Mutran, E., see Rabiner, D.J. 962
Morrison, K.R., see Rice, T. 429n Mutran, E., see Weissert, W.G. 960
Morse, R., see Leaver, S.G. 137n Muurinen, J. 23, 68n, 365, 377, 383, 389, 392, 1680
Mort, E.A. 484 M.V.A. Consultancy 1227
Mortimer, E., see Plotkin, S. 1768n Mwabu, G. 282
Morton, E.S. 1324n, 1328 Myerberg, C., see Myerberg, D.Z. 1077
Moscovice, I. 975 Myerberg, D.Z. 1077
Moscovice, I., see Lurie, N. 735, 925 Myers, G., see Collishaw, N.E. 1575
Moser, J.W., see Willke, R.J. 602 Myint, H. 107
Moses, K.E., see Moulton, B.R. 138n, 148n Mynttinen, A., see Sintonen, H. 1277
Moskowitz, R.W., see Katz, S. 975 Myrdal, G. 1665
Moulton, B.R. 138, 138n, 148n
Moulton, B.R., see Abraham, K.G. 123n, 130n
Moulton, B.R., see Armknecht, PA. 158n Nabi, I., see Hammer, J.S. 1806n
Moulton, L., see Manning, W.G. 271, 1643 Nagey, D., see Kogan, M. 1064
Moy, E. 477 Nagi, S.Z. 999n
Mroz, T.A., see Bollen, K.A. 303 Nakahara, T., see Bhattacharya, J. 325, 326, 583
Mroz, T.A., see Guilkey, D.K. 303 Nanda, P. 272
Mroz, T.A., see Lee, R.H. 1417 Naon, D. 963
Msall, M.E., see Hoyme, E. 1066 Napier, K. 1545, 1576, 1576n
Mueller, C. 1255, 1259, 1261-1265, 1267, 1269-1272 Napolitano, D.M., see Harwood, H.J. 1664
Mueller, C., see Mark, T. 735 Nasar, S. 896n
Mulcahy, M., see Ashmore, M. 2, 3 Nash, K. 1276, 1280, 1282
Mullahy, J. 197, 268, 269, 290, 291, 300, 315-321, 323, National Advisory Mental Health Council 916
324, 788, 788n, 898, 1549n, 1558, 1656, 1657, 1681, National Audit Office 866
1708, 1794 National Cancer Institute 1703
Mullahy, J., see Grossman, M. 404 National Center for Addiction and Substance Abuse 1079
Mullahy, J., see Kopp, R.J. 333 National Governor's Association 1068, 1077, 1078
Mullen, P. 1242, 1243 National Household Survey on Drug Abuse 1633n
Muller, C., see Bolduc, D. 282 National Oceanic and Atmospheric Administration
Muller, L.S. 1015 1226n, 1227n
Mulley Jr., A.G., see Barry, M.J. 216 National Vaccine Advisory Committee 1066
Mulley Jr., A.G., see Fowler Jr., EJ. 216 Navin, J., see Montgomery, E. 683, 684
Mulligan, C.B., see Becker, G.S. 398, 401, 403, 404, Naylor, C.D. 1240
1563, 1682, 1713 Naylor, C.D., see Mark, D.B. 215
Mullner, R., see Sloan, EA. 1160 Nazarian, L., see Szilagyi, PG. 737
Mulvey, E.P., see Steadman, H.J. 898, 899 Neale, M., see Mannix, E. 95
Mumford, L., see Starfield, B.H. 249, 249n Neary, I. 1299
Mumford, L., see Weiner, J.P. 249, 249n, 798, 814 Needleman, H.L. 1066
Munch, N., see Malin, H. 1660 Needleman, J. 1124, 1166
Mundlak, Y. 314 Neipp, J. 1415
Murphy, B., ee Catron, B. 140n, 149, 149n Nelson, A.F, see Hart, L.G. 714
1-28 Author Index
Nelson, C., see Kessler, R.C. 897 Newhouse, J.P., see Norton, E.C. 959, 978
Nelson, C.L., see Mark, D.B. 215 Newhouse, J.P., see O'Grady, K. 440
Nelson, E.C., see Greenfield, S. 492, 734 Newhouse, J.P., see Phelps, C.E. 229, 234, 429, 430, 434,
Nelson, J.P. 1645 580-583, 1222
Nelson, L., see Ohsfeldt, R. 493 Newhouse, J.P., see Ware Jr., J.E. 739
NelsonP. 1182 Newhouse, J.P., see Wasserman, J. 299, 1547, 1550,
Nelson, R., see Cohen, W.M. 1318, 1318n 1552, 1596, 1597
Nelson, R., see Levin, R.C. 1317 Newhouse, J.P., see Wells, K.B. 911
Nelson, R., see McGuire, T. 252n Newman, R., see Moore, W.J. 1327
Nelson, W. 1286-1288 Newton, J. 1186, 1228
Neslusan, C.A., see Coulson, N.E. 315 Newton, P., see Bailey, J. 1186
Nestel, G., see Chirikos, T. 101 In Newton, P., see King, N. 1186
Netter, N., see Mann, J. 1763n Ng, Y.C., see Hunt-McCool. J. 289
Neuburger, H., see Parsonage, M. 1209 Ng, Y. 62n, 1732
Neuhauser, D., see Weinstein, M. 478n Nguyen, N.X. 504n, 515
Neuman, P. 83 Nicholl, J.P., see Brazier, J. 1744,. 1745
Neuman, P.. see Lamphere, J.A. 766 Nichols, A. 610n
Neumann, L., see Lynk, W. 1130, 1426, 1449 Nichols, L.M., see Marsteller. J.A. 721
Neumann, P.J. 1752 Nichols, P., see Broman, S. 1062
Neun, S., see Santerre, R. 1, 464n Nicholson, E.M., see Cordes, J.J. 1583
Neutra, R., see Weinstein, M. 478n Nicolas, P., see Warner, K.E. 1600n, 1607 1609
New, B.,seeDixon, J. 1241 Niessen, L.C. 1285-1289
Newacheck, P. 1065 Niskanen, T., see Nordblad, A. 1278, 1283
Newacheck, P., see Halfon, N. 1066 Nitzan, S. 1416n
Newacheck, P., see St. Peter, R. 1060 Nocera, S. 960, 964, 972, 974
Newbold, K.B., see Birch. S. 1835n, 1839 Noel, G.L., see Reed, W.W. 1745
Newbold, R.C., see Angus, D.C. 733 Noether, M. 471n, 1128, 1168n, 1449
Newcomer, R. 736 Noll, R.G. 1492, 1493
Newell, C., see McDowell. I. 1743 Nord, E. 88, 89, 94, 1733, 1899, 1903-1905, 1905n
Newhouse, J.P. 1, 2,1 1, 13, 17, 18, 21, 22, 22n, 23, 24n, Nordblad, A. 1278, 1283
25, 46, 74, 83, 123, 162n, 163,249. 267,268, 286,
Nordhaus, W.D. 148n
334, 335, 421, 429n, 433-435, 4361, 437n, 438, 441. Nordquist, G. 1684
442, 449, 451,453,474, 475,487,491n, 499n, 501,
Norman, G.R.. see Streiner, D.L. 1743, 1744
504, 505,512, 520, 522n, 542, 554, 555,557, 580, Norman, P., see Grogan, S. 1181
582-584, 587,596, 599, 600, 604, 607, 629, 715-717,
Norman, V., see Dixit, A. 1558
724, 726, 731,742, 759, 760n, 763,771,772, 772n,
Normand, S.-L., see Gatsonis., C.A. 306
773,775, 778, 780, 786, 787,790, 791n, 792, 792n,
Norquist, G.S. 736
793,793n, 797,798, 805,806, 814, 817-819, 821,
NoIr; R. 1593
822, 826, 826n, 827, 829n, 832, 854-856, 862, 864,
Norstrm, T. 1635n
867, 870,91 1,919, 920, 926, 928,929, 932, 933, 935,
Northcraft, G., see Mannix. E. 95
985,1059n, 1105, 1113n, 1121, 1153,1157, 1262,
Norton, E.C. 274, 306, 325, 327, 863, 959, 967, 972, 978,
1280, 1264, 1271, 1301, 1413, 1415, 1436, 1529n,
1708, 1709, 1817n 980,982, 1160, 1167, 1644
Norton, E.C., see Fisher, W.H. 873
Newhouse, J.P., see Bailit, H. 1274
Norton, E.C., see Lindrooth, R.C. 981
Newhouse, J.P., see Berndt, E.R. 119
Newhouse, J.P., see Carter, G. 782 Norton, E.C., see Morris, C.N. 325, 327, 975, 978
Newhouse, J.P. see Cutler, D. 135, 159, 163, 173, 237n, Norton, E.C., see Phillips, C.D. 961
630n Norton, E.C., see Sloan, EA. 979
Newhouse, J.P., see Duan, N. 286, 287, 290. 291, 1265, Notman, E.H., see Ettoer, S.L. 778, 810. 815, 933, 934
1267, 1273 Notman, E.H., see Jacobsen, K. 940
Newhouse, J.P., see Ettner, S.L. 778, 810, 815, 933, 934 Novotny, T.E., see Shultz, J.M. 1575
Newhouse, J.P., see Frank, R.G. 609n, 924, 926 Nozick, R. 89n, 1808, 1890, 1891
Newhouse, J.P., see Gatsonis, C.A. 306 NTC Publications 1633n
Newbouse, J.P., see Helms, L.J. 431 Nunnally, J. 896
Newhouse, J.P., see Keeler, E.B. 422-425, 441, 579n, Nutting, P.A., see Franks, P. 1177
595,772n, 773,777n, 825,826, 911, 913 Nyamete, A., see Mwabu, G. 282
Newhouse, J.P., see Leibowitz, A. 286, 440, 1301 Nye,B.E 1354, 1360
Newhouse, J.P., see Lindsey, P.A. 717 Nyman, J. 960, 964, 965, 968 970, 1121
Newhouse, J.P., see Manning, W.G. 23, 249, 286, 288, Nyman, J.A., see Harrington, C.A. 958, 965, 966
438,439,728,729,732,735,911,922, 1010n, 1255, NZBR 1393
1261, 1263-1265, 1267, 1271, 1273, 1285, 1286,
1288, 1289, 1575, 1577, 1579, 1580, 1580n, 1581, O'Brien, B. 100, 434
1581n, 1582, 1650. 1651n, 1660, 1687, 1707, 1841 O'Brien, B., see Drummond, M. 100, 102n, 103, 105,
Newhouse, J.P., see McClellan, M. 270 188, 1284
Author Index I-29
O'Connell, B.. see Kennedy, E 1222 Pan, L., see Bruun, K. 1649
O'Donnell, O. 304, 1235 Pannarunothai, S. 1806n
O'Flaherty, B. 900 Pappas, G. 1852n
O'Flynn, R., see Gudex, C. 1241 Paranjpe, N.V., see Strumwasser, I. 730, 737
O'Grady, K. 440 Parente, S.T., see Phelps, C.E. 70n, 73n, 240n, 254
O'Hara, M., see Easley, D. 1149 Parfit, D. 1898
O'Neil, A., see Garnick, D.W. 734 Park, B.U. 334
O'Rourke, C.A. 1288 Park, R.E., see Leape, L.L. 244
Oates, W.E., see Cropper, M.L. 1679n Parker, L. 1066
Oaxaca, R.L., see Auster, R.D. 510 Parkin, D. 18, 19, 22-24, 1182, 1255. 1261, 1267, 1270,
OECD 13, 16, 17, 17n, 18, 20, 22, 33, 47, 57, 962, 963, 1273
1256, 1677, 1693, 1807, 1810n Parkin, D., see McGuire, A. 19, 23, 48
Office of Mental Health, State of New York 932 Parkin, D., see Yule, B. 1254, 1257, 1261, 1262
Ohsfeldt, R. 493, 1550, 1564, 1565, 1596, 1597 Parks, C., see Szilagyi, P.G. 737
Ohsfeldt, R., see Antel, J.J. 1515, 1525, 1527, 1528 Parloff, M.B., see Elkin, I. 904
Ohsfeldt, R., see Lanning, J.A. 1504, 1514, 1527 Parmley, W.W., see Glantz, S.A. 1546, 1581
Okunade, A.A. 23 Parshardes, P., see Dunne, J.P. 48
Olds,D. 1066, 1081, 1082 Parsonage, M. 1209
Olekalns, N., see Bardsley, P. 1561, 1594 1596 Parsons, D. 1000n, 1022n, 1024, 1026, 1026n
Olsen, D.M. 229 Pashardes, P. 1559
Olsen, E.O. 89, 90, 1812, 1893, 1894 Pass, C.L., see Witt, S.F. 1587, 1589
Olsen, E.O., see Holtman, A.G. 1258, 1261-1263, 1265, Paterson, M., see Banks, D.A. 1167
1268, 1270, 1271 Patrick, D.L. 216, 1732
Olsen, J.A. 1734 Patrick, D.L., see Andresen, E.M. 1745
Olsen, J.A., see Donaldson, C. 278 Pattison, R., see Melnick, G.A. 1429, 1448, 1449, 1463,
Olsen, R.J., see Hay, J. 286, 788n 1468
Olson, C. 657n, 688, 691, 694 Pattison, R.V. 1156
Ong, L.M. 1179 Pauker, S.G., see McNeil, B.J. 1741
Onstad, L., see Horbar, J. 1068 Paul-Shaheen, P 70n
Oppenheim, G.L. 484n Paulman, P.M., see DiFranza, J.R. 1592
Ordover, J.A. 1474, 1474n Pauly, M. 18, 48, 63, 67, 68, 72, 75, 76, 78, 84, 86n,
Oregon Health Services Commission 96 125n, 134, 134n, 160n, 164n, 186, 210, 256, 414,463,
Orphanides,A. 1562, 1584 468, 475,476, 477n, 494, 495,495n, 497, 501,502,
Orvis, B.R., see Polich, M.J. 1639 505, 507, 509, 509n, 512, 513, 518, 526, 541, 542,
Osiewalski, .,see Koop, G. 334, 1157 550-552, 576, 587, 596, 605, 627, 628n, 712, 714,
Oss, M. 906 715, 722, 725, 759, 771n, 772, 772n, 777, 778, 826n,
827,896,908,913,979, 1064, 1096, 1099, 1103n,
Oster,G. 1544,1575
lllln, 1114,1133,1144,1145,1152, 1155, 1218,
Osterberg, E. 1665
1347, 1364, 1409, 1413, 1415, 1416n, 1436, 1460n,
Osterberg, E., see Bruun, K. 1649
1463, 1464, 1464n, 1465, 1466, 1492,1686, 1767,
Ouchavev, V.K., see Calnan, M. 1181
1778n
Outreville, JF., see Cummins, J.D. 1362
Pauly, M., see Danzon, PM. 1366
Over, A.M., see Ainsworth, M. 1778n
Pauly, M., see Escarce, J. 474, 476, 478n, 495
Owens, P., see Hunt, H.A. 1044n
Pauly, M., see Friedman, B. 1125n, 1453
Owings,M.,seeGruber, J. 507, 508, 510, 511, 518, 525
Pauly, M., see Gaynor, M. 333
Oxley, H., see Gerdtham, U.-G. 11, 18, 32, 38, 39, 45, 46,
Pauly, M., see Grannemann, T.W. 1157, 1452
48, 49
Pauly, M., see Held, P.J. 556
Ozminkowski, R.J., see Experton, B. 734 Pauly, M., see Hillman, A.L. 557, 715, 716
Pauly, M., see McGuire, T.G. 79n, 507, 507n, 508, 509n,
Paarsch, H.J., see Hamilton, B.H. 1840 518, 524, 725, 1417
Paci, P., see Joshi, H. 1836n Pauly, M., see Ramsey, S. 596, 605, 709, 723n, 724, 913,
Paci, P, see Wagstaff, A. 1822, 1835, 1841n, 1842, 914, 917, 918
1849n, 1850n, 1852n, 1853n Pavlova, Z., see Werman, B.G. 516
Pacula, R. 1642 Pearce. I.E, see McKenzie, G.W. 252n
Pacula, R., see Chaloupka, F.J. 1550, 1552, 1596, 1597 Pearl, R. 1545
Pacula, R., see Cook, P.J. 1639 Pearson, M., see Blondal, S. 1039
Padgett, D. 926 Pearson, S.D. 737
Paffrath, D., see Breyer, F. 1157 Peat Marwick, K.PM.G. 1122
Pagan, A.R., see Breusch, T.S. 30 Pedersen, K.M. 1258, 1258n, 1261, 1262, 1268
Pagani, L., see Tremblay, R. 1080 Pedersen, L., see Gosden, T. 1188
Pai, M., see Barr, R.D. 1745 Pedersen, P.A., see Krasnik, A. 1189
Pakes, A., see Berry, S. 1431, 1432, 1434 Pedulla, J., see Eng, C. 961, 962
Palmer, R.M, 600, 630 Pekurinen, M. 1558, 1561, 1565, 1589, 1591n,
Pamuk, E., see Preston, S.H. 1852n 1593-1595
I-30 Author Index
Peles, Y. 1588 Phibbs, C., tee Garnick, D.W. 1431. 1433

Peltzman, D.J., see Garnick. D.W. 1431, 1433 Phibbs, C., see Luft, H.S. 1434
Peltzman, D.J.,see Luft, H.S. 1434 Phibbs, C., see Robinson, J.C. 873, 1167
Peltzman, S. 252n, 1300, 1311, 1315, 1446 Philipson, T. 71, 325, 326, 1065. 1695, 1700, 1707. 1712,
Penchansky, R., see Berki, S.E. 728, 729 1713, 1777n, 1786, 1787n, 1794, 1796n
Penner, S., see Scheffler, R.M. 912 Philipson, T., see Ahituv, A. 1790. 1793
Penrod, J.. see Holtz-Eakin, D. 668, 672 Philipson, T., see Boozer, M. 1786
Pereira Gray, D., see Maynard, A. 1187 Philipson, T., see Dow, W. 1774, 1775
Pereira, J. 89n Philipson, T., see Geoffard, P. 1695. 1696, 1763n,
Perkins, A., see Croxson, B. 1187 1769-1771, 1771n, 1778n. 1779n, 1783n
Perkins, N.A.K. 249, 249n Philipson, T., see Goldstein, K.P. 1794n
Perkins, NA.K., see Phelps, C.E. 252 Philipson, T., see Lakdawalla, D. 984, 986. 1152, 1166,
Perkoff, G.T. 728 1436, 1438
Perloff, J.M., see Carlton, D.W. 1473 Phillips, C.D. 961
Perman, R. 40 Phillips, C.V. 1710
Perneger, T.V. 927 Phillips, D.R., see Bartlett, H.P. 964
Perozek, M., see Coughlin, T.A. 960 Phillips, PC.B. 40,41,43
Peri, T.J. 1062 Phillips, P.R., see Jones-Lee, M.W. 1225
Perrin, J.M., see Hickson, G.B. 255, 1189 Phillips, R.S., see Udvarhelyi, I.S. 737
Perrin, J.M., see Sloan, FA. 1161 Phlips, L. 1556, 1558, 1559
Perron, P. 41 PhRMA 1299, 1299n, 1301, 1302, 1304, 1305, 1314.
Perron, P. see Campbell, JY. 41 1314n, 1322, 1328
Perron, P., see Phillips, P.C.B. 41, 43 Physician Payment Review Commission 476, 478n, 504,
Perry, G. 1321 710, 719, 720, 728, 805,905, 926
Perry, M.K. 1192 Pickard, S., see Ham. C. 1389
Persky, J. 130n, 136n Pickett, G. 1712
Pesando, J., see Gunderson, M. 690n Pickett, G., see Warner, K.E. 1611
Pesaran, M.H. 30, 39,44 Picone, G. 395n, 1680
Pesaran, M.H., see Im. K.S. 41, 44 Picone, G., see Hoerger, T.J. 282, 975, 977
Pescosolido, B.A. 463n Picone, G., see Sloan, FA. 960, 972, 981, 1162, 1165
Peter, R., see Halfon, N. 1066 Pierce, J., see Flewelling. R.L. 1548
Peters, L.L. 1169 Pierce, J.P. 1592
Petersen, J., see Hornberger, J.C. 1746 Pietila, I., see Pietila, T. 1283
Petersen, PRE.,see Pedersen, K.M. 1258, 1258n, 1261, Pietila, T. 1283
1262, 1268 Piggott, J., see Cameron, A.C. 315, 319
Peterson, C., see Lillard, L. 1708 Pigou, A.C. 1579n
Peterson, D.E. 1548 Pilkonis. P.A., see Elkin, I. 904
Peterson, J., see Fossett, J. 1075 Pinch, T., see Ashmore, M. 2, 3
Peterson, M.A., see Shanley, M.G. 1355, 1359 Pinsky, M.R., see Angus, D.C. 733
Peterson, M.R., see Bray, R.M. 1657n Piper, J.M. 1072
Peto, R. 1546 Piro, P., see Worthington, N.L. 1517, 1525, 1526n
Pezzin, L.E. 962, 975 Pirzada, S., see Hart, L.G. 714
Pezzin, L.E., see Kemper. P. 962 Pitchik, C. 1416n
Pfaff, M. 18, 22 Pitt, M.M. 296
Pfleiderer. P., see Bulow, J.I. 1568 Pitt, M.M., see Lee, L.-E 274, 295. 296
Phelan. C., see Rust, J. 676, 678-680 Pittman, D.J., see Heien, D.M. 1661
Phelps, C. 1, 70n, 73n, 80n, 103, 185, 228, 229, 234, 240, Plant, R. 1869
240n, 243-245, 252, 254, 256, 429,430, 434, 464n, Pless, I., see MacArthur, C. 1056
475, 503,511,513n, 518,557,580-583,904, 1222, Pliskin, J. 102n, 1729, 1741
1284, 1349,1663,1684, 1689,1694,1706, 1777n Plisldn, J., see Johannesson, M. 102n. 103, 212, 1735
Phelps, C., see Garber. A.M. 104, 199, 236, 256n, 261, Plotkin, S. 1768n
395n Plotnick, R. 89n, 1829, 1831n
Phelps, C., see Helms, L.J. 431 Plott, C.R. 1416n
Phelps, C., see Keeler, E.B. 422425, 441,579n, 911 Plummer, M.L., see Mann, C.C. 1306
Phelps, C., see Leung, S. 1638n Png, I.P.L. 1476
Phelps, C., see Lillard, L. 1708 Pogany, J.,seeBallance, R. 1299n, 1300, 1304. 1318,
2
Phelps, C., see Manning, W.G. 1259, 1261, 1 61n, 1262, 1321. 1329
1263, 1265-1267, 1269, 1269n, 1272 Poggio, E.L., see Gaumer, G.L. 599
Phelps, C., see Marquis, M.S. 437n, 620 Pogue, T.F. 1662,1663
42
Phelps, C., see Newhouse, J.P 23, 249, 286, 9n, 433, Pohlmeier, W. 315, 319. 320, 323, 1182, 1183
434, 449, 580, 582, 1262 Polich, C.L., see Gruber, L.R. 718-720
Phelps, C., see Olds, D. 1081 Polich, M.J. 1639
Phelps, C., see Thorpe, K.E. 1161, 1516, 1518, 1525 Pollak, R.A. 129n, 131, 132n, 138n, 1557, 1558
Phibbs, C. 1132n, 1429 Pollard, J.W., see Arnould, R.J. 733
Author Index I-31
Pollay, R.A. 1587, 1588 Quinn, K. 709, 710

Polochek, S., see Gaynor, M. 73n Quinn, R.J., see Read, J.L. 1746
Polsky, D., see Escarce, J. 474
Pommerehne, W.W., see Borcherding, T.E. 1169 Rabiner, D.J. 962
Pope, C. 1228, 1232n Radcliff, T.A., see Connor, R.A. 1454
Pope, G. 495, 798. 800, 805, 811n, 812, 814, 857, 858. Radfar, M. 1587
1119, 1167 Rahn, G.J., see Kralewski, J.E. 727
Pope, G., see Ellis, R.P. 143n, 784, 786, 798, 800, 806, Raiffa, H., see Keeney, R.L. 194
807, 810, 81n, 814, 822, 835, 932 Rakhmanova, G., see Leon, D.A. 1646
Pope, G., see Hurdle, S. 525, 1417 Ramey, J., see Shewry, S. 831
Popham, R.E. 1640n Ramic, H., see Calnan, M. 1181
Popham, R,E., see Bruun, K. 1649 Ramsey, J. 1113
Popkin, M.K., see Lurie, N. 925 Ramsey, S. 596, 605, 709, 723n, 724, 913, 914, 917, 918
Porell, F.E 412n, 798. 1131 Ramsey, S., see Pauly, M. 596, 605
Porell, , see Ash, A.S. 798, 800, 810, 822 Ramseyer, M., see Rasmusen, E. 1475
Porell, F, see Gruenberg, L. 806 Rand, E., see Grossman, M. 1680
Porter. A.M.D., see Howie, J.G.R. 1180 Randolph, L., see Roback, G. 1411n
Porter, R.H. 1549, 1587, 1588, 1593, 1594 Rangarajan, S., see Holahan, J. 720
Portney, P.R., see Mullahy, J. 300 Rao, B., see Mangen, S.P 936, 940
Posner, R. 1343, 1474, 1476, 1492, 1493 Rao, R. 1587
Posner, R., see Philipson, T. 1700, 1707, 1712, 1777n, Rapoport, J. 737
1786 Rappoport, P., see Cooter, R. 63n
Posnett, J., see Dixon, P 1181 Rasell, E. 1827. 1827n
Posnett, J., see Hitiris, T. 11, 18-20, 31, 49 Rasmusen, E. 1475
Posnett, J., see Jones, A. 1572n Ratcliffe, J., see Silcock, J. 1188
Poterba, J., see Gruber, J. 651, 65 In, 696 Ravallion, M., see Bidani, B. 1854n
Potter, D.A., see McKinlay, J.B. 506 Raw, M., see Buck, D. 1608
Potter, D.E.B.,seeCohen,J.W. 134n Raw, M., see Joossens, L. 1570
Potter, D.E.B., see Rhoades, J. 958 Rawls, J. 89n, 186, 1807, 1809, 1867, 1890, 1892, 1896,
Poullier, J.-P. 19 1897, 1897n, 1898, 1903
Poullier, J.P., see Getzen, T.E. 20 Ray, W.A. 602, 630
Pourtney, P.R. 1225 Rea,S. 1374
Read, J.L. 1746
Powe, N.R., see Anderson, G.E 800
Reagan, P., see Currie, J. 1060
Powell, J.L., see Ahn, H. 293, 294
Reber, S., see Cutler, D. 614, 616, 616n, 621, 622, 624.
Powers, S., see Cowing, T.G. 1452
719, 722, 728,729,773
Pratt, J. 228, 229, 236, 497n
Rebick, M. 1039
Preker, A., see Chinitz, D. 761
Rebitzer, J.B., see Encinosa, W.E. 1191
Prelec, D., seeLoewenstein, G. 1743
Reddy, S.K., see Holak, S.L. 1588
Prescott, N.M., see Griffiths, D.A.T. 1
Redelmeier, D.A. 1742
Preston, I., see Besley, T. 1230, 1238
Redelmeier, D.A., see Hornberger, J.C. 1746
Preston, S. 1056, 1852n
Reder, M. 133
Preuss, W., see Breyer, E 1157 Reder, M., see Benham, L. 469, 474
Price, J.R. 616, 620, 773
Redisch, M., see Pauly, M. 477n, 494,495, 111 In, 1152,
Price, K.E, see Martin, D. 1177 1155, 1436
Price, K.E, see Martin, D.P. 736 Redmayne, S., see Klein, R. 1891
Price Waterhouse 1600n, 1606 Reed,L. 909,926
Priest, G. 1362 Reed, S.K. 737
Prihoda, R., see Eggers, P. 730 Reed, S.K., see Cole, R.E. 734
Prihoda, R., see Lubitz, J. 984 Reed, W.W. 1745
PrimoffVistnes, J. 315 Regidor, E. 1222n
Prokhorskas, R., see Calnan, M. 1181 Regier, D.A., see Frank, R.G. 901
Propper, C. 875, 1098, 1222, 1226, 1242, 1819n, 1844, Reid, W. 68n
1853 Reiffen, D. 1474n
Propper, C., see Burgess, S.M. 325 Reiley, D.R., see Berndt, E.R. 1301, 1303n
Propper, C., see Croxson, B. 1187 Reilly. B.A., see Sloan, EA. 1647, 1648, 1663
Propper, C., see O'Donnell, 0. 1235 Reinhardt, U. 62, 107, 109, 473, 482, 495, 504, 510, 1276
Pudney, S. 305, 1638 Reinhardt, U.. see Berry, C. 526
Puffer, F. 1812n, 1839 Reiss, A., see Earls, F. 1082
Pui, C.H. 1075 Reiss, P, see Bresnahan, T. 1117, 1447, 1455
Pulice, R., see Schlesinger, M. 873 Relies, D., see Carter, G. 782
Relman, A.S. 1156
Quade, D., see Riportella-Muller, R. 1060 Remington, P.L., see Peterson, D.E. 1548
Queen, S., see Pappas, G. 1852n Remler, D. 488,493, 716, 740
I-32 Author Index
Remler, D., see Cutler, D.M. 135, 159, 163, 173 Ristow,W. 151n
Ren, Q.-F, see Hu, T.-W. 1555 Rittenhouse, B.E. 1737
Rendall, M., see Gudex, C. 1241 Rivera, M., see Beinecke, R.H. 906
Reno, V., see Mashaw, J. 1002n, 1008 Rivers, D. 302
Reschovsky, J. 975, 977 Rivlin, A.M. 984
Reschovsky, J., see Pezzin, L.E. 962, 975 Rizzo, J. 476, 525, 525n, 1303n, 1417
Resnick, M.B. 1081 Rizzo, J., see Calem, P.S. 1165
Ressler, R.W., see Saba, R.P. 1570 Rizzo, J., see Reynolds, R.A. 1366
Restuccia, J.D., see Gertman, P.M. 440n Rob, R., see Kandori, M. 237
Reti, M., see Lee, A.J. 506n Roback, G. 1411n
Reuijl, J.C. 1587 Robbins, L. 107
Reuijl, J.C., see Leeflang, P.S.H. 1587 Robbins. P.C., see Steadman, H.J. 898, 899
Reuter, J., see Merrill., J. 618 Roberts, J. 11, 18-20, 39, 44, 45, 47, 48
Revicki, D., see Sobal. J. 1682 Roberts, J., see Milgrom, P. 1122n
Reynolds, K.J., see Crocker, K.J. 1473 Roberts, J.S. 1653
Reynolds, R.A. 1366 Roberts, M.J. 1586, 1587
Rhoades, J. 958 Robine, J.-M., see Cambois, E. 963, 987
Rhoads, G.G., see Yano, K. 1651 Robinson, J. 105,495, 711, 714, 728 730, 742, 829n,
Rhodes, E.. see Charnes, A. 1278 873, 1114,1116,1167. 1168n, 1412,1428, 1450,
Ribar, D., see Kenkel, D.S. 1655, 1656, 1658 1451, 1522
Ricci, E.M. 430n Robinson, J., see Giacomini, M. 83
Rice, D. 151n. 898, 899, 1010n, 1575, 1660, 1664 Robinson, J., see Luft, H.S. 1450
Rice, D., see Bartlett, J.C. 1575n, 1580n Robinson, J., see Phibbs. C.S. 1429
Rice, D., see Cooper, B.S. 1010n Robinson, L., see Hoyme, E. 1066
Rice, D., see Miller, L.S. 1575n. 1580n Robinson, P. 274, 294
Rice, D., see Shultz, J.M. 1575 Robinson,R. 1193
Rice, N. 306, 307 Robinson, R., see Godber, E. 1192
Rice, N., see Carr-Hill, R.A. 1182 Robinson, R., see Ham, C. 20
Rice, T. 17, 58, 62, 63, 78, 79, 85, 107, 108, 429n, 464n, Robinson, W.T. 1324n
465, 475, 504, 515,515n, 518,523,525, 526, 557, Rochaix, L. 228, 501, 514, 525, 1184
913,978,982, 1126n, 1814n, 1815n Roche, M., see Daly, E. 1748
Rice, T., see Gabel, J. 526, 719 Rochefort, D. 939, 940
Rice, T., see Hurley, J. 514 Rochefort, D., see Dill, A.P. 937
Rice, T., see Labelle, R. 63, 78. 1259, 1259n Rockefeller, J.D. 1810
Rice, T., see Sullivan, C. 717 Roddy,P.C. 418
Richards, J.W., see Goldstein. A.O. 1592 Roderick, P., see Townsend, J.L. 1547, 1549, 1553, 1573
Richards, S.C., see Freeman, G.K. 1179, 1183 Rodgers, D.L., see Olsen, E.O. 1812. 1893, 1894
Richardson, J. 1185, 1734, 1735, 1743 Rodgers, J. 618
Richardson, J., see Chapman, S. 1548 Rodgers, J.D., see Hochman, H.M. 1807
Richardson, J., see Cook, J. 1729 Rodriguez, C., see Regidor, E. 1222n
Richardson, J., see Nord, E. 88, 89, 94, 1905n Rodriguez, J.,see Rosenberg, M. 1056
Richardson, L., see Riportella-Muller, R. 1060 Rodriguez, T., see Schauffler, H.H. 1689
Richardson, W.C., see Martin, D. 736, 1177 Rodstein, B., see Rosenberg, S. 717
Rickman, N., see Fenn, P. 874, 875 Rodwin, M.A. 520
Ried, L.D., see Harris, B.L. 432 Roemer, J. 1878, 1892n
Ried, M. 351, 357, 361, 364, 392n Roemer, M.I. 431, 1506n
Ried, W., see Erbsiand, M. 300, 383, 387 Roemer, R. 1544, 1567
Riedel, D.C., see Heaney, C.T. 429 Roeper, P., see Grunewald, P.J. 1640
Riess, J.A., see Needleman, H.L. 1066 Rogal, D. 836
Rigoni, R., see Griffiths, D.A.T. I Rogal, D., see Lee, C. 831, 834
Rigotti, N.A. 1597 Rogers, D.L., see Olsen, E.O. 89, 90
Riley, G. 832 Rogers, J. 1178
Riley, G., see Kasper, J.D. 730 Rogers, W. 442
Riley, G., see Lubitz, J. 806, 984 Rogers, W., see Greenfield, S. 492, 734
Riley, G., see McCombs. J.S. 736 Rogers, W., see Keeler, E.B. 1162, 1163
Riley, J.G. 1415 Rogers, W., see Manning, W.G. 728, 729. 732, 735
Riley, J.G., see Hirshliefer, J. 1700 Rogers, W., see Sturm, R. 737
Ringel, J.S., see Evans, W.N. 1578 Rogers, W., see Valdez, R.B. 443
Riordan, M. 1473, 1474, 1474n, 1475 Rogers, W., see Ware Jr., J.E. 739
Riordan, M., see Ma, C.A. 477n, 522, 544 Rogerson, W. 490, 491, 857, 859, 1119n
Riordan, M., see McGuire, T.G. 941 Roghmann, K.J., see Szilagyi, P.G. 737
Riphahn, R. 1026n, 1039 Rogowski, J., see Karoly, L. 676, 677
Riportella-Muller, R. 1060 Rohman, M., see Wechsler, H. 1712
Rishards, J.W., see DiFranza, J.R. 1592 Roizen, R.,see Boland, B. 1640n
Author Index I-33
Roland, M. 1228 Rowsell, K.V., see James, M. 1241

Roland, M., see Carr-Hill, R.A. 1182 Royal College of Physicians 1542
Rolph, J.E. 1354, 1360, 1361 Royalty, A. 1415
Rolph, J.E., see Buchanan, J.L. 587 Royalty, A., see Baker, L. 1073
Rolph, J.E., see Keeler, E.B. 256, 419, 441, 1269, 1271 Rubenstein, A. 1179
Romeo, A.A. 1161,1429,1522 Rubin, A. 1061
Rongen, G., see Grytten, J. 1840 Rubin, D.B., see Angrist, J.D. 336
Ronis, D.L., see Strumwasser, 1. 730, 737 Rubin, D.B., see Rosenbaum, P.R. 293
Room, R. 1657 Rubin, J. 918, 942, 943
Room, R., see Bruun, K. 1649 Rubin. J., see Deb, P. 927
Roos, L.L., see Roos, N.P. 240, 244 Rubin, J., see Lambrinos, J. 943
Roos, N.P. 240, 244, 572, 1177, 1278, 1279, 1282 Rubin, R.J. 1366, 1369
Rosen, B. 1184, 1187 Rubinstein, L.V., see Keeler, E.B. 1162, 1163
Rosen, H. 1659, 1660, 1663 Rubio-Stipec, M., see Swendsen, J.D. 898
Rosen, H., see Holtz-Eakin, D. 668, 672 Ruchlin, H.S. 961
Rosen, S. 655, 656, 690n Ruddle, S., see Higgings, J. 1240, 1240n
Rosenbach, M.L., see Freiman, M.PR 934 Ruggles, P., see Blank, R. 1072n
Rosenbach, M.L., see Frisman, L.K. 941 Ruhm, C.J. 1647, 1648
Rosenbaum, P., see Feeny, D. 1745 Rumbarger, J.J. 1653
Rosenbaum, PR. 293 Rupp, A. 920
Rosenbaum, S., see Rowland, D. 1061 Rupp, A., see Frank, R.G. 778, 815
Rosenberg, C.E. 1147 Rupp, A., see Lave, J.R. 598
Rosenberg, J.D. 1376 Rupp, A., see Salkever, D.S. 1515, 1518
Rosenberg, M. 1056 Rupp, K. 1013n, 1014n
Rosenberg, P., see Ehrenberg, R. 692, 695 Russell, I., see Bombardier, C. 1745
Rosenberg, S. 717 Russell, L.B. 184, 600, 712, 1161. 1493, 1521, 1526,
Rosenblatt, A., see Dunn, D.L. 786, 808, 810 1529, 1685, 1704, 1705, 1710, 1713
Rosenblatt, R.A., see Hart, L.G. 714 Russell, L.B., see Davis, K. 548, 581
Rosenkranz, S., see Diehr, P. 242 Russell, L.B., see Gold, M. 183, 188, 212, 213, 1284,
Rosenqvist, G. 1261, 1262, 1267 1705, 1739, 1752
Rosenqvist, G., see Arinen, S. 315, 319, 1261-1263, Rust, J. 676, 678-680
1265, 1267, 1270 Rutten, F. 1828
Rosenqvist, G., see Hakkinen, U. 315, 319. 320
Rutten, E, see van Doorslaer, E. 87n, 88, 1824n, 1843n
Rosenthal, G. 581
Rutten, F., see van de Ven, W.P.M.M. 16
Rosenthal, G., see Klarman, H. 101
Ryan, M. 1181, 1225, 1259
Rosenthal, M. 474, 522
Ryan, M., see Healey, A. 1186
Rosenzweig, M. 299, 310, 381,1062, 1084, 1681, 1684,
Ryan, M., see Mooney, G. 75, 499, 1181
1685,1811, 1840, 1841n
Ryan, M., see Mooney, G.H. 414
Rosenzweig, M., see Lee, L.-F. 274, 295, 296
Ryan, M.P. 1319
Rosett, J.G., see Berndt, E.R. 150n, 151
Ryan, S. 692, 694
Rosett, R.N. 433, 580, 582
Ryder, H.E. 1559
Rosko, M.D. 1507n
Rosko, M.D., see Broyles, R.W. 429n
Ross, D., see Scherer, FM. 1301, 1302, 1427n, 1460n, Saba, R.P. 1570
1467n Sackett, D. 1741, 1742, 1748
Ross, S., see Scott, A. 1181 Sackett, D., see Torrance, G. 101
Ross-Degnan, D., see Soumerai, S.B. 1327 Sadanand, A. 227, 230-233
Rosser, R. 101, 102, 1748 Saffer, H. 1587, 1588, 1591, 1645, 1648, 1662n
Rossiter, L.E 512 Saffer, H., see Chaloupka, FJ. 1548, 1570, 1596, 1648
Rossiter, L.E, see Adamache, K.W. 719 Saffer, H., see Grossman, M. 1595, 1641
Rossiter, L.F., see Wilensky, G.R. 1184, 1185 Sager, M.A. 600
Roth, A. 522, 883 Sagoff, M. 63n
Roth, L.H., see Steadman, H.J. 898, 899 Sahn, D.E. 18
Rothman, K. 1842 Salek, S., see Edgar, A. 1739
Rothschild, M. 608, 609, 634, 772, 772n, 929, 1414 Salganicoff, A., see Rowland, D. 1060
Rothschild, M., see Deneckere, R. 1412n Salinger, M.A. 1474
Rothstein, R. 1036 Salisbury, C.J. 1181
Rotondi, A.J.,see Angus, D.C. 733 Salkever, D. 827, 1033n, 1417n, 1494, 1506n, 1507n,
Rotte, R., see Geil, P. 315, 319 1508, 1515, 1517-1519, 1525, 1526
Rougemont, A., see Pemeger, T.V. 927 Salkever, D., see Frank, R.G. 474, 477n, 522n, 1084,
Rowland, D. 1060, 1061 1160, 1322, 1323, 1445n
Rowland, D., see Davis, K. 1498 Salkever, D., see Rupp, A. 920
Rowlands. O. 1178 Salkfeld, G., see Richardson, J. 1743
Rowley, D.. see Schoendorf, K. 1083 Salleras, L., see Mananu C. 1285, 1286, 1288, 1289
1-34 Author Index
Saloner, G., see Ordover, J.A. 1474, 1474n Schlesinger, H. 1688

Salop, S. 228, 727, 1411, 1475, 1476 Schlesinger, H., see Cuffel, B. 901
Salop, S., see Krattenmaker, T.G. 1475 Schlesinger, M. 743, 873, 1160, 1161, 1163
Salop, S., see Ordover, J.A. 1474, 1474n Schlesinger, M., see Gray, B.H. 1144, 1168
Salop, S., see Riordan, M.H. 1473-1475 Schlesinger. M., see Mechanic, D. 521n, 912
Salter, B., see Khayat, K. 1181 Schmalensee, R. 1446, 1586, 1587, 1593
Salter, V., see Enterline, P.E. 430n Schmidt, P. 1554
Saltman, D.C., see Veale, B.M. 1183 Schmidt, P., see Aigner, D. 333, 1277
Sammartino, F. 1634, 1659, 1659n Schmidt, R., see Breyer, F 1157
Sammartino, F.J., see Cordes, J.J. 1583 Schmidt, R.C., see Park, B.U. 334
Samuelson, L., see Roberts, M.J. 1586, 1587 Schmidt, W., see Bruun, K. 1649
Samuelson, P.A. 1205 Schmidt, W., see Pophamn,R.E. 1640n
Sander, W. 400, 401 Schmitz, R.J., see Gaumer, G.L. 599
Sanderson, D., see Akehurst, R. 1289 Schnabel, M. 1588
Sanderson. W., see Gertler, P. 280, 283, 1222 Schneider, B. 830
Sangl, J., see Stone, R. 960 Schneider, D., see Fries, B.E. 968
Sant, D., see Ippolito, R.A. 1547, 1588, 1593, 1594 Schneider, E.C. 75
Santerre, R. 1, 464n Schneider, E., see Borcherding., T.E. 1169
Santoro, M. 1319 Schneider. L. 1586-1589, 1593, 1594, 1699
Santos Silva, J. 316, 320, 1183 Schnitzler, M.A.,see Woodward, R.S. 498
Santos Silva, J., see Windmeijer, F. 316, 317, 324, 1182, Schnopp-Wyatt, D.L.. see Jason, L.A 1597
1183 Schoeffski, 0., see Von der Schulenburg, J.M. 548
Santos, A.B., see Taube, C.A. 937 Schoemaker, PJ.H. 1734
Sappington, D.E. 75 Schoemaker, P.J.H., see Hershey, J.C. 1747
Sappington, D.E., see Glass, D. 864 Schoenbaum, M. 1583, 1595, 1698. 1699n
Sappington, D.E., see Lewis, T.R. 867, 869 Schoendorf, K. 1083
Sarachek, D., see Auster, R. 299, 353, 381, 396 Schoeni, R., see Dow, W. 1055
Sarin, R.K., see Currim, I.S. 1748 Sch6ffski, O. 1331
Sarin, R.K., see Dyer, J.S. 1735, 1737 Schokkaert, E. 769n, 806, 830, 834n, 835
Sass, T.R., see Mitchell, J.M. 517, 1193 Scholten, P., see Krasnik, A. 1189
Sato, M., see Boadway, R. 999n, 1022n Schone,B. 688
Satterthwaite, M. 502, 1105, 1113, 1114, 1410, 1411 Schotter, A., see Pitchik, C. 1416n
Satterthwaite, M., see Dranove, D. 475, 743, 858, 887, Schrio-Geist, C.A., see Hendricks. W. 1010n
1097,1107, 1109n, 1133,1407n. 1413n. 1414, 1450n Schroeder, S. 504
Satterthwaite, M., see Pauly, M. 475, 476, 502, 518, 1114 Schulte, B. 962, 964
Sattinger, M.A., see Klass, M.W. 1473n Schultz, T., see Rosenzweig, M.R. 299, 381, 1084, 1681,
Saunders, K., see Clark, D.O. 798n, 802, 803n 1684, 1685, 1811, 1840, 1841n
Saunders, K, see von Korff, M. 802 Schumacher, D., see DesHarnais, S.1. 601, 864, 865
Savoca, E., see Haas-Wilson. D. 280 Schumann, L., see Vita, M.G. 1453
Saywell, R.M. 1072n Schumann, P., see Ehrenberg, R. 692, 695
Saywell, R.M., see Palmer, R.M. 600, 630 Schut, FT. 772, 791
Scanlon, D., see Chernew, M. 714, 860, 1133 Schut, ET., see van de Ven, W.P.M.M. 16, 764n, 770,
Scanlon, T. 63n 770n, 826n, 833n
Scanlon, W. 966, 1493n Schwab, R.M., see Lyon, A.B. 1573, 1660n
Scanlon, W., see Feder, J. 965 Schwartz, A. 227
Scanlon, W.. see Holahan, J. 514n Schwartz, A., see Wilde, L. 227
Scarlett, S., see Leaver, S.G. 137n Schwartz. M. 1475
Schachtel, B.P. 1300 Schwartz, W. 1361
Schaub, T., see Leu, R.E. 33, 1577. 1580n Schwartz, W., see Aaron, H.J. 1240
Schauffiler, H.H. 1689 Schwartz, W., see Newhouse, J.P 742. 1113n
Scheffler, R. 431, 909n, 912, 1123, 1276. 1277, 1280, Schwartzman, D. 1305, 1307, 1307n, 1320n, 1328
1282, 1283 Schweitzer. S.O., see Maryniuk, G.A. 1264, 1290
Scheffler, R., see Foreman, S.E. 1463 Scitovsky, A. 162n, 163, 430, 439, 582, 617, 985
Scheffler, R., see Haas-Wilson, D. 279 Scitovsky, T. 1558
Scheffler, R., see Watts, C.A. 909 Scotchmer, S., see Farrell, J. 494
Scheffman, D.T. 1427, 1430n Scott Morton, FM. 1478
Scheffman, D.T., see Salop, S.C. 1475 Scott, A. 307,512, 1181-1183. 1188, 1189. 1192
Schelling, T. 99, 185, 1546, 1556, 1557, 1600n Scott, A.,seeVick,S. 1183
3
Schenzler, C., see Sloan, F.A. 1647, 1648, 1663 Scott, C., see Rupp, K. 101 n
Scherer, F.M. 473, 493, 517, 1301, 1302. 1304, 1305n, Scott, E. 1577
1308,1309n, 1311n, 1315, 1316, 1319,1320n, 1323. Sculpher, M., see Briggs, A. 197
1326n, 1328,1331, 1407n, 1427n, 1460n, 1467n Scutchfield, ED., see Freiberg, L. 433
Schleiter, M.K., see Wells, K.B. 1712 Secretary of the Department of Health and Human Services
Schlesinger, E., see Schlesinger, M. 743 900
Author Index I1-35
Segal, S. 898 Sher, S.I., see Billi, J.E. 729

Seibring, M., see DiMasi, J.A. 1310 Sherboume, C.D., see Shapiro, M.E 442
Seidman,B.,seeRoback,G. 1411n Sherboume, C.D., see Ware Jr., J.E. 739, 803, 1732
Seiford, L.M. 1278, 1278n Sherman, D., see Lamphere, J.A. 766
Selby, J.V. 433 Shemn, M.A., see Yelin, E.H. 738
Selby-Harrington, M., see Riportella-Muller, R. 1060 Sheshinski, E., see Brito, D. 71, 1694, 1706, 1778n
Selden, T. 383, 393, 394, 715, 724, 827, 1680 Sheshinski, E., see Diamond, P. 1021n, 1022n
Selden, T., see McCoskey, S.K. 44, 48 Shewry, S. 831
Seldon, A. 1233 Shiboski, S., see Kupperman, M. 1743
Seldon, B.J. 1548, 1587, 1589 Shickle, D., see Edgar, A. 1739
Seldon, B.J., see Boyd, R. 1586, 1662 Shields, M., see Pudney, S. 305
Selvanathan, E.A., see Selvanathan, S. 48 Shiell, A. 64n, 1754
Selvanathan, S. 48 Shiell, A., see Scott, A. 307, 512, 1182, 1189
Sen, A. 63n, 66, 88, 1728, 1816, 1871n, 1877, 1906 Shilts, R. 1781
Sennett, C.S., see Gaumer, G.L. 599 Shin, R., see Simpson, J. 1448, 1449
Sensenig, A. 175 Shin, Y. 43
Sensenig, A., see Levit, K.R. 959, 960 Shin, Y., see Im, K.S. 41, 44
Settle, R., see Long, S.H. 619 Shin, Y.,see Pesaran, M.H. 44
Sgontz, L.G., see Pogue, T.F. 1662, 1663 Ship, .I., see Bentley, J.M. 1280
Shackley, P. 77, 77n Shirom, A. 830
Shackley, P., see Ryan, M. 1181 Shkolnikov, V.M., see Leon, D.A. 1646
Shaddy, J., see Juba, D.A. 617 Shleifer, A. 876, 1118
Shadle, M., see Gruber, L.R. 718-720 Shleifer, A., see Bemheim, B.D. 978, 979
Shaffer, PA., see Haddix, A.C. 1686, 1705 Shleifer, A., see Hart, O0. 852, 1101, 1150
Shakotko, R. 1062 Shmanske, S. 1238, 1239
Shakotko. R., see Grossman, M. 1551, 1552 Shoemaker, P. 102n
Shalit, S.S. 1152 Short, P. 681n, 1061, 1067
Shanley, M., see Dranove, D. 155n, 473, 477n, 493, 726, Shortell, S. 1162, 1451
1112,1117,1126,1128,1129, 1168n, 1429, Shortell, S., see Conrad, D. 1478
1448-1450 Shoven, J.B. 1577, 1582
Shanley, M.G. 1355, 1359 Showalter, M.H. 1549, 1563, 1569
Shaper, A.G. 1651, 1651n Showstack, J., see Braveman, P.A. 733
Shapiro, E., see Roos, N.P. 572 Shulman, S.R. 1314, 1329
Shapiro, 1. 163 Shultz, J.M. 1575
Shapiro, J., see Leon, D.A. 1646 Sickles, R., see Behrman, J.R. 325-327, 329, 330
Shapiro, L., see Billi, J.E. 729 Siddarth, S., see Pollay, R.A. 1587, 1588
Shapiro, M.D. 162n, 163 Siddiqui, S. 325
Shapiro, M.D., see Barsky, R.B. 1682, 1692 Siegel, J.E. 194
Shapiro, M.D., see Shapiro, I. 163 Siegel, J.E., see Gold, M.R. 183, 188, 212, 213, 1284,
Shapiro, M..E 442 1705, 1739, 1752
Shapiro, S. 936 Siegel, J.E., see Luce, B.R. 189
Sharfstein, S.S., see Hustead, E. 909 Siegel, M., see King, C. 1592
Shavell, S. 81, 1122n, 1343-1347, 1360, 1364, 1375, Siegel, M., see Pollay, R.A. 1587, 1588
1663n, 1687 Silbumrn, S., see Hayward, L. 1653
Shaw, G.B. 589n Silcock, J. 1188
Shaw, K., see Montgomery, E. 690n Silver, E., see Steadman, H.J. 898, 899
Shayne, M.W., see Sloan, FA. 961 Silverman, C., see Segal, S. 898
Shea, J.A., see Escarce, J.J. 714 Silverman, W., see Upton, C. 1261
Shea, T., see Elkin, I. 904 Silvers, J.B. 1165
Sheffer. M.J., see Kopp, S.W. 1324, 1328 Simar, L., see Park, B.U. 334
Shefrin, H.M., see Thaler, R. 1557 Simon, C. 469n, 474
Sheiman, 1. 830 Simon, C., see Dranove, D. 1117, 1128, 1168n, 1429.
Sheiner, L. 662, 664, 674n, 691, 694, 697, 698 1450
Sheiner, L., see Cutler, D.M. 741, 743 Simon, G.E., see Clark, D.O. 798n, 802, 803n
Sheingold, S. 597, 600 Simon, H. 522
Sheingold, S., see Brown, R.E. 1685n Simon, L., see Rowland, D. 1061
Sheldon, T. 830, 833 Simone, B., see Lichtenstein, R. 829n
Sheldon, T., see Bevan, G. 833 Simonich, W.L. 1548, 1589, 1593, 1594
Shell, K., see Fisher, FM. 139n Simoons, M.L., see Mark, D.B. 215
Shelton, R.M., see Wennberg, J.E. 240 Simpson, C.A., see Vuchinich, R.E. 1557n
Shen, Y. 499n, 816, 933 Simpson, H., see Culyer, A. 71-73
Shepard, D.S., see Beinecke, R.H. 906 Simpson, J. 1448, 1449
Shepard, D.S., see Callahan, J.J. 904, 906, 912, 924 Sinclair, A.J., see Wolfson, A.D. 1746
Shepard, D.S., see Pliskin, J.S. 102n, 1729, 1741 Sindelar. J. 1639
I-36 Author Index
Sindelar, J., see Grossman, M. 404 Smith, M.D., see Kahn, J.G. 814
Sindelar, J., see Mullahy, J. 300, 898, 1656, 1657 Smith, P., see Haveman, R. 1015n, 1029
Singer, B., see Heckman, J. 313, 322, 329. 1773n Smith, P., see Martin, S. 1229, 1234
Singer, P., see Nord, E. 88, 89.94 Smith, P., see Sheldon, T.A. 830, 833
Singleton, N., see Rowlands, O. 1178 Smith, R. 690n, 1352, 1387
Sinsheimer, P., see Rice, D.P 1575 Smith, R., see Cox, H. 1589
Sintonen, H. 429n, 1260-1263, 1265, 1267, 1270, 1271, Smith, R., see Dunne, J.P. 48
1273. 1276, 1277, 1280-1282, 1732, 1745 Smith, R., see Ehrenberg, R. 662n, 690n
Sintonen, H., seeArinen, S. 315, 319, 1261-1263, 1265, Smith, R., see Olsen, J.A. 1734
1267, 1270, 1291 Smith, R.. see Pesaran, M.H. 30, 39.44
Sintonen, H., see Pietila, T. 1283 Smith, R.J. 302
Sintonen, H., see Rosenqvist, G. 1261, 1262, 1267 Smith, R.J., see Blundell, R.W. 301,302, 324
Sintonen, H., see Utriainen, P. 1282 Smith, V.K. 1701
Sirio, C.A.,see Angus, D.C. 733 Smith, W,L., see Cox, D.R. 1210
Sirtalan, I., see Grossman, M. 1642, 1685 Smits, A., see Wensing, M. 1181
Sisk, J.E. 737 Smucker, D.R., see Carey, T.S. 733
Siu, A.L. 440 Snaith, A.H., see Buttery, R.B. 1228
Sivarajan, L., see Levit, K.R. 959, 960 Snyder, N.M., see Scitovsky, A.A. 430, 439. 582
Sjaastad, J., see Korenman, S. 1063 Sobal, J. 1682
Skegg, J.L., see Atkinson, A.B. 1589, 1593 Sobel, J., see Crawford, V. 1416n
Skinner. A., see Goldman, H.H. 903 Soderholm-Difatte, V., see Gerety. M.B. 599, 630
Skinner, A., see Shapiro, S. 936 Sdderlund, N., see Propper, C. 1098
Skinner, J. 334, 1277, 1280 Sdderqvist, T., see Johannesson, M. 1227
Skinner, J., see Hubbard, R.G. 981 Soest, A. van 1005, 1005n
Skinner, J., see Lee, R. 984 Sofaer. S., see Hibbard, J.H. 75
Skog, O. 1653 Sogaard, J., see Gerdtham, U.-G. 11, 18, 22, 26, 27, 37,
Skog, O., see Bruun, K. 1649 46
Slade, EP. 670n Sohrab, L., see Hudes, J. 728, 729
Slade, F.P. 1024 Soldo. B.J., see Wolf, D.A. 974
Slade, J., see Warner, K.E. 1701 Solomon, D.H., see Leape, L.L. 244
Slemrod, J. 1663 Solomon, N., see Royalty, A.B. 1415
Sloan, F.A. 465n, 472, 476, 498n, 525, 552, 557, 599, Solomon, N.A., see Garber. A.M. 193, 195
854, 960,961.972, 979, 981, 1075, 1154, 1156, 1157, Somers, S.A. 978
1160-1162, 1164-1166, 1350,1354, 1357, 1358, Sommers, J.P.. see Cohen, J.W. 134n
1360,1362, 1370. 1373n, 1417, 1436, 1494, 1502, Sonnenberg, F., see Siu, A.L. 440
1506n, 1509,1511, 1512,1517, 1520, 1526-1528, Sonnenreich, P.C., see Chayet 1494
1647, 1648, 1663 Sonnenstuhl, W.J., see Trice, H.M. 1657n
Sloan, F.A., see Adamache, K.W. 1463 S6rbom, D., see Joreskog, K.G. 383
Sloan, F.A., see Becker, E.R. 1147, 1156 Sorensen, A.A., see McCusker, J. 736
Sloan, EA., see Bovbjerg, R.R. 1373 Sorensen, R., see Grytten, J. 1181, 1184, 1185, 1840
Sloan, F.A., see Conover, C.J. 961 Sorenson, J., see Freed, G. 1065
Sloan, EA., see Ginsburg, P.B. 1158 Sostrin, S.V.,seeWerman,B.G. 516
Sloan, FA., see Feldman, R. 75, 78, 109, 463n, 489n, Sotsky, S.M., see Elkin, . 904
510, 511, 514n, 525 Soumerai, S.B. 1327
Sloan, EA., see Hoerger, T.J. 282, 975, 977 Southam, J.A. 1243
Sloan, EA., see Morrisey, M.A. 596, 598, 1423, 1428n, Sparer, M., see Glied. S. 728
1512, 1525 Spavins, T., see McGuire, T. 252n
Sloan, EA., see Newhouse, J.P. 520 Spector, W.D., see Cohen, M.A. 972
Sloan, F.A., see Steinwald, B. 465n Speizer, E.E., see Stampfer, M.J. 1651
Sloan, F.A., see Wedig, G.J. 1163-1165 Spence, D.A., see Rivlin, A.M. 984
Sloss, E., see Manning, W.G. 1575, 1577, 1579, 1580, Spence, M. 491n, 576, 858, 1121, 1346, 1351, 1372,
1580n, 1581,1581n, 1582,1650,1651n, 1660, 1687, 1399, 1411, 1415
1707 Spengler, J.J. 1473
Sloss, E., see Newhouse, J.P. 437n. 438, 441, 442, 790, Spicer, M.W. 1214
791n, 792, 792n, 793,793n, 797, 805,814, 932, 933 Spicer, R.S., see Miller, T.R. 1664
Slovic, P. 1753 Spier, K., see Dranove, D. 652
Smalls, M., see Donaldson, C. 1285, 1288, 1290 Spiker, D., see Brooks-Gunn, J. 1062
Smart, C., see Hartunian, N. 1010n Spiller, P.T. 1473, 1475
Smart, J.J.C. 1871n, 1873 Spiller, P.T., see Scheffman, D.T. 1427, 1430n
Smith, A. 576n Spillman, B.C. 982
Smith, C.H. 1181 Spillman, B.C., see Murtaugh, C.M. 959, 975-979
Smith, C.M. 484n Spinnewyn, E 1559
Smith, D.G. 739 Spinnewyn, F., see Phlips, L. 1559
Smith, K.E., see Rodgers, J. 618 Spitzer, J.J. 29
Author Index I-37
Spitznagel, E.L.. see Griffith, M.J. 617 Stigler, G. 228, 1562

Spolsky, V., see Bailit, H. 1274 Stiglitz, J. 1414, 1659, 1763n, 1777n
St. Peter, R. 1060 Stiglitz, J., see Amott, R. 1706
Stahl, I. 27n Stiglitz, J., see Atkinson, A.B. 1887n
Staiger, D. 381, 603, 630 Stiglitz, J., see Dixit, A. 1411
Staiger, D., see Norton, E.C. 1160, 1167 Stiglitz, J., see Rothschild, M. 608, 609, 634, 772, 772n.
Stallard, E., see Manton, K.G. 864, 986, 987 929, 1414
Stalmeier, P.EM. 1741, 1742 Stiglitz, J., see Salop, S. 228
Stampfer, M.J. 1651 Stiller, J.M., see Levit, K.R. 959. 960
Stampfer, M.J., see Camargo Jr., C.A. 1651, 1652 Stinson, FS. 1632
Stano, M. 244n, 507, 512, 1416, 1464n Stinson, ES., see DeBakey, S.F 1649, 1649n
Stano, M., see Folland, S. 1, 67, 70n, 78, 464n, 475, 477n Stinson, F.S., see Williams, G.D. 1632
Stapleton, D., seeRupp, K. 1014n Stock, J., see Lumsdaine, R. 677, 679, 680
Starfield, B. 249, 249n, 1177, 1178 Stock, J., see Staiger, D. 381
Starfield, B., see Casanova, C. 1066 Stoddard, A.M., see McCusker, J. 736
Starfield, B., see Gervas, J. 1177, 1188 Stoddard, J., see Newacheck, P. 1065
Starfield, B., see Weiner, J.P. 143n, 249, 249n, 798, 814, Stoddart, G. 26, 85
932 Stoddart, G., see Drummond, M.F 100, 102n, 103, 105,
Stark, A., see Lane, D. 964 188, 205, 1284, 1285
Starr,P. 470,471n,718,719, 1095, 1147 Stoddart, G., see Evans, R.G. 67n, 68n
Starr, R., see Black, M. 1081 Stoddart, G., see Labelle, R. 63, 78, 1259, 1259n
Stason, W.B., see Weinstein, M.C. 101. 184, 188, 189, Stokey, N., see Green, J. 1416n
205, 1285, 1726 Stone, J., see Eberts, R. 690, 691, 694
Staten, M. 1126, 1130, 1449, 1463, 1464n Stone, R. 960
Stavrinos, V.G. 1587, 1594, 1595 Stoter, G., see Stiggelbout, A.M. 1741, 1746
Steadman, H.J. 898, 899 Stoyle, T.FE,see Donaldson, L.J. 1241n
Steams, FE., see Bradbury, R.C. 733 Stratford-Devai, F., see Hutchinson, B. 1189
Steams, S. 492, 716, 985, 1189 Stratford-Devai, F., see Hutchison, B. 83n
Steams, S., see Rabiner, D.J. 962 Stratmann, T. 383, 388
Steel, M.FJ., see Koop, G. 334. 1157 Stratton, S., see Rosenberg, S. 717
Steele 92n Strauss, J. 1055, 1063, 1678
Steenland, K. 1596 Strauss, J., see Dow, W. 1055
Stein, L.I. 904, 937 Strautmanis, I., see Barr, R.D. 1745
Steinberg, E.P, see Anderson, G.F 800 Streatfield, J., see Baker, R. 1181
Steinberg, EP., see Davis, K. 1498 Street, A., see Cook, J. 1729
Steiner, A., see Godber, E. 1192 Street, A., see Duckett, S. 1234, 1235
Steiner, P., see Dorfman, R. 1109n, 1303 Street, A., see Nord, E. 88, 89, 94, 1904, 1905
Steiner, R.L. 1324 Streiner, D.L. 1743, 1744
Steiner, R.L., see Masson, A. 1321, 1324 Str6Liski, D., see Lanoie, P. 1017n
Steinmeier, T., see Gustman, A. 665n, 676, 677, 679 Strobino, D., see Frank, R. 1084, 1709
Steinwachs, D., see Frank, R.G. 474, 477n Strong, P.M., see McPherson, K. 243
Steinwachs, D., see Rupp, A. 920 Strotz, R.H. 1557
Steinwachs, D., see Salkever, D.S. 1515, 1518, 1525 Strouse, T.B., see Cassileth, B.R. 1728
Steinwachs, D., see Starfield. B.H. 249, 249n Strumwasser, I. 730, 737
Steinwachs, D., see Weiner, J.P. 249, 249n, 798, 814 Striwe, W., see Zweifel, P. 979
Steinwald, 1B. 465n Stuart, B.C., see Coulson, N.E. 315
Steinwald, B., see Sloan, EA. 1154, 1157, 1502, 1509, Sturm, P., see Ford, I.K. 155n, 156n, 161, 161n
1520, 1526, 1527 Sturm, R. 737, 916, 924, 927
Stephen, K.W., see Donaldson, C. 1285, 1288, 1290 Sturm, R., see Goldman, W. 906, 912, 924
Stephens, S., see Wrightson, W. 619 Sugden, R. 64, 1809, 1877, 1892
Stergachis, A., see Harris, B.L. 432 Suits, D. 1822n, 1824
Stem, N.H., see Atkinson, A.B. 298 Sulkunen, P. 1633
Stem, R.S. 737 Sulkunen, P., see Bruun, K. 1649
Stern, S. 274, 294, 295, 676n, 960, 972, 974 Sullivan, C. 717
Steven, I.D. 1181 Sullivan, D. 1461n, 1463, 1568
Stevens, S.S. 1733 Sullivan, D., see Ashenfelter, 0. 1568
Stevenson, M.A., see Kasper, J.D. 730 Sullivan, D., see Coelen, C. 1510, 1517, 1520, 1522
Stewart, K.J,, see Armknecht, P.A. 158n Sullivan, S., see Scheffler, R. 1123
Stewart, K.J., see Moulton, B.R. 138 Sullum, J. 1545n
Stewart, M.A. 1179 Sulvetta, M.B., see Miller, M.E. 864
Stewart, M.J. 1589-1591 Sumaranayake, V.A., see Bryant, R.R. 1655
Stewart, M.W., see Levit, K.R. 959, 960 Summers, L. 134n, 658, 663, 695
Stiggelbout, A.M. 1741, 1746, 1747 Summers, L., see Bemheim, B.D. 978, 979
Stiggelbout, A.M., see Wakker, P. 1750 Sumner, D.A. 1568, 1603, 1604n
1-38 Author Index
Sumner, M.T. 1568, 1593 Teres, D., see Rapoport, J. 737

Sundberg, J.O., see Shoven, J.B. 1577, 1582 Terza, J.V. 323, 324
Sung, H.-Y. 1548, 1549, 1561, 1569 Terza, J.V, see Coulson, N.E. 315
Sung, H.-Y., see Hu, T.-W. 1079, 1544, 1549, 1594, 1595 Terza,J.V.. see Kenkel, D.S. 283, 284, 316, 1711
Sung, H.-Y., see Keeler, T.E. 1548, 1549, 1569, 1604n Test, M.A., see Stein, L.I. 904, 937
Supina, D., see Kenkel. D.S. 1690 Teutsch, S.M., see Haddix, A.C. 1686, 1705
Suranovic, S.M. 1562, 1584 Thaler, R. 1557
Sussman, E., see Langa, K. 1120 Thanassoulis, E., see Dyson. R. 1280
Sutherland, H.J. 1741 Tharp, M., see Wilcox, G.B. 1587
Sutherland, H.J., see Llewellyn-Thomas, H. 1734, 1748 The Insurance Experiment Group, see Newhouse, J.P.
Sutton, J. 1447 123, 162n, 249, 504, 542, 583,584, 587,604, 629,
Sutton, M. 278 715, 724, 731, 926, 1059n, 1105, 1264, 1271, 1301,
Sutton, M., see Buck. D. 1608 1708, 1709, 1817n
Sutton, M., see Dolan, P. 1747 The National Health Accounts Team, see Levit, K.R.
Svejnar, J. 1467n 125n, 159
Swain, B.E., see Selby, J.V. 433 Theil. H. 1586
Swan, J.H., see Hanington, C.A. 958, 965, 966 Theodossiou, I. 277
Swartz, K. 822, 828 Thimann, C., see Koch, M. 1010n
Swartz, K., see Hadley, J. 1513, 1518, 1526, 1527 Thisse, J.F, see Anderson, S.P 1411. 1436
Sweanor, D.T. 1567, 1569, 1571 Thomas, D., see Currie, J. 1060, 1075
Sweanor, D.T., see Warner, K.E. 1701 Thomas, D.. see Dow, W. 1055
Swendsen, J.D. 898 Thomas, D., see Strauss, J. 1055, 1063, 1678
Swint, M.J., see Nelson, W. 1286, 1288 Thomas, J.W. 780, 803, 806
Szilagyi, PG. 737 Thomas, J.W., seeLichtenstein, R. 829n
Thomas,L.G. 1313, 1314, 1330
Thomas, L.G., see Grabowski, H.G. 1310
Tabacosmos, Ltd., see Agro-Economic Services, Ltd.
Thomas, W., see Tonrrance, G. 101
1599
Thompson, G.B. 1707
Tacier, P., see Griffiths, D.A.T. I
Thompson, M., see Hartunian, N. 10On
Tai-Seale, M. 516n, 525
Thompson, M.E. 1565, 1593
Takagi, Y., see Ikegami, N. 968
Thompson, M.E., see Forbes, W.F. 1575
Takeuchi, D., see Bruce, M.L. 899
Thompson, R.S. 1709
Talbot, R.W., see Southam, J.A. 1243
Thorpe, K. 557,968,696, 1117n, 1161, 1516, 1518, 1525
Tang, K., see Rasell, E. 1827, 1827n
Thrall, R.M., see Seiford, L.M. 1278, 1278n
Tansel, A. 1594, 1595
Throop, EB., see Palmer, R.M. 600. 630
Tapon, F., see McRae, JJ. 1325 Thun, M., see Steenland, K. 1596
Tarantola, D., see Mann, J. 1763n Thurlow, M.L., see Ysseldyke, J.E. 1036n
Tarlov, A.R., see Greenfield, S. 492, 734 Thursby, J.G. 1570
Task Force on Prescription Drugs 1304, 1328 Thursby, M.C.,seeThursby, J.G. 1570
Tate, R., see Roos, N.P 572 Thurston, N.K. 698
Tatelbaum, R., see Olds, D. 1066 Tiaira, D.A., see Dunn, D.L. 786, 808. 810
Taube, C.A. 909, 911, 937 Tibshirani, R., see Llewellyn-Thomas, H. 1734. 1748
Taube, C.A., see Frank, R.G. 920 Tierney, W.M., see Fitzgerald, J.F. 598, 630
Taube, C.A., see Lave, J.R. 598, 920 Till, J.E., see Llewellyn-Thomas, H. 1734, 1748
Taubman, P., see Barel, A. 898, 899 Till, J.E., see Sutherland, H.J. 1741
Taubman, P., see Behrman, J.R. 325-327, 329, 330 Timothy, R.P., see Fowler Jr.. EJ. 216
Tauchen, G., see Cook, P.J. 1646n, 1647, 1648, 1650 Tirole, J., see Hart, 0. 1474
Tauras, J.A. 1553, 1555 Tirole, J., see Laffont, J.-J. 499n, 852, 867-870, 872, 874,
Tauras, J.A., see Chaloupka, F.J. 1565, 1597 876, 878, 879
Tavakoli, M., see Goddard, J. 1240, 1243 Tobacco Institute 1548
Taylor, A.K., see Cohen, JW. 134n Tobacco Merchants Association 1600n, 1606
Taylor, D., see Sloan, F.A. 1162, 1165. 1166 Tobin, J. 1812, 1887
Taylor, J.O., see Wechsler, H. 1712 Tobin, M.J., see Needleman. H.L. 1066
Taylor, L., see Guilkey, D.K. 303 Tolan, P. 1080
Taylor, L.D. 421 Tolley, G.S. 1687, 1706
Taylor, L.D., see Houthakker, H.S. 1558 Tolley, H.D. 804
Taylor,P. 1545n, 1601, 1603 Tollison, R.D. 1542n, 1580
Teel, JE., see Abernethy, A. 1587, 1594 Tompkins, C., see Gruenberg, L. 806. 817, 819
Tegene, A. 1587 Tonigan, J.S., see Bien, T. 1711
Telser, L.G. 1588 Toole, A.A. 1308
Temin, P 1300, 1305 Topel, R. 673
Tenaglia, A.N., see Cassileth, B.R. 1728 Toren, M., see Vitaliano, D.F. 333, 1157
Tengs,T.O. 1080, 1705, 1706 Torgerson, D., see Gosden, T. 1187, 1188
Tennant, R.B. 1542 Torrance, G. 101, 105, 1732-1734. 1739, 1745 1747
Author Index 1-39
Torrance, G., see Barr, R.D. 1745 Umbeck, J., see Staten, M. 1126, 1130, 1449, 1463,
Torrance, G., see Bennet, K. 1735 1464n
Torrance, G., see Drummond, M. 100, 102n, 103, 105, Umland, B. 907
188, 205, 1284, 1285 Umland, B., see Buck, J.A. 908, 926
Torrance, G., see Feeny, D. 102n, 105, 1732, 1745 Unic, .J., see Stalmeier, P.EM. 1741, 1742
Torrance, G., see Garber, A.M. 104 United Nations 983
Torrance, G., see Sackett, D.L. 1741, 1742, 1748 United Nations Development Programme 1817, 1906
Torrance, G., see Siegel, J.E. 194 University of Michigan News and Information Services
Torrance, N., see Scott, A. 1181 1592
Torrey, FE. 899 Upton, C. 1261
Toshiko, I., see Ikegami, N. 968 Upward, R., see Propper, C. 1844, 1853
Townsend, J.L. 1547, 1549, 1553, 1573, 1595, 1596 Urban, G., see Berndt, E.R. 1301, 1303n
Townsend, J.L., see Atkinson, A.B. 1575, 1583 Urban, G., see Robinson, W.T. 1324n
Townsend, P. 1075, 1810n Uribe, M., see Picone, G. 1680
Towse, A. 1344, 1387 US Bureau of the Census 1001nOO, 1002n, 1304, 1325
Tracy, J., see Gyourko, J. 690n US Centers for Disease Control and Prevention 1066,
Trajtenberg, M. 133n, 170 1078
Trapnell, G.R., see Price, J.R. 773 US Congress, Joint Economic Committee 151n
Trauner, J.B., see Luft, H.S. 620, 730, 735, 741 US Congress Office of Technology Assessment 205, 209
Travis, K. 1065 US Congressional Budget Office 604, 1124n, 1322,
Travis, K., see Wickizer, T. 1123 1322n, 1324-1326,1328,1572-1574
Tremblay, C.H. 1549, 1587, 1589, 1594, 1786n US Congressional Research Service 650-652, 654, 959,
Tremblay, R. 1080 966, 1604
Tremblay, V.J., see Kao, K. 1587, 1588, 1593, 1594 US Department of Education 1036
Tremblay, V.J., see Tremblay, C.H. 1549, 1587, 1589, US Department of Health and Human Services 143n,
1594 1078,1079,1542, 1543n, 1545, 1546,1547n, 1556,
Treyz, G.I. 1607 1575n, 1576, 1581n, 1583-1585, 1587, 1588, 1591,
Trice, H.M. 1657n 1592, 1595, 1596, 1688, 1690, 1698
Trinh, C.N., see Hudes, J. 728, 729 US Department of Health, Education and Welfare 122,
Triplett, J. 122n, 124n, 128, 139n, 142n, 144n, 165, 165n, 151n, 162,523, 1542, 1593
173, 663,690n, 898 US Department of Justice 1412, 1420n, 1424n, 1426,
Triplett, J., see Bemdt, E.R. 119 1427
Trivedi, PK., see Cameron, A.C. 315, 318, 319 US Department of Labor, Bureau of Labor Statistics
Trivedi, P.K., see Deb, P. 315, 321 123n, 136n, 138, 138n, 140n
US Federal Trade Commission 1302, 1302n, 1303, 1306
Trognon, A., see Gourieroux, C.A. 271,317
Trude, S., see Keeler, E.B. 817, 819 US Federal Trade Commission, see US Department of
Justice 1420n, 1424n, 1426, 1427
Truman, B.I., see Lowry, R. 900
US General Accounting Office 149n, 157n, 652, 664,
Truong, P. 1226
664n, 770, 780, 783, 1013n, 1016, 1018n, 1030n,
Tu, E., see Liang, J. 959
1034,1061, 1064, 1066, 1073n, 1074,1077, 1377,
Tucker, A.M., see Fowles, J.B. 805
1578, 1796n
Tudor, C., see Riley, G. 832
US House of Representatives 1014, 1016. 1017, 1020n,
Tufts University Center for the Study of Drug Development
1028, 1311
1322n
US Medicare Payment Advisory Commission 514, 832,
Tugwell, P., see Bennet, K. 1735 1158
Tulchinsky, T.H. 1678, 1679
US National Center for Health Statistics 1078
Tulloh, J.F.C., see Antczak-Bouckoms, A.A. 1288 US National Commission on Childhood Disability
Tunson, S.L., see Williams, G.D. 1632
1015n, 1034
Tuohy, C. 77 US National Science Foundation 1307, 1308n
Turner, W.M., see Porell, F.W. 798 US Office of Technology Assessment 429n, 1062n,
Tussing, A.D. 742, 1182, 1185 1068n, 1072n, 1301, 1302, 1308, 1316, 1320, 1328.
Tversky, A. 1750 1329
Tye, J.B. 1592 US Preventive Services Task Force 1677, 1683, 1684,
Tyroler, H.A., see Criqui, M.H. 1651 1696, 1711
Tzur, J., see Nitzan, S. 1416n US Prospective Payment Assessment Commission 143n,
920, 1160, 1427
Ubel, P. 88 US Senate Finance Committee 123n, 130n, 155n
Udvarhelyi, I.S. 737 Usher, D. 364n
UK Department of Health 1587, 1588, 1590 1592 Utriainen, P. 1282
Ullah, A. 272, 276 Uyenuo, D., see Lane, D. 964
Ullrich, K.. see Hoffmeyer, U. 1145
Ulrich, V., see Erbsland, M. 300, 383, 387 Vacker, B.,seeWilcox, G.B. 1587
Ulrich, V., see Pohlmeier, W. 315, 319, 320, 323, 1182, Valdes, B. 1587, 1589
1183 Valdez, R.B. 443
I-40 Author Index
Valdez. R.B., see Manning, W.G. 911, 922 Varavikova, E.A., see Tulchinsky, T.H. 1678, 1679
Vallee. B.L. 1646 Varey, C., see Kahneman, D. 88
Valletta, R., see Buchmueller, T. 668, 670, 672, 688 Varian, H. 89n, 477n, 480, 1275n, 1276, 1892
Valvona, J., see Morrisey. M.A. 596, 598, 1423, 1428n Vamik, A., see Wasserman, D. 1653
Valvona, J.,seeSloan, EA. 599, 1160. 1161, 1528 Varrela, J., see Pietild, T. 1283
van Amerongen, B.M., see Yule, B.F. 1285, 1287, 1288 Vassin, S., see Leon, D.A. 1646
van Barneveld, E.M. 771, 790, 820n, 821, 821n, 823. Veale,B.M. 1183
824, 824n, 825, 826, 828, 828n Vehmanen, R. 1288,1289
van Barneveld, E.M., see van de Ven, W.P.M.M. 764n, Vella, F. 270
770, 770n, 793n, 794, 820n
Venezian, E., see Schlesinger, H. 1688
van Busschbach, J. 1747
Vergara, R. 699n
van Daal, W.A.J., see Verhoef, L.C.G. 1741
Verhoef, L.C.G. 1741
vandeKar, A. 1181, 1182
van de Ven, W.P.M.M. 16, 83, 283, 287, 300, 429n, 621, Vernon, J.M. 1474
625. 744, 764n, 770 770n, 771, 780, 784, 785, 793n. Vernon, J.M., see Grabowski, H.G. 1310, 1316, 1317,
794, 797, 806, 820. 820n, 826n, 829n, 833n. 929n, 1322, 1323n, 1324, 1702, 1703
1177. 1180, 1188 Venilli, D.K., see Marsteller, J.A. 721
van de Ven, W.PM.M., see Lamers, L.M. 802 Vertrees, J.C., see Manton, K.G. 864
van de Ven, W.PM.M., see van Barneveld, E.M. 771, Vessey, M., see Daly, E. 1748
790, 821,821n, 823,824, 824n, 825,826, 828 Vick, S. 1183
van de Ven, W.P.M.M., see van Vliet, R.C.J.A. 798, 801, Vick, S., see Scott, A. 1183
805 Vincent, D.R., see Manelli, A.M. 873
van de Voorde, C., see Schokkaert, E. 769n, 806, 830, Vining, A.E., see Boardman, A.R. 1169
834n. 835 Vining, A.R. 1169
van de Walle, D. 1806n Viscusi, W.K. 11, 18, 20, 31, 49, 99, 414n, 690n, 1080,
van den Berg, J., see Kunst, A.E. 1853n 1542n, 1577,1580,1582, 1583n, 1595, 1697, 1697n,
van den Heuvel, J., see Widstr6m, E. 1255, 1257 1698, 1699. 1734
van der Burg, H., see Wagstaff. A. 1821, 1824, 1846 Vishny, R.,seeHart, O. 852, 1101, 1150
van der Burg, H., see van Doorslaer, E. 1832, 1833, 1845 Vistnes, G. 1167
van der Gaag, J. 300, 301 Vistnes, J., see Schone, B. 688
van der Gaag, J., see Baker, J.L. 1806n Vita, M.G. 1452, 1453
van der Gaag, J., see Dor, A. 280 Vitaliano, D.F. 333, 1157, 1452
van der Gaag, J., see Wolfe, B. 300 Vitaro, F., see Tremblay, R. 1080
van der Gaag, J., see van de Ven. W.P.M.M. 300
Vogt, W., see Abraham, J. 1456
van der Maas, P.J., see Barendregt, J.J. 1577
Vogt, W., see Bhattacharya, J. 325, 326, 583
van der Merwe, R. 1548, 1608
Vogt, W., see Gaynor, M. 743. 1103n, 1130, 1413, 1413n,
van der Merwe, R., see Joossens, L. 1570
1465
van der Po], M., see Cairns, J.A. 307
van der Weide, J. 1412 Volpp, K. 1451
van der Zee, J., see Boerma, W.G.W. 1177 von der Schulenburg, J.M. 548, 1828n
van Doorslaer, E. 87n, 88, 300, 377, 383, 388-391, 508n, von Korff, M. 802, 909
1832, 1833,1839, 1843, 1843n, 1844, 1845, 1853 von Korff, M., see Clark, D.O. 798n, 802, 803n
van Doorslaer, E., see Culyer, A.J. 64, 65, 1814n von Neumann, J. 1734
van Doorslaer, E., see Humphries, K.H. 1854 von Weizsacker, C.C. 1557
van Doorslaer, E., see Janssen, R. 1827 Vraciu, R.A., see Sloan, F.A. 1156
van Doorslaer, E., see Kakwani, N.C. 1843. 1849n, 1851, Vuchinich, R.E. 1557n
1852, 1852n Vuong, Q.H., see Rivers, D. 302
van Doorslaer, E., see Wagstaff., A. 89, 568, 1182, 1807,
1810n, 1821, 1822, 1822n, 1824, 1827, 1831, 1832,
Wachsman, L., see Zambrana, R. 1060
1835, 1841n, 1842-1846,1849n, 1850n, 1852n,
Wadeson, N., see Casson, M. 1179
1853n, 1865n, 1877n, 1894
van Montfort, P., see Wensing, M. 1181 Wadsworth, M.E. 1062
van Praag, B.M.S., see van Vliet, R.C.J.A. 300 Wadsworth, Y.J., see Veale, B.M. 1183
van Praag, B.M.S., see van de Ven, W.P.M.M. 283, 287 Wagner, E., see Cherkin, D. 420, 432, 583, 713, 1708
van Schaik, M.C.M.,see Yule, B.F. 1285, 1287, 1288 Wagner, E., see Hart, L.G. 714
van Vliet, R,C.J.A. 300, 791-793, 793n, 798, 801, 804, Wagner, E., see von Korff, M. 802
805,820n, 823,825 Wagner, H.M. 1210
van Vliet, R.C.J.A., see Lamers, L.M. 801, 802, 827n Wagner, J., see Romeo, A.A. 1161, 1429, 1522
van Vliet, R.C.J.A., see van Barneveld. E.M. 771, 790, Wagner, R.E., see Tollison, R.D. 1542n
821,821n, 823,824, 824n, 825,826, 828 Wagstaff. A. 23, 65, 66, 89, 93, 300, 313, 332, 334, 365,
van Vliet, R.C.J.A., see van de Ven, W.P.M.M. 621, 764n, 375,383,387-391,395,568, 1182, 1807, 1810n,
770, 770n, 771,780, 784, 785, 793n, 794, 797,806, 1821,1822,1822n, 1824. 1827,1831,1832, 1835,
820, 820n, 829n 1841n, 1842-1846, 1849n, 1850n, 1852n, 1853n,
Vandergoot, D., see Hunt, H.A. 1044n 1865n, 1877n, 1894
Author Index I-41
Wagstaff, A., see Culyer, A.J. 64, 65, 91, 92, 95, 103, Wassernan, J., see Manning, W.G. 1575, 1577, 1579,
1727, 1731, 1736, 1813, 1814, 1814n, 1815, 1869, 1580, 1580n, 1581, 1581n, 1582, 1650, 1651n, 1660,
1880, 1889, 1894 1687. 1707
Wagstaff, A., see Dardanoni, V. 383, 393, 394, 1680 Waterhouse, J.C., see Harper, R. 1745
Wagstaff. A., see Janssen, R. 1827 Watkins, J.T., see Elkin, I. 904
Wagstaff, A., see Kakwani, N.C. 1843, 1849n, 1851, Watson, S. 1031n
1852, 1852n Watson, S., see Scott, A. 1181
Wagstaff, A., see van Doorslaer, E. 87n, 88, 1824n, 1832, Watts, C.A. 909
1833, 1839, 1843, 1843n, 1844, 1845, 1853 Watts, C.A., see Scheffler, R.M. 909n
Wai, H.S., see McCall, N. 792 Webb, S., see Disney, R. 1001, 1002n
Waidman, T., see Bound, J. 1027 Wechsler, H. 1712
Wakker, P 197, 1737, 1750 Wechsler, H., see Chaloupka, F.J. 276, 1550, 1552, 1596,
Wakker, P., see Bleichrodt, H. 1730 1641
Waldfogel, J., see Berry, S. 1412n Wedig, G. 435, 525, 548, 1163-1165
Waldfogel, J., see Volpp, K. 1451 Wedig,G.,seeMorrisey,M.A. 1160, 1165
Weeks, E., see Hibbard, J.A. 1414
Waldman, D., see Gertler, P. 556, 968
Weeks, W.B. 472
Waldo, D.R., see Lazenby, H.C. 174n
Wehner. P., see Dranove, D. 510,511,518
Wales, T.J. 1638
Weichselbaum, R., see McNeil, B.J. 1741
Walker, A., see Blades, C.A. 1 Weiler, P.C. 1344, 1351, 1352n, 1353n, 1354, 1358,
Walker, M., see Shaper, A.G. 1651 1361n, 1368, 1370, 1371, 1373, 1377-1380, 1380n,
Wallace, R.B., see Criqui, M.H. 1651 1381, 1390, 1391
Wallack, S.S. 937 Weiler. P.C., see Abraham, K.S. 1377, 1381
Wallack, S.S., see Gruenberg, L. 817, 819 Weiner, J.M., see Rivlin, A.M. 984
Wallen, J., see Roddy, P.C. 418 Weiner, J.P. 143n, 249, 249n, 711,712, 714, 718, 798,
Walsh, D.C. 1654 799, 81 ln, 814, 932
Walsh, E.G., see Pope, G.C. 798, 800, 805, 81 In, 814 Weiner, J.P., see Fowles, J.B. 805
Walsh,J.,seeCohen, W.M. 1318, 1318n Weiner, J.P., see Frank, R.G. 474, 477n
Walston, S.L. 1478 Weiner, J.P., see Starfield, B.H. 249, 249n
Walter, E., see Alegria, M. 897 Weinstein, H. 1079
Walter, S.D., see Hutchison, B. 83n Weinstein, M.C. 101, 184. 188. 189, 205, 478n, 917,
Walters, S.J., see Brazier, J. 1744, 1745 1285, 1726, 1751
Walters, S.J., see Harper, R. 1745 Weinstein, M.C., see Antczak-Bouckoms, A.A. 1287,
Walzer, M. 95, 1887, 1888, 1906n 1288, 1290
Wang, P., see Kenkel, D.S. 1656 Weinstein, M.C., see Garber, A.M. 104
Wannametheee, G., see Shaper, A.G. 1651 Weinstein, M.C., see Gold, M. 183, 188, 212, 213, 1284,
Warburton, C, 1649n, 1653 1705, 1739, 1752
Ward, M., see Topel, R. 673 Weinstein, M.C., see Johannesson, M. 102n, 103, 194,
Ward, N.B., see Lurie, N. 431 212, 1735
Ward, R., see Sumner, M.T. 1568 Weinstein, M.C., see Pliskin, J.S. 102n, 1729, 1741
Wardell,W. 1311 Weinstein, M.C., see Siegel, J.E. 194
Ware, J.E. 739, 803, 1732 Weinstein, M.C., see White, B.A. 1285, 1286, 1288
Weinstein, M.L., see Read, J.L. 1746
Ware, J.E., see Bombardier, C. 1745
Weintraub, M. 1300
Ware, J.E., see Greenfield, S. 492
Weisbrod, B.A. 18, 71, 72, 74, 86, 93, 98. 452, 856, 904,
Ware, J.E., see Kaplan, S.H. 1179
1149, 1150, 1233, 1702, 1712, 1763n, 1767n
Ware, J.E., see Manning, WG. 249, 1841
Weisbrod, B.A., see Goddeeris, J.H. 1167
Ware, J.E., see Shapiro, M.E. 442
Weisburst, S., see Scherer, FM. 1319
Ware, J.E., see Valdez, R.B. 443 Weiss, L. 1129n, 1179
Ware, J.E., see Wells, K.B. 911, 1712 Weiss, S. 136n
Warner, K.E. 1505n, 1542n, 1544, 1548, 1550, 1567, Weissert, W.G. 960, 967
1574-1578. 1583-1585, 1587-1589, 1591-1595, Weissman, J. 570, 629
1600n, 1601-1603,1606-1609,1611,1701,1704, Weitzman, S., see Feeny, D. 1745
1713 Welch,H. 1128
Warner, K.E., see Chaloupka, F 1679, 1699, 1707 Welch, I., see Bikhchandani, S. 237
Warner, K.E., see Tye, J.B. 1592 Welch, W. 620, 718, 730, 737, 740, 793
Warner, PA., see Warner, K.E. 1704 Welch, W., see Frank, R.G. 740
Warren-Boulton, F, see Woodward, R.S. 76, 255, 521n, Welch, W., see Hillman, A.L. 715
1185 Welch, W., see Welch, H. 1128
Wartman, S.A. 1179 Wellington, A.J. 688
Wasem, J., see Chinitz, D. 761 Wells, E., see Chetwynd, J. 1587
Washington, A.E., see Kupperman, M. 1743 Wells, K.B. 738, 911, 925, 936, 1712
Wasserman, D. 1653 Wells, K.B., see Hosek, S.D. 729, 735, 739
Wasserman, J. 299, 1547. 1550, 1552, 1596, 1597 Wells, K.B., see Keeler, E.B. 316, 325, 911, 934
I-42 Author Index
Wells, K.B., see Manning. W.G. 911,922 Wilcox, G.B. 1587

Wells. K.B., see Norquist. G.S. 736 Wilcox, P.M., see Jachuk, S.J. 1726, 1732
Wells, K.B., see Sturm, R. 737 Wilcox-Gok, V., see Deb, P. 927
Wen, Q., see Anis, A.H. 1330n Wilde, L.L. 227
4
Wendel, J., see Banks, D.A. 1167 Wilde, L.L., see Plott, C.R. 1 16n
Wendling, W. 1504 Wilde, L.L., see Sadanand, A. 227,230-233
Wennberg, J.E. 240, 496 Wilde, L.L., see Schwartz, A. 227
Wennberg, J.E., see Barry, M.J. 216 Wilensky,G.R. 1184, 1185
Wennberg, J.E., see Fowler Jr., E.J. 216 Wilensky, G.R., see Rossiter, L.F 512
Wennbelg, J.E., see McPherson. K. 243 Wiley, J., see Rasmusen. E. 1475
Wensing,M. 1181 Wilhite, A., see Bryant, R.R. 1655
Werden, G. 1428, 1433, 1436, 1447 Wilkerson, J.C., see Brownson, R.C. 1595
Werden, G., see Evans, W. 1446 Wilkin,D. 1185, 1186
Werden. G., see Froeb, L.M. 1427, 1430 Wilkinson, F., see Burchell, B. 1179
Werman, B.G. 516 Willams, A., see Gudex, C. 1241
Werner, J., see Wendling, W. 1504 Willemain, T.R., see Gruenberg, L.W. 967
West, P.A., see Cullis, J.G. 1, 1835n Willett, W.C., see Stampfer, M.J. 1651
West, R.R., see Frankel, S.J. 1233, 1234, 1242, 1243 Williams, A. 3 63, 64, 70, 87n, 89n, 91. 94, 98, 105. 109,
Westerberg, 1. 1282 505, 568, 1241, 1241n, 1728, 1733, 1743. 1806, 1876,
Westerberg, I., see J6nsson, B. 1281 1899, 1901, 1906
Westfall, G., see Rosenberg, S. 717 Williams, A.F, see DuMouchel, W.A. 1646n, 1648
Westfield, FM. 1428, 1474 Williams,A.H. 1809, 1813, 1814n, 1815n
Wetenschappelijke Raad voor het Regeringsbeleid 1810n Williams, A.H., see Culyer, A.J. 1809n
Wetle, T., see Bonnano, J.B. 719 Williams, A.H., see Dolan, P. 1742, 1745-1747, 1752,
Wharton Applied Research Center 1606 1753
Wheeler, J.R.C. 717 Williams, A.H., see Gudex, C. 1241, 1740
Wheeler, J.R.C., see Feldstein, P. 717, 1123 Williams, AH., see Maynard, A. 1809n, 1815n
Wheeler, J.R.C., see Wickizer, T. 717, 1123 Williams, A.H., see Newhouse, J.P.742, 1113n
Whetten-Goldstein, K., see Sloan. FA. 1161 Williams, A.H., see Sugden, R. 64
Whinston, M., see Bolton, P 1474 Williams, B. 1056, 1076, 1077, 1080. 1887
Whinston, M., see Caves, R.E. 1322 Williams, B., see Sen, A. 1871n
White House 767 Williams, B., see Smart, J.J.C. 1871n, 1873
White, B.A. 1285, 1286, 1288 Williams, D., see Paul-Shaheen, P. 70n
White, B.A., see Antczak-Bouckoms, A.A. 1288 Williams, G.D. 1632
White, H.. see Mark, D.B. 215 Williams, G.D., see Stinson, F.S. 1632
White, L. 1412 Williams, S. 1181
White, M.J. 1349, 1358 Williams, S., see Calnan, M. 1181
White, M.J.,seeFarber,H.S. 1357, 1358 Williams, V., see Boardman, A.E. 1184, 1185
White, W., see Dranove, D. 73n. 155n, 473, 477n, 493, Williamson, M.E., see Evans, R.G. 74
499,726,854, 1112, 1120.1124, 1126, 1128, 1129, Williamson, O.E. 77n, 1179. 1192, 1473
1131, 1131n, 1132, 1445, 1448, 1449,1450n Willig, R. 252n, 427n
White-Means, S.l. 974, 976 Willits, D., see Grogan, S. 1181
White-Means, S.I., see Chang, C.F. 960, 972, 974 Willke, R.J. 602
Whitehead, M., see Townsend, P. 1075 Wilson, A., see Sloan, EA. 1161
Whitelaw, N., see Liang, J. 959 Wilson, C. 608,772, 1415
Whitney,C. 1116n Wilson, D. 1190
Whittle, J., see Anderson, G.E 800 Wilson, D., see Propper, C. 1098
Whitton, A.C., see Barr, R.D. 1745 Wilson, D.B. 1062
Wholey, D. 624, 744 Wilson,G.W. 1157
Wholey, D., see Bu s, L. 1132. 1132n, 1431, 1433 Wilson, J., see Nash, K. 1276, 1280, 1282
Wholey, D., see Christianson, J.B. 742, 744 Wilson, J.A., see Foreman, S.E. 1463
Whynes, D., see Ennew, C. 1193 Wilson, K., see Haveman, R. 1015n, 1029
Whynes, D.K. 865, 1187 Wilson, R.M., see Picone, G. 395n, 1680
Wiclizer, T.M. 717, 1123 Winder, A., see Bates, 1. 1709
Wickizer, T.M., see Feldstein. P.J. 717, 741, 1123 Windmeijer, F. 316, 317, 324, 1182, 1183
Wickizer, T.M., see Wheeler, J.R.C. 717 Windmeijer, E,see Blundell, R.W. 308
Widstr6m, E. 1255, 1257 Winrmeijer, F., see Cameron, A.C. 315, 319
Widstrbm, E., see Pietil, T. 1283 Windmeijer, F., see Dustmann, C. 313, 323
WidstrOm, E., see Utriainen, P. 1282 Windmeijer, F., see Santos Silva, J. 316, 320, 1183
Wiens, M., see Hertzman, C. 1062 Wingert, T.D., see Kralewski, J.E. 727
Wiggins, D. 1815n Winkler, A. 683, 684
Wilcox, D.W., see Shapiro, I. 163 Winkler, J.D., see Wasserman, J. 299, 1547, 1550, 1552,
n
Wilcox, D.W, see Shapiro, M.D. 162 , 163 1596, 1597
Wilcox, E., see Sensenig, A. 175 Winograd, C.H., see Gerety. M.B. 599. 630
Author Index I-43
Winston, G.C. 1546. 1556, 1557, 1562 Wright, E., see Horbar, J. 1068
Winter, R.A. 1362, 1363 Wright, G., see Adams, E. 1128
Winter, R.A., see Mathewson, EJ. 727 Wright, J.C., see Neumann, P.J. 1752
Winter, R.A., see Mathewson, G.F 1475 Wrightson, C.W., see Long, S.H. 619
Winter, S.G.,seeLevin, R.C. 1317 Wrightson, W. 619
Wise, C.G., see Billi, J.E. 729 Wroblewski, R., see DesHamais, S.I. 601, 864, 865
Wise, D., see Eichner, M. 572, 1415 Wu, D.-M. 400
Wise, D., see Kotlikoff, L. 690n Wu,L.,seeDranove,D. 1132
Wise, D., see Lumsdaine, R. 677, 679, 680 Wu, S.Y., see Nordquist, G. 1684
Wise, D., see Pratt, JW. 228, 236 Wyckoff, J., see Lankford, H. 1036
Wiseman, J., see Blades, C.A. I Wynder, E.L. 1545, 1593
Wiseman, V. 1703 WyszewJanski, L., see Thomas, J.W. 780
Witsberger, C., see Newhouse, J.P. 742
Witt, S.F. 1587, 1589 Xu, X. 1548, 1778n
Witte, A.D., see Schmidt, P. 1554
Wittehen, H., see Kessler, R.C. 897
Yaari, M. 88,95
Wittenburg, D., see Burkhauser, R. 1017
Yaari, M., see Rubenstein, A. 1179
Wohlgenant, M.K., see Sumner, D.A. 1568
Yan, D.Y., see Jin, S.G. 1575
Wojtowycz, M.A.,seeTussing,A.D. 742, 1182, 1185
Yang, K.-T., see Wilcox, G.B. 1587
Wolak, F, see Hay, JW. 1767n
Yang, W., see Clements, K.W. 1638
Wolf, C. 1703
Wolf, D.A. 974 Yannopoulos, A., see McAvinchey, I.D. 412n, 583, 1220,
1230
Wolfe, B. 300, 1003-1005, 1008n, 1062
Yano,K. 1651
Wolfe, B., see Behrman, J.R. 300
Wolfe, B., see Burkhauser, R. 1002n, 1008, 1009, 1009n, Yarrow, G. 1330n
1010, 1029n Yasser, R., see Rosenberg, S. 717
Wolfe, B., see Eze, P. 602 Yates, J. 1228, 1232, 1232n, 1234
Wolfe, B., see Gottschalk, P. 1811n, 1822 Yawn, B.P., see Smith, C.M. 484n
Wolfe, B., see Haveman, R. 1000, 1001, 1002n, 1009n, Yazbeck, A., see Behrman, J.R. 325-327, 329, 330
1011, 1011n, 1015n, 1024, 1025n, 1029 Yeates, W.K. 1234
Wolfe, B., see Moffitt, R. 682, 683, 684n Yelin, E. 738, 1007
Wolfe, B.,seeSteams, S. 492,716, 1189 Yellowlees, H. 939, 941
Wolfe, B.L., see van der Gaag, J. 300, 301 Yelowitz, A. 681, 683, 684, 686, 1032n
Wolfson, A. 1746 Yelowitz, A., see Gruber, J. 982
Wolfson, A., see Evans, R.G. 72 Yen, L.-L., see Hsieh, C.-R. 1594
Wolfson, A., see Tuohy, C. 77 Yen, S.T. 297-299
Wolfson, M., see Forster, J.L. 1597 Yen, S.T.,seeJones,A.M. 1711
Wolinsky, A. 500, 1416n Yi, S., see Choi, J.P. 1474, 1474n
Wolpin, K., see Rosenzweig, M. 310, 1062 Yip, W. 515, 516n, 518, 525, 526
Won, D.K., see Levit, K.R. 959-960 Yip, W., see Sturm, R. 737
Wong,H. 476,502n, 1131 You, J.H., see Bishop, J.A. 1549, 1568, 1587, 1588, 1593
Wong, H., see Brooks, J. 1130, 1160, 1449, 1467n, 1469 Yoshikawa, A., see Bhattacharya, J. 325, 326, 583
Wong, W.L., see Chapman, S. 1599, 1608 Young, C.A., see Hudes, J. 728, 729
Wood, D., see Halfon, N. 1066 Young, D.A., see Guterman, S. 601
Woodbury, M.A., see Manton, K.G. 864 Young, G.J. 1160
Woodbury, S. 690, 691, 696 Young, P. 237
Woodland, A.D., see Wales, T.J. 1638 Young, T. 1558
Woodward, A., see Frank, R.G. 901 Ysseldyke, J.E. I036n
Woodward, R.S. 76,255,498,521n, 1185 Yu, S., see Leung, S.F 288, 289
Woolhandler, S. 82 Yu, W., see Ash, A.S. 784, 786, 797, 798, 800, 810, 811,
Woolhandler, S., see Himmelstein, D. 1060 811n, 812, 814
Woolley, J.M. 1453 Yudkowsky, B. 1060
World Bank 1678, 1709, 1768n, 1867 Yule, B. 1254, 1257, 1261, 1262, 1285, 1287, 1288
World Health Organization 67n, 1000n, 1567, 1570, Yule, B., see Parkin, D. 18, 19, 22-24, 1261, 1267, 1270,
1594, 1596, 1597, 1694n 1273
World Health Organization Brief Intervention Study Group Yuskavage, R.E. 167n, 169
1711 Yuskavage,R.E.,seeLum, S.K.S. 169n
Worrall, J.D. 1017n, 1027n, 1028, 1029n, 1045n
Worrall, J.D., see Butler, R.J. 1356, 1374 Zador, P., see DuMouchel, W.A. 1646n, 1648
Worthington, D. 1221, 1239, 1240 Zakhari, S., see Lands, W.E.M. 1651n
Worthington, N.L. 1517, 1525, 1526n Zakharov, S., see Leon, D.A. 1646
Wouters, A.V. 738, 790 Zalkind, J., see van der Weide, J. 1412
Wozniak, G., see Emons, D.W. 474, 488, 493 Zambrana. R. 1060
Wozniak, G., see Escarce, J. 474 Zarkin, G.A. 1655, 1690
I-44 Author Index
Zarkin, G.A., see French, M.T. 1655 Zimmerman, D., see Gravelle, J.G. 1580, 1582
Zarkin, G.A., see Norton, E.C. 306 Zimmerman, Y.A., see Branch. L.G. 961
Zarmebka, P. 29 Zimmerman Murphy, M. 287, 289
Zeager, L.A., see Baldwin, M.L. 1009n Zimmermann, K.F., see Geil, P. 315, 319
Zeckhauser, R. 18, 85, 256, 575, 576, 588. 627, 775n, Zinner, D.E., see Neumann, P.J. 1752
908, 925,929, 1413 Zollinger, T.W., see Palmer, R.M. 600, 630
Zeckhauser, R., see Altman, D. 618, 797n Zollinger, T.W., see Saywell, R.M. 1072n
Zeckhauser, R., see Cutler, D. 614, 616n, 621, 759, 932 Zubkoff, M., see Greenfield, S. 492, 734
Zeckhauser, R., see Neipp, J. 1415 Zubrick, S.R., see Hayward, L. 1653
Zeckhauser, R., see Nichols, A. 610n Zuckerman, S. 333, 475, 485, 1124, 1157, 1356, 1372
Zeckhauser, R., see Phillips, C.V. 1710 Zuckerman, S., see Feder, J. 597
Zeckhauser, R., see Pratt, J.W. 228, 236 Zuckerman, S., see Hadley, J. 600
Zeckhauser, R., see Rizzo, J. 476, 525n
Zuckerman, S., see Holahan, . 720
Zeckhauser, R., see Spence, M. 576
Zuger, A. 1303
Zeckhauser, R., see Weinstein, M.C. 917
Zupp, P., see Maibach, E. 623
Zedlewski, S., see Holahan, J. 1827, 1827n
Zeger, S.L., see Peterson, D.E. 1548 Zupp, P., see Neuman, P. 83
Zeitzer, I. 1040, 1040n Zurekas, S.H., see Buchmueller. T.C. 278
Zeldes, S.P., see Hubbard, R.G. 981 Zwanziger, J. 730, 738, 742, 1116, 1451, 1527, 1529
Zellner, B.B. 821n Zwanziger, J., see Keeler, E. 1130, 1159, 1448, 1449
Zelman, W.A. 716, 721, 726 Zwanziger, J., see Mann, J. 1160
Zervos, D., see Orphanides, A. 1562, 1584 Zwanziger, J., see Manning, W.G. 1263, 1264, 1285.
Zhang, P. 1601n, 1603, 1604 1286, 1288, 1289
Zhang, X., see Miller, L.S. 1575n, 1580n Zwanziger, J., see Melnick, G.A. 742, 1429, 1448, 1449,
Zhao, J., see Hashimoto, M. 661n 1451, 1463, 1468
Zhao, S., see Kessler, R.C. 897 Zweifel, P. 1,353,375,381, 382n, 383, 385, 387, 415,
Zheng, S.W., see Clements, K.W. 1638 426, 435,450,451,453,464n, 498n, 499,499n, 507,
Zhou, X.H., see Morris, C.N. 325, 327, 975, 978 507n, 511n, 580n, 584, 759, 979, 984, 985, 1153,
Zhou, Z., see Kronick, R. 790, 791, 798, 814, 834 1154, 1182, 1185, 1264, 1686
Zill, N. 1064 Zweifel, P., see Beck, K. 776n, 793, 804, 822n
Zimmerman, D. 686 Zweifel, P., see Nocera, S. 960, 964, 972, 974
SUBJECT INDEX
ability to pay in health care delivery 1817-1818 - in long-term care 978-979

access - in managed care 722
- and child health 1058 - in mental health 897, 907, 925-936, 941
- to coverage in health plan markets 776 - - carveouts 935-936
- to health care - - evidence of 926-927
- - equality of - - and managed care 928-931
- - - clinical services 1707-1709 - policy responses to 927-928, 931-936
- - - in finance and delivery 1812-1813 - - risk adjustment 932-935
-- - in health sector 89-90 advertising
- - - opportunity set ethically constrained - of alcohol 1644-1645
1893-1894 - in pharmaceutical industry 1303
- - long-term models of 966-968 - of tobacco and smoking 1543, 1584-1593
- to smoking by youths 1597-1598 - - and child health 1079
accidents - - econometric evidence 1585-1591
- in child health programs 1078 -- noneconomic literature, findings from
- and disability programs, USA 1017
1591-1593
- in medical malpractice 1345
-- theory 1585
Acquired Immune Deficiency Syndrome (AIDS)
Advisory Council on Breakthrough Drugs 1329
814, 1763, 1766, 1768, 1772
age/ageing
- assortative matching 1773, 1777
- and disability 1002-1003, 1004, 1005
- information on 1785-1786, 1789-1791
- and equity in health 1876, 1904
-interventions, timing 1781-1782
- and health care spending 128
-prevention of 1695, 1700, 1707, 1712
- health risks by 675
Activities of Daily Living (ADLs) 804
- health spending by 795-796
- and disability 1001
addiction models of smoking 1543, 1556-1563 - in long-term care 976-977
- imperfectly rational models 1556-1557 - - macroeconomic effect of 986
- myopic models 1557-1559 - long-term care for 957
-rational models 1559-1561 - and SSDI beneficiaries 1014
-- critiques of 1561-1563 - and SSI beneficiaries 1016
addictive disorders 899 agency
administered waiting 1222 - in health care markets 1415-1417
administrative costs - in health sector 75-76
- of long-term care 980 - long-term relationships, in general practice
- in medical malpractice 1344 1178
administrative data sets in child health research - problems of in health insurance 566-567
1083 - in waiting lists 1219
adverse events in medical malpractice 1351 - see also physician agency
adverse selection Agricultural Adjustment Act (USA, 1933) 1601
- antitrust in health care markets 1414-1415 agriculture, tobacco (USA) 1544
- in health insurance 567, 607, 614, 616, 623 - economic contribution 1606-1610
- in health insurance markets 608-609, - regulatory system on 1601-1605
616-623, 634-637 - - farm programs 1601-1603
- in health plan markets 771-773, 774 - - relevance to health 1603-1605
1-45
I-46 Subject Index
- size and nature 1599-1601 antidiscrimination legislation against disability

Aid to Families with Dependent Children 1032-1033
(AFDC) 681, 682, 684-685 antipoverty effectiveness of cash transfers to
alcohol consumption 1629-1673 disabled 1029
- consequences of 1645-1653, 1697 antitrust and competition in health care markets
--cirrhosis 1649-1650 1405-1487
-- crime and suicide 1652-1653 - background to 1408-1409
- heart disease 1651-1652 - differences with other markets 1409-1418
- - intertemporal 1634 -- differentiated product 1410-1412
- - medical care 1650-1651 - - government regulation 1417-1418
- - motor-vehicle mortality 1646-1649 - - imperfect information 1413-1417
- - - studies of 1648 - - - adverse selection 1414-1415
-demand for 1638-1645 - -- agency problems 1415-1417
-- advertising 1644-1645 -- - moral hazard 1413-1414
- - heavy drinking 1642-1643 -- not-for-profit firms 1418
- - measurement 1638-1639 - foreclosure 1470-1478
- - social influence 1643-1644 -- anti-competitive concerns 1473-1476
- - - most favoured nation clauses 1475-1476
-- youth 1641-1642
- -- vertical integration 1473-1475
- dependence and abuse 1655-1656
- - efficiencies 1471-1473
-productivity, impact on 1653-1658
-- vertical restraints
- - direct effects 1654-1657
- - - cases 1472
- - and human capital 1657-1658
- - - research on 1476-1478
- research agenda 1634-1638
- hospital mergers 1418-1456
- taxation, evaluation of 1658-1665
- - conversions and ownership status 1436-1445
- - background to 1658-1659
- - - model, identifying 1443-1445
- economic efficiency 1662-1663
- - - model, implications 1439-1443
-- equity criteria 1659-1662
- - empirical evidence 1445-1455
- - - horizontal, and user fees 1660-1662 - - - economies of scale and efficiency
- - - incidence 1659-1660
1451-1453
--total social cost 1664-1665 - - - entry studies 1455-1456
- trends and patterns 1632-1634 - --event studies 1453-1455
- - aggregate data 1632-1633 --- structure-conduct-performance
--individual differences 1633-1634 - - - - non-price 1450-1451
allocative efficiency in human capital model 396 - - - - price 1448-1450
alternative dispute resolution in medical - - enforcement 1418-1426
malpractice 1376 - - - framework for analysis 1419-1420
alternative models in health plan markets - - - outcomes 1420-1426
794-807 - - -process 1418-1419
altruism in health care delivery 1807 - - market definition 1426-1436
Ambulatory Care Group (ACG) 798, 799, 808, - - - background 1426-1429
813 - - - and managed care 1435-1436
American Cancer Society 1685 - - - proposed method 1429-1435
American Hospitals Association 149 - monopsony 1456-1478
American Medical Association 465, 469-471 - - bilateral market power 1466-1470
Americans with Disabilities Act (1990) 1032 - - - identifying market power 1466-1467
anthropometric measures of child health 1062 - - - relative bargaining power 1468-1470
anti-substitution laws on branded drugs 1321 - - - welfare implications 1468
anti-ulcer agents 1300, 1301 - - monopsony power 1457-1466
antibiotics 1300, 1307 - - - and antitrust enforcement 1457-1459
anticipation effects in moral hazard 412, 441 - - - detecting 1461-1462
Subject Index I-47
- - - studies of 1463-1466 - caps on 1355, 1372, 1375

- - - and welfare 1459-1461 -punitive 1372-1373
antitrust in pharmaceutical industry 1327 - scheduled 1373
'any willing provider'
- in health plan markets 777 baby boom generation and long-term care 983
- in managed care 721 balance billing in insurance reimbursement
arc price elasticity in health insurance 430 547-548
aspirin 1306 Balanced Budget Agreement in long-term care
asymmetric information 971
-in general practice 1180, 1184 'bang-bang' equilibrium in human capital model
- in government purchase of health services 364
867-874 bargaining within household, in child health
- - competitive bidding 872-873 1058
- - mechanism design approach 868-870 barriers to care in child health 1060
- - payment mechanisms 870-872 bed supply models of long-term care 964-966
- in health care markets 1095 Belgium
- in health insurance markets 626 - equity in health care utilization, indices 1846
- health expenditure 14, 850
- in health plan markets 759
- health plans in 833
- in health sector 73-79
- risk adjustment and sharing in 830
- in managed care 721-723
bequests in long-term care 974
- in medical malpractice 1344, 1395
best practice adoption 223-264
- in not-for-profit hospitals 1148, 1149
- market equilibrium 228-237
- in physician agency 463, 466, 496-503
---incomplete information models 230-237
-- irreducible uncertainty 497-498
- - - general model 230-236
- - unobservable physician actions 498-502
- - - - application 234
- - unobservable physician characteristics
- - - - insurance and costs of search 235-236
502-503
- - - - price variability 236
- in supplier-induced demand 78-79 - -- - search with unknown quality 234-235
- in waiting lists 1236 - - - herd and cascade models 237
Australia - - search and 228-230
- dental services, inputs and outcomes 1256 - production and dissemination of information
- expenditure on mental health and substance 257-261
abuse 898 - - costs of 258-261
- health expenditure 14, 850 - - property rights 257-258
- long-term care for elderly 963 - production function, disagreement 237-256
- medical expenditure 171 - - healer's dilemma 237-239
- smoking and tobacco - physician-specific variations 248-251
- - advertising, banned 1589 --regional variations 239-248
- -counter-advertising 1595 - - - and individual variations 251-252
-taxes to finance health care delivery 1828 - - welfare losses 252-256
- waiting lists, incentives to reduce 1234 best price in pharmaceutical industry 1326
Austria bidding models in insurance reimbursement
- dental services, inputs and outcomes 1256 557-558
- health expenditure 14, 850 binary choice in waiting lists 1226
- long-term care for elderly 963 binary responses in health econometrics
- payment systems in general practice 1188 275-276
authority, delegation of in moral hazard 411, 414 biological science and pharmaceutical industry
average causal effects in health econometrics 1304
269 birth rates, teenage 1079
awards in medical malpractice 1345, 1358, birthweight as measure in child health 1062,
1371, 1383 1066
1-48 Subject Index
bivariate probit model in health econometrics capital cost recovery in not-for-profit hospitals
283 1153
bivariate relationships in waiting lists 1228 capital funds in not-for-profit hospitals
block contracts in NHS 866 1163-1165
blockbusters in new drug development 1317, capital stock measures of hospital regulation
1331 1519-1521
Blue Cross 482, 575-576 capitation payments
- and health care markets 1130 - in general practice 1188
- and managed care 720 - in managed care 715-716, 723
- monopsony in health care markets 1469 - in medical malpractice 1348
- and not-for-profit hospitals 1159 - for mental health 921-925
board and care homes, long-term 961 - - early experiences 922-923
booking appointments in waiting lists 1243 -- recent evidence 923-925
bootstrap data in health econometrics 272 - and waiting lists 1233
branded drugs 1321 capping in pharmaceutical industry 1329
- and generic drugs 1322-1324 cartel theory 1152
British Social Attitudes Survey (BSAS) 1231 carveouts
broad-spectrum antibiotics 1307 - in health plan markets 778, 815
burden-of-illness (BOI) 1703 - in mental health 897, 905-907, 925-926,
Bureau of Economic Analysis (USA) 168, 169 935-936
Bureau of Labour Statistics (USA) 122, cascade models of information diffusion 237
135-136, 138, 147, 170, 174 cash benefits in disability policy 1028-1030
causal relationships in human capital model 396
California Office of Statewide Health Planning caveat emptor in medical malpractice 1345,
and Development (OSHPD) 1116, 1129 1346
California study of medical malpractice 1351, Centers for Disease Control (CDC, USA)
1354 - alcohol-related mortality 1646
Canada - on infectious diseases 1766, 1789, 1795
- dental services, inputs and outcomes 1256 - prevention, spending on 1685, 1686
- expenditure on mental health and substance cephalexin 1322-1323
abuse 898 certificate of need (CON)
- generic substitution in drugs 1325 - in hospital investment
- health expenditure 14, 850 --controls 1491, 1493-1495, 1498, 1501-1502
- long-term care for elderly 963 - - effects of 1526-1527
- medical expenditure 172-173 - - impact assessment of 1508-1511, 1517,
- medical malpractice in 1344, 1357 1521-1522, 1526-1527
- - collateral source effect 1374 - - model of 1504
-- contingent fees 1375 - and long-term care 965
- - negligence claims for 1387 Channeling demonstration 962
- national health insurance in 698 chargeback in pharmaceutical industry 1326
- not-for-profit hospitals in 1145 children
- outpatient drug reimbursement 1330 - in developed countries 1053-1090
-physicians' home visits 1839-1840 ---'at risk' 1081
- population projections of elderly 983 - - government's role in 1067-1083
- QALYs in measuring health outcomes 101 - - - crowdout 1074
- smoking - - - direct provision of services 1077
-- economic contribution of industry 1608 - - - institutional background 1067-1068
-- social costs of 1575 - - - and insurance 1068-1071
-- taxation of 1570-1571 - - - -coverage 1075-1076
- universal health insurance 1069 - - - public health programs 1078-1083
capacity structures and waiting lists 1207 - - - - community-based 1082
Subject Index I-49
- - - - education and information campaigns - - - increasing taxes, health benefits of

1078-1080 1577-1579
- - - - home visiting 1081-1082 - - - social costs 1575-1577
---- incentives 1082 cirrhosis 1649-1650, 1664
- - - - product safety 1080-1081 claims frequency in medical malpractice 1343,
- - - utilization of care, efficiency 1071-1073 1354, 1355-1356, 1360
-- market failure in 1063-1067 claims severity in medical malpractice 1343,
- -- externalities 1065-1067 1355
- - - imperfect information 1063-1065 class and child health 1075
- - model of demand for health inputs Clayton Act (USA, 1914) 1408, 1419
1056-1058 cleanings, in dental services, prices 1262, 1269
- - - health as investment 1061-1062 clinical guidelines for waiting lists 1240
- - - law of demand 1058-1061 clinical testing in new drug development 1322
- - - utility of model 1058-1063 cluster market approach in health care markets
- and disability 1033-1037 1423
- special education 1035-1037 cognitive development in child health 1062
- - SSI benefits for 1033-1035 coinsurance
- and health insurance 443 - and moral hazard 579, 580, 584-586
China - rate of 421-424, 428-429, 437, 446
- health expenditure 15 cold surgical cases in waiting lists 1219
- smoking, social costs of 1575 collateral offset in malpractice awards 1390
Chronic Disease Score (CDS) 802, 803 collateral source effect in malpractice awards
chronic illness and long-term care 957 1374
churning, in child health 1061 Colombia 830
cigarettes commitment in government purchase of health
- demand for and price 1547-1556 services 874-878
- - addiction models 1556-1563 - investment and hold-up 875-877
- - - imperfectly rational models 1556-1557 - ratchet effect 877-878
- - - myopic models 1557-1559 - supply assurance 874-875
- - - rational models 1559-1561 communicable diseases, targeting 71
- - - - critiques of 1561-1563 community-based health programs in child
-- aggregate data 1547-1549 health 1082
- - behavioural economic analyses of community benefits in not-for-profit hospitals
1563-1564 1144, 1160
- - econometric studies 1564-1565 compensation
-- individual data 1549-1556 - in cost-effectiveness of health interventions
- in human capital model 399-400, 403 187
- restrictions 1596-1597 - and equity in health 1892-1893
-taxation 1565-1584 -in medical malpractice 1343, 1345,
- -comparisons 1566-1571 1357-1358, 1369, 1372, 1387, 1390
- - - country variations 1569-1571 competing risks in health econometrics 330-331
- - - effects on retail price 1567-1569 competition
- - - purposes and methods 1566-1567 - and antitrust in health care industry 1411, 1415
- - - smuggling 1569-1571 - in managed care 726-727, 743-744
--fairness standards 1571-1574 - in not-for-profit hospitals 1167-1168
- - - benefit principle 1574 - and selection in health insurance markets 624
- - - horizontal and vertical equity 1572-1574 competitive bidding in government purchase of
- - optimal, and economic efficiency 1579-1584 health services 872-873
- - - negative externalities 1579-1583 competitive equilibrium and antitrust in health
-- - other considerations 1583-1584 care industry 1411-1412
- - public health standards 1574-1579 competitive markets, nursing home care in 964
1-50 Subject Index
complementary relationships in human capital - of alcohol see alcohol consumption

model 397 - in human capital model 374
complements in insurance reimbursement 548 - tobacco, consequences 1544
complex output in not-for-profit hospitals 1148, contingent fee in medical malpractice 1357,
1149 1387
concentrated output in not-for-profit hospitals - limits 1375-1376
1159 contingent valuation
conditional demand function in human capital - in health sector 100
model 386 - in waiting lists 1222, 1225, 1227
conditional logit in health econometrics continuing care retirement community (CCRC)
- estimator 310-311 961
- model 278 contract length in health insurance 626-629
condom, use of 1695, 1701, 1790, 1791 contractarianism and equity in health 1892
congestive heart failure, quality of care 1162 contracted physicians in health care markets
Connecticut Study on health insurance 429 1125-1134
consequentialism and equity in health 1872 - measuring quality 1133-1134
Consolidated Omnibus Reconciliation Act - multinomial choice models 1131-1133
(COBRA) 671 - payer-driven competition 1126-1131
consultants, and waiting lists 1228 - - effect of 1128-1131
Consumer Expenditure Survey (CEX) 136-138, contracting out in not-for-profit hospitals 1150
161-162, 694 contribution margin in new drug development
consumer incentives in health care 409-459 1316
- dimensions of 412-414 control function in health econometrics 270
- and ex ante moral hazard 415-420 copayment in health care 420, 430, 432-433
-- background 415-418 - for drugs 434, 1301, 1325
-- empirical evidence 418-420 - in long-term care 977
- ex post moral hazard 420-451 coronary heart disease 1162, 1651
- - background 420-429, 444-446 cost-benefit analyses (CBA)
- - - deductible and coinsurance effects on - in dental services 1285, 1287-1288
demand 423-426 - of health interventions 184-187, 219
- - - insurance plan generosity 420-423 - - decision making in 198-199
- - - pure price effects 426-429 - - distributional considerations 211
- - dynamic 451-453 - - fixed costs 207
- - - background 451-453 - - measurement of 205
- - - empirical evidence 453 - in health sector 99, 100, 103-105
- - empirical evidence 429-444 -of prevention 1706
- - - assessment 443-444 cost-effective analysis (CEA)
- - - Health Insurance Experiment 436-443 - in child health 1076
- - - natural experiments 429-433 - in dental services 1285, 1286-1287
- - - observational comparisons 433-436 - efficiency in health sector 60
- - full price effects 444-449 - of health interventions see health
- - background 444-446 interventions, cost-effectiveness of
- - - empirical evidence 446-448 - in health sector 97, 98-99, 100, 103-104
- - - methodology 448-449 - of prevention 1705-1706
- - rationing by physician 449-451 - in regulation of hospital prices and investment
- - - background 449-450 1499-1503
- - - empirical evidence 450-451 - in waiting lists 1231
consumer information cost of adjustment health models 390, 392
- in health plan markets 779 cost-of-illness (COI)
- on prevention, lack of 1696-1701 - and alcohol consumption 1664
consumption - to infectious diseases 1787-1789
Subject index I-51
- and prevention 1703 - in public assistance programs 686-687

cost per case contracts 853 crowns, in dental services, prices 1262
cost reimbursement by government 853 Current Population Survey (CPS) 648, 652, 679
cost sharing customary care in medical malpractice 1343,
- in child health 1059 1347, 1348, 1383
- demand side 856 Czech Republic
- in health care markets 1119 - health expenditure per capita 14
- in managed care 712-713, 722-724 - risk adjustment and sharing in 830
- supply side 856
- - in physician agency 487-492 data envelopment analysis (DEA)
cost shifting - in dental services 1276-1280
- dynamic, in health care markets 1123-1124 - in health econometrics 331
- in medical malpractice 1376 deadweight time price in waiting lists 1214
- in not-for-profit hospitals 1158-1160 death in human capital model 350
-in PPS 864 decay rate in waiting lists 1215
decision making in health interventions 187-198
cost-utility analyses
- in health sector 104-105 - average ratios 192
- and health policy 197-198
- in quality of life measurements 1725
- incremental ratios for multiple alternatives
costs
193-196
- acceptable, in health plan markets 767, 779
- optimal 198-204
- administrative, of long-term care 980
- sensitivity analysis 196-197
- of equity capital in not-for-profit hospitals
- time horizons 191-192
1165
decreasing block pricing in health insurance
- of managed care, growth in 740-743
policies 421, 454
- monitoring, in general practice 1179
deductibles in health insurance policies
- in new drug development 1310 421-423, 425, 428, 451-452
- in not-for-profit hospitals 1155-1157 - and medical malpractice 1361
- of waiting per family 1224
- and moral hazard 584, 585
count data regressions in health econometrics defensive medicine 516-517, 1368-1369, 1395
315-324 defensive practice in medical malpractice 1343,
-basic model 316-318 1365
- count data 315-316 demand
- overdispersion and excess zeros 318-323 - for alcoholic beverages 1638-1645
- unobservable heterogeneity and simultaneity - - advertising 1644-1645
biases 323-324 -- heavy drinking 1642-1643
counter-advertising of tobacco and smoking - - measurement 1638-1639
1544, 1593-1596 - - social influence 1643-1644
Cox model in health econometrics 326 -- youth 1641-1642
cream skimming - for child health inputs 1056
- in government purchase of health services 871 - for dental services 1268-1274
- in health plan markets 773-774, 775 - induced, in health care markets 1416
crime and alcohol consumption 1652-1653 - for regulation in US hospitals 1491-1493
crises in medical malpractice insurance 1343, -- models 1503-1504
1355, 1359, 1361 demand elasticity for drugs 1301
critical periods in child health 1062 demand function in human capital model 378,
cross-price elasticities for NHS private care 1230 394
cross-subsidies in health insurance markets 609, demand-side cost sharing 856
771 demand for dental services 1261
crowdout demographic change and moral hazard 451
- in child health programs 1074 demographic composition of disability
- in not-for-profit hospitals 1161 1002-1006
I-52 Subject Index
demographic models in health plan markets - markets for 1254-1257

794-797 - - inputs and outputs 1255-1257
demographics of long-term care 983-987 - - and other medical services 1254-1255
- exacerbation of problems 984-987 -productivity 1274-1284
- trends 983-984 - - economies of scale 1276, 1282-1283
denial of coverage, claims in medical -- efficiency 1275
malpractice 1383-1384 - - measurement and results 1280-1284
denial of payment, claims in medical - - methods 1276-1280
malpractice 1389 --productivity 1280-1282
Denmark - - technical efficiency 1283
- birth rates, teenage 1079 --theoretical concepts 1275-1276
- dental services, inputs and outcomes 1256 dentures, in dental services, prices 1262
- equity in health care delivery depreciation rate in human capital model 357,
--financing 1821, 1825, 1834 377
- -private health insurance in 1820, 1826 deprivation and child health programs 1083
- - taxes to finance 1824 detailing in pharmaceutical industry 1303, 1327
- utilization, indices 1844, 1846 deterrence in medical malpractice 1343, 1345,
- health expenditure 14, 850 1348, 1350, 1358, 1377, 1380, 1388,
- long-term care for elderly 963 1390, 1392
- payment systems in general practice - benefits in liability 1370-1371
1188-1189 developed countries, health insurance in
dental services 1251-1296 568-571
- demand for 1257-1274 developing countries
- - empirical work on 1260-1274 -pharmaceutical industry patents in 1318-1319
- - -models 1265-1268 - prevention, policies to encourage 1709-1710
- - - - cross-section studies 1265-1267 diagnosis-based models in health plan markets
- - -- time series 1267-1268 798-802
- - - results 1268-1274 Diagnostic Cost Groups (DCGs) 798, 800-801,
- -- -insurance 1255, 1271-1272 808-809, 813
- - - -prices and income in demand 1268-1270 Diagnostic Related Groups (DRGs)
- - - -supplier inducement 1272-1273 - in government purchase of health services 851,
- - - -time cost 1270-1271 862, 866
-- - variables, measurement 1261-1265 - in health care markets 1097, 1118, 1119, 1121
- - --insurance 1263-1265 - and hospital mergers 1455
- - - money price 1261-1262 - in health plan markets 799, 817
---- socioeconomic variables 1265 - in mental health 933
- -- - supplier inducement 1263 differential optimal waiting lists 1209
- - - - time price 1262-1263 differentiated oligopoly in pharmaceutical
- - theoretical considerations 1257-1260 industry 1320
- - - individual demand, modelling 1257-1259 direct provision of services in child health
- - - supplier inducement 1259-1260 programs 1077
- economic evaluation 1284-1291 direct utility function in waiting lists 1226
- costs, measurement of 1285-1286 directors, board of in not-for-profit hospitals
- - outcomes, measurement of 1286-1288 1147
-- results 1288-1291 disability 995-991051
- - - fluoridation programmes 1288-1289 - of children 1033-1037
- - - periodontal programmes 1290 - special education 1035-1037
- - - restorative treatments 1290-1291 - - SSI benefits for 1033-1035
- - - sealant programmes 1289-1290 - defined 998-999
- externalities in 1254 - demographic composition 1002-1006
and insurance 440, 442, 443, 1255, 1271-1272 - disease patterns underlying 1006
Subject Index I-53
- economic costs of 1010-1012 - - variations in procedures 243, 244-246,

- economic well-being of 1008-1010 247-251, 254-255
- economics of policies toward 1020-1033 - see also physician agency; physicians
---antidiscrimination legislation 1032-1033 donations for not-for-profit hospitals 1151
- - cash benefits 1028-1030 -private 1147
- - economic well-being 1028-1030 Dorfman-Steiner theorem in pharmaceutical
- - effects 1012-1023 industry 1303
- - in-kind benefits 1031-1032 double hurdle model in health econometrics 297
- labor supply effects of benefits 1023-1028 drug lag 1311, 1314
-- - studies of 1024 drugs
- - poverty reduction 1028-1030 - demand for 440
--public health care 1031-1032 - exposure to in child health 1066
-- rehabilitation 1030-1031 -formulary 1319, 1325-1327
--return-to-work 1030-1031 -generic 152, 158
-- training 1030-1031 - - competition 1321-1322
- employment patterns of 1006-1008 - illicit 1066
- and health insurance 442 -prescription 158
- identification of 997-991012 -pricing 1319-1328
- - branded drugs 1320
- policy in OECD countries 1037-1040
- -branded vs generic competition 1322-1324
-policy toward working-age, USA 1012-1020
- - generic drug competition 1321-1322
- - accident and injury prevention program 1017
- - and Medicaid 432
- - health insurance 1019-1020
- - new institutions 1325-1328
-- rehabilitation 1018-1019
-- stimulating generic substitution 1324-1325
- - return-to-work programs 1018-1019
- tort liability risks and 1320-1321
- - SSDI program 1013-1015, 1042-1044
- proprietary (over the counter) 1300,
- - SSI program 1015-1016, 1044-1045
1303-1304
-- vocational training 1018-1019 - rational design 1307
-- Workers' Compensation program 1017, - see also substance abuse
1045-1047 duration analysis in health econometrics
- prevalence of 999-991002 325-331
- unresolved issues 1040-1042 - competing risks and multiple spells 330-331
- years lived with, causes 1007 - methods 326-330
disability-adjusted life years (DALYs) 102 - - parametric models 327-328
- and equity in health 1867 - - semiparametric models 326-327
- and prevention 1678 - - unobservable heterogeneity 328-330
- and quality of life measurements 1726 - survival and duration data 325-326
Disability Payment System (DPS) 798, 801, 814 dynamic demand for health models 388
discounted marginal benefits in human capital dynamic moral hazard 413, 451
model 360, 365
disease patterns underlying disability 1006 earnings of disabled 1008-1009, 1011
dissemination of information 225, 257-261 econometrics see health econometrics
- costs of 258-261 economic contribution of smoking taxation 1544
- property rights 257-258 economic effects of alcohol consumption
distributional equity in health sector 87 1634-1635
distributive justice and equity in health 1869, economic evaluation
1877, 1880, 1888-1891, 1893 -of dental services 1284-1291
doctors - -costs, measurement of 1285-1286
- and information diffusion 226-227, 229 - -outcomes, measurement of 1286-1288
- - and property rights of 257-260 --results 1288-1291
- - search models 235-239 - - - fluoridation programmes 1288-1289
I-54 Subject Index
-- - periodontal programmes 1290 efficient levels of care and medical malpractice

- - -restorative treatments 1290-1291 1343
- - - sealant programmes 1289-1290 elasticity
- of health care 67-87 - in human capital model 369, 372, 374, 382
- - derived demand for 68-70 - of numbers on waiting lists 1228
- - externalities 70-73 elderly
- informational asymmetry 73-79 - long-term care for 963, 988
- - - supplier-induced demand 78-79 - population projections of 983
- - uncertainty 79-87 - see also age/ageing
- - - economies of scale 81-82 eliciting preferences and waiting lists 1225
- - - insurance, welfare improving effects 80-81 emergency room
- - - moral hazard 83-87 - care in child health 1066
- - - risk selection 82-83 - utilization of 440
economic models of hospital behaviour 1153 - visits 433
Economic Stabilization Program (ESP), employer-provided health insurance and
regulation by 1497-1498, 1502 mobility 654-658
-effects of 1509, 1511, 1516-1517, 1527 Employer-Retirement Income Security Act
economic stabilization programs (ESP) in (ERISA) 1386
hospital regulation 1527 employment
economic well-being and disability 1008-1010, - discrimination, and equity in health care
1028-1030 delivery 1837
- patterns of, and disability 1006-1008
economics of equity in health care 1866-1870
endogeneity
economies of scale
- of demand in waiting lists 1230
- in dental services 1276, 1282-1283
- in not-for-profit hospitals 1170
- health care as economic commodity 81-82
enforcement of antitrust in health care markets
- in health econometrics 331
- hospital mergers 1418-1426
- in hospital mergers 1451-1453
- - framework for analysis 1419-1420
- - outcomes 1420-1426
economies of scope in not-for-profit hospitals
--process 1418-1419
1157
- monopsony power 1457-1459
education English rule in medical malpractice 1376, 1387
- in child health programs 1078 enrollment
- and disability 1002 - biased, in insurance markets 616-623
- and not-for-profit hospitals 1168 - in health plan markets 778
- in public health programs in child health - - periodic open 770
1078-1080 enterprise liability in medical malpractice
- special, for disabled children 1035-1037 1377-1378, 1381, 1388
- see also schooling entry barriers in health plan markets 778, 1103
Education for All Handicapped Children Act environmental tobacco smoke (ETS)
(USA, 1975) 1035 1580-1582, 1596
effective price of medical care 424-425, 452 epidemics
efficacy in pharmaceutical industry 1308 -behaviour of 1767-1768
efficiency - dynamics of 1769-1773
- alcohol consumption, excise taxes on epidemiology of infectious diseases see
1662-1663 infectious diseases, epidemiology of
- in dental services 1275, 1283 episode treatment costs, index based on 163-166
- in health plan markets 763 EQ-SD index of well-being 1745, 1748
- and monopsony in health care markets equality of access in health sector 89-90
1471-1473 equality of health 1846-1855
- in not-for-profit hospitals 1168 - and economics of health sector 92-94
Subject Index I-55
- pure inequalities 1847-1849 - - - linear maximand: equal weights

- - empirical work on 1848-1849 1884-1885
- - measurement of 1847-1848 - - - one side condition 1884
- socioeconomic inequalities 1849-1855 - - - other permutations 1885
-- empirical work on 1852-1855 - - without side conditions 1871-1883
-- measuring 1849-1852 - - - linear maximand 1871-1875, 1875
equilibrium in information diffusion 228-237 - - - no side conditions 1871
- incomplete information models 230-237 - - - non-linear and kinked maximand
- - general model 230-236 1880-1883
- - - application 234 - --non-linear maximand 1876-1880
- - - insurance and costs of search 235-236 - philosophy and economics 1866-1870
- - - price variability 236 - reducing inequalities 1865
-- - search with unknown quality 234-235 eradication of infectious diseases, welfare
- - herd and cascade models 237 effects 1784-1785
- search and 228-230 estimation in health econometrics 268-274
equity-efficiency tradeoffs and equity in health - estimation strategies 269-274
- - evaluation 268-269
care 1901-1905
- - and model specification 270-272
equity in health care 1863-1910
- - nonparametric and semi parametric
- and alcohol taxation 1659-1662
estimators 272-274
- - incidence 1659-1660
- - treatment effects 269-270
- - and user fees 1660-1662
ethical codes in health plan markets 779
- in child health 1076
European Medicinal Products Evaluation
- and economics of health sector 87-96
Agency (EMEA) 1314
- - allocation according to need 90-92
evaluation in health econometrics 268-269
- - equality of access 89-90
ex ante moral hazard in health care 415-420
- - equality of health 92-94
- background 415-418
- - rival notions of 95-96
- empirical evidence 418-420
- economist's role 1899-1901 ex post moral hazard in health care 420-451
-equity-efficiency tradeoffs, estimating - background 420-429, 444-446
1901-1905 - - deductible and coinsurance effects on
-health, measured 1899 demand 423-426
- opportunity set ethically constrained, theories -- insurance plan generosity 420-423
1889-1898 - - pure price effects 426-429
- - with a maximand 1895-1898 - dynamic 451-453
- - - equality of opportunity 1895 - - background 451-453
-- - fitting population change into a framework - - empirical evidence 453
1898 - empirical evidence 429-444
- - - Rawlsian theory 1896-1898 - - assessment 443-444
- - with no maximand 1890-1895 - - Health Insurance Experiment 436-443
- - - contractarianism 1892 - - natural experiments 429-433
- - - equality of access 1893-1894 - - observational comparisons 433-436
- - - ethics and rule of rescue 1894 - full price effects 444-449
-- -libertarianism 1890-1891 - - background 444-446
--- no-envy principle 1892-1893 - - empirical evidence 446-448
- - - participatory democracy 1891 - - methodology 448-449
- opportunity set ethically unconstrained, - rationing by physician 449-451
theories 1871-1889 - - background 449-450
-- assumed general properties 1871 - - empirical evidence 450-451
- needs-based 1887-1889 examinations, in dental services, prices 1262,
- with side conditions 1884-1885 1269
I-56 Subject Index
excess capacity in hospital regulation 1523-1524 fiduciary relationships in not-for-profit hospitals

excess zeros in health econometrics 318-323 1148
exchequer budget costs in waiting lists 1236 15-D index of well-being 1745, 1748
excise taxes on alcohol 1658-1661 fillings, in dental services, prices 1261, 1269
- economic efficiency of 1662-1663 finished consultant episode and health service
exercise treadmill testing, cost-effectiveness of quality 853
193 finite density estimator in health econometrics
existence value and waiting lists 1225 313
exit barriers in health plan markets 778, 1103 Finland
Expected Utility Theory (EUT) and quality of - dental services
life measurements 1734-1735 - -demand for 1270
expected value of optimal wait 1211 - - inputs and outcomes 1256
experience rated malpractice insurance 1350, --productivity 1281, 1282
1359-1360 - equity in health care delivery
explicit subsidies --financing 1821, 1825, 1834
- in health plan markets 763 --utilization, indices 1846
- in not-for-profit hospitals 1148 - health expenditure 14, 850
externalities - long-term care for elderly 963
- in dental care 1254 - smoking and tobacco
- health care as economic commodity 70-73 - advertising, banned 1589
- in market failure - - counter-advertising of 1595
- - in child health 1065-1067 first come first served basis of waiting lists 1212
first mover advantage in pharmaceutical industry
- and prevention 1694-1696
1324
- and not-for-profit hospitals 1144
fixed effects in health econometrics 308
- public system of mental health 941-943
fixed price contracts in government purchase of
- in smoking taxation 1571
health services 858-859
-negative 1579-1583
Flexner Report (USA) 472
extra-welfareism in economics of health sector
fluoridation programmes in dental services
62-66
1288-1289
extractions, in dental services, prices 1262, 1269
Food and Drug Administration (USA) 1309,
1311, 1314, 1700, 1702
fair innings approach to life expectancy 1876 Food, Drug, and Cosmetic Act (USA, 1938)
fairness in health plan markets 763, 783-784 1309
fairness standards in taxation on smoking Food Stamps 681, 686
1571-1574 for-profit hospitals 1144, 1146
- benefit principle 1574 foreclosure, antitrust in health care markets
- horizontal and vertical equity 1572-1574 1470-1478
feasibility in health plan markets 784-785 - anti-competitive concerns 1473-1476
Federal Employees Health Benefit Program - - most favoured nation clauses 1475-1476
(FEHBP) 926-927 - vertical integration 1473-1475
Federal Trade Commission (FTC) 1419, 1700 -efficiencies 1471-1473
Federal Trades Commission Act (USA, 1914) - vertical restraints, research on 1476-1478
1408 formal long-term care 960
fee-for-service payments formula pricing in pharmaceutical industry 1329
- in dental services 1260 France
- in general practice 1188, 1189 - dental services, inputs and outcomes 1256
-in health care markets 1118 - equity in health care delivery
- in managed care 715-716 -- financing 1821, 1825, 1834
- and medical malpractice 1348, 1383 - private health insurance in 1820, 1826
fertility rates and long-term care 983 - taxes to finance 1827
Subject Index I-57
- health expenditure 14, 850 generalized Tobit model in health econometrics

- international drug sales 1313 285, 286
- long-term care for elderly 963 generic drugs 152, 158
- not-for-profit hospitals in 1145 - and branded drugs 1322-1324
- outpatient drug reimbursement 1329 - competition 1321-1322
- population projections of elderly 983 generic substitution in drugs 1324, 1325
Frisch demand function in human capital model geographical identifiers in child health research
384 1083
frontier production function in not-for-profit Germany
hospitals 1156, 1157 - dental services, inputs and outcomes 1256
full price effects in ex post moral hazard in - and disability
health care 444 449, 452 -- policy 1038, 1039
- background 444-446 - - transfer recipients 1003
- empirical evidence 446-448 - equity in health care delivery 1828
- methodology 448-449 --financing 1821, 1825, 1834
full price of medical care 446-447 - - taxes to finance 1824, 1826
full wealth constraint in human capital model -health care costs 1331
354 - health expenditure 14, 850
fundholding in general practice, UK 1187 - international drug sales 1313
future stocks in health model 392 - long-term care for elderly 963
- not-for-profit hospitals in 1145
games of persuasion in health care markets 1416 - population projections of elderly 983
gatekeepers - risk adjustment and sharing in 830
- in child health 1061 - vocational rehabilitation programs 1019
- in general practice 1177, 1185 government
- in managed care 717, 722 - in child health programs 1067-1083
- in waiting lists 1219 - - crowdout 1074
gender - - direct provision of services 1077
- health spending by 795-796 - - institutional background 1067-1068
- in long-term care 976 - - and insurance 1068-1071
- and SSDI beneficiaries 1014 - - - coverage 1075-1076
- and SSI beneficiaries 1016 - - public health programs 1078-1083
General Household Survey (UK) 1549 - - - community-based 1082
general practice 1175-1200 - - - education and information campaigns
- as firm 1190-1193 1078-1080
- - group practice 1190-1191 - - - home visiting 1081-1082
--partnerships 1190-1191 --- incentives 1082
--vertical integration 1191-1193 - - -product safety 1080-1081
-fundholding, UK 1187 -- utilization of care, efficiency 1071-1073
- GP-patient relationship 1178-1180 - health services, purchase of 847-890
- models of GP behaviour 1184-1185 - - asymmetric information 867-874
-payment systems 1187-1190 - - - competitive bidding 872-873
-referrals, UK 1186 - - - mechanism design approach 868-870
-referrals, variation in 1185-1187 - - - payment mechanisms 870-872
-utilization of services 1180-1184 - -commitment 874-878
-- GP, choice of 1183 - - - investment and hold-up 875-877
--practice, choice of 1181 - -- ratchet effect 877-878
-- seeking medical care 1182-1183 - - - supply assurance 874-875
-- treatment, choice of 1183-1184 - - demand and fixed price contracts 857-862
generalized method of moments (GMM) in - - - efficiency 858-859
health econometrics 271 --- efficient treatment numbers 860-861
1-58 Subject Index
- -- quality and effort 859-860 - - - economies of scale 81-82

- - payment mechanisms 866-867, 870-872 --- insurance, welfare improving effects 80-81
-- quality and costs 853-857 - - - moral hazard 83-87
- - - framework for 853-855 - - - risk selection 82-83
- - - prospective payment 862-866 - equity in finance and delivery 1803-1862
- - -c- osts 863-865 --defined 1806-1819
- - - -quality 865-866 - - - ability to pay principle, justifying
- - - quality and effort 855-857 1817-1818
- - reputations 878-883 - - - and access 1812-1813
- - - for characteristics 879 - - - and altruism 1807-1808
- - - for past behaviour 880-883 - - - conflicts 1813-1814
- public sector provision of health services -- ideology and health care systems
25-26 1809-1810
- regulation of antitrust in health care markets --- literature on 1810-1811
1417-1418 - - - principles, justifying 1814-1817
- regulation of drug introductions 1308-1316 - - - social justice and ideology 1808-1809
- sovereignty in medical care 1204 - - equality of health 1846-1855
Greece - - - pure inequalities 1847-1848
- counter-advertising of smoking 1595 - - - socioeconomic inequalities 1849-1855
- health expenditure 14, 850 - financing typologies 1819-1822
Gross Domestic Product and medical -- health care utilization 1833-1846
expenditure 166-167 -- - inequality, measuring 1841-1846
gross investments in human capital model 350, -- -- direct standardization-based index
358 1842-1844
- marginal cost of 355, 391 - -- - indirect standardization-based index
- production function 378 1844-1846
gross margins in pharmaceutical industry 1304 - - - inequality tests in 1835-1841
Group Health Cooperative (GHC), Puget Sound - - horizontal equity 1828-1833
432-433, 922, 1709 - - vertical equity 1822-1828
group practice in general practice 1190-1191 - -- Kakwani's progressivity index 1822-1824
grouped data regression in health econometrics - expenditure on long-term care 984
276 - as investment 1058, 1062-1063
- moral hazard in see moral hazard in health care
Harmonization of the Methodology for Health Care Financing Administration (HFCA)
Economic Evaluation of Health 142-143, 150, 174
Technology (HARMET) 183 health care markets
Harvard study of medical malpractice 1351, - antitrust and competition in see antitrust and
1354 competition in health care markets
Health and Retirement Survey (HRS) 680 - industrial organization of 1093-1139
- and prevention 1692, 1698 -- agency and physician-patient relationship
health capital model of health demand 394-404 1098-1101
health care - - asymmetric information in 1095
- consumer incentives see consumer incentives - - costly search in 1095-1096
in health care - - hospital PPS and regulated physicians
- costs and medical malpractice 1343 1117-1124
- as economic commodity 67-87 -- - dynamic cost-shifting 1123-1124
- - derived demand for 68-70 - - - quality choice under 1118-1121
-- externalities 70-73 - - - utilization review 1121-1123
- informational asymmetry 73-79 - - independent physicians and hospital cost
- - supplier-induced demand 78-79 reimbursement 1111-1117
- uncertainty 79-87 - - - medical arms race 1114-1117
Subject Index 1-59
- - - price competition among 1112-1114 -- second-generation studies 29-44

- - - quality competition among 1114-1117 -- panel data analyses 29-40
- - managed care and contracted physicians - -- error-correction models (Gerdtham)
1125-1134 30-31
--- measuring quality 1133-1134 - - - -heterogeneous models (Roberts) 39-40
- - -multinomial choice models 1131-1133 - - - -Hitiris and Possnett on 31
- - -payer-driven competition 1126-1131 - --- institutional variables (Gerdtham) 32-38
- - marketability of risk 1096 - - - - levels and growth rates (Barros) 38-39
-- monopolistic competition in 1102-1110 - - - - methods 29-30
- - - demand under 1104-1107 - - -- risk-risk analysis (Viscusi) 31-32
- - - equilibrium price 1107-1110 - - - unit root and cointegration analyses 40-44
-- -information 1107-1110 - - - - empirical results 43-44
--- quality 1107-1110 - - - -methods 40-43
health care utilization see utilization of services - limited dependent variables 285-299
and care - - hurdle models 297-299
health conditional demand function in human - - selectivity models 292-297
capital model 386 - - - identification by covariance restrictions
health econometrics 265-344 296-297
- count data regressions 315-324 - - - Manski bounds 292-293
-- basic model 316-318 - -- semiparametric estimators 293-296
- - count data 315-316 - - two-part, selectivity, and hurdle models
- - overdispersion and excess zeros 318-323 285-289
- - unobservable heterogeneity and simultaneity - -- empirical evidence 289
biases 323-324 - - - Monte Carlo evidence 288-289
- duration analysis 325-331 - - - taxonomy 285-286
-- competing risks and multiple spells 330-331 -- - two-part versus selectivity models 286-288
- methods 326-330 - - two-part models and retransformation
- - - parametric models 327-328 289-292
-- - semiparametric models 326-327 - longitudinal and hierarchical data 305-314
--- unobservable heterogeneity 328-330 - - multilevel models 305-308
- - survival and duration data 325-326 - panel data, individual effects in 309-314
- on ESP program impact 1509 - - - conditional logit estimator 310-311
- and full price effects in moral hazard 449 - - - linear models 309-310
- identification and estimation in 268-274 - - - pantob estimator 314
- - estimation strategies 269-274 - - - parameterizing 311-314
- - - evaluation 268-269 - - random versus fixed effects 308
- - - and model specification 270-272 - qualitative dependent variables 275-284
- - - nonparametric and semi parametric - - binary responses 275-276
estimators 272-274 -- bivariate models 283-284
-- -treatment effects 269-270 - - multinomial and ordered responses 276-283
- on international expenditure comparisons - - - and grouped data regression 276-278
21-44 - - - multinomial logit 278-280
-- first-generation studies 21-29 - - - multinomial probit model 282-283
-- - cross-section bivariate regressions 21-25 - - - nested multinomial logit 280-282
---- conversion factor instability 24-25 - smoking, advertising and promotion of
- - - - health care as luxury good 22-23 1585-1591
- - - - Newhouse on 21-22 - stochastic frontiers 331-335
--- cross-section multivariate regressions - - cost function studies 331-332
25-29 - - frontier models 332-335
- - - - Gerdtham on 26-29 - unobservable heterogeneity 299-305
- - - - Leu on 25-26 - - linear models 299-301
1-60 Subject Index
- - - instrumental variables 299-300 - preference assessment 215-216

- - - MIMIC model 300-301 -- preference heterogeneity 216
- - nonlinear models 301-305 - QALYs 212-214
- - - applications 302-303 -- - and CE analysis 217
- - - framework 301-302 -- survival and probability of health states
--- simulation estimators 305 214-215
- -- switching regressions 303-304 - perspective and cost measurement 204-211
health economic models of prevention - - distributional considerations 211
1680-1685 - - and fixed costs 207-211
- contribution of 1712-1714 --societal 205-207
- human capital models 1680-1683 - recommendations 218-219
- insurance models 1683-1684 health maintenance organizations (HMOs)
- supply of 1684-1685 -in health care markets 1125-1126, 1131
health expenditure, international 11-53 - in health insurance markets 591-593, 623
- empirical results, summary 45-48 - in health plan markets 783, 829
- - explanatory variables, effects 45-47 - and hospital regulation 1514-1515
- - unit root and cointegration analysis 47-48 in long-term care 971
- first-generation studies 21-29 - in managed care 709, 710, 713-714, 718-719
-- cross-section bivariate regressions 21-25 - in mental health 922
- - - conversion factor instability 24-25 - in pharmaceutical industry 1325-1326
- - - health care as luxury good 22-23 - prevention, investment in 1690, 1707-1709
- - - Newhouse on 21-22 - in private health insurance 650
- - cross-section multivariate regressions 25-29 health plan markets, risk adjustment in 755-845
- - - Gerdtham on 26-29 - efficiency and fairness 758-760
- - - Leu on 25-26 - models 779-817
- methodology 19-20 - - alternative models 794-807
- in pharmaceutical industry 1299 -- - demographic 794-797
- reasons for 16-19 - - - diagnosis-based 798-802
- second-generation studies 29-44 - - - mortality 804-806
- - panel data analyses 29-40 - - - other information 806-807
--- error-correction models (Gerdtham) 30-31 - - - prescription drugs, information from
- - heterogeneous models (Roberts) 39-40 802-803
- - - Hitiris and Possnett on 31 - - - prior year expenditures 797-798
- - - institutional variables (Gerdtham) 32-38 - - - self-reported information 803-804
--- levels and growth rates (Barros) 38-39 - - choosing, criteria for 780-785
- - - methods 29-30 - - fairness 783-784
- - - risk-risk analysis (Viscusi) 31-32 - -- feasibility 784-785
- - unit root and cointegration analyses 40-44 - - - incentives 780-783
Health Insurance Experiment 419, 435-438, 444 - designing or implementing 785-794
health insurance see insurance - - - functional form 787-789
health interventions, cost-effectiveness of -- individual versus contract level adjustment
181-221 785-786
decision making in 187-198 - -- information, use of 786-787
-- average ratios 192 -- maximum variance 791-793
-- and health policy 197-198 --- partial eligibility 789
- - incremental ratios for multiple alternatives -- - success of 793-794
193-196 --- variance, determinants of 790-791
--optimal 198-204 - - ongoing development 813-817
- - sensitivity analysis 196-197 - -- carveouts 815
- - time horizons 191-192 - - - conventional versus optimal adjustment
- measuring outcomes 211-217 815-817
Subject Index I-61
-- -subpopulations 814 - - quality and effort 859-860

- - - timing information 814 - - uncertainty 861-862
- - predictive power 807-813 - payment mechanisms 866-867, 870-872
- - - comparisons other than variance 810-813 - quality and costs 853-857
- - - variance measures 807-810 - - framework for 853-855
- payment flows 764-767 - - prospective payment 862-866
- - different modalities of 766-767 - - quality and effort 855-857
-- premium subsidies 765-766 - reputations 878-883
- - risk-adjusted premium subsidies 764 - - for characteristics 879
-- solidarity contributions 765 - - for past behaviour 880-883
- policy relevance 762-763 health-state valuations and quality of life 1740,
- practice of 828-836 1742, 1744
- - international comparisons 828-834 health status 442
- - problems 834-836 - in long-term care 975
-regulation 769-771 - and Medicaid 432
-risk sharing 763, 817-828 health years equivalent (HYEs) 103-104
- - empirical results 823-826 Healthcare Employers Data Information Set
--forms 819-823 (HEDIS) 1133
- - - all members 819-820 healthy time in human capital model 368, 381
- - - condition-specific 822-823 healthy-years equivalent (HYE) 1729, 1731
- - - high-risks 820-822 heart disease and alcohol consumption
-- optimal forms 826-827 1651-1652
- - proportional or prior costs 827-828 heavy drinking, alcohol 1642-1643
- selection 771-774 - external costs 1660-1661, 1663
- - adverse 771-773 Heckman-Singer approach in health
-- cream skimming 773-774 econometrics 331
- - effects of 774-776 hedonic prices, and waiting lists 1225
- - preventing 776-779 herd models of information diffusion 237
- spending on Herfindahl-Hirschmann Index (HHI) 1421, 1424
- - by chronic conditions 811 heterogeneity
- - by gender and age 795-796 - in not-for-profit hospitals 1170
- sponsor, role of 761-762 - unobservable variables in health econometrics
- subsidy formula 767-769 299-305
- supply and demand prices 760-761 -- bias in 299
health possibility frontier 1867 - - and duration analysis 328-330
health production 299 - - linear models 299-301
health production function of alcohol - - - MIMIC model 300-301
consumption 1638 - - nonlinear models 301-305
health services, government purchasing of - - - applications 302-303
847-890 - - - framework 301-302
- asymmetric information 867-874 --- simulation estimators 305
- - competitive bidding 872-873 - - - switching regressions 303-304
- - mechanism design approach 868-870 heteroscedasticity in health econometrics 291
-- payment mechanisms 870-872 hierarchical data in health econometrics
- commitment 874-878 305-314
- - investment and hold-up 875-877 hip fracture, quality of care 1162
- - ratchet effect 877-878 hold-up in government purchase of health
-- supply assurance 874-875 services 875-877
- demand and fixed price contracts 857-862 home care, long-term 960
- - efficiency 858-859 home visiting in public health programs in child
-- efficient treatment numbers 860-861 health 1081-1082
1-62 Subject Ide.
homelessness and mental health 900 - - - hospital investment 1519-1523

horizontal equity in health care delivery 1819, - - - - capital stocks 1519-1521
1828-1833 - - - - specialized services 1521-1523
Horizontal Merger Guidelines: in health care - - - impact on costs 1507-1516
markets 1426-1427, 1446-1448 -- - - intra-program variations 1515-1516
Hospital Cost Containment Act (USA, 1977) - - -- research on 1507-1515
1498 - - - mandatory rate-setting programs 1503,
hospital PPS in health care markets 1117-1124 1509-1510, 1524-1526
- dynamic cost-shifting 1123-1124 - - - modelling 1506-1507
- quality choice under 1118-1121 - - - reserve margins 1523-1524
- utilization review 1121-1123 - - - utilization 1516-1519
hospitalization costs 1302 - - federal controls 1497-1498
hospitals - - Medicare Prospective Payment System
-cases 1422 1498-1499
- cost studies in health econometrics 332-333 - -- research on 1527-1530
- for-profit 1144, 1146 - - rate regulation 1495-1497
-mergers 1418-1456 - - review programs 1494
- - conversions and ownership status 1436-1445 - scope and nature of 1493-1503
- - - model of 1439-1445
- - - economic trends 1500
--empirical evidence 1445-1455
- - strategic and incentives 1499-1503
- - - economies of scale and efficiency
- quality audits of 1100
1451-1453
- services in Medical Producer Price Index
- - - entry studies 1455-1456
148-150
- - - event studies 1453-1455
hours worked
- -- structure-conduct-performance
- and health insurance 687-689
- - - - non-price 1450-1451
- in long-term care 974
---- price 1448-1450
- and wages 695-696
-- enforcement 1418-1426
household production function in human capital
- - - framework for analysis 1419-1420
model 350
- - -outcomes 1420-1426
housing for long-term care 981
- - - process 1418-1419
- - market definition 1426-1436
human capital model
- - - background 1426-1429
- of health demand 347-408
- - basic model 352-367
- - - and managed care 1435-1436
- - - proposed method 1429-1435 - - - assumptions 352-354
- - of not-for-profit hospitals 1159 --- "bang-bang" equilibrium 363-366
- not-for-profit see not-for-profit hospitals - - - equilibrium conditions 354-356
- price and investment regulation, USA - - - optimal length of life 356-363
1489-1535 - - - special cases 366-367
- - certificate of need controls 1491, - empirical testing 377-383
1493-1495, 1498. 1501-1502 -- - structure and reduced form 378-381
- - - impact assessment of 1508-1511, 1517, - - extensions to 383-395
1521-1522, 1526-1527 - - - with cross-sectional data 383-388
-- model of 1504 - - - with longitudinal data 388-392
- - demand for 1491-1493 - - - theoretical 392-395
-- -models 1503-1504 - - health and schooling 394-404
- - effects of 1504-1530 - - pure consumption model 374-377
- - - CON programs 1526-1527 -- pure investment model 367-374
- - - dependent variables 1504-1506 - - - depreciation rate effects 369-371
- - - ESP program 1527 - - - market and nonmarket efficiency 371-374
- - - excess capacity 1523-1524 - of prevention 1680-1683
Subject Index 1-63
human capital parameter in production function - and disability 1031-1032

for health 376 incentives
Human Immunodeficiency Virus (HIV) 814, - direct, to reduce waiting 1234
1763, 1768, 1772 - for disabled, measures for 1020
- assortative matching 1773-1776 - in health care markets 1118-1119
-information on 1785, 1787, 1790-1791 - in health plan markets 780-783
- new drug development 1315 - in pharmaceutical industry 1327
- prevention of 1695, 1700, 1707, 1712 - referrals, in general practice 1187
- and smoking 1576 - in regulation of hospital prices and investment
- see also Acquired Immune Deficiency 1499-1503
Syndrome (AIDS) income
Hungary 14 - and dental services 1268-1270
hurdle model in health econometrics 286 - and health insurance 439, 451
- and Medicaid 432
iatrogenic injuries 1343, 1352, 1387 income effect
- no-fault programs 1378-1381, 1391 - in moral hazard 417, 422
Iceland - of people spending on medical care 577
- dental services, inputs and outcomes 1256 income elasticity of medical care 383
- health expenditure per capita 14 incomplete information
identification in health econometrics 268-274 - and best practice adoption 230-237
- estimation strategies 269-274 - - general model 230-236
- evaluation 268-269
- - - application 234
- and model specification 270-272 - - - insurance and costs of search 235-236
- nonparametric and semi-parametric estimators
- - - price variability 236
272-274
- -- search with unknown quality 234-235
- treatment effects 269-270
- - herd and cascade models 237
illness and stock of health in human capital
model 393
inconvenience costs of illness in human capital
immunity and infectious diseases 1776-1777
model 387
immunization, and child health 1059-1060,
indemnity
1076
- benefits of long-term care 980
imperfect information
- in insurance reimbursement 541-542
-in health care markets, differences 1413-1417
- -adverse selection 1414-1415 - policy 575, 586
- - agency problems 1415-1417 Independent Practice Associations (IPAs) 593,
-- moral hazard 1413-1414 922
- in liability for medical malpractice 1344, - in managed care 709, 713-714, 719
1347, 1349, 1382 India 15
- - custom-based standard of care, bias in Individuals with Disabilities Act (USA) 1035
1347-1349 inducement in general practice 1185
- - tort liability theory 1347-1350, 1399 inequities in child health 1075
- uncertain standards 1349-1350 infant health 1068
- and market failure in child health 1063-1065 infectious diseases, epidemiology of 1761-1799
- over-consumption of drugs 1312 - and condom use 1792-1793
imperfectly rational models of cigarette demand - predictions about disease occurrence
1556-1557 1767-1777
implicit subsidies - - assortative matching 1773-1776
- in health plan markets 763 - epidemics
- in not-for-profit hospitals 1148 - -- behaviour of 1767-1768
in-kind benefits ---dynamics of 1769-1773
- in child health 1077 -- immunity, effect of 1776-1777
1-64 Subject Index
- - implications for disease eradication - see also asymmetric information, imperfect

1768-1769 information; incomplete information
- prevalence elastic behaviour, evidence initial assets in human capital model 393
1789-1795 injuries, liability for medical malpractice in
-public health interventions 1777-1787 1351-1359
-- eradication, welfare effects 1784-1785 - adverse events and negligent injuries
- information, allocation of 1785-1787 1351-1354
- - mandatory vaccination 1780-1781 - causes of 1353-1354
- - Pareto-optimal timing of 1781-1784 - disposition of 1357-1359
-- subsidies 1777-1780 - malpractice claims
-welfare loss 1787-1789 - - and negligent injuries 1354
informal long-term care 960 --trends in 1355-1357
- supply of 972-975 injury prevention program., USA 1017
informal models of long-term care 972-975 innovation in pharmaceutical industry
information 1316-1319
- campaigns in public health programs in child - importance of 1317-1318
health 1078-1080 - and Uruguay round agreement 1318-1319
- diffusion and best practice adoption 223-264 inpatient waiting lists 1205
- - market equilibrium 228-237 input factor productivity in dental services 1275
- - - general model of 230-236 instantaneous adjustment in health models 389
Instrumental Activities of Daily Living (IADLs)
- - - - application 234
804
- - - - insurance and costs of search 235-236
instrumental variables
- - - - price variability 236
- in health econometrics 270, 299-300
- -- - search with unknown quality 234-235
- in human capital model 399-400
- - - incomplete information models 230-237
insurance 563-643
- - - - herd and cascade models 237
- benefits in reimbursement 541-542
- - - search and 228-230
- and child health in developed countries
-- production and dissemination of information
1068-1071
257-261
- - coverage 1075-1076
- - - costs of 258-261
- contract length 626-629
-- - property rights 257-258 - coverage
- production function, disagreement 237-256 - - sources 570
- - - healer's dilemma 237-239 - - universal, in child health 1060, 1069
-- - medical practice, variations 241 - demand- and supply-side controls 604-606
- - - physician-specific variations 248-251 -for dental services 1255, 1271-1272
- - - procedures and admission rates, - in developed countries 568-571
correlations 243 - and disability, USA 1019-1020
- - - regional variations 239-248 - health care as economic commodity 80-81
- - - - and individual variations 251-252 - and health outcomes 629-631
- - welfare losses 252-256 - in human capital model 372, 387
- -- from care variations 254 - implications 631-634
- and games of persuasion in health care - and information diffusion 225-226
markets 1416 - - incomplete search, model of 230-236
- market failure in new drug development 1315 - - and market equilibrium 228-230
- in medical malpractice 1346 - - with unknown product quality 234-235
- new, cost of waiting for 1221 - insurers and providers, relationships 590-604
- in not-for-profit hospitals 1148 -- equilibrium treatment decisions 594-596
on prevention, providing 1710-1712 - supply-side payment and medical treatment
- and public intervention on infectious diseases 596-604
1785-1787 - and labor market see labor market
Subject Index I-65
- markets for 606-625 intertemporal consequences of alcohol

--biased enrollment 616-623 consumption 1634
- - competition and selection, tradeoff 624 intertemporal utility function in human capital
- - continuous risk groups 614-616 model 352
-- cross-subsidy 609 Investigational New Drug (IND) 1309
- - equilibria with multiple individuals 612-613 investment
-- equilibrium with adverse selection 608-609, - in government purchase of health services
634-637 875-877
- - plan manipulation 609-612 - health care as 1058, 1062-1063
- - - importance of 623-624 - in human capital model 398-399
-- risk adjustment 624-625 - in not-for-profit hospitals 1163-1165
- and mental health 895-896, 925-936 - regulation of in hospitals 1519-1523
- - carveouts 897, 905-907, 925-926, 935-936 - see also gross investments; net investment;
- - evidence of 926-927 pure investment model
-- and managed care 928-931 Ireland
- - policy responses to 927-928, 931-936 - dental services, inputs and outcomes 1256
- - risk adjustment 932-935 - equity in health care delivery
-models of prevention 1683-1684 -- financing 1821, 1825, 1834
- moral hazard in 83, 576-588 - - private health insurance in 1820
- - taxes to finance 1824
- - coinsurance 584-586
-- utilization, indices 1844
- - in health care 411, 415, 421
- health expenditure 14, 850
- - plans, generosity 420-423
- health plans in 833
-- optimal 586-588
- long-term care for elderly 963
-- price elasticity of medical care demand
- payment systems in general practice 1188
580-584
- person-specific pricing 626-629
Israel 830
- and pharmaceutical industry 1301
Italy
- and physician agency 464, 466-467
- equity in health care delivery
- - demand and supply of services 471, 473,
-- financing 1821, 1825, 1834
475,478, 482, 484, 486, 492, 494,
- - private health insurance in 1820, 1826
498-490 - -utilization, indices 1844
- - and physician-induced demand 504, 506, - health expenditure 14, 850
512-515, 518-519, 522 - international drug sales 1313
- - and uncertainty 499 - long-term care for elderly 963
-premium uncertainty 626-629 - outpatient drug reimbursement 1329-1330
-prevention, models of 1683-1684 - pharmaceutical industry patents in 1319
- and principal-agent problems 588-590 - population projections of elderly 983
- principles of 571-576 Iverson's approach to waiting lists 1215-1218
-- with fixed spending 572-576
- reimbursement see reimbursement in insurance Japan
- status, and prevention 1691 - dental services, inputs and outcomes 1256
- tax-financed 82 - direct provision of services in child health
- in USA 569-571 programs 1077
internal markets - health expenditure 14, 850
- in health sector 77 - international drug sales 1313
- and waiting lists 1242 - long-term care for elderly 963
international health expenditure see health - outpatient drug reimbursement 1330
expenditure, international - population projections of elderly 983
intersectoral effects of waiting lists 1230 jawboning in drugs market 1329
intertemporal comparisons of alcohol job lock and private health insurance 654-657,
consumption 1640 665, 670, 671
1-66 Subject Index
job mobility see mobility latent variables in health econometrics 270

joint production function in human capital Latin America: health expenditure 15
model 382 lead poisoning 1066
joint production in human capital model 353 Learned Hand standard in medical malpractice
Justice, Department of (USA) 1419 1346
length of life in human capital model 356, 362
Kaiser Permanente 433, 592, 718 length of stay and health service quality 853
Kefauver-Harris Act (USA, 1962) 1321 less developed countries see developing
kernel density estimator in health econometrics countries
272 liability for malpractice see under malpractice
knowledge capital 352, 372 libertarianism and equity in health 1890-1891
Korea 13-14 life expectancy and long-term care 983
lifestyles in child health programs 1078
labor force nonparticipation among disabled Lindsay model of waiting lists 1214-1215
1025 linear cost sharing 872
labor force participation linear programming techniques 1157
- and disability 1006, 1008, 1025 LISREL model in health econometrics 300
- in labor markets 674-689 litigation costs, reducing in medical malpractice
labor market, health insurance in 645-706 1374-1377, 1380
- costs, rising 696-699 logit model in health econometrics 275
- equilibrium theory 654-664 long-term care 955-994
close substitutes 977
- costs 658-663
- demand for 975-982
- employer-provided insurance
- - empirical evidence on 975-978
- - - characteristics of 653
- - and private insurance 978-981
- - - and mobility 654-658
-- spend-down on savings 981-982
- - mandates 663-664
- demographics of 983-987
- health insurance coverage 648-650
- - exacerbation of problems 984-987
- sources 649
--trends 983-984
- and job-job mobility 664-674
- for elderly 963
- - research on 668-669
- supply of 964-975
-- selection 665-666
- - access models 966-968
-- self-employment decisions 672
- -bed supply models 964-966
- - variation in value of insurance 666-672 - informal care models 972-975
- - and welfare 673-674 - - quality of care models 968-972
- and participation 674-689 - taxonomy of 958-964
- - and hours worked 687-689 - - international comparisons 962-964
- - prime-age, research on 688 -- nursing home 958-959
-- and public assistance 681-687 - - other forms 959-962
- - and retirement 674-681 long-term insurance 626-627
- - - health risks, by age 675 longitudinal data
- price elasticity of demand 696-697 - in health econometrics 269
- private health insurance 650 - in human capital model 388-392
- and wages 690-695 lung cancer, risks 1698
- - and hours worked 695-696 Luxembourg
- - research on 691-692 - dental services, inputs and outcomes 1256
- - time series patterns 689-690 - health expenditure per capita 14
- workplace, role of 651-654
labor slack in not-for-profit hospitals 1153, 1167 malpractice
labor supply effects of benefits in disability insurance 1347, 1359-1364, 1400-1401
1023-1028 --contracts 1360-1362
lagged health variables 392 - crises in 1343, 1355. 1359, 1361
Subject Index I-67
- experience rated 1350, 1359-1360 - -- - contingent fees, limits 1375-1376

-- premium levels 1343, 1362-1364 - - - quality and standards of care 1376-1377
-liability for 1339-1404 - -- awards, structure of 1371-1374
- - effects of 1364-1369 - - -- litigation costs, reducing 1374-1376
- - - defensive medicine 1368-1369 - - in UK 1386-1389
- - - evidence 1366-1368 - in managed care 744
- - - formulating analysis 1364-1366 managed behavioural health care (MBHC) 922,
- - evaluation of system 1369-1371 923
- - - costs 1369-1370 managed care 707-753
- - - deterrence benefits 1370-1371 - agency in 463-464, 467, 469, 474, 486-488,
-- injuries and claims, evidence on 1351-1359 490, 492-495
- - - adverse events and negligent injuries - - networks 496
1351-1354 - - objectives 527
- - - - causes of 1353-1354 - - and physician-induced demand 505, 518-520
- - - disposition of 1357-1359 - antitrust in health care markets 1435-1436
- - - malpractice claims - in child health 1061
- - - - and negligent injuries 1354 - competition among plans 743-744
- - - - trends in 1355-1357 - consumer cost-sharing 712-713
-- insurance 1350, 1359-1364, 1400-1401 - costs, studies of 741-742
- - - contracts 1360-1362 - covered benefits 712
- - - premium levels 1343, 1362-1364 - defined 711-717
- - under managed care 1382-1386 equilibrium treatment decisions in 594-596
- - - denial of coverage, claims 1383-1384 - growth of, economic issues on 743-744
- - - ERISA pre-emption 1386 - - malpractice 744
-- - liability for negligent care 1385 - - risk adjustment 744
- - in New Zealand 1392-1395 - in health care markets 1125-1134
--non-USA experience 1386-1395 - measuring quality 1133-1134
- - optimal rules 1396-1399, 1398-1399 -- multinomial choice models 1131-1133
- - - first party 1397 - - payer-driven competition 1126-1131
- - -negligence 1398-1399 -- effect of 1128-1131
-- - strict third party 1397-1398 - and health insurance 438, 449
-- radical alternatives 1378-1382 - history of 717-721
- - - no-fault programs 1378-1381 - by HMOs 593
- - -private contracting 1381-1382 - liability for medical malpractice under 744,
- -in Sweden 1389-1392 1348, 1382-1386
- - tort liability theory 1345-1351 - - denial of coverage, claims 1383-1384
- - - custom-based standard of care, bias in - - ERISA pre-emption 1386
1347-1349 - - liability for negligent care 1385
- -- with imperfect information 1347-1350, - - and tort reform 1369
1399 - and market failure 721-727
- - - other control mechanisms 1350-1351 - - asymmetric information 721-723
- - - with perfect information 1345-1347 - - industry competitiveness 726-727
- - - uncertain standards 1349-1350 - - information 725-726
- - tort reform - - moral hazard 723-725
--- awards, structure of - in mental health 928-936
- - - - collateral source effect 1374 - - in managed behavioural health care
- - - - limits 1371-1372 organizations (MBHC) 912-918
- - - -periodic payments 1374 - - and capitation 921-925
- -- enterprise liability 1377-1378 - -hospital payment systems 920-921
- - - litigation costs, reducing - - and moral hazard 918-925
- - - - alternative dispute resolution 1376 - in not-for-profit hospitals 1159
1-68 Subject Index
- provider selection and organization 713-714 - plan manipulation 609-612

- providers, paying 714-716 - - importance of 623-624
- research on 727-743 - risk adjustment 624-625
-- cost growth 740-743 Marrakech Treaty 1319
- quality 739-740 marriage and health insurance 687-688
-- selection 728-731 maternally reported measures of child health
- - studies 729-730 1063
-- spillover effects 740 maximum allowable cost in pharmaceutical
-- utilization 731-739 industry 1324, 1329
- - - studies of 733-738 maximum likelihood (ML) estimators in health
- service utilization 716-717 econometrics 271
mandating in health plan markets 777 maximum treatment curve in waiting lists 1217
manipulated equilibrium in market competition McCarran-Ferguson Act (USA, 1948) 1408
610 McMaster Health Utility Index (HUI) of
Manski bounds in health econometrics 292 well-being 1745, 1748
marginal benefits in human capital model 355 measles 1066
marginal cost of gross investments in human measured determinants in human capital models
capital model 355, 391 391
marginal efficiency of capital schedule in human Medicaid 125, 431-432
capital model 367-369, 370, 374 - in child health 1059-1060, 1067, 1069-1071
marginal product of health capital 355 - generic substitution in drugs 1324-1325, 1327
marginal utility in human capital model 374, 375 - health insurance
market areas of not-for-profit hospitals 1159 - - coverage 648
market data on cost of administered waiting - - sources 570
1222-1225 - health spending, by gender and age 795
market efficiency in human capital model - and long-term care 962, 977
371-374 - - as payer of last resort 979
market failure and managed care 718, 720
- in child health programs 1063-1067 - and physician agency 482
--extemalities 1065-1067 - and public assistance 681, 685-686
- imperfect information 1063-1065 medical-appropriateness of care 440
- in managed care 721-727 medical arms race (MAR)
-- asymmetric information 721-723 - in health care markets 1114-1115, 1117, 1129
- - industry competitiveness 726-727 - and hospital mergers 1450-1453
- - information 725-726 medical care: prices and output 119-180
- - moral hazard 723-725 - construction of indexes at BLS 135-162
- in new drug development 1315 - - CPI (Consumer Price Index), summary of
- and prevention 1694-1703 136-138
- - consumer information, lack of 1696-1701 - - PPIs (Producer Price Indices)
- externalities 1694-1696 -- - hospital services 148-150
- - prevention as public good 1701-1703 -- medical products 150-153
market wage in long-term care 974 - for medical-related goods and services
marketability of risk in health care markets 1096 141-153
markets for health insurance 606-625 - - - output measurement 142-153
- biased enrollment 616-623 --- physicians' services 144-148
- competition and selection, tradeoff 624 --- summary of 138-141
- continuous risk groups 614-616 - - products in CPI and MCPI 153-162
- cross-subsidy 609 - - item structure of MCPI 155-158
- equilibria with multiple individuals 612-613 - -- weighing issues 158-162
equilibrium with adverse selection 608-609, - episode treatment costs, index based on
634-637 163-166
Subject Index I-69
- market for CPIs and PPIs 124-135 - as monopsonist 552

- - price index, forming 133-135 - and physician agency 482, 485, 513-515
--pricing services 129-133 - price index and capitation payment 807
-- US medical care marketplace 124-129 - and retirement 679
- price indexes Medicare Prospective Payments System
-- in national income and product accounts - equilibrium treatment decisions under
166-175 596-597
- - - issues outside US 171-174 - fixed price reimbursement 853
- - - medical expenditures in 166-168 - government purchase of health services 862
-- - national health accounts 174-175 - in health care markets 1116
-- US context 168-171 - - regulated physicians in 1117-1124
- - related research 162-163 - - - dynamic cost-shifting 1123-1124
medical care and alcohol consumption - - - quality choice under 1118-1121
1650-1651 - - - utilization review 1121-1123
medical care consumer price index (MCPI) 120, - in health insurance markets 623
122 - and hospital price and investment regulation
- major product groups in 160 1498-1499, 1513
-price inflation in 154 - - research on 1527-1530
- services in 153-162 - on mental health 919, 920, 933
- - item structure of 155-158 - and not-for-profit hospitals 1159
-- weighing issues 158-162 mental ability in human capital model 379
- summary of 136-138 mental health 893-954
medical care expenditure price index 164-165 - adverse selection 897, 907, 925-936, 941
medical devices, manufacture of 1299 - - carveouts 897, 905-907, 925-926, 935-936
medical ethics - - evidence of 926-927
- and equity in health 1894, 1899 -- and managed care 928-931
- and physician agency 520-521 - -policy responses to 927-928, 931-936
medical knowledge in marketplace 127 - - risk adjustment 897, 932-935
medical practice, variations 241 - and health insurance 443
- welfare loss from 254 - institutional context
medical producer price index (MPPI) 120, 122 - managed behavioural health care 905-907
- construction of 135-162 - - payment for care 901-902
- - hospital services 148-150 - - persons with mental health 897-900
-- medical products 150-153 - - supply of services in 902-904
- - for medical-related goods and services - - who is treated 900-901
141-153 - insurance markets 925-936
-- output measurement 142-153 - - carveouts 935-936
-- physicians' services 144-148 - - evidence of 926-927
- sub-indices of 145-146 -- and managed care 928-936
- summary of 138-141 - -policy responses to 927-928, 931-936
medical staff in not-for-profit hospitals 1155 - - risk adjustment 932-935
MediCal Studies on health insurance 431-432 - moral hazard 897, 907-925, 926
medical treatment, waiting lists for see waiting --evidence of 908-912
lists - - managed care 918-925
Medicare 125, 144 - - - and capitation 921-925
- balanced billing by 547 - - - hospital payment systems 920-921
- coverage 648 --rationing 912-918
- expenditure in last year of life 74, 984-985 - private insurance markets 907-925
- health insurance, sources 570 - - and capitation 921-925
- and long-term care 962, 968, 971 - - hospital payment systems 920-921
- and managed care 718, 719-720 - - and rationing 912-918
1-70 Subject Index
-- supply-side incentives 918-925 - as economic commodity 83-87

- public system 936-943 - ex ante 415-420
- - externalities 941-943 - - background 415-418
- - fiscal federalism 938-941 - - empirical evidence 418-420
-- technology for 937-938 - on prevention 1686, 1688, 1691
- and substance abuse treatment system 936-943 - ex post see ex post moral hazard in health care
mergers - in health care markets 1096, 1413-1414
- of hospitals see under hospitals - in health insurance 576-588
- in pharmaceutical industry 1305 - - coinsurance 584-586
Mexico 13-14 -- optimal 586-588
micro data on waiting lists 1231 - - price elasticity of medical care demand
Middle East: health expenditure 15 580-584
misspecification in health econometrics 268 - and income support for disabled 1021
mixed investment-consumption health model and insurance reimbursement 542-545
392 - in long-term care 979
mobility and health insurance 664-674 - in managed care 723-725
- employer-provided 654-658 - in medical malpractice 1361
- selection 665-666 and mental health 897, 907-925
- self-employment decisions 672 -- evidence of 908-912, 926
- variation in value of insurance 666-672 - - managed care 918-925
- and welfare 673-674 - - - and capitation 921-925
Monitoring the Future (MTF) survey 1639 - - - hospital payment systems 920-921
monopolistic competition - rationing 912-918
- in industrial organization of health care - and prevention 1684, 1685-1694
markets 1102-1110 -- extent of 1690-1693
- - demand under 1104-1107 - health insurance status 1691
- equilibrium price 1107-1110 - - in public sector insurance systems
- - information 1107-1110 1693-1694
--quality 1107-1110 - - solutions to 1687-1690
- in pharmaceutical industry 1328 morbidity
- and physician agency 475-478, 481, 485, 490, - among children (USA) 1068
502 - from infectious diseases 1763
monopoly - and smoking 1576
- in health care markets 1415 mortality
- power of pharmaceutical industry 1301 alcohol-related 1646-1647, 1650, 1697
monopsony - among children (USA) 1068, 1071, 1076
- in health care markets 1456-1478 - -excess 1080-1081
- - bilateral market power 1466-1470 - from infectious diseases 1763
- - - identifying market power 1466-1467 - models in health plan markets 804-806
- - - relative bargaining power 1468-1470 - rates in human capital model 383
- -- welfare implications 1468 - and smoking 1575, 1576, 1697
- monopsony power 1457-1466 most favored customer incentives in
--- and antitrust enforcement 1457-1459 pharmaceutical industry 1327
- - - detecting 1461-1462 most favored nation clauses in hospital mergers
- - - studies of 1463-1466 1456-1457, 1475-1476
- - and welfare 1459-1461 motor-vehicle mortality and alcohol
- in insurance reimbursement 551-553 consumption 1646-1649
Monte Carlo simulation in health econometrics multilevel models in health econometrics
272 305-305
moral hazard in health care 409-459 multinational enterprises, and pharmaceutical
- delegation of authority to physician 414-415 industry 1304-1305
Subject Index I-71
multinomial logit model in health econometrics need in health care delivery 1818
278 needs-based theories of equity in health care
multinomial probit model in health econometrics 1887-1889
282 negative binomial in health econometrics 318
multiple indicators-multiple causes (MIMIC) negbin model in health econometrics 318
model negligence and liability for medical malpractice
- in health econometrics 270, 300-301 1398-1399
- of human capital 383-384, 389 - adverse events and negligent injuries
myopic models of cigarette demand 1557-1559 1351-1354
- causes of 1353-1354
naive time price in waiting lists 1214 -injuries 1351, 1354, 1358, 1387
narcotic substances 1306 -malpractice claims 1354
National Cancer Institute 1703 - under managed care 1385
National Childhood Vaccine Injury Act (USA, -responsibility for 1352
1986) 1320 negligence rule in medical malpractice 1343,
National Committee for Quality Assurance 1344, 1346, 1378
(NCQA) 1100 neonatal care 1068-1069
National Health Interview Survey (NHIS) 1001 neonatal mortality 1068
- on alcohol consumption 1643, 1655, 1697 nested multinomial logit model in health
- on cigarette smoking 1549, 1553, 1555, 1579, econometrics 278
1697 net investment in human capital model 365
- on infectious diseases 1792-1794 Netherlands
- on prevention 1692, 1708 - birth rates, teenage 1079
National Health Service (UK) 25 - dental services
- access to health care 1812 - - inputs and outcomes 1256
- consumers' search in 229 - - markets for 1255
- cost sharing in 866 - and disability 1005
- as health plan market 833 -- policy 1038, 1039
- hold-up in 876-877 - - transfer recipients 1003
- medical expenditure 171 - equity in health care delivery 1828
-payment contracts in 851, 852, 853, 878 --financing 1821, 1825, 1834
- prevention, incentives for 1693 - utilization, indices 1844, 1846
- reforms to 1098 - health expenditure 14, 850
- waiting lists in 1205-1207 - health spending by gender and age 796
-- time treatment 1234 - international drug sales 1313
National Household Survey on Drug Abuse 1633 - long-term care for elderly 963
National Institute on Alcohol Abuse and - not-for-profit hospitals in 1145
Alcoholism (NIAAA) 1710 - premium contributions 766
National Institutes of Health (NIH) 1307-1308 - risk adjustment and sharing in 830
National Long-Term Care Demonstration 962 Network Model HMOs 593
National Longitudinal Survey of Youth (NLSY) networks and physician agency 493, 496
- on AIDS 1766, 1790-1791 New Drug Approval (NDA) 1310, 1311, 1314,
- on alcohol consumption 1639, 1642, 1315
1655-1657 new drug chemical entities 1316
- in child health 1060, 1063 New York Prospective Hospital Reimbursement
- and health insurance 670, 672 Methodology (NYPHRM) 1516,
National Medical Care Expenditure Survey 1518-1519
(NMES) 667, 670, 972 New York study of medical malpractice 1351
National Science Foundation 1307 New Zealand
natural experiments on moral hazard 418, 429, - accident rehabilitation compensation and
431,439, 443 insurance scheme 1392-1395
1-72 Subject Index
- dental services, inputs and outcomes 1256 - - fiduciary relationships 1148-1i50

- health expenditure 14, 850 - - implicit subsidies 1151
- long-term care for elderly 963 --inertia 1151
- medical malpractice in 1344 - - ownership rights 1148
- - liability for 1392-1395 -- profitability 1151-1152
- - no-fault programs 1379-1380 - -public goods 1150
- prioritisation of waiting lists 1241 - - transaction costs 1148
no-envy principle and equity in health -legal distinctions 1146-1147
1892-1893 -models of 1153-1155
No-Net-Cost Tobacco Program Act (USA, 1982) --objectives 1153-1154
1602 - - as physicians' cooperative 1154-1155
nonmarket efficiency in human capital model -- statics analysis 1154
371-374, 382, 391 - ownership conversions 1166-1167
nonparametric estimators in health econometrics - performance of 1155-1166
270, 272-274 - - capital funds and investment 1163-1165
nonprice competition in not-for-profit hospitals --cost 1155-1157
1167 - - and for-profit performance 1165-1166
normative economics of health sector 55-118 - pricing patterns 1158-1160
- efficiency and normative framework 59-66 -- profitability 1158
- - neo-classical welfare framework 60-62 - - quality of care 1161-1163
- - welfare economics and extra-welfareism - technology, diffusion of 1161
62-66 - uncompensated care 1160-1161
- equity in health sector 87-96 nursing home care, long-term 958, 964, 967
-- allocation according to need 90-92 - supply of 965
- - equality of access 89-90 nursing homes and long-term care 958-959
Nutrition Labeling and Education Act (USA,
- - equality of health 92-94
1990) 1700
- - rival notions of 95-96
- health care as economic commodity 67-87
observational comparisons in health insurance
- - derived demand for 68-70
429, 433,443
- - externalities 70-73
obstetrical procedures 1073
- - informational asymmetry 73-79 Occupational Safety and Health Administration
- - - supplier-induced demand 78-79 (OSHA) 1013, 1017
- - uncertainty 79-87 Office of Statewide Health Planning and
- - - economies of scale 81-82 Development (OSHPD, California) 1116,
- -- insurance, welfare improving effects 80-81 1129
- - - moral hazard 83-87 oligopolistic rivalry in pharmaceutical industry
- - - risk selection 82-83 1301
- programs and interventions, evaluation 96-107 Omnibus Budget Reconciliation Act (USA)
- - equity and methods 104-107 1605
Norway opportunity cost
- dental services, inputs and outcomes 1256 - in human capital model 366
- health expenditure 14, 850 - of time and moral hazard 416, 444, 446-447
- long-term care for elderly 963 optimal inpatient treatment, and waiting lists
- payment systems in general practice 1188 1207
not-for-profit firms 1418 optimal insurance 85, 578, 586-588
not-for-profit hospitals 1141-1174 - estimates of 587
- competition, role of 1167-1168 optimal length of life 366
- dominance of 1148-1153 optimal reimbursement 542-545
- - best rationale 1153 optimal stock of health 362
- - cartel theory 1152 optimal waiting lists 1207, 1231
- explicit subsidies 1151 option demand and waiting lists 1233
Subject Index I-73
ordered probits in health econometrics 276 patents, pharmaceutical 1302, 1307, 1316-1319,
Organization for Economic Cooperation and 1328
Development (OECD) - importance of 1317-1318
- disability policy 1037-1040 -protection 1318
- health expenditure in 28 - and Uruguay round agreement 1318-1319
- medical expenditure 171 patients
organizational goals in not-for-profit hospitals - best interest 521-522
1147 - competition for 487-492
orthodontia, in dental services 1262 -and GP relationship 1178-1180
outcomes - - agency and 1098-1101
- best, in quality of life 1743-1747 - and information diffusion 226-227, 229-230
- - criteria for assessing 1743-1744 - - property rights of 257-260
- - evaluating methods 1745-1747 - - treatment patterns 244-245, 247-251, 253
--evaluating systems 1744-1745 Patient's Charter (UK) 1243
- and CPI 165-166 payer-driven competition in health care markets
- for dental services 1256, 1286-1288 1126-1131
- and health insurance markets 629-631 - effect of 1128-1131
- in hospital mergers 1420-1426 payments
- measuring 211-217 - denial of claims in medical malpractice 1389
- - preference assessment 215-216 - flows in health plan markets 764-767
- - preference heterogeneity 216 - - different modalities of 766-767
- - QALYs 212-214 - - premium subsidies 765-766
- - - and CE analysis 217 - risk-adjusted premium subsidies 764
- - survival and probability of health states - - solidarity contributions 765
214-215 - in government purchase of health services
outreach programs, in child health 1073 866-867, 870-872
over the counter drugs 1300, 1303-1304 - periodic, in malpractice awards 1374
overdispersion in health econometrics 318 - systems in general practice 1187-1190
overhead costs in medical malpractice 1369, - see also capitation payments; copayment;
1371, 1380, 1391, 1393 Medicare PPS
own-price short run elasticities for NHS private pediatric hospitalizations 1064, 1066
care 1230 Peer Review Organizations (PROs) 1121-1122
ownership rights in not-for-profit hospitals 1148 periodic open enrollment in health plan markets
-conversions 1166-1167 770
periodic payments in malpractice awards 1374
panel data in health econometrics 270-272 periodontal programmes in dental services 1290
- individual effects in 309-314 person-specific pricing in health insurance
- - conditional logit estimator 310-311 markets 626-629
- - linear models 309-310 person trade-off (PTO) of quality of life
- - pantob estimator 314 measurements 1733, 1903
- - parameterizing 311-314 pharmaceutical industry 1297-1336
Panel on Cost-Effectiveness in Health and -characteristics 1300-1303
Medicine 183, 207 - government regulation of drug introductions
Panel Study of Income Dynamics (PSID) 667 1308-1316
pantob estimator in health econometrics 314 -imports 1304
partial adjustment in health models 389 -'me-too' variants 1312
partial reimbursement 541-542 -patents and innovation 1316-1319
partially linear model in health econometrics 272 --importance of 1317-1318
participatory democracy and equity in health - - and Uruguay round agreement 1318-1319
1891 -pricing 1319-1328
partnerships in general practice 1190-1191 --branded drugs 1320
1-74 Subject Index
- branded versus generic competition - - unobservable physician characteristics

1322-1324 502-503
-- generic drug competition 1321-1322 physician determined demand in moral hazard
-- new institutions 1325-1328 411
- - stimulating generic substitution 1324-1325 physician induced demand 465, 467, 490,
- -tort liability risks and 1320-1321 503-519
- products in MPPI 150-153 - in child health 1064, 1065
-profits and price controls 1328-1331 - demand inducement theory 507-509
- research and development 1305-1308 - fees and inducement 513-516
-structure 1303-1305 physician-owned mutual companies 1363
pharmacoeconomic cost-effectiveness of health physician-patient relationship
interventions 183 - agency in health care markets 1098-1101
pharmacy benefit management firms (PBMs) - and delegation of authority 414
1325-1326 - in general practice 1178
Pharmacy Cost Groups (PCGs) 803 Physician Review Organizations (PROs) 1528
philanthropy, in not-for-profit hospitals 1164 physicians
physician agency 461-536 in industrial organization of health care
- behaviour with complete information 475-496 markets
- -administered demand and supply prices - -contracted, and managed care 1125-1134
481-486 --- measuring quality 1133-1134
-- competition for patients 487-492 - - - multinomial choice models 1131-1133
- - efficient production of services 494-495 -- payer-driven competition 1126-1131
- - monopolistically competitive firm 475-481 - independent, and hospital cost
- - network incentives in managed care 492-494 reimbursement 1111-1117
- - noncontractible quality 487-492 - - medical arms race 1114-1117
- supply-side cost sharing 487-492 - - - price competition among 1112-1114
- - third-party payer 481-486 - - quality competition among 1114-1117
- demand and supply for services 468-475 - - physician-patient relationship 1098-1101
- - competition among physicians 473-475 - regulated, and hospital PPS 1117-1124
- - entry conditions and monopoly profits - - - dynamic cost-shifting 1123-1124
469-473 - - quality choice under 1118-1121
- income and expenses 470 - - - utilization review 1121-1123
- - prices and quantities 469 - liability in medical malpractice 1364
- panel income data 516-519 - - response to 1366
-- defensive medicine 516-517 - rationing by in ex post moral hazard 449-451
- - self-referrals 517 - - background 449-450
- physician induced demand 503-519 - - empirical evidence 450-451
- - demand inducement theory 507-509 - services in MPPI 144-148
- - fees and inducement 513-516 - utility, in waiting lists 1218
- - income effects 509-513 physicians' cooperatives, not-for-profit hospitals
-- physician-to-population ratios 509-513 as 1154-1155
- physician objectives 520-526 Physicians' Health Study on alcohol
- - and medical ethics 520-521 consumption 1652
-- revenue targeting 525-526 planar thallium radionuclide imaging,
-- target incomes 522-526 cost-effectiveness of 193
- - - income effect 523-525 Poisson model in health econometrics 317
- - utility and patient's best interest 521-522 Poland 14
- uncertainty and asymmetric information policy
496-503 - on disability
- - irreducible uncertainty 497-498 - - OECD countries 1037-1040
- - unobservable physician actions 498-502 - - toward working-age, USA 1012-1020
Subject Index I-75
- accident and injury prevention program - restrictions 770

1017 - and uncertainty 626-629
- - - health insurance 1019-1020 prenatal care 1072
- - -rehabilitation 1018-1019 prescription drugs, information from in health
- - - retum-to-work programs 1018-1019 plan markets 802-803
- - - SSDI program 1013-1015, 1042-1044 prescriptions 1300
- - - SSI program 1015-1016, 1044-1045 Preventive Services Task Force (USA) 1685,
- - - vocational training 1018-1019 1696, 1711
- -- Workers' Compensation program 1017, prevention 1675-1720
1045-1047 - costs of and moral hazard 416
- in equity in health 1900 - and cure, costs of 1704-1706
- instruments in long-temi care 968 - in health economic models 1680-1685
- on medical malpractice 1344 - - human capital models 1680-1683
-on waiting lists 1231-1243 - - insurance models 1683-1684
- - efficient waiting 1239-1243 - supply of 1684-1685
-- -prioritisation, indices 1240-1242 - and market failure 1694-1703
- - - quasi (internal) markets 1242 - consumer information, lack of 1696-1701
- - - reducing uncertainty 1242-1243 --extemalities 1694-1696
-- options, taxonomy of 1232-1239 - -prevention as public good 1701-1703
- - - demand rationing 1233 - meaning and scope 1677-1679
- - - private provision 1238-1239 - and medical malpractice 1353
- - - subsidies 1235-1238 - and moral hazard 1685-1694
--- supply expansion 1234-1235 -- extent of 1690-1693
political coalition and waiting lists 1238 - health insurance status 1691
pooling in health insurance 610, 612 - - in public sector insurance systems
poor, health insurance for 430, 442 1693-1694
Portugal - - solutions to 1687-1690
- equity in health care delivery -policies to encourage 1706-1712
--financing 1821, 1825, 1834 - - access to clinical services 1707-1709
- private health insurance in 1820, 1826 - - in developing countries 1709-1710
-- taxes to finance 1827 - - information for 1710-1712
-health expenditure 14, 850 - -taxes and subsidies 1706-1707
positron emission tomography (PET), - research, and health economics 1712-1714
cost-effectiveness of 193-194 preventive effort in moral hazard 411, 415, 418,
poverty reduction, and disability 1028-1030 419
predictable losses in health plan markets 763 preventive services
Preferred Provider Organizations (PPOs) 592 - in child health 1071
-in health care markets 1125-1126, 1131 - in managed care 712
- in managed care 709, 710, 714 price controls in pharmaceutical industry 1329
- in mental health 923 price-cost margin in pharmaceutical industry
- in private health insurance 650 1301
pregnancy and child health programs 1079 price discrimination in pharmaceutical industry
Pregnancy Discrimination Act (1979) 712 1323
premiums, health insurance price elasticity of demand
- contributions 760-761 - for health care 412, 433-435, 437, 441
- in health plan markets - - and moral hazard 580-584
- - community-rated 765 - in health insurance 696-697
- - risk rate consumers 777 price indexes in medical care
-- subsidies 761, 765-766 -forming 133-135
- - - risk-adjusted 764 - in national income and product accounts
- and liability for malpractice 1362-1364 166-175
1-76 Subject Index
-- issues outside US 171-174 private contracting in medical malpractice

- medical expenditures in 166-168 1381-1382
- - national health accounts 174-175 private donations to not-for-profit hospitals
- US context 168-171 1147, 1164
- related research 162-163 private health insurance
price regulation and physician agency 464, 467, - and demand for long-term care 978-981
481 - in long-term care 957-958, 978-981
prices/pricing - markets for in mental health 907-925
- competition - - and capitation 921-925
- - among physicians 1112-1114 - hospital payment systems 920-921
- - in not-for-profit hospitals 1161 - - and rationing 912-918
- in dental services 1268-1270 - - supply-side incentives 918-925
- measures 1261-1262 - policies, features 650
- of health inputs 1058 private practice and waiting lists 1220
- in pharmaceutical industry 1319-1328 probability in human capital models 393
--branded drugs 1320 orobit model in health econometrics 275
- - branded versus generic competition process innovation in health care 453
1322-1324 producer sovereignty in medical care 1204
- generic drug competition 1321-1322 product differentiation
- - new institutions 1325-1328 - and antitrust in health care industry 1411
- stimulating generic substitution 1324-1325 - in pharmaceutical industry 1302-1303, 1328
product innovation in health care 452
- - tort liability risks and 1320-1321
product safety in public health programs in child
- of tobacco and smoking 1542
health 1080-1081
--impact on demand of 1546-1565
production
- - - addiction models 1556-1563
in human capital model 372
-- - -critiques of 1561-1563
- of information 225, 257-261
- - - - imperfectly rational models 1556-1557
- costs of 258-261
- - - -myopic models 1557-1559
-- property rights 257-258
-- - -rational models 1559-1561
production function
- - behavioural economic analyses of - in child health 1057
1563-1564 disagreement on best practice 237-256
- - - conventional studies of 1547-1556
- healer's dilemma 237-239
--- -aggregate data 1547-1549 - physician-specific variations 248-251
- - - individual data 1549-1556 - regional variations 239-248
--- econometric studies 1564-1565 - - and individual variations 251-252
- variability of 228-237 - in human capital model 367, 381
- incomplete information models 230-237 in not-for-profit hospitals 1154
- - - general model 230-236 production possibility frontier and equity in
- -- application 234 health 1867
- - -- insurance and costs of search 235-236 productive efficiency in insurance
- - - - price variability 236 reimbursement 553-554
- - - search with unknown quality 234-235 productivity
- - - Herd and cascade models 237 - alcohol, impact on 1653-1658
- - search and 228-230 - - direct effects 1654-1657
primary care beds, UK 1192 - - and human capital 1657-1658
primary prevention 1677 - in dental services 1275, 1280--1282
principal-agent problems in insurance 588-590 - loss in waiting lists 1223
prior year expenditures models in health plan profitability
markets 797-798 in not-for-profit hospitals 1151-1152
prioritisation, indices for in waiting lists 1240 - in pharmaceutical industry 1307, 1328
Subject Index 1-77
Program of All-inclusive Care for the Elderly -- taxation on 1574-1579

(PACE) 961-962 - - - increasing, health benefits of 1577-1579
property rights - - - social costs 1575-1577
- to drugs, devices and ideas 257-258 public hospital organizations 1147
- and equity in health 1890 public mental health system 936-943
- in not-for-profit hospitals 1148 - externalities 941-943
proportional hazards model in health - fiscal federalism 938-941
econometrics 326 - technology for 937-938
proprietary drugs 1300, 1303-1304 purchaser in quasi markets 1242
prospective payment in government purchasing purchasing power parity in health expenditure
of health services 862-866 21, 22-23, 24
- costs 863-865 pure consumption model of health demand
- quality 865-866 374-377, 394
provider in quasi markets 1242 Pure Food and Drug Act (USA, 1906) 1309
provider market power in insurance pure health inequalities 1847-1849
reimbursement 551-553 - empirical work on 1848-1849
provider-owned insurers 1363 - measurement of 1847-1848
pseudo-maximum likelihood (PML) estimators - socioeconomic 1849
in health econometrics 271 pure investment model of health demand
psychiatric hospitals 1163 367-374, 377, 385
public assistance and health insurance 681-687 - depreciation rate effects 369-371
- research on 683 - market and nonmarket efficiency 371-374
pure price effects and moral hazard in health
public goods
care 426-429, 441
- in health sector 71
- not-for-profit hospitals as 1150
qualitative dependent variables in health
public health interventions in infectious diseases
econometrics 275-284
1777-1787
- binary responses 275-276
- dynamic welfare effects 1784-1785
- bivariate models 283-284
- information, allocation of 1785-1787
- multinomial and ordered responses 276-283
- mandatory vaccination 1780-1781 - - and grouped data regression 276-278
Pareto-optimal timing of 1781-1784 - multinomial logit 278-280
- subsidies 1777-1780 - multinomial probit model 282-283
- welfare loss 1787-1789 - - nested multinomial logit 280-282
public health service and care quality-adjusted life years (QALYs)
- child health, programs in 1078-1083 - and cost-effectiveness of health interventions
- - community-based 1082 185, 189-191, 194
-- education and information campaigns - - and decision making 197, 200-201
1078-1080 - - measurement of 204, 206, 212, 217
- - home visiting 1081-1082 -- survival, estimating 214-216
- - incentives 1082 - costs of coronary heart disease 193, 195
--product safety 1080-1081 - and equity in health 1865, 1867, 1871, 1904
- and disability 1031-1032 - and general practice 1184
- on infectious diseases 1763-1764, 1767, 1785 - in measuring health outcomes 97, 101-102,
- means-tested, in child health 1072, 1076-1077 105-106, 212-214
- mental health - - and cost effectiveness analysis 217
- - externalities and 941-943 - and mental health 917
- - fiscal federalism and 938-941 - model in quality of life measurements 1726,
- and moral hazard 413 1729, 1749
- prevention, experts on 1677 - - additive separability 1743
- and smoking 1543 - - assumptions 1740-1743
--counter-advertising 1594 - - constant proportional trade-off 1741
1-78 Subject Index
- lifetime preferences, stability 1742-1743 quality of service in government purchase of

-- risk attitude over life years 1741-1742 health services 853-857
- - utility dependence 1740-1741 - and effort 855-857
- - zero rate of time preference 1742 in fixed price contracts 859-860
- and prevention 1705 - framework for 853-855
- and waiting lists 1209 - prospective payment 862-866
- - maximize total predicted 1241 Quality of Well-Being Scale (QWB) 1745
quality models of long-term care 968-972 quantile regression in health econometrics 271
-measuring 1133-1134 quantity setting by physicians 467, 469, 474,
quality of care 477, 481,484, 486, 490, 496, 501
- in child health 1061 objectives 523, 527
- in long-term care 968-972 - and physician-induced demand 516-519
- in not-for-profit hospitals 1161-1163 quasi markets and waiting lists 1242
quality of life measurements 1723-1760 queuing model of waiting lists 1210, 1243
- best outcome measures 1743-1747
- - criteria for assessing 1743-1744 race
- evaluating methods 1745-1747 - and child health 1075, 1083
- -evaluating systems 1744-1745 - and disability 1003, 1004, 1005
- in long-term care 976
-economic validity of 1751-1752
RAND Health Insurance Experiment (HIE)
- generation of values for health states 1748
161-162, 286
- - aggregation of 1748-1749
- on alcohol consumption 1650, 1655
- harmonization 1752
and child health 1059, 1064
-
- health
- in dental services 1264, 1271, 1273
- - aggregation of values 1750
- and health outcomes 629
- calculation of values 1750
- and managed care 723-724, 728, 732
- - description of 1749
- and medical care demand 584, 588
- - source of values 1750
in mental health 910-911, 926
- -valuation of 1750
- and prevention 1708
health or health gain 1751
- and smoking 1579
- measuring health outcomes 1725-1726 random effects in health econometrics 308
- -requirements 1727-1729 randomized trials on moral hazard 418, 429, 437
- profiles or states 1751 ratchet effect in government purchase of health
- QALY model 1749 services 877-878
- -assumptions 1740-1743 rate of return regulation in pharmaceutical
- - - additive separability 1743 industry 1329, 1330
- -- constant proportional trade-off 1741 rate regulation in hospitals 1495-1497
- - - lifetime preferences, stability 1742-1743 rate-setting programs in hospitals 1503,
- - - risk attitude over life years 1741-1742 1509-1510, 1524-1526
- - - utility dependence 1740-1741 rational drug design 1307
- - - zero rate of time preference 1742 rational models of cigarette demand 1559-1561
- and respondent characteristics 1747-1748 - critiques of 1561-1563
- and socioeconomic inequalities in health 1854 rationing
- time preferences 1751 - of health services 444, 449, 454
- using different values 1752 - in mental health 912-918
- valuations, eliciting 1752-1754 - by physician 449-451
- values considered 1729-1731 rebate for no claims in health insurance 434
- - aggregation 1740 reduced form equations in human capital model
-- description of 1731-1732 379, 384
- - how generated 1739-1740 referrals in general practice 1185-1187
--how valued 1732-1737 regulation
- - who values 1738-1739 - of alcohol consumption 1658
Subject Index I-79
- in health plan markets 769-771 - research on 677-678

- mechanisms in medical malpractice 1350 retransformation problem in health econometrics
-in not-for-profit hospitals 1154 290
- of prices in pharmaceutical industry 1329 return-to-work and disability 1018-1019,
rehabilitation 1030-1031
- and disability economics 1030-1031 revealed preference in waiting lists 1222
- working-age population, USA 1018-1019 revenue targeting and physician agency 525-526
reimbursement in insurance 537-560 risk-adjusted premium subsidies in health plan
- alternatives to 548-551 markets 764
- and balance billing 547-548 risk adjustment
- bidding models 557-558 - in health insurance 624-625
- complements 548 - in health plan markets see health plan markets
- empirical results on 556-557 - in managed care 744
- heterogeneity in non-competitive markets - in mental health 897, 932-935
554-556 risk-aversion
- indemnity and health insurance benefits - in human capital models 393, 394
541-542 - and moral hazard 417
- in insurance theory 539-540 risk hedging in pharmaceutical industry 1305
- medical services, nature of 540 risk selection in health care as economic
- monopsony in 551-553 commodity 82-83
- and moral hazard 542-545 risk sharing in health plan markets 763, 777,
817-828
- optimal 542-545
- partial 541-542 - condition-specific 822
- empirical results 823-826
- and productive efficiency 553-554
- forms 819-823
- provider market power in 551-553
- - all members 819-820
- and service benefit insurance 545-547
- - condition-specific 822-823
substitutes 548
-- high-risks 820-822
rental price of health capital 356
- optimal forms 826-827
reputations in government purchase of health
- outlier 819
services 878-883
- proportional or prior costs 819, 827-828
- for characteristics 879
risks
for past behaviour 880-883
- of alcohol consumption 1647
research and development
- and quality of life measurements 1741-1742
- on infectious diseases 1787-1789 roll-back in pharmaceutical industry 1331
- in pharmaceutical industry 1302, 1307-1308 Roy's identity in health insurance 427
- on prevention 1702, 1703 rule of rescue and equity in health 1894, 1899
reservation wage in long-term care 973 Russia 830
reserve margins in hospital regulation
1523-1524 Saskatchewan Study on health insurance 430
residual claimants in not-for-profit hospitals safety in pharmaceutical industry 1308
1152 sample selection model in health econometrics
restorative treatments in dental services 272
1290-1291 San Francisco Home Health Study (SFHHS)
restricted-activity days in human capital model 1786
381 San Francisco Men's Health Study (SFMHS)
restrictions in health plan markets 770 1771-1772, 1774-1775
retail pharmacies 1304 sanitation, as public good 71
retail price, effect of taxation on smoking savings and demand for long-term care 981-982
1567-1569 Say's law of hospital beds 1221
retirement and health insurance 674-681 schooling
- health risks, by age 675 - and alcohol consumption 1657-1658
1-80 Subject Index
- in human capital model of health demand sexually transmitted diseases (STDs) 1707,
394-404 1773, 1790
- - formal 373, 379, 389, 395-397, 399, 400, Sherman Act (USA, 1890) 1408, 1419
402 sick leave
-- coefficients 382, 387 - in human capital model 387
- and prevention 1681-1682 - pay, in moral hazard 411, 416, 418, 421, 422,
- see also education 445
Scotland: dental services in 1281 sick time in human capital model 353
sealant programmes in dental services simulation estimators in health econometrics 305
1289-1290 simultaneous equations in health econometrics
search 299
- costly. in health care markets 1095-1096 simultaneous model profit demand and supply in
- and information diffusion in health care 225, waiting lists 1229
227 single photon emission computed tomography
- - incomplete, model of 230-236 193, 194
-- - applications 234 skew distribution in new drug development 1316
-- - and insurance 235-236 smearing estimator in health econometrics 290
- - - with unknown product quality 234-235 smoking 1539-1627
- and market equilibrium 228-230 - advertising and promotion 1584-1593
- in managed care 725-726, 728-731 - - econometric evidence 1585-1591
secondary prevention 1677 - noneconomic literature, findings from
selection 1591-1593
- and competition in health insurance markets - theory 1585
624 - agricultural policy and macroeconomic
- health plan markets, risk adjustment in implications 1598-1610
771-774 - - contribution of to economy 1606-1610
- - cream skimming 773-774 -- - farming communities 1609-1610
- - effects of 774-776 - - states and nations 1606-1610
- - preventing 776-779 - tobacco industry, size and nature 1598-1601
- and job-job mobility 665-666 -- globally 1598
- in markets for health insurance 624 -- - in USA 1598-1601
- model and moral hazard 447 - tobacco regulatory system, USA 1601-1605
- of provider in managed care 713-714, - - nature of system 1601-1603
728-731 - - - relevance of to smoking and health
-- studies of 729-730 1603-1605
- of risk in health care 82-83 - and child health 1078-1079
selective contracting in managed care 719, 727 control policies and demand 1593-1598
self-employment and health insurance 672 - - health information and counter-advertising
self-enforcing agreements in government 1593-1596
purchase of health services 881 -- smoking restrictions 1596-1597
self-referrals to physicians 517 -- youth access to, limits 1597-1598
self-reported information models in health plan - in human capital model 399-400, 403
markets 803-804 - management implications of consumption
semi parametric estimators in health 1544-1546
econometrics 270, 272-274 price of, impact on demand 1546-1565
sensitivity analysis in cost-effectiveness of -- addiction models 1556-1563
health interventions 196-197, 204 -- critiques of 1561-1563
separating in health insurance 610-612 - - - imperfectly rational models 1556-1557
service benefit insurance 545-547 - - myopic models 1557-1559
settlement process in medical malpractice 1357 ---rational models 1559-1561
Subject Index I-81
- - behavioural economic analyses of solidary contributions in health plan markets

1563-1564 760-761, 765
- - conventional studies of 1547-1556 sophisticated block contracts in NHS 866
- - - aggregate data 1547-1549 South Africa 1608
- - -individual data 1549-1556 Spain
- -econometric studies 1564-1565 - dental services, inputs and outcomes 1256
-restrictions 1552 - equity in health care delivery
-taxation 1543, 1565-1584 -- financing 1821, 1825
-- comparisons 1566-1571 - -private health insurance in 1820, 1826
--- country variations 1569-1571 -- utilization, indices 1844
- - - effects on retail price 1567-1569 - health expenditure 14, 850
- - - purposes and methods 1566-1567 - long-term care for elderly 963
- - - smuggling 1569-1571 -payment systems in general practice 1188
- - and environmental tobacco smoke (ETS) -tobacco advertising, banned 1589
1580-1582, 1596 spend-down behaviour in long-term care
- - fairness standards 1571-1574 981-982
- - - benefit principle 1574 sponsors, role of in health plan markets 760,
- - - horizontal and vertical equity 1572-1574 761-762
- - optimal, and economic efficiency 1579-1584 Stackelberg equilibria in waiting lists 1217
- - - negative extemalities 1579-1583 standard gamble (SG) of quality of life
- - - other considerations 1583-1584 measurements 1733-1735, 1737,
- - public health standards 1574-1579 1745-1747
-- - increasing taxes, health benefits of standard merger case in health care markets
1577-1579 1448
- - - social costs 1575-1577 standard of care
smuggling, and taxation on smoking 1569-1571 -bias in 1347-1349
social consequences of alcohol consumption - and medical malpractice 1353, 1376, 1383
1645-1653 Stanford University Study on health insurance
- cirrhosis 1649-1650 430
- crime and suicide 1652-1653 stated preference in waiting lists 1222
- intertemporal 1634 static moral hazard 413, 420, 429, 451, 454
- medical care 1650-1651 statutes of limitations in medical malpractice
- motor-vehicle mortality 1646-1649 1355, 1363
- - studies of 1648 steroids 1300
social contract theory 1892 stochastic frontiers in health econometrics
social costs 331-335
- of alcohol consumption 1664-1665 - cost function studies 331-332
- of smoking 1574, 1575-1577 - frontier models 332-335
social influences on alcohol consumption stock/flow inventory approach to waiting lists
1643-1644 1210
social insurance in health care delivery 1826 stock of health in human capital model 358,
social justice in health care delivery 1807, 361-364, 377, 378
1808-1809 stop loss in health insurance 437, 584, 585
Social Security Act (USA, 1972) 1494 streptomycin 1306
Social Security Disability Income (SSDI) stress echocardiography, cost-effectiveness of
program, USA 1012-1015, 1042-1044 193, 194
Social Security tax, for health insurance 569 strict liability in medical malpractice 1378
social welfare function (SWF) stroke, quality of care 1162
- in health sector 93 structural equations in human capital model 378
- in quality of life measurements 1728 subsidies
socioeconomic inequalities in health 1849-1855 - formula in health plan markets 767-769
1-82 Subject Index
- in not-for-profit hospitals 1148 - policy 1038

- - explicit 1151 - - transfer recipients 1003
- -implicit 1151 - equity in health care delivery
- prevention, policies to encourage !706-1707 --financing 1821, 1825, 1834
in public health interventions in infectious - taxes to finance 1824
diseases 1777-1780, 1782-1784 -utilization, indices 1846
- in waiting lists and medical treatment - health expenditure 14, 850
1235-1238 -international drug sales 1313
substance abuse 899-900 long-term care for elderly 963
alcohol 1654 - medical expenditure 172
- in child health programs 1078 medical malpractice in 1344
- treatment system 936-943 - - collateral source effect 1374
- see also mental health - -liability for 1389-1392
substantive equality of opportunity in health care -- no-fault programs 1379-1380
1878-1879 - vocational rehabilitation programs 1019
substitutes in insurance reimbursement 548 waiting lists
substitution effect - - contingent valuation of 1227
- in moral hazard 417 - - incentives to reduce 1235
- in spending on medical care 577 Swedish Patient Compensation Insurance (PCI)
substitution in human capital model 372, 374 1389-1392
Sudden Infant Death Syndrome (SIDS) 1078 switching regression model in health
suicide and alcohol consumption 1652-1653 econometrics 303-304
sulfanilamide 1306 Switzerland
dental services, inputs and outcomes 1256
sum-ranking and equity in health 1872
equity in health care delivery
Supplementary Security Income (SSI) program,
-- financing 1821, 1825, 1834
USA
- -private health insurance in 1820
- and disability 1015-1016, 1034, 1044-1045
- taxes to finance 1827
- and health insurance 681, 685-686
- -utilization, indices 1844, 1846
- for long-term care 976
health expenditure 13-14, 850
supplier-induced demand
- international drug sales 1313
- in dental services 1259-1260, 1263,
- not-for-profit hospitals in 1145
1272-1273
- in insurance 589
synthetic wards, and waiting lists 1229
- in waiting lists 1228, 1230, 1236
systematic review in health econometrics 331
supply assurance in government purchase of
health services 874-875 Taiwan Studies on health insurance 433
supply side cost sharing 856 target incomes
Survey of Income and Program Participation - and physician agency 522-526
(SIPP) - by physicians 465, 468, 481, 485-486, 496,
- on disability 1001 520, 522-526
- in health insurance 670, 672, 676, 679, 694 Tax Equity and Responsibility Act (USA, 1982)
survival analysis in health econometrics 1513
325-326 tax-exempt bond financing in not-for-profit
Sweden hospitals 1147
- dental services tax-financed health care 453
-- inputs and outcomes 1256 taxation
- markets for 1255 - of alcohol 1658-1665
- -productivity 1281-1282 - -background to 1658-1659
- direct provision of services in child health - - economic efficiency 1662-1663
programs 1077 - equity criteria 1659-1662
- and disability - - horizontal, and user fees 1660-1662
Subject Index I-83
- - - incidence 1659-1660 - - relevance of to smoking and health

- - total social cost 1664-1665 1603-1605
- and equity in health care delivery 1820, 1824 - size and nature 1598-1601
- and not-for-profit hospitals 1156, 1163-1164 --globally 1598
-prevention, policies to encourage 1706-1707 - - in USA 1598-1601
-of smoking 1543, 1565-1584 - see also cigarettes
- comparisons 1566-1571 tort liability 1345-1351
--- country variations 1569-1571 - economic models of 1364-1365
- - - effects on retail price 1567-1569 - with imperfect information 1347-1350, 1399
- - - purposes and methods 1566-1567 - - custom-based standard of care, bias in
- - - smuggling 1569-1571 1347-1 349
- - fairness standards 1571-1574 --uncertain standards 1349-1350
- - - benefit principle 1574 - other control mechanisms 1350-1351
- - - horizontal and vertical equity 1572-1574 - with perfect information 1345-1347
-- optimal, and economic efficiency 1579-1584 -reform 1355, 1371
- - negative externalities 1579-1583 - - awards, structure of
- - - other considerations 1583-1584 - - collateral source effect 1374
- public health standards 1574-1579 -- - limits 1371-1372
-- - increasing taxes, health benefits of - - -periodic payments 1374
1577-1579 - enterprise liability 1377-1378
- - - social costs 1575-1577 - litigation costs, reducing
teaching hospitals 1152, 1157, 1218 --- alternative dispute resolution 1376
technical efficiency in dental services 1275, - - - contingent fees, limits 1375-1376
1283 - - quality and standards of care 1376-1377
technology - - - awards, structure of 1371-1374
- administrative, in managed care 726 - - - litigation costs, reducing 1374-1376
- change, in medicine 451 - risks and prices, pharmaceutical industry
- - and medical care marketplace 127 1320-1321
- diffusion, in not-for-profit hospitals 1161 total cost of queuing in waiting lists 1211
- in treatment in mental health 937-938 total time lost in waiting lists 1211
Temporary Assistance for Needy Families 681 toxicity risks of drugs 1300
tertiary prevention 1677 tracking systems in child health programs 1077
thalidomide 1309 training and disability 1030-1031
therapeutic gains in new drug development 1320 - vocational (USA) 1018-1019
third variables in human capital model 396-397 transaction costs
time costs - in general practice 1179
- in dental services 1270-1271 - and moral hazard 590
- and moral hazard in health care 411, 447, 453 - in not-for-profit hospitals 1144, 1146, 1148,
time inputs 1152
- in human capital model 384 transfer recipients, disabled as 1026, 1027, 1028
- in waiting lists 1222 transitory surge in demand for health insurance
time preference in human capital model 374, 439
397-399, 401-405 travel time in human capital model 372, 390
time prices in dental services 1262-1263 trust and not-for-profit hospitals 1144
time trade-off (TTO) of quality of life trusts for long-term care 981
measurements 1733-1737, 1742, Turkey
1745-1747, 1854 - counter-advertising of smoking 1595
tobacco industry health expenditure per capita 14
- agriculture, USA 1544, 1601-1605 two-episode model of health insurance 423, 450
-regulatory program, USA two-part model in health econometrics 285
- - nature of system 1601-1603 two-part model of demand 434
1-84 Subject Index
uncertainty - - collateral source effect 1374

- and antitrust in health care markets 1413 - - contingent fees 1375
- in government purchase of health services - liability for 1386-1389
861-862 -- negligence claims for 1387
- health care as economic commodity 79-87 - mental health, expenditure on 898
- - economies of scale 81-82 - not-for-profit hospitals in 1145
-- insurance, welfare improving effects 80-81 - outpatient drug reimbursement 1330
- - moral hazard 83-87 population projections of elderly 983
- - risk selection 82-83 - primary care beds in 1192
- in human capital models 393, 394 - QALYs in measuring health outcomes 101
- in new drug development 1315 - risk adjustment and sharing in 830
- and physician agency 496-503 smoking
- - irreducible uncertainty 497-498 - advertising
- - unobservable physician actions 498-502 -- banned 1589
-- unobservable physician characteristics -- - and consumption 1587
502-503 - counter-advertising of 1595
- premium, in health insurance markets 626-629 -- economic contribution of industry 1608
- reducing, and waiting lists 1242-1243 -- social costs of 1575
- of standards, in medical malpractice 1349 - vocational rehabilitation programs 1019
uncompensated care - waiting lists in 1205-1207
- in child health 1066 - incentives to reduce 1235
- and health insurance 570 United Mine Workers Study on health insurance
- in not-for-profit hospitals 1160-1161 431
under-provision of waiting lists 1234 United States
uniform distribution of waiting values 1227 - dental services
United Kingdom - - inputs and outcomes 1256
- administered waiting in 1222 - markets for 1255
- birth rates, teenage 1079 and disability
- class, and child health 1075 -- policy 1038
- dental services - transfer recipients 1003
- inputs and outcomes 1256 - disability policy toward working-age persons
- - markets for 1255 1012-1020
- and disability - - accident and injury prevention program 1017
--policy 1038 -- health insurance 1019-1020
- - transfer recipients 1003 - -rehabilitation 1018-1019
- equity in health care delivery - - return-to-work programs 1018-1019
--financing 1821, 1825, 1834 - - SSDI program 1013-1015, 1042-1044
- -private health insurance in 1820, 1826 - - SSI program 1015-1016, 1044-1045
-- taxes to finance 1824 - - vocational training 1018-1019
- -utilization, indices 1844, 1846 - - Workers' Compensation program 1017,
- general practice in 1045-1047
--fundholding 1187 - equity in health care delivery
- - payment systems 1188 -- financing 1821, 1825, 1834
- - referrals in 1186 - - private health insurance in 1820
- health expenditure 14, 850 - utilization, indices 1844, 1846
- health plans in 833 - health expenditure
- internal markets in health sector 77 - - on mental health 898, 902
- international drug sales 1313 - - per capita 13-14
- long-term care for elderly 963 - - in public sector 850
- medical expenditure 171 - - on speciality care organizations 903
- medical malpractice in 1344, 1357 - international drug sales 1313
Subject Index 1-85
- long-term care for elderly 963 - in medical practice 241

- medical care marketplace, features 124-129 -physician-specific 248-251
- new drug approvals 1311 - regional 239-248
- not-for-profit hospitals in 1145 - - and individual variations 251-252
- payment systems in general practice 1189 - welfare loss from 254
- population projections of elderly 983 verdict in medical malpractice 1358
- QALYs in measuring health outcomes 101 vertical equity in health care delivery 1819,
- risk adjustment and sharing in 831 1822-1828
smoking, economics of - empirical work on 1824-1828
-- advertising, and consumption 1587, 1588 - Kakwani's progressivity index 1822-1824
- tobacco industry, size and nature 1598-1601 vertical integration in general practice
-- tobacco regulatory system 1601-1605 1191-1193
-- - nature of system 1601-1603 Viagra 1324
- - - relevance of to smoking and health violence
1603-1605 - and alcohol consumption 1646, 1652
unmeasured determinants in human capital - in child health programs 1078, 1079
models 391 violent death 1646
unpaid caregivers in long-term care 957 visual analogue scale (VAS) of quality of life
unverifiable quality of health services 851 measurements 1733-1734, 1747, 1854
user cost of health capital 356 Vocational Rehabilitation program (USA) 1013
utiliarianism and equity in health 1872-1873 vocational training and disability, USA
utility possibility frontier and equity in health 1018-1019
1867 von Neumann-Morgenstern utility in health
interventions 199
utilization of services and care 1833-1846
voucher model of premium contributions 766
- in child health 1070, 1071-1073, 1075, 1083
- in dental services 1273
wage coefficient in human capital model 384
- in general practice 1180-1184
wage effects in human capital model 387
-- GP, choice of 1183
wage elasticity of medical care 372, 375, 380
- -practice, choice of 1181 wage rate in human capital model 371-372, 382
- seeking medical care 1182-1183 wages
--treatment, choice of 1183-1184 - of disabled 1008-1009, 1011
- and hospital regulation 1516 - and health insurance 690-695
-inequality, measuring 1841-1846 - - and hours worked 695-696
- direct standardization-based index - - - research on 691-692
1842-1844 - - time series patterns 689-690
-- indirect standardization-based index - in not-for-profit hospitals 1157
1844-1846 waiting lists 1201-1249
- inequality tests in 1835-1841 - empirical matters 1221-1231
- in managed care 731-739 - costs of administered waiting 1222-1228
- - monitoring 716-717 - - - contingent valuation methods 1225-1228
utilization review - - - market data on 1222-1225
-in health care markets 1121-1123 - - costs of physical waiting 1221-1222
- in managed care 709, 717, 727 - inter-sectoral effects 1230-1231
- supply variables, impact of 1228-1230
Vaccines for Children Program 1702 existing, managing 1239
vaccines/vaccination -policy issues 1231-1243
- in infectious diseases 1794-1795 - efficient waiting 1239-1243
--mandatory 1780-1781 - - -prioritisation, indices 1240-1242
- and pharmaceutical industry 1299 -- quasi (internal) markets 1242
- and prevention 1694-1696, 1702-1703, 1709 - - - reducing uncertainty 1242-1243
variations in information diffusion -- options, taxonomy of 1232-1239
1-86 Subject Index
- - - demand rationing 1233 - - neo-classical framework 60-62

- --private provision 1238-1239 welfare lock in AFDC program 682
- - - subsidies 1235-1238 welfarism and equity in health 1872
-- - supply expansion 1234-1235 well care 420
theoretical issues 1205-1221 wholesalers in pharmaceutical industry 1304
- demand and consumer surplus dissipation willingness-to-pay (WTP)
1212-1214 - in dental services 1288
- - demand side 1214-1215 - and quality of life 1734
-- supply side 1215-1221 women
- - - fostering private practice 1220 costs of alternative strategies for coronary
- - - least cost supply 1221 heart disease 193
---managing 1218-1220 - and disability
waiting time in human capital model 372, 390 - - employment patterns 1006, 1008
Waxman-Hatch Act (USA, 1984) 1321-1322 -- status of 1004, 1005
wealth - health spending on 795-796
- in human capital model 375 - long-term care for 960
and long-term care 977 - and SSDI beneficiaries 1014
wealth elasticity of medical care 380 - and SSI beneficiaries 1016
Weibull model in health econometrics 327 work-loss days in human capital model 381, 388
welfare work-related training and education services for
- gains from health insurance 575 disability 1019
- and health insurance 673-674 Workers' Compensation program, USA 1017,
- and information diffusion 1045-1047
- - losses from 252-256 - and disability 1012-1013
- - and medical care variations 254 - and medical malpractice 1356, 1378-1379
- reform, in child health 1061 World Health Organization 1546, 1597, 1711
welfare economics - on infectious diseases 1766, 1789
- and equity in health 1874
- of health care X-efficiency in waiting lists 1231
- cost-effectiveness of 183-185, 203,
211-212, 219 youth
-- decision making in 188, 198 - alcohol, demand for 1641-1642
- - uncertainty social influences on 1643-1644
- - - economies of scale 81-82 cigarette demand 1552
- - - improving effects 80-81 smoking, access to, limited 1597-1598
- - - moral hazard 83-87
- - - risk selection 82-83 zero inflated Poisson model in health
- of health sector, efficiency in econometrics 319
- and extra-welfareism 62-66 Zimbabwe: tobacco industry 1599, 1608

(A.J. Culyer, J.P. Newhouse) Handbook of Health Ec

Uploaded by

Copyright:

Available Formats

(A.J. Culyer, J.P. Newhouse) Handbook of Health Ec

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(A.J. Culyer, J.P. Newhouse) Handbook of Health Ec

Uploaded by

Copyright:

Available Formats

INTRODUCTION TO THE SERIES

KENNETH J. ARROW and MICHAEL D. INTRILIGATOR

PART 1 - OVERVIEWS AND PARADIGMS

PART 2 - DEMAND AND REIMBURSEMENT FOR MEDICAL SERVICES

PART 3 - INSURANCE MARKETS, MANAGED CARE, AND CONTRACTING

PART 5 - THE MEDICAL CARE MARKET

PART 6 - LAW AND REGULATION

PART 7 - HEALTH HABITS

ECONOMICS AND MENTAL HEALTH*

Handbook of Health Economics, Volume 1, Edited by A.J. Culyer and J.P.Newhouse

2. The institutional context

2.1. Mental illness and persons with mental illness

2.2. Who is treatedfor mental illness

2.3. Who pays for mental health care

Payer MH/SA %* All health %*

Private insurance 25.8 31.0

*Percentages may not sum due to rounding.

2.4. The supply of mental health services

State mental hospitals $1,814 million $7,824 million

Source: Mental Health, United States (1998).

2.5. Managed behavioralhealth care

Traditional- Carve ' '

Enrollees / managed care plan 1 . -- ,

3. Private insurance markets, moral hazard, and mental health care

3.1. Evidence ofmoral hazard

Study Type of elasticity MH/SA estimate General health estimate

McGuire (1981) Point -1.00

Source: McGuire (1989) and authors' calculations.

3.2. Rationing in managed care

* Writing contracts to providers that include financial incentives to limit care.

E NtBt(mlt(ql), m2t(q2), ... , mst(qs)) (1)

with a total cost of

E Nt ( mst (s)). (2)

Maximizing benefits less costs with respect to each qS implies

Nt(Btsmt -mt) = 0 Vs. (3)

Bts = 1 Vs. (4')

3.3. Managedcare, supply side incentives and moral hazard: evidence

the evidence on the impact of payment and organizational arrangements on utilization

3.3.1. Research on hospitalpayment systems

3.3.2. Research on managed care and capitation

Figure 6. Payment systems for managed care.

Case Risk Impact Studies

Mass GIC RS Pre-post outpatient spending Ma and McGuire (1998a)

ASO = Administrative Service Only; RS = Risk Sharing; Cap = Capitation.

4. Insurance markets, adverse selection and mental health care

4.1. Evidence of selection in MH/SA

4.2. Policy responses to selection:fee-for-service-indemnity contracts

4.3. Selection and managed care: distorting "quality"

NBstmstRt - N BstmsCt- E Ntmst =0 Vs.

25 See also Van de Ven and Ellis (2000) on risk adjustment.

4.4. Policy responses and managed care

4.4.1. Risk adjustment

Sit = a + XitB + Ii +-sit. (7)

data from a population of approximately 450,000 enrollees were examined to evaluate

4.4.2. Behavioralhealth carve-outs

5. The public mental health and substance abuse treatment system

5.1. The technology of treatmentfor the severely mentally ill

This characterization suggests important differences in the technology required for

5.2. Fiscalfederalismandpublic mental health care