Journal of Public Economics 79 (2001) 375–398

www.elsevier.nl / locate / econbase

Do doctors respond to financial incentives? UK family

doctors and the GP fundholder scheme
B. Croxson a , C. Propper b , *, A. Perkins c
a
Institute of Child Health, University College London, London, UK
b
Department of Economics and CMPO, University of Bristol, CASE and CEPR, 8 Woodland Road,
Bristol, BS8 1 TN, UK
c
Norwich Union, London, UK
Received 1 August 1998; received in revised form 1 December 1999; accepted 1 December 1999

Abstract

The 1991 reforms to the UK NHS created a group of buyers of hospital care from
amongst primary care physicians. The implementation of the reforms was such that these
buyers had incentives to increase their use of hospital services prior to entering the scheme
in order to inflate their budgets. It has been argued that non-financial motives would limit
such behaviour. The paper shows that these health care providers did respond to the
financial incentives offered by the scheme, increasing hospital-based activity prior to entry,
and so inflating their budgets upwards for the duration of the fundholding scheme.
 2001 Elsevier Science B.V. All rights reserved.

Keywords: Physicians and financial incentives; Health care reform

JEL classification: Public sector labour markets (J450); Analysis of health care markets (I110)

1. Introduction

In 1990 the NHS internal market reforms changed the incentives of participants
in the UK health care market. The tax-financed system was retained, but the
functions of insurance and supply were separated. The reforms created a set of

*Corresponding author.
E-mail address: [email protected] (C. Propper).

0047-2727 / 01 / $ – see front matter  2001 Elsevier Science B.V. All rights reserved.
PII: S0047-2727( 00 )00074-8
376 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

buyers of care, known as purchasers, who received budgets from general tax
finance, and a set of hospital suppliers, known as providers. It was argued that
separating the two roles would improve both productive and allocative efficiency
in the NHS. Perhaps the most contentious aspect of these reforms was allowing a
subset of family doctors to act as purchasers under the GP fundholding scheme.
Under the previous arrangements family doctors had been gatekeepers to all forms
of medical care. They provided primary care in their surgeries, referred patients to
hospital for further treatment or diagnostic tests, and prescribed pharmaceuticals.
But they were not responsible for the costs of either hospital treatment or their
prescribing. Under the reforms, the fundholder scheme gave family doctors
budgets for these two activities. It was argued that the scheme would increase the
efficiency of family doctors by making them responsible for the financial costs of
their health care decisions and that family doctors would be better purchasers of
hospital care than third party purchasers who only bought but did not provide any
health care.
The outcome of this scheme has been hotly debated. On the one hand, it has
been argued that fundholders have been better purchasers because they have better
information on patients pre- and post-hospital treatment. They have been able to
innovate, to change methods of treatment, and to improve the efficiency of hospital
care suppliers. This has benefited their own patients but may also have had
positive spillover effects for other patients. On the other hand, it has been argued
that the scheme has resulted in a two-tier service with more resources available to
the patients of fundholders, leading to better treatment for this group at the
expense of all other patients and possibly also higher incomes for fundholders.
Much of the evidence used to support these claims is, however, based on
small-scale case studies, many of which are unable to distinguish between
differences due to self-selection of a particular type of family doctor into the
scheme and their behaviour in response to the scheme.
The essence of the problem is that fundholders were given budgets based on
their activity before they became fundholders, and were subject to relatively little
monitoring in how they used these funds. They therefore had unintended
incentives to increase activity in the statutory waiting period before becoming a
fundholder, and to decrease activity after becoming a fundholder to retain the
surplus from the fund. The policy concern is whether they responded to these
incentives. To address this question, the paper exploits the experiment implicit in
the way fundholding was introduced, plus part of a unique and important data set
scarcely used by researchers which contains information on every inpatient
episode in the UK since the advent of the internal market reforms. The fundholder
scheme is one of the few natural policy experiments in the UK welfare state: not
all family doctors became fundholders and those that did so became fundholders at
different dates. The data used is all admissions to hospital from all family doctor
practices in one geographic area in the UK during 4 years, matched to characteris-
tics of the practices including their fundholding status and their patients. These
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 377

data are used to control for observed characteristics of the practice and populations
which are associated with hospital admissions, to isolate a fixed effect associated
with those family doctors who ever became fundholders, and to identify whether
there was any effect on behaviour at the time these doctors became fundholders.
The results provide clear evidence that fundholders have responded to the financial
incentives of the scheme. Fundholders have increased admission activity in the
year before becoming a fundholder where it will bring them financial benefits.
The paper contributes to the growing literature that examines the power of
professional norms versus financial incentives as rewards for employees. Wilson
(1989) argues that the nature of government is such that goals of government
employees must differ from employees in the private sector, but notes that this
does not mean that government will maximise the public interest. The material
presented to support this argument is, however, mostly derived from case studies.
In one of the few non-case study empirical tests, Heckman et al. (1996) examine
whether bureaucrats who are rewarded on the basis of measured outcomes respond
to these financial incentives. In an examination of case workers who place job
applicants under the Job Training Partnership Act they find that case workers
prefer to accept the least employable applicants rather than the most employable as
suggested by a cream skimming argument and conclude that financial incentives
may be a countervailing force against the preferences of case workers. It has long
been argued that to understand the choices of doctors it is necessary to understand
their objective functions (e.g., Pauly, 1980). Hellinger (1996) reviews the evidence
and finds that financial incentives do affect the referrals of physicians working in
managed care plans in the US, which share some features of the fundholder
scheme. More specifically, the paper contributes to the literature on the effect of
fundholding in the NHS reforms. To date most of this literature has not been able
to separate out fundholder fixed effects from the impact of the scheme (Glenners-
ter et al., 1994; LeGrand et al., 1998). We find that family doctors contracted to the
UK public sector do respond to financial incentives. The changes in admissions
resulting from fundholding might have benefited family doctors who were
fundholders, or their patients or both, but probably did so at the expense of other
patients.
The paper is organised as follows. Section 2 examines the incentives under the
GP scheme, reviews the existing knowledge and puts forward our hypotheses.
Section 3 presents the data we use. Section 4 presents the results and the final
section discusses their significance.

2. Fundholder incentives

Family doctors, known in the UK as general practitioners or GPs, play a central

role in determining access to hospital care and specialist services in the UK NHS.
There are essentially two routes into hospital: as an elective or an emergency
378 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

patient. Elective patients are usually admitted to hospital only if they have been
referred by a GP, and they are admitted only to scheduled appointments. By
contrast emergency cases can be admitted directly, at any time, and people can
self-refer themselves. A large number of people are also referred to hospital as
emergency cases by their GPs. GPs therefore affect both how many people are
admitted to hospital and whether they are admitted as emergency or elective cases.
A number of factors affect GPs’ referral decisions, including not only a patient’s
health but also the prevailing economic and legal institutions. In the UK, a key
component of the institutional structure affecting their decisions has been the
advent of the GP fundholding scheme, introduced as part of the 1990 Internal
Market reforms to the NHS.
The Internal Market reforms divided organisations within the NHS into
purchasers and providers. Purchasers comprised health authorities and a self-
selected subgroup of GPs, called GP fundholders (GPFHs). Health authorities are
government agencies responsible for the health of the population of a particular
geographic district. They are given an annual budget allocation by central
government, and are required to purchase health care on behalf of their residents.
In theory they can purchase care from any health care provider: in practice, they
purchase most care from within-district NHS providers, which, after the reforms,
were formed into self-governing NHS trusts.
GPs are self-employed but receive almost all their income from the government.
Some of this income is dependent on the number of patients registered in their
practice. GPFHs were given an additional source of income, a budget from which
they were required to purchase a specific range of services on behalf of their
registered patients. These services include hospital-based elective care, some
diagnostic services, and pharmaceutical drugs. Non-fundholding GPs do not hold
budgets for secondary care. They are responsible for the provision of primary care
(as are GPFHs). Health authorities meet non-fundholders’ prescribing and sec-
ondary care costs. Health authorities also meet the costs of that hospital care of
GPFHs that is not covered by the fund (some elective care and all emergency
admissions). The fundholding scheme also had an element of ‘stop-loss’ insurance:
health authorities paid any costs incurred on a single patient in a year that were
above £5000.
The fundholding scheme was voluntary, GPs joining as they wished in
successive years. To be eligible, practices had to have a certain minimum number
of registered patients (in the first year of the scheme this was 11 000, then reduced
to 9000 in the second year and subsequently to 7000). GPs who wished to become
fundholders applied for fundholding status at least a year before they were granted
this status. During their preparatory period the health authority assessed their
suitability for the scheme as well as their prescribing and referral activity. Very few
practices were rejected from the scheme if they met the practice size criteria. The
scheme has expanded over time as successive governments have followed a policy
of expanding the purchasing role of GPs. This is evident in increases in the
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 379

number of hospital services included in the scheme, in the number of fundholders,

and finally in the piloting of ‘total fundholding’, which expands the scheme so that
some GPs hold a unified budget from which they are required to purchase all
health care for their patients, including emergency care.1 We do not examine total
fundholding here. Fundholders who underspent their budgets were allowed to keep
their surpluses, provided that the monies were spent on improving patient care.
There was little monitoring of this requirement, and as GPs (and GPFHs) own their
practices, improvements to the practice can be translated into future income.
GPFHs who overspent their funds by more than £5000 p.a. could have their
fundholding status removed by the health authority.
Fundholding was designed to close the gap between the fundholders’ decisions
over referral and prescribing and the financial consequences of these decisions.
The aim was that GPFHs would improve the efficiency of service provision. A
limited budget was supposed to give them an incentive to maximise the health-care
return to any given expenditure as they were believed to be in a position to get
information on the quantity and quality of hospital services. But as enacted the
fundholding scheme gave incentives for certain types of behaviour. Fundholders’
elective care budgets were based on their activity in the period before they became
a fundholder. In the scheme’s early years, most fundholders’ budgets were based
on their activity in the year before they became fundholders. This gave them an
incentive to increase referrals in that ‘preparatory’ year.2 The scheme also gave
GPFHs budgets that covered only a subset of elective procedures. This altered
their relative prices of emergency and elective activity, making emergency
admissions cheaper than non-emergency admissions, and of different types of
elective activity, as not all elective admissions came under the scheme.
Fundholding may affect not only GPs’ referral behaviour, but also providers’
behaviour with respect to admissions, because of the relative size of fundholder
and health authority budgets and the form of contracts between the two types of
buyers and the sellers of services. Health authorities are the larger buyers and tend
to use lump sum payments for a total amount of activity (known as block
contracts) (Propper, 1996). GPFHs, at least in the first few years of the scheme,
tended to use cost per case reimbursement. If GPFHs use cost-per-case contracts
and health authorities use block contracts for elective activity, providers have an
incentive to admit fundholders’ patients as elective cases. There is also evidence
that, particularly in the early years of the scheme, fundholders were given
generous funds (Petchey, 1995), which may have increased their bargaining power
with respect to sellers of health care compared to non-fundholders, allowing them
better access for their patients than those of non-fundholding GPs.

1
Under recent government policy, the fundholding scheme has been replaced by local Primary Care
Groups, comprising groups of GPs and other health care professionals, eventually holding a budget
from which they will have to purchase all services for their patients.
2
In later years, some fundholders were assessed on their activity over a number of preceding years.
380 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

There is a small, but growing, body of evidence on the behaviour of

fundholders. But the picture that emerges from this literature is far from clear.
There are several studies that test for differences between fundholders and
non-fundholders. Some find a negative association between fundholding and
services utilisation. Whynes et al. (1995) found fundholders prescribe more generic
drugs, have a lower overall volume of prescriptions, and have a slower rate of
increase in prescriptions, compared to non-fundholders. Howie et al. (1994) found
that fundholding status has a negative association with the utilisation of elective
services and Fear and Cattell (1994) found that fundholders are less likely to
request domiciliary visits from psychiatrists (visits which are paid for by
fundholders). But other studies have found little difference between referrals made
by fundholders and non-fundholders, and some have found that fundholders are
unlikely to alter their referral patterns by switching activity from one provider to
another (Coulter and Bradlow, 1993).
Toth et al. (1997) examine whether fundholders substitute emergency for
elective activity using data on four procedures to which patients may be referred as
either elective or emergency admissions. They find no evidence that fundholders
substitute emergency for non-emergency referrals after becoming fundholders, but
find some, though not overwhelming, evidence that fundholders have a lower ratio
of emergency to total referrals. Healey and Reid (1994) examined whether
fundholders inflate preparatory year prescribing expenditure and found no evi-
dence of this. On the other hand, Surender et al. (1995) found a significant
increase in preparatory year referrals comparing 10 fundholders with six non-
fundholders.
The conclusions that can be drawn from these studies are limited. It could be the
case that the mixed results are because the incentives of the fundholder scheme
described are not be strong enough to outweigh other factors that drive fundholder
behaviour. One study found that GPs joining the scheme in its early years were
attracted by the prospect of freedom and autonomy, and not necessarily solely by
personal financial incentives (Newton et al., 1993). If exercising freedom and
autonomy are consistent with changing the level and pattern of referrals in a
direction that is the opposite to the financial incentives given by the fundholding
scheme then studies of fundholders versus non-fundholders may find little
difference between them. On the other hand, these non-financial goals may not be
incompatible with changing referral patterns for financial gain so we might still
expect to see an impact of a major change in financial incentives.
But it is more likely the case that many of the studies carried out to date are not
able sufficiently to disentangle fundholder and non-fundholder behaviour in
response to the incentives of the scheme from differences between the practices
that became fundholders from those that did not (Dixon and Glennerster, 1995;
Petchey, 1995; Gosden and Torgerson, 1997; LeGrand et al., 1998). This may be
because the data on differences between fundholders and non-fundholders prac-
tices which are known to affect referrals was not available to the researchers, or
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 381

because the studies are small, or because they use data from very early in the
scheme, when the only fundholders were a group who are acknowledged to be
rather different from other GPs.
The present study overcomes many of these problems by utilising the experi-
ment that occurred in fundholding. Not all GPs became fundholders and those that
did became so at different dates. Using all referrals to hospital over four years
from a large sample of GPs, some of whom became fundholders at different dates
in the four year window and others who did not, we can isolate both a fixed effect
associated those who were ever fundholders and any effect on behaviour at the
time practices became fundholders. We can observe whether fundholders changed
their referral patterns before becoming a fundholder to inflate their fund and
whether they changed them afterwards. By matching this data to measures of
factors other than GP fundholder status known to determine referrals we can
further isolate whether there is a fundholder effect over and above these observed
differences between practices and practice populations. So we allow for the fact
that fundholder status may measure characteristics other than the financial
incentives that fundholding gives, and that fundholding status may be correlated
with other features of the practice which determine referral levels.
Referral decisions to emergency and non-emergency treatments are modelled as:

r j 5 f j (Z, X, F ) (1)

where r j is the number of referrals, j indexes whether the referral is emergency or

non-emergency, Z is a vector of population characteristics which determine the
level of referrals,3 X is a vector of GP characteristics other than fundholding and F
is a vector of fundholding status and timing dummies.
Our focus is on the effect the timing of fundholder status has on referrals,
holding constant the fixed ‘fundholder ever’ effect. We hypothesise that fundhol-
ders will increase non-emergency admissions the year before becoming fundhol-
ders. In this year they will either not alter emergency admissions, or will at the
margins substitute non-emergency for emergency admissions to the extent that
they can re-classify in this way. So the level and the ratio of non-emergency
admissions to total will rise in the year prior to fundholding. We have less clear
priors about the level of non-emergency admissions once a GP becomes a

3
There is evidence that referral rates tend to decrease with distance from hospital; increase with
patients’ age; are higher for women; and are usually higher for people coming from deprived areas
(e.g., Hull et al., 1997). Holding health status constant, there is evidence that different GPs have
significantly different ‘referral thresholds’ (Roland and Coulter, 1993). There is evidence that referrals
differ according to the number of partners in a practice (in which case in-house referrals might
substitute for out-patient referrals), the length of time elapsed since qualified; the ‘social distance’
between the GP and patient (Scott et al., 1996) and the level of risk the GP is willing to bear (Lowy et
al., 1994). There is also evidence that women use health services more than men (e.g., Social Trends,
1995).
382 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

fundholder. Some fundholders may maintain non-emergency admissions at the

higher level. Others may use the larger fund to provide different types of care or to
save the surpluses, either of which will result in a fall of non-emergency
admissions after becoming a fundholder. We would expect a rise in the level of
emergencies as fundholders seek to get cases treated for free. So there will either
be a fall or no change in the level of non-emergencies, and a fall in the ratio of
non-emergencies to total admissions.

3. The data

The data we use are from the Contract Minimum Data Set (CMDS). The CMDS
and its derived data sets are an important resource almost unused by economists.
The CMDS contains information about every inpatient episode in the UK since the
creation of the internal market in 1990. Martin and Smith (1996) analysed one
year of these data from a sample of Health Authorities to explain the determinants
of hospital length of stay, and these data aggregated to small area level to model
waiting times for elective surgery (Martin and Smith, 1999). We use 4 years of
data from one Health Authority enabling us to exploit the panel nature of the data
and undertake analyses of a type not previously conducted to examine fundholder
and non-fundholder behaviour.
We use the CMDS for North West Anglia Health Authority (NWA). This is a
record of all hospital admissions for individuals resident in the geographical area
covered by NWA. (There are approximately 100 HAs in the UK: each resident of
the UK falls under one HA.) Each admission is classified by details of the date and
type of admission and discharge, the speciality, the diagnosis, the patient’s GP, and
the patient’s age, sex, and postcode. In our statistical analysis we use data relating
to admissions from all the GP practices within NWA during the four financial
years 1993 / 4 to 1996 / 7, to two hospitals, Peterborough Hospitals Trust (PH) and
the King’s Lynn and Wisbech Hospitals Trust (KL). As shown in Appendix A, this
gives almost 350 000 records for 324 000 admissions. Of these admissions,
303 200 (99%) could be associated with a GP practice in NWA.4 This is not the
whole dataset as there are at least 6000 admissions to local hospitals by GPs
outside NWA,5 and NWA GPs can refer people to other hospitals outside the NWA
district. Analysis is confined to admissions to the two local hospitals, PH and KL,

4
Those records that could not be associated with a practice were evenly distributed across specialty,
hospital, and admission method.
5
Admissions to any particular hospital comprise the following different categories: first, admissions
from local GPs of local residents; second, admissions of local residents by out-of-district GPs; and
third, admissions of out-of-district residents by out-of district GPs. Using our data-set, derived from one
district, we can estimate the first two categories but not the last. Hospital sources confirm that the
majority of admissions to PH and KL are from NWA GPs.
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 383

Table 1
Number of fundholders, 1993 / 4 to 1996 / 7
1993 / 94 1994 / 95 1995 / 96 1996 / 97
Total number of fundholders 12 (20.34%) 14 (23.73%) 17 (28.81%) 25 (42.37%)
Number becoming fundholders 8 (13.56%) 2 (3.39%) 3 (5.08%) 8 (13.56%)
in year
Number of fundholders in their 2 (3.39%) 3 (5.08%) 8 (13.56%) 5 (8.47%)
preparatory year

given the likelihood of a different set of constraints governing admissions to

distant hospitals, comprising as they do a lower proportion of emergencies.6
In each of the four years covered by our data there were 58 general practices in
NWA.7 Table 1 shows the fundholding status of these practices over the sample
period. The patient and GP characteristics of the GPFH practices which are ever
fundholding in the sample period are compared to those of the non-fundholder
practices in Table 2. The table shows that fundholders on average were located in
more deprived parts of the district, were closer to one of the two hospitals in
NWA, had larger list sizes per GP, and had a higher median age of the GPs in the
practice than non-fundholders. The population characteristics of the Health
Authority are described in Appendix A. Within NWA GP services are administered
in four discrete areas, with boundaries roughly coterminous with those of local
district councils: Peterborough, Fenland, King’s Lynn, and Swaffham. Their
geographical boundaries were partly defined by a natural east / west polarisation
within the district, since prior to 1992 NWA was two separate health authorities 8 ,
and during our 4 years the east and west sides of the district were served by two
different social services departments. GPs in Fenland also differ from those in

Table 2
Characteristics of fundholding and non-fundholding practices
Variable Fundholders Non-fundholders
(n530) (n529)
Mean (S.E.) Mean (S.E.)
List size per GP 1865 (370) 1662 (383)
Median age of GP 45.5 (6.2) 43.1 (5.3)
Distance to nearest hospital (King’s 9.4 (9.2) 16.8 (11.1)
Lynn or Peterborough)
Deprivation (Jarman) score for practice 17.7 (14.8) 7.76 (18.3)
Ratio of male patients to all patients 0.49 (0.02) 0.49 (0.2)

6
Eighty-seven percent of all admissions from NWA GPs were to these two hospitals.
7
These 58 practices contained, in total, 226 individual general practitioners.
8
Peterborough DHA and West Norfolk DHA.
384 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

other areas in that they are also almost equi-distant from the two local acute
hospitals.
Thirty-six percent of the admissions in our dataset are classified as emergency,
and the remainder comprise elective day cases, elective in-patients, maternity-
related cases, and inter-hospital transfers. Within this CMDS dataset 39 specialities
are defined. The boundary between some specialities is unclear and defined
differently in the two hospitals in our sample, so we grouped together General
Medicine and Care of the Elderly as Integrated Medicine, and General Surgery and
Urology as Surgery, and also grouped together specialities where the number of
admissions into each was small.

4. Empirical specification and results

Fig. 1 shows all non-emergency admissions by financial year for four groups of
fundholders in the data. These are those who became fundholders in the third
(1993 / 4) to sixth (1996 / 7) year of the scheme. (Each cohort of fundholders is
known as a ‘wave’.) The vertical bars indicate referrals to the NWA hospitals by
financial year, beginning in 1993 / 4. So for the first group of fundholders, this year
was the year in which they became fundholders. For the second group, this year
was the preparatory year. The figure shows a clear pattern for the first three
groups. Admissions to NWA hospitals fall in the year in which the practice is first
a fundholder, and for the second and third groups depicted in the figure,

Fig. 1. Non-emergency admissions to KL and PH, by fundholding wave and year, waves 3 to 6. (No
shading indicates year before becoming a fundholder: dark shading indicates year of becoming a
fundholder.)
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 385

admissions rise in the preparatory year. We do not observe the preparatory year for
the practices who became fundholders in the third wave (or for those who became
fundholders in the first and second year of the scheme although they are part of
our data set). The figure indicates that the behaviour of fundholders in the third to
the sixth wave was to increase referrals in the preparatory year and decrease them
in the year of becoming a fundholder. There is less of a rise for the final group.
But this pattern is consistent with the change in the way fundholding budgets were
set for this group. Their budgets were dependent on activity in a number of
proceeding years and not just in the preparatory year. The aim of the econometric
analysis is to examine whether these results are robust to differences between and
within fundholders and other GPs, and whether they are most evident in those
specialities in which GPFHs have the greatest incentive and opportunity to alter
their referral patterns.
We model the count of admissions, emergency or non-emergency, per practice,
financial year, and speciality. To allow for the impact of the number of patients in
the practice, we divide the number of admissions by list size. We sum admissions
across the two hospitals, as many of the counts are zero by hospital as GPs
generally refer to either one or the other for each type of treatment. We model
these admissions as a function of population characteristics, characteristics of the
GPs in the practice, and practice fundholder status. The population characteristics
are intended to measure demand for health care in the practice population. They
are from either the CMDS or the Census, in which case they are matched to the
GP identifier in the CMDS. The characteristics we examine are the proportion of
males in the practice population and a measure of socio-economic deprivation of
the ward in which the practice is located (the Jarman score).9 It is well established
that hospital use is negatively associated with distance to medical facilities, so
distance from the practice to the nearest hospital was included as a measure of
demand. Characteristics of the GPs in the practice were included to allow separate
identification of the effect of being a fundholder from other characteristics that
might be associated with both fundholder status and referral behaviour. These
included median age of the GPs in the practice and the list size per GP.10 Three
fundholder dummy indicators were constructed for each practice: the first
indicating whether the practice was ever a fundholder, the second the financial
year in which the practice became fundholding and the third indicating the
financial year immediately prior to the practice becoming fundholding. Finally, we
allow for differences in the levels of admissions across time, speciality and area.
The means of these variables are presented in Table 3 and further details of their
construction are provided in Appendix A.
9
Other commonly used measures of practice population need for health care are the Standardised
Mortality and Morbidity rates of the practice area and its age distribution. However as these measures
are strongly correlated with the Jarman scores they were not used as covariates in the final regressions.
10
A number of insignificant variables with no theoretical association with referrals were dropped,
including dummy indicators of dates when the number of GPs in the practice rose or fell and ratio of
male to female partners.
386 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

Table 3
Means of regressors
Variable Mean S.D. Min Max
List size 6714.00 4475.43 1127.75 17 823.75
Number of GPs 3.71 2.34 0.75 10.00
List size / number of GPs 1846.34 418.92 793.23 2769.13
Distance to nearest hospital (King’s 10.77 9.66 0.10 31.20
Lynn or Peterborough)
Deprivation(Jarman) score for practice 12.89 17.21 217.75 62.59
Ratio of male patients to all patients 0.50 0.02 0.45 0.56

4.1. Pooled estimates

We begin by estimating the number of referrals by practice i in year t pooling

across specialities:

r jist 5 b1 1 b2 Zit 1 b3 Xit 1 b4 Fit 1 b5 S 1 b6 T 1 b7 A 1 ´ist (2)

where r jist is the (log of) the number of referrals by practice i in speciality s and
year t, j indexes whether the referral is emergency or non-emergency, Zit is a
vector of population characteristics including the ratio of men to women and level
of deprivation of the practice population, Xit is a vector of GP characteristics, Fit a
vector of fundholding status dummies which identify whether a practice was ever a
fundholder, the year before the practice became a fundholder and the year in which
the practice was first a fundholder. S, T and A are speciality, time, and area
dummies, respectively, and ´ist is white noise error.
We estimate (2) using OLS with robust standard errors to allow for heteros-
cedasticity, and robust regression where large outliers from the OLS regression are
given less weight (also allowing for heteroscedasticity).11 We report only robust
estimates.12 To examine whether fundholders are shifting to emergency from
non-emergency admissions after becoming fundholders we also estimate the ratio
of emergency to total admissions using the same estimators.
The results for the level of admissions of each type are in Table 4. The estimates
in column (1) are for emergency admissions, those in columns (2)–(4) are for

11
All estimates were undertaken using STATA version 5. The robust estimator is the rreg command,
and the OLS allowing for heteroscedasticity is the newey command. The observations which were
given less weight in the robust regression were predominantly those from specialities where the level of
admissions was very small (paediatrics for non-emergency, and orthopaedics and oral surgery for
emergency admissions). Observations from three (of the 58) practices were weighted less more
frequently than from other practices. One of these practices is known to have unusually low referral
(and prescribing) rates; the other two are located in the lower part of Fenland and have high referrals to
hospitals outside the area throughout the period.
12
OLS results are available from the authors.
Table 4
Robust OLS and Fixed effect regressions of admissions
Pooled OLS Fixed effect estimates Pooled OLS

Dependent variable: (1) Emergency (2) Non-emergency (3) Non-emergency (4) Emergency/ (5) Emergency (6) Non-emergency (7) Non-emergency (8) Emergency/ (9) Non-emergency

B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

admissions/list size admissions/list size admissions, elective total admissions admissions/list size admissions/List size admissions, elective total admissions admissions/List size
subset/list size subset/list size

Fundholder in preparatory 20.044 0.113*** 0.147*** 20.014** 20.05 0.07* 0.10** 20.010 0.125***
year (0.047) (0.041) (0.044) (0.007) (0.05) (0.09) (0.046) (0.007) (0.041)
Became a Fundholder this 20.109** 20.143*** 20.109*** 0.008 20.02 20.048 20.018 0.008 20.134***
year (0.044) (0.038) (0.044) (0.006) (0.04) (0.07) (0.04) (0.006) (0.039)
Fundholder at some point 0.026 0.017 20.058** 0.002 – – – – 0.038
(0.028) (0.025) (0.028) (0.004) (0.024)
List size3100 per GP 20.001 20.006** 20.007*** 0.001** – – – – –
(0.003) (0.002) (0.003) (0.0005)
Median GP age 20.003 20.007*** 20.006** 0.0004 – – – – –
(0.003) (0.002) (0.003) (0.0004)
Log distance to nearest 0.001 20.039*** 20.048*** 0.004 – – – – –
NWA hospital (KL (0.015) (0.013) (0.015) (0.002)
or PHT)
Deprivation (Jarman) score 0.012*** 0.005*** 0.003** 0.001*** – – – – –
of practice (0.001) (0.001) (0.001) (0.0004)
Ratio of male patients to 0.276 20.540 20.038 0.055 – – – – –
total (0.813) (0.704) (0.813) (0.112)
Constant 24.397*** 23.128*** 23.362*** 20.293*** 24.5*** 3.99*** 23.96*** 20.36*** 23.868***
(0.373) (0.323) (0.327) (0.0052) (0.43) (0.32) (0.33) (0.007) (0.041)
Area effects Yes Yes Yes Yes No No No No Yes
Year effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Specialty effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
2
R 0.893 0.795 0.611 0.934 0.72 0.64 0.49 0.90 0.79
N 2070 2070 1160 2070 2088 2088 1160 2088 2070

387
Robust standard errors in parentheses *P,0.1; **P,0.05; ***P,0.01.
388 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

non-emergency admissions. Column (2) is all non-emergencies. Column (3)

examines the subset of specialities in which we would a priori expect GPFHs to
have most effect in altering timing and location of treatment, since the proportion
of both non-emergency admissions and admissions paid for by fundholders vary
across specialities. We would expect to find the greatest impact of fundholder
incentives in the five specialities in which elective care is important and
fundholder procedures make up a higher proportion of activity. These specialities
are Ear Nose and Throat (ENT), Gynaecology, General Surgery, Ophthalmology
and Orthopaedics. The estimates in column (4) are for the ratio of emergency to
total admissions.
Table 4 shows that practice population and GP characteristics affect admissions.
Neither list size per GP nor the median age of GPs within the practice is
consistently associated with emergency admissions, but both are negatively
associated with non-emergency admissions. Distance to medical facilities is
negatively associated with non-emergency referrals. Deprivation is positively
associated with the level of both emergency and non-emergency admissions. The
sex distribution of the practice population is not significantly associated with
admissions, though practices with higher ratios of male to female patients have
lower levels of non-emergency referrals, consistent with the fact that women have
greater contact with health service providers (Social Trends, 1995).
Fundholder status per se (practices that were fundholders or were in their
preparatory year at some point in the 4-year window) is not associated with either
emergency or non-emergency admissions across all specialities. For the subset of
specialities in which we expect fundholders to alter their behaviour (column 3),
being a fundholder is associated with lower admissions.13 But the timing of
fundholder status is associated with admissions. Column (1) shows that emergency
admissions fall the year the practice became a fundholder. The next two columns
show the timing of fundholder status has the hypothesised association with
non-emergency admissions. Column (2) indicates there is a statistically significant
rise in non-emergency admissions in the preparatory year and a statistically
significant fall in the year of becoming a fundholder. The rise in the preparatory
year is of a similar magnitude to the fall in the year immediately after. Column (3)
indicates that the pattern of a rise followed by a fall is clear in the speciality subset
where we would expect GPFHs to have greatest ability to alter behaviour: in other
words, the rise and fall appear where we would expect behavioural change.
The analysis of the ratio of emergency to total admissions (column 4) shows the
net effect of fundholding timing on changes in non-emergency and emergency
admissions. There is a significant fall in the ratio of emergency to total admissions
the year before a practice becomes a fundholder. This indicates that not only do
practices raise their non-emergency admissions the year before they become

13
As discussed below, the last group of fundholders differs from those in earlier waves. It has higher
admissions, controlling for other variables in Table 4.
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 389

fundholders, but they raise them as a share of all admissions from their practice.
There is also a rise in the share of emergency admissions after becoming
fundholders, although this is not statistically significant.

4.2. Allowing for GP and specialty effects

The above results are from data aggregated across specialities, and ignore the
fact we have data for each practice over 4 years. We can explore the impact of
timing of becoming a fundholder relative to an individual GP mean and can go
below aggregate level to speciality level. Columns (5)–(8) in Table 4 report fixed
effects estimates corresponding to columns (1)–(4), respectively. These results
show that relative to each GP’s mean, the pattern in admissions prior to and after
entry into the fundholder scheme is the same as that seen in the pooled regressions,
though the estimates are smaller and less well-defined (which is not unexpected
given we have a small number of years of observations in the data). Column (6)
shows that there is a significant increase in all non-emergency admissions in the
year prior to entry into fundholding. Column (7) shows there is a significant
increase in admissions in the subset of specialties where GPs are most likely to be
able to influence activity. There are falls in the year immediately post entry for
both these sets of admissions, but the coefficients are not well defined.
We also address the possibility that fundholder status might be endogenous. As
outlined above, fundholder status was voluntary, and it is possible (and plausible)
that early entrants might be those who were best able to extract rent from the
scheme. If this was the case we might expect the estimates on fundholder status to
be upwardly biased. This might also bias the fundholder timing variables (the
focus of this paper) since timing is, by definition, relative to the date at which a
practice became a fundholder.
We examine possible bias in two ways. First, in the final column of Table 4 we
report pooled OLS estimates without those controls which may be associated with
fundholder status. The results indicate that the estimates on the timing of
fundholder status are robust to exclusion of these controls: the coefficients on
preparatory and first year of fundholder status change little. Second, we use the
pooled data set to estimate admissions for fundholders only, controlling for
possible endogeneity of fundholder status by standard Heckman 2SLS methods.
Identification in the Heckman model other than by functional form requires that
there are some variables which affect the probability of becoming a fundholder,
but which do not affect the level of admissions conditional on being a fundholder.
With the data available, one candidate for exclusion from the determinants of the
level of admissions is age of the GPs in the practice. This may well determine the
desire of the practice to take on more management, and as the capital value of
practices is realised on retirement, will affect the value of gains from becoming a
fundholder, but should not affect the level of admissions. We therefore estimate the
probability of becoming a fundholder as a function of all practice characteristics,
390 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

excluding calendar year and speciality, and admissions as a function of practice

characteristics other than average age of the GP, allowing for year and speciality
effects. We repeat this using the data as a panel. In this case, the IMR term is
collinear with GP fixed effects, so cannot estimate the effect of GP characteristics
in the second stage of the regression.
The results in Table 5 indicate that the estimated impact of the two timing
variables appear to be robust to the correction for self selection. The IMR term is
not significant in either columns (2) or (4) of Table 5, nor are the coefficients
different from those in columns (2) and (6), respectively, of Table 4.
Table 4 shows differences between those specialties where GPs may be able to
alter their referral behaviour and those where they have less ability to do so. To
investigate this further we estimate fixed effects estimates within specialty for

Table 5
Two step estimates
Pooled data Panel data

Probit estimate Non-emergency Fixed effects Non-emergency

of being a admissions/list probit estimator admissions/list
fundholder size with IMR term of being a size with IMR term
(fundholders only) fundholder (fundholders only)

Fundholder in preparatory year – 0.111** – 0.074*

(0.04) (0.04)
Became a fundholder this year – 20.109** – 20.04
(0.038) (0.038)

List size3100 per GP 0.004 20.0039 0.04 –

(0.005) (0.003) (0.06)
Median GP age 0.067** – 0.07 –
(0.005) (0.005)
Distance to nearest NWA hospital 20.432** 20.005 20.34* –
(KL or PHT) (0.027) (0.024) (0.027)
Deprivation(Jarman) score of 20.004* 0.008** 20.004 –
practice (0.002) (0.001) (0.002)
Ratio of male patients to total 215.533** 21.336 216.5 –
(1.58) (0.91) (13.8)

IMR – 20.016 – 20.15

(0.0692) (0.11)
Area effects Yes Yes No No
Specialty effects No Yes No Yes
Year effects No Yes No Yes

Constant 5.067** 23.077** 4.5 23.7***

(0.707) (0.463) (6.2) (0.09)

Psuedo R 2 /R 2 0.19 0.819 0.20 0.819

N 4140 1062 58 1062

Standard errors in parentheses *P,0.1; **P,0.05; ***P,0.01.

 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398
Table 6
Non-emergency admissions by specialty, fixed effects regression
Elective specialties/Specialties with more GPFH procedures Non-elective specialties/Specialties with few GPFH procedures

ENT Gynaecology Ophthal- Orthopaedics General Integrated Oral Paediatrics Other

mology surgery medicine surgery

Fundholder in 0.08 0.18** 0.21 0.21** 0.002 0.01 0.03 20.09 0.14
preparatory year (0.11) (0.08) (0.22) (0.09) (0.08) (0.19) (0.09) (0.26) (0.10)
Became a fundholder 0.02 20.03 0.02 20.06 20.12* 20.26 20.03 20.17 0.10
this year (0.10) (0.08) (0.2) (0.08) (0.07) (0.18) (0.08) (0.24) (0.10)
Year effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

R2 0.734 0.87 0.902 0.77 0.83 0.94 0.663 0.93 0.93

N 230 230 230 230 230 230 230 230 230

Robust standard errors in parentheses *P,0.1; **P,0.05; ***P,0.01,

391
392 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

elective admissions. This is quite a robust test as there are only four observations
per GP. The results presented in Table 6 show significant timing effects in
non-emergency admissions in those specialities where GPFHs have ability and
incentives to alter behaviour. In Gynaecology and Orthopaedics there is a
significant rise in admissions in the year before becoming a fundholder. In
Ophthalmology there is a similar rise, though the coefficient is not well defined. In
Surgery there is a significant fall in the year of becoming a fundholder. In the four
specialities in which GPs have less ability and fewer incentives to alter behaviour
(the right-hand panel) there are not significant timing effects. There is a fall in
admissions in the year of becoming a fundholder, which although not well defined
is interesting, since it may reflect changes in behaviour in this non-surgical
speciality (Integrated Medicine) if GPFHs are substituting non-hospital-based
treatment for hospital treatment. The same analysis repeated for emergency
admissions indicates little association between the timing of admissions and
fundholding status.14 These disaggregated analyses therefore confirm the patterns
in the aggregate data: the changes are in non-emergency admissions and occur in
those specialities where ex ante behaviour would be predicted to be changeable.

5. Discussion

In this paper we have examined the impact of the GP Fundholding scheme on

the behaviour of fundholders. We argue that the way the scheme was implemented
gave fundholders a financial incentive to increase non-emergency referrals to
hospital in the year prior to that in which they became fundholders, and to lower
them or to substitute emergency for non-emergency admissions thereafter. We use
a data set containing all admissions by 58 GPs over a 4-year period and focus on
admission to the two main hospitals used by these GPs. The size and panel
structure of the data enables us to control for observed and unobserved difference
between practices which became fundholders and those that did not, and allows us
to identify the association between the timing of fundholder status and admissions
behaviour. In line with the literature, we have focused on the effect of the
fundholding incentive structure on GP behaviour, not on the factors affecting
whether GPs joined the fundholding scheme.
We find that fundholders do appear to respond to financial incentives. They have
a rise in elective admissions prior to becoming a fundholder, and a smaller fall in
elective admissions immediately on assuming fundholding status. There is little
evidence that they substitute free emergency for elective hospital treatment. So
fundholders raise those admissions over which they have most control and which
determine their budgets in the year before they becoming fundholders and lower

14
Results available from authors.
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 393

them immediately afterwards. These results are robust to controlling for possible
endogeneity of fundholding. Holding constant for practice effects that may affect
the level of admissions, there is not a clear difference in the average admission
level of fundholders and other GPs.15 Our clear findings are in contrast to the
rather confused picture that has emerged to date from case studies and studies
unable to control for factors over and above fundholding which may affect GP
referral behaviour.
Our statistical analysis examines referrals to the two hospitals within the health
authority. The fall in referrals in the fundholding year might be the result of
fundholders switching away from the two local hospitals to others outside the
district. While fundholders do have a slightly lower referral rate to local hospitals
over the 4-year window (83.7% of all referrals by fundholders were to local
hospitals, compared to 84.4% of referrals by non-fundholders) there is no evidence
that the drop in referrals in the year of becoming a fundholder is accompanied by
an increase in referrals to other hospitals. On the contrary, both local and distant
NHS hospital referrals fall in the year practices become fundholders, and the
likelihood of being referred to a local hospital actually rises.
Our results are from the analysis of data from one district health authority out of
about 100 in the UK. This health authority is fairly typical and we observe a large
number of GPs and their behaviour over 4 years. Our finding that hospital
admissions increase in the preparatory year is consistent with studies that have
found that prescribing may be inflated in the year prior to fundholding and that
prescribing patterns change once the practice holds a budget. Our results are
generalisable to all fundholders to the extent that fundholders in other health
authorities had their budgets based on preparatory year activity.
Our results show clearly that fundholders have responded to financial incentives.
But in terms of welfare does this matter? In the absence of outcome data we
cannot establish welfare outcomes, but we can speculate as to what might be the
impact of the changes in behaviour we observe for patients and for doctors. First,
the increase in admissions the year prior to fundholding benefits the patients of
fundholders. Either more patients received treatment, or some patients got
treatment sooner than they otherwise would have. Waiting is one of the main ways
in which UK health care is rationed and waiting for treatment has a welfare cost
for patients (Propper, 1996). So this increase in admissions is beneficial to
fundholders’ patients.
Second, the rise in preparatory year referrals means that fundholders’ budgets

15
Early fundholders (those who became fundholders before wave 6) have lower rates of admission
for both emergency and non-emergency treatment than other GPs. Later fundholders have higher rates.
This result fits with earlier studies that have examined only early wave fundholders and found they are
different. It also illustrates the issue of selection and the problem of inference from studies of early
scheme joiners: as the scheme expanded the later entrants were more like the ‘average GP’.
394 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

are inflated upwards for the whole of the period that they are fundholders.16 This
represents a gain for either the fundholders’ patients and / or for the fundholders.
The extra money does not appear to have been spent on referrals, since elective
referrals do not rise in the year after entry into the fundholding scheme. But the
money could have been spent on better patient care in the form of more primary
care, or referrals to the private sector. It could be used to provide better premises
or to improve those services provided to patients directly by fundholding GPs.
There is evidence showing faster growth in out-of-hospital clinics among
fundholders than non-fundholders, suggesting that they may be substituting away
from in-patient care. There is, however, no evidence about whether this substitu-
tion led to more appropriate care for patients (LeGrand et al., 1998). The
fundholding scheme was certainly intended to change the way in which primary
care doctors treated their patient: if these shifts away from NHS hospital care are
not to treatments of lower value then fundholders’ patients have only gained from
the scheme. Alternatively, fundholders might simply be retaining the financial
surplus resulting from a higher budget and lower admissions for their own gain. In
this case the only gain to their patients is the once-off increase in admissions.
The third welfare effect arises because fundholders’ budgets are deducted from
the fixed total allocation given to the health authority to cover all the population in
its area. The larger budgets for fundholders therefore mean fewer funds for
non-fundholders and their patients. As the rise in referrals in the preparatory year
is not negligible (in the order of 7%), this represents a real shift of resources away
from non-fundholding to fundholding practices. This means less money per capita
for non-fundholding patients. But the welfare of these patients has not necessarily
fallen. Some commentators have suggested that one effect of fundholding has been
to improve the treatment for all patients. They argue that fundholders have been
able to use their superior purchasing power to make general improvements in
access to care and in the process of service delivery, which have benefited
non-fundholders’ patients (Glennerster et al., 1994). If this is true then even
though non-fundholders have less money per patient, the money they have will
buy their patients better care. On the other hand, a recent survey of the evidence
concluded that there is a so-called ‘two-tier’ system, with fundholders’ patients
having better access to hospital services (LeGrand et al., 1998). Research on
waiting times also suggests that fundholders patients have gained at the expense of
non-fundholder patients: fundholder patients in one area waited less time for
elective care than non-fundholder patients (Dowling, 1997). If similar effects are
happening in the area studied here, then is it likely that the loss in relative budgets
for non-fundholder patients reduced the welfare of these patients.

16
Although the scheme was disbanded (from 1999 onwards) following a change of government in the
UK, the scheme was introduced as a permanent change. It ran for 8 years, and so for early groups of
fundholders, the period of fundholding was several years.
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 395

In summary, our analysis shows that GPs have responded to the financial
incentives in the fundholding scheme. To some extent this may not be a problem:
the scheme was designed to alter incentives in order to improve the efficiency with
which health care is delivered. However, the scheme clearly has had unintended
equity consequences. Fundholding GPs have been able to increase their budgets
for hospital care by bringing referrals forward. Whether it also means there was
better care available for patients in these practices depends on how fundholders
used their additional funds. The cash constraints on the NHS mean this referral
behaviour leaves less money for the hospital care of patients not in fundholder
practices. Unless there have been positive spillovers from the hospital treatment of
fundholder patients to the treatment of non-fundholder patients, the welfare of this
latter group has fallen.

Acknowledgements

The authors gratefully acknowledge the support and co-operation of North West
Anglia Health Authority, and discussions with Simon Burgess. The project was
conducted while two of the authors were based in the School of Health, University
of East Anglia. The project was funded by a grant from the Anglia and Oxford
NHS Executive Health Services & Public Health Research Sub-Committee. Part of
Propper’s time was funded by the Leverhulme Trust under a grant given to the
CMPO. This paper is entirely a reflection of the views and interpretation of the
authors and does not in any way reflect the views of the NHS or health authority.

Appendix A

Table A.1
Admissions in the NWA CMDS data set
Number of
records
All records (all Finished Consultant Episodes) 349 887
All admissions (First episode Finished Consultant Episodes) 323 955
All admissions April 1 1993–March 31 1997 322 826
Admissions with Registered GP defined 303 218
Admissions to PH and KL 263 001
Admissions to PH and KL by NWA GPs 257 174
Admissions to hospitals other than PH and KL by NWA GPs 38 135
Admissions to PH and KL by non-NWA GPs, for NWA residents 5827
Admissions of NWA residents to hospitals other than PH and KL 2082
by non-NWA GPs
396 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

A.1. The characteristics of NWA

The district covers about 1200 square miles, and includes three urban centres,
Peterborough, King’s Lynn, and Wisbech, as well as a number of small towns and
substantial rural area. Population characteristics also vary considerably within the
district. Using the Jarman index as a measure of deprivation, some electoral wards
appear very deprived: seven wards fall within the 10% most deprived electoral
wards in England using the Jarman index, contributing to NWA’s position as the
least affluent area in the Anglia and Oxford Region. Comparing conditions in the
four areas wards in central Peterborough are very deprived, and the Peterborough
area overall is significantly more deprived than the other three areas. Housing
conditions are also worse in Peterborough than in the other areas: a survey
undertaken by local authorities in NWA found that 20% of all private sector
dwellings in Peterborough were ‘unfit’, compared with 13% in Breckland (which
covers the Swaffham area) and 4% in Fenland and King’s Lynn & West Norfolk.
One of the legacies of Peterborough’s status as a ‘new town’ is a relatively high
number of older people living alone, isolated from community support networks, a
factor that has been linked with rising emergency admissions in other areas.

A.2. The definition and construction of the variables

The NHS financial year runs for the twelve months from 1 April. Admissions
are recorded in the Contract Minimum Dataset (CMDS) only after a patient has
been discharged: this means that, in any one financial year, there will be records
for patients admitted before its start but discharged during that year, and patients
admitted but not recorded as discharged will not be included. We controlled for
this by excluding from the dataset patients discharged after 1 April 1993 but
admitted before that date, and by including records for patients who were
discharged after 31 March 1997 but admitted earlier.
Data in the CMDS is recorded by episodes (called Finished Consultant
Episodes), and any one admission might have associated with it a number of
different episodes, as a patient is passed between different hospital consultants. We
used only the records associated with first episodes to focus on admissions.

A.2.1. Independent variables

The information used to construct the variables was derived from the CMDS,
the census, and from other information held at the health authority. Median GP age
was constructed using the GP’s date of birth. The number of GPs in each practice
is derived from a daily count at each practice. Distance is calculated using the
logarithm of the distance (in kilometres) from the GP surgery to the nearest district
general hospital, calculated as straight-line from ordnance survey grid references
(derived from practice postcodes).
 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398 397

A.2.2. Dependent variables

Analyses in levels use the log of the number of (emergency or non-emergency)
admissions per practice, financial year, speciality, hospital and admission method.
Observations of zero counts were created where there were no admissions. This
was divided by list size where list size is the smallest of the four quarterly list
sizes for the year in question. Where analyses of the ratio of emergency to total
admissions is analysed the dependent variable is the ratio of admissions
(emergency or non-emergency depending on value of admission method) to total
admissions for that practice-year-speciality combination.

References

Coulter, A., Bradlow, J., 1993. Effect of NHS reforms on general practitioners’ referral patterns. British
Medical Journal 306, 433–437.
Dixon, J., Glennerster, H., 1995. What do we know about fundholding in general practice? British
Medical Journal 311, 727–730.
Dowling, B., 1997. Effect of fundholding on waiting times: database study. British Medical Journal
315, 290–292.
Fear, C.F., Cattell, H.R., 1994. Fund-holding general practices and old age psychiatry. Psychiatric
Bulletin 18, 263–265.
Glennerster, H., Matsaganis, M., Owens, P., Hancock, S., 1994. Implementing GP fundholding: Wild
card or winning hand? Oxford University Press, Buckingham.
Gosden, T., Torgerson, D.J., 1997. The effect of fundholding on prescribing and referral costs: a review
of the evidence. Health Policy 40, 103–114.
Healey, A., Reid, J.P., 1994. Do prospective fundholders inflate their prescribing costs? A study of
Grampian fundholding and non-fundholding practices. Health Bulletin 52, 282–284.
Heckman, J.J., Smith, J.A., Taber, C., 1996. What do bureaucrats do? The effect of performance
standards and bureaucratic preferences on acceptance into the JTPA program. NBER working paper
5535.
Hellinger, F., 1996. The impact of financial incentives on physician behaviour in managed care plans: a
review of the evidence. Medical Care Research Review 53, 294–314.
Howie, J.G.R., Heaney, D.J., Maxwell, M., 1994. Evaluating care of patients reporting pain in
fundholding practices. British Medical Journal 309, 705–710.
Hull, S.A., Jones, I.R., Moser, K., 1997. Factors influencing the attendance rate at accident and
emergency departments in East London: the contributions of practice organisation, population
characteristics and distance. Journal Health Service Research and Policy 2, 6–13.
LeGrand, J., Mays, N., Mulligan, J., 1998. Learning From the NHS Internal Market. A Review of the
Evidence, Kings Fund Publishing, London.
Lowy, A., Kohler, B., Nicholl, J., 1994. Attendance at accident and emergency departments:
unnecessary or inappropriate? Journal of Public Health Medicine 16, 134–140.
Martin, S., Smith, P., 1996. Explaining variations in inpatient length of stay in the UK National Health
Service. Journal of Health Economics 15 (3), 257–286.
Martin, S., Smith, P., 1999. Modelling waiting times for surgery. Journal of Public Economics 71,
141–164.
Newton, J. et al., 1993. Fundholding in Northern Region: the first year. British Medical Journal 306,
375–378.
Pauly, M.V., 1980. Doctors and Their Workshops. NBER Monograph, University of Chicago Press,
Chicago.
398 B. Croxson et al. / Journal of Public Economics 79 (2001) 375 – 398

Petchey, R., 1995. General practitioner fundholding: weighing the evidence. Lancet 346, 1139–1142.
Propper, C., 1996. The effect of competition on prices in the NHS internal market. Journal of Public
Economics 61, 307–335.
Roland, M.O., Coulter, A., 1993. Hospital Referrals, Oxford University Press, Oxford.
Scott, A., Shiell, A., King, M., 1996. Is general practitioner decision making associated with patient
socio-economic status? Social Science and Medicine 42, 35–46.
Social Trends, 1995. London: HMSO.
Surender, R., Bradlow, J., Coulter, A., Doll, H., Brown, S.S., 1995. Prospective study of trends in
referral patterns in fundholding and non-fundholding practices in the Oxford region, 1990–4. British
Medical Journal 311, 1205–1208.
Toth, B., Harvey, I., Peters, T., 1997. Did the introduction of general practice fundholding change
patterns of emergency admission to hospital? Journal Health Service Research and Policy 2, 71–74.
Whynes, D., Baines, D.L., Tolley, K.H., 1995. GP fundholding and the costs of prescribing. Journal of
Public Health Medicine 17, 323–329.
Wilson, J., 1989. Bureaucracy: What Government Agencies Do and Why They Do It, Basic Books,
New York.