Single Case Experimental Designs: Strategies For Studying Behavior Change (Barlow & Hersen)

~
ZJ 'JVL
b 1
1^
1
>
1 %T
STRATEGIES FOR STUDYING
BEHAVIOR CHANGE
SECOND EDITION
^
^
HHBBI I
Digitized by tine Internet Arciiive
in 2013
http://archive.org/details/vidhOOdavi
Single Case
Experimental Designs
(PGPS-56)
Pergamon Titles of Related Interest
Barlow/Hayes/Nelson THE SCIENTIST PRACTITIONER: Research

and Accountability in Clinical and Educational Settings
Bellack/Hersen RESEARCH METHODS IN CLINICAL PSYCHOLOGY
Hersen/Bellack BEHAVIORAL ASSESSMENT: A Practical
Handbook, Second Edition
Ollendlck/Hersen CHILD BEHAVIORAL ASSESSMENT: Principles
and Procedures
Related Journals'"
BEHAVIORAL ASSESSMENT
PERSONALITY AND INDIVIDUAL DIFFERENCES
Free specimen copies available upon request.

PERGAMON GENERAL PSYCHOLOGY SERIES
EDITORS
Arnold P. Goldstein, Syracuse University
Leonard Krasner, SUNY at Stony Brool<
Single Case Experimental

Designs
strategies for Studying
Behavior Change
Second Edition
David H. Barlow
SUNY at Albany
IVIichel Hersen
University of Pittsburgii School of Medicine
With invited chapters by
Donald P. Hartmann
University of Utah
and
Alan E. Kazdin
University of Pittsburgt) Sctiooi of IVIedicine
PERGAMON PRESS
NEW YORK OXFORD BEIJING FRANKFURT

SAO PAULO SYDNEY TOKYO TORONTO

U.S.A. Pergamon Press Inc., Maxwell House, Fairview Park,
Elmsford, New York 10523, U.S.A.
U.K. Pergamon Press pic, Headington Hill Hall,
Oxford 0X3 OBW, England
PEOPLE'S REPUBLIC Pergamon Press, Room 4037, Qianmen Hotel, Beijing,
OF CHINA People's Republic of China
FEDERAL REPUBLIC Pergamon Press GmbH, Hammerweg 6,
OF GERMANY D-6242 Kronberg, Federal Republic of Germany
BRAZIL Pergamon Editora Ltda, Rua Ega de Queiros, 346,
CEP 04011, Paraiso, Sao Paulo, Brazil
AUSTRALIA Pergamon Press Australia Pty Ltd., P.O. Box 544,
Potts Point, N.S.W. 2011, Australia
JAPAN Pergamon Press, 5th Floor, Matsuoka Central Building,

1-7-1 Nishishinjuku, Shinjuku-ku, Tokyo 160, Japan
CANADA Pergamon Press Canada Ltd., Suite No 271,

253 College Street, Toronto, Ontario, Canada M5T 1 R5
Copyright () 1984 Pergamon Press, Inc.
Library of Congress Cataloging in Publication Data
Barlow, David H.
Single case experimental designs, 2nd ed.
(Pergamon general psychology series)

Author's names in reverse order in 1st ed., 1976.
Includes bibliographies and indexes.
1. Psychology-Research. 2. Experimental design.
1. Hersen, Michel. II. Title. III. Series. [DNLM:
1. Behavior. 2. Psychology, Experimental. 3. Research
design. BF 76.5 H572s]
BF76.5.B384 1984 150'.724 84-6292
ISBN 0-08-030136-3
ISBN 0-08-030135-5 (soft)
A// Rights reserved.No part of this publication may be

reproduced, stored in a retrieval system or transmitted in
any form or by any means: electronic, electrostatic,
magnetic tape, mechanical, photocopying, recording or
otherwise, without permission in writing from the
publishers.
Printing: 456789 Year: 1234567890
Printed in the United States of America

Contents
Preface ix
Epigram xi
1. The Single-case in Basic and Applied Research: An Historical

Perspective 1
1.1. Introduction 1
1.2. Beginnings in Experimental Physiology and Psychology 2

1.3. Origins of theGroup Comparison Approach 5
1.4. Development of Applied Research: The Case Study
Method 8
1.5. Limitations of the Group Comparison Approach 14
1.6. Alternatives to the Group Comparison Approach 17
1.7. The Scientist-Practitioner Split 21
1.8. A Return to the Individual 23
1.9. The Experimental Analysis of Behavior 29
2. General Issues in a Single-case Approach 32
2.1. Introduction 32
2.2. Variability ^ 33
2.3. Experimental Analysis of Sources of Variability Through
Improvised Designs 39
2.4. Behavior Trends and Intrasubject Averaging 45
2.5. Relation of Variability to Generality of Findings 49
2.6. Generality of Findings 50
2.7. Limitations of Group Designs in EstabHshing Generality of
Findings 51
2.8. Homogeneous Groups Versus Replication of a Single-Case
Experiment 56
2.9. Applied Research Questions Requiring Alternative Designs 62
2.10. Blurring the Distinction Between Design Options 64
vi Contents
3. General Procedures in Single-case Research 67
3.2. Repeated Measurement 68
3.3. Choosing a Baseline 71
3.4. Changing One Variable at a Time 79
3.5. Reversal and Withdrawal 88
3.6. Length of Phases 95
3.7. Evaluation of Irreversible Procedures 101
3.8. Assessing Response Maintenance 105
4. Assessment Strategies 107

by Donald P. Hartman
4.2. Selecting Target Behaviors 109
4.3. Tracking the Target Behavior Using Repeated Measures 110
4.4. Other Assessment Techniques 131
5. Basic A-B-A Withdrawal Designs 140

5.2. A-B Design 142
5.3. A-B- A Design 152
5.4. A-B-A-B Design 157
5.5. B-A-B Design 166
5.6. A-B-C-B Design 170
6. Extensions of the A-B-A Design, Uses in Drug Evaluation, and

Interaction Design Strategies 174
6.1. Extensions and Variations of the A-B-A Withdrawal

Design 174
6.2. A-B-A-B-A-B Design 175
6.3. Comparing Separate Therapeutic Variables, or Treatments 177
6.4. Parametric Variations of the Basic Therapeutic Procedures
A-B-A-B '-B"-B'" Design 179
6.5. Drug Evaluations 183
6.6. Strategies for Studying Interaction Effects 193
6.7. Changing Criterion Design 205
7. Multiple Baseline Designs 209

7.2. Multiple Baseline Designs 210
7.3. Variations of Multiple Baseline Designs 244
7.4. Issues in Drug Evaluation 249
Contents vii
8. Alternating Treatments Design 252

8.2. Procedural Considerations 256
8.3. Examples of Alternating Treatments Designs 265
8.4. Advantages of the Alternating Treatments Design 280
8.5. Visual Analysis of the Alternating Treatments Design 281
8.6. Simultaneous Treatment Design 282
9. Statistical Analyses for Single-case Experimental Designs 285

by Alan E. Kazdin

9.2. Special Data Characteristics 287
9.3. The Role of Statistical Evaluation in Single-Case Research 290
9.4. Specific Statistical Tests 293
9.5. Time Series Analysis 296
9.6. Randomization Tests 302
9.7. The R Test of Ranks 308
9.8. The Split-Middle Technique 312
9.9. Evaluation of Statistical Tests: General Issues 319
9.10. Conclusions 321
10. Beyond the Individual: Replication Procedures 325

10.2. Direct Replication 326
10.3. Systematic Replication 347
10.4. Clinical Replication 366
10.5. Advantages of Replication of Single-Case Experiments 370
Hiawatha Designs an Experiment 372
References 374
Subject Index 405
Name Index 409
About the Authors 419

TO THE MEMORY OF
Frederic I. Barlow
and
to Members of the Hersen Family
who died in World War II
Preface
In the preface to the first edition of this book we said:
We do not expect this book to be the final statement on single-case designs. We

learned at least as much as we already knew in analyzing the variety of innovative
and creative applications of these designs to varying applied problems. The
unquestionable appropriateness of these designs in applied settings should ensure
additional design innovations in the future.
At the time, seemed a reasonable statement to make, but we think that

this
few of us involved in applied research anticipated the explosive growth of
interest in single-case designs and how many methodological and strategical
innovations would subsequently appear. As a result of developments in the 8
years since the first edition, this book can be more accurately described as new
than as revised. Fully 5 of the 10 chapters are new or have been completely
rewritten. The remaining five chapters have been substantially revised and
updated to reflect new guidelines and the current wisdom on experimental
strategies involving single-case designs.
Developments in the field have not been restricted to new or modified
experimental designs. New thinking has emerged on the analyses of data from
these designs, particularly with regard to use of statistical procedures. We
were most fortunate in having Alan Kazdin take into account these develop-
ments in the revision of his chapter on statistical analyses for single-case
experimental designs. Furthermore, the area of techniques of measurement
and assessment relevant to single-case designs has changed greatly in the years
since the first edition. Don Hartmann, the Editor of Behavioral Assessment
and one of the leading figures in assessment and single-case designs, has
strengthened the book considerably with his lucid chapter. Nevertheless, the
primary purpose of the book was, and remains, the provision of a source-
book of single-case designs, with guidelines for their use in applied settings.
To Sallie Morgan, who is very tired of typing the letters A-B-C over and
over again for the past 10 years, we can say that we couldn't have done it
without you, or without Mary Newell and Susan Capozzoli. Also, Susan
SCEI>-A
ix
X Preface
Cohen made a significant contribution in searching out the seemingly endless

articles on single-case designs that have accumulated over the years. And
Susan, as well as Janet Klosko and Janet Twomey, deserves credit for compil-
ing for what we hope is a useful index, a task for which they have developed
considerable expertise. Finally, this work really is the creation of the commu-
ways to alleviate human suffering and
nity of scientists dedicated to exploring
enhance human potential. These intellectual colleagues and forebears are now
too numerous to name, but we hope that this book serves our colleagues as
well as the next generation.
David H. Barlow
Albany,New York
Michel Hersen
Pittsburgh, Pennsylvania
Epigram
Conversation between Tolman and Allport
TOLMAN: "I know I should be more idiographic in my research, but I
just don't know how to be."

ALLPORT: "Let's learn!"
CHAPTER 1
The Single-case in Basic and Applied

Research: An Historical Perspective
1.1. INTRODUCTION
The individual is of paramount importance in the cUnical science of human
behavior change. Until recently, however, this science lacked an adequate
methodology for studying behavior change in individuals. This gap in our
methodology has retarded the development and evaluation of new procedures
in clinical psychology and psychiatry as well as in educational fields.
Historically, the intensive study of the individual held a preeminent place in
the fields of psychology and psychiatry. In spite of this background, an
adequate experimental methodology for studying the individual was very
slow to develop in applied research.* To find out why, it is useful to gain some
perspective on the historical development of methodology in the broad area
of psychological research.
The purpose of this chapter is to provide such a perspective, beginning with
the origins of methodology in the basic sciences of physiology and experimen-
tal psychology in the middle of the last century. Because most of this early
work was performed on individual organisms, reasons for the development
of between-group comparison methodology in basic research (which did not
occur until the turn of the century) are outlined. The rapid development of
inferential statistics and sampling theory during the early 20th century
enabled greater sophistication in the research methodology of experimental
psychology. The manner in which this affected research methods in applied
areas during the middle of the century is discussed.
*In this book applied research refers to experimentation in the area of human
behavior change relevant to the disciplines of clinical psychology, psychiatry, social
work, and education.
1
2 Single-case Experimental Designs
In the meantime, applied research was off to a shaky start in the offices of
early psychiatrists with a technique known as the case study method. The
separate development of applied research is traced from those early begin-
nings through the grand collaborative group comparison studies proposed in
the 1950s. The subsequent disenchantment with this approach in applied
research forced a search for alternatives. The rise and fall of the major
alternatives
process research and naturalistic studies
is outlined near the
end of the chapter. This disenchantment also set the stage for a renewal of
interest in the scientific study of the individual. The multiple origins of single-
case experimental designs in the laboratories of experimental psychology and
the offices of clinicians complete the chapter. Descriptions of single-case
designs and guidelines for their use as they are evolving in applied research
comprise the remainder of this book.
1.2. BEGINNINGS IN EXPERIMENTAL PHYSIOLOGY

AND PSYCHOLOGY
The scientific study of individual human behavior has roots deep in the
history of psychology and physiology. When psychology and physiology
became sciences, the initial experiments were performed on individual or-
ganisms, and the results of these pioneering endeavors remain relevant to the
world today. The science of physiology began in the 1830s, with
scientific
Johannes Miiller and Claude Bernard, but an important landmark for ap-
plied research was the work of Paul Broca in 1861. At this time, Broca was
caring for a man who was hospitalized for an inability to speak intelligibly.
Before the man died, Broca examined him carefully; subsequent to death, he
performed an autopsy. The finding of a lesion in the third frontal convolution
of the cerebral cortex convinced Broca, and eventually the rest of the scien-
tific world, that this was the speech center of the brain. Broca's method was
the clinical extension of the newly developed experimental methodology

called extirpation of parts introduced to physiology by Marshall Hall and
,
Pierre Flouren in the 1850s. In this method, brain function was mapped out
by systematically destroying parts of the brain in animals and noting the
effects on behavior.
The importance of this research in the context of the present discussion lies
in the demonstration that important findings with wide generality were

gleaned from single organisms. This methodology was to have a major
impact on the beginnings of experimental psychology.
Boring (1950) fixed the beginnings of experimental psychology in 1860,
with the publication of Fechner's Elemente der Psychophysik. Fechner is
most famous for developing measures of sensation through several psy-

chophysical methods. With these methods, Fechner was able to determine
sensory thresholds and just noticeable differences (JNDs) in various sense
The Single-case in Basic and Applied Research 3
modalities. What is common to these methods is the repeated measurement

of a response at different intensities or different locations of a given stimulus
in an individual subject. For example, when stimulating skin with two points
in a certain region to determine the minimal separation which the subject
reliably recognizes as two stimulations, one may use the method of constant
stimuli. In this method the two points repeatedly stimulate two areas of skin
at five to seven fixed separations, in random order, ranging from a few
millimeters apart to the relatively large separation of 10 mm. During each
stimulation, the subject reports whether he or she senses one point or two.
After repeated trials, the point at which the subject "notices" two separate
points can be determined. It is interesting to note that Fechner was one of the
firstto apply statistical methods to psychological problems. Fechner noticed
that judgments of just noticeable differences in the sensory modalities varied
somewhat from trial to trial. To quantify this variation, or "error" in judg-
ment, he borrowed the normal law of error and demonstrated that these
errors were normally distributed around a mean, which then became the
"true" sensory threshold. This use of descriptive statistics anticipated the
application of these procedures to groups of individuals at the turn of the
century, when traits or capabilities were also found to be normally distributed
around a mean. The emphasis on error, or the average response, raised issues
regarding imprecision of measurement that were to be highlighted in between-
group comparison approaches (see below and chapter 2). It should be noted,
however, that Fechner was concerned with variability within the subject, and
he continued his remarkable work on series of individuals.
These traditions in methodology were continued by Wilhelm Wundt.
Wundt's contributions, and those of his students and followers, most notably
Titchener, had an important impact on the field of psychology, but it is the
scientific methodology he and his students employed that most interests us.
To Wundt, the subject matter of psychology was immediate experience,

such as how a subject experiences light and sound. Since these experiences
were private events and could not be directly observed, Wundt created a new
method called introspection. Mention of the procedure may strike a respon-
sive chord in some modern-day clinicians, but in fact this methodology is
quite different from the introspection technique of free association and
others, often used in clinical settings to uncover repressed or unconscious
material. Nor did introspection bear any relation to armchair dreams or
reflections that are so frequent a part of experience. Introspection, as Wundt
employed was a highly specific and rigorous procedure that was used with
it,
individual subjects who were highly trained. This training involved learning
to describe experiences in an objective manner, free from emotional or
language restraints. For example, the experience of seeing a brightly colored
object would be described in terms of shapes and hues without recourse to
aesthetic appeal. To illustrate the objectivity of this system, introspection of
emotional experiences where scientific calm and objectivity might be dis-

rupted was not allowed. Introspection of this experience was to be done at a
later date when the scientific attitude returned. This method, then, became
retrospection, and the weaknesses of approach were accepted by Wundt
this
to preserve objectivity. Like Fechner's psychophysics, which is essentially an
introspectionist methodology, the emphasis hinges on the study of a highly
trained individual with the clear assumption, after some replication on other
individuals, that findings would have generality to the population of individu-
als. Wundt and his followers comprised a school of psychology known as the
Structuralist School, and many topics important to psychology were first
studied with this rather primitive but individually oriented form of scientific
analysis. The major subject matter, however, continued to be sensation and
perception. With Fechner's psychophysical methods, the groundwork for the
study of sensation and perception was laid. Perhaps because of these begin-
nings, a strong tradition of studying individual organisms has ensued in the
fields of sensation and perception and physiological psychology. This tradi-
tion has not extended to other areas of experimental psychology, such as
learning, or to the more clinical areas of investigation that are broadly based
on learning principles or theories. This course of events is surprising because
the efforts to study principles of learning comprise one of the more famous
examples of the scientific study of the single-case. This effort was made by
Hermann Ebbinghaus, one of the towering figures in the development of
psychology. With a belief in the scientific approach to psychology, and heavily
influenced by Fechner's methods (Boring, 1950), Ebbinghaus established
principles of human learning that remain basic to work in this area.
Basic to Ebbinghaus 's experiments was the invention of a new instrument
to measure learning and forgetting
the nonsense syllable. With a long list of
nonsense syllables and himself as the subject, he investigated the effects of
different variables (such as the amount of material to be remembered) on the
efficiency of memory. Perhaps his best known discovery was the retention
curve, which illustrated the process of forgetting over time. Chaplin and
Kraweic (1960) noted that he "worked so carefully that the resuhs of his
experiments have never been seriously questioned" (p. 180). But what is most
relevant and remarkable about his work is his emphasis on repeated measures
of performance in one individual over time (see chapter 4). As Boring (1950)
pointed out, Ebbinghaus made repetition the basis for the experimental
measurement of memory. It would be some 70 years before a new approach,
called the experimental analysis of behavior, was to employ repeated mea-
surement in individuals to study complex animal and human behaviors.
One of the best known scientists in the fields of physiology and psychology
during these early years was Pavlov (Pavlov, 1928). Although Pavlov consid-
ered himself a physiologist, his work on principles of association and learning
was his greatest contribution, and, along with his basic methodology, is so
well known that summaries are not required. What is often overlooked,
however, is that Pavlov's basic findings were gleaned from single organisms
and strengthened by replication on other organisms. In terms of scientific
yield, the study of the individual organism reached an early peak with Pavlov,
and Skimjer would later cite this approach as an important link and a strong
bond between himself and Pavlov (Skinner, 1966a).
1.3. ORIGINS OF THE GROUP COMPARISON

APPROACH
Important research in experimental psychology and physiology using single

cases did not stop with these efforts, but the turn of the century witnessed a
new development which would have a marked effect on basic and, at a later
date, applied research. This development was the discovery and measurement
of individual differences. The study of individual differences can be traced to
Adolphe Quetelet, a Belgian astronomer, who discovered that human traits
(e.g., height) followed the normal curve (Stilson, 1966). Quetelet interpreted
these findings to mean that nature strove to produce the "average" man but,
due to various reasons, failed, resulting in errors or variances in traits that
grouped around the average. As one moved further from this average, fewer
examples of the trait were evident, following the well-known normal distribu-
tion. This approach, in turn, had its origins in Darwin's observations on
individual variation within a species. Quetelet viewed these variations or
errors as unfortunate since he viewed the average man, which he termed
rhomme moyen, as a cherished goal rather than a descriptive fact of central
tendency. If nature were "striving" to produce the average man, but failed
due to various accidents, then the average, in this view, was obviously the
ideal. Where nature failed, however, man could pick up the pieces, account
for the errors, and estimate the average man through statistical techniques.
The influence of this finding on psychological research was enormous, as it
paved the way for the application of sophisticated statistical procedures to
psychological problems. Quetelet would probably be distressed to learn,
however, that his concept of the average individual would come under attack
during the 20th century by those who observed that there is no average
individual (e.g., Dunlap, 1932; Sidman, 1960).
This viewpoint notwithstanding, the study of individual differences and the
statistical approach to psychology became prominent during the first half of
the 20th century and changed the face of psychological research. With a push
from the American functional school of psychology and a developing interest
in the measurement and testing of intelligence, the foundation for comparing
groups of individuals was laid.
Gallon and Pearson expanded the study of individual differences at che

many of the descriptive statistics still in use
turn of the century and developed
today, most notably the notion of correlation, which led to factor analysis,
and significant advances in construction of intelligence tests first introduced
by Binet in 1905. At about this time, Pearson, along with Galton and
Weldon, founded the journal Biometricka with the purpose of advancing
quantitative research in biology and psychology. Many of the newly devised
statistical tests were first published there. Pearson was highly enthusiastic
about the statistical approach and seemed to believe, at times, that inaccurate
data could be made to yield accurate conclusions if the proper statistics were
applied (Boring, 1950). Although this view was rejected by more conservative
colleagues, it points up a confidence in the power of statistical procedures that
reappears from time to time in the execution of psychological research (e.g.,
D. A. Shapiro & Shapiro, 1983; M. L. Smith & Glass, 1977; G. T. Wilson &
Rachman, 1983).
One of the best known psychologists to adopt this approach was James
McKeen Cattell. Cattell, along with Farrand, devised a number of simple
mental tests that were administered to freshmen at Columbia University to
determine the range of individual differences. Cattell also devised the order
of merit method, whereby a number of judges would rank items or people on
a given quality, and the average response of the judges constituted the rank of
that item vis-a-vis other items. In this way, Cattell had 10 scientists rate a
number of eminent colleagues. The scientist with the highest score (on the
average) achieved the top rank.
It may seem
ironic at first glance that a concern with individual differences
led to an emphasis on groups and averages, but differences among individu-
als, or intersubject variability, and the distribution of these differences neces-
sitate a comparison among individuals and a concern for a description of a

group or population as a whole. In this context observations from a single
organism are irrelevant. Darwin, after all, was concerned with survival of a
species and not the survival of individual organisms.
The invention of many of the descriptive statistics and some crude statisti-
cal tests of comparison made it easier to compare performance in large
groups of subjects. From 1900 to 1930, much of the research in experimental
psychology, particularly learning, took advantage of these statistics to com-
pare groups of subjects (usually rats) on various performance tests (e.g., see
Birney & Teevan, 1961). Crude statistics that could attribute differences
between groups to something other than chance began to appear, such as the
critical ratio test (Walker & Lev, 1953). The idea that the variability or error
among organisms could be accounted for or averaged out in large groups was
a commonsense notion emanating from the new emphasis on variability
among organisms. The fact that this research resulted in an average finding
from the hypothetical average rat drew some isolated criticism.For instance,
in 1932, while reviewing research in experimental psychology, Dunlap pointed

out that there was no average and Lewin (1933) noted that "... the only
rat,
situations which should be grouped for statistical treatment are those which
have for the individual rats or for the individual children the same psycholog-
ical structure and only for such period of time as this structure exists" (p.
328). The new emphasis on variability and averages, however, would have
pleased Quetelet, whose slogan could have been "Average is Beautiful."
The influence of inferential statistics
During the 1930s, the work of R. A. Fisher, which subsequently exerted

considerable influence on psychological research, first appeared. Most of the
sophisticated statistical procedures in use today for comparing groups were
invented by Fisher. It would be difficult to pick up psychological or psychia-
tric journals concerned with behavior change and not find research data
analyzed by the ubiquitous analysis of variance. It is interesting, however, to
consider the origin of these tests. who was a
Early in his career, Fisher,
mathematician interested in genetics, made an important decision. Faced
with pursuing a career at a biometrics laboratory, he chose instead a relatively
obscure agricultural station on the grounds that this position would offer him
more opportunity for independent research. This personal decision at the
very least changed the language of experimental design in psychological
research, introducing agricultural terms to describe relevant designs and
While Fisher's statistical
variables (e.g., split plot analysis of variance).
innovations were one of the more important developments of the century for
psychology, the philosophy underlying the use of these procedures is clearly in
line with Quetelet 's notion of the importance of the average. As a good
agronomist, Fisher was concerned with the yield from a given area of land
under various soil treatments, plant varieties, or other agricultural variables.
Much as in the study of individual differences, the fate of the individual plant
is irrelevant in the context of the yield from the group of plants in that area.
Agricultural variables are important to the farm and society if the yield is
better on the average than a similar plot treated differently. The implications
of this philosophy for applied research will be discussed in chapter 2.
The work of Fisher was not limited to the invention of sophisticated

statistical tests. An equally important contribution was the consideration of
the problem of induction or inference. Essentially, this issue concerns general-
ity of findings. If some data from a group or a plot of land, this
are obtained
information is not very valuable if it is group
relevant only to that particular
or plot of land because similar data must be collected from each new plot.
Fisher (1925) worked out the properties of statistical tests, which made it
possible to estimate the relevance of data from one small group with certain
characteristics to the universe of individuals with those characteristics. In
other words, inference is made from the sample to the population. This work
and the subsequent developments in the field of sampling theory made it
possible to talk in terms of psychological principles with broad generality and
applicability
a primary goal in any science. This type of estimation, how-
ever, was based on appropriate statistics, averages, and intersubject variabil-
ity in the sample, which further reinforced the group comparison approach in
basic research.
As the science of psychology grew out of its infancy, its methodology was
largely determined by the broad generality of findings made possible

lure of
through the brillant work of Fisher and his followers. Because of the empha-
sis on averages and intersubject variability required by this design in order to
make general statements, the intensive study of the single organism, so

popular in the early history of psychology, fell out of favor. By the 1950s,
when investigators began to consider the possibility of doing serious research
in applied settings, the group comparison approach was so entrenched that
anyone studying single organisms was considered something of an oddity by
no less an authority than Underwood (1957). The Zeitgeist in psychological
research was group comparison and statistical estimation. While an occa-
sional paper was published during the 1950s defending the study of the single-
case (S. J. Beck, 1953; Rosenzweig, 1951), or at least pointing out its place in
psychological research (duMas, 1955), very little basic research was carried
out on single-cases. A notable exception was the work of B. F. Skinner and his
students and colleagues, who were busy developing an approach known as
the experimental analysis of behavior, or operant conditioning. This work,
however, did not have a large impact on methodology in other areas of
psychology during the 1950s, and applied research was just beginning.
Against this background, it is not surprising that applied researchers in the
1950s employed the group comparison approach, despite the fact that the
origins of the study of clinically relevant phenomena were quite differen.t
from the origin of more basic research described above.
1.4. DEVELOPMENT OF APPLIED RESEARCH:

THE CASE STUDY METHOD
As the sciences of physiology and psychology were developing during the

late 19th and 20th centuries, people were suffering from emotional and
behavioral problems and were receiving treatment. Occasionally, patients
recovered, and therapists would carefully document their procedures and
communicate them to colleagues. Hypotheses attributing success or failure to
various assumed causes emanated from these cases, and these hypotheses
gradually grew into theories of psychotherapy Theories proliferated, and
procedures based on observations of cases and inferences from these theories

grew in number. As Paul (1969) noted, those theories or procedures that
could be communicated clearly or that presented new and exciting principles
tended to attract followers to the organization, and schools of psychotherapy
were formed. At the heart of this process is the case study method of
investigation (Bolger, 1965). This method (and its extensions) was, with few
exceptions, the sole methodology of clinical investigation through the first
half of the 20th century.

The case study method, of course, is the clinical base for the experimental
study of single-cases and, as such, an important function in present-
it retains
day applied research (Barlow, 1980; Barlow, Hayes, & Nelson, 1983; Kazdin,
1981) (see section 1.7). Unfortunately, during this period clinicians were
unaware, for the most part, of the basic principles of applied research, such
as definition of variables and manipulation of independent variables. Thus it
is noteworthy from an historical point of view that several case studies
reported during this period came tantalizingly close to providing the basic
scientific ingredients of experimental single-case research. The most famous
of these, of course, is the J. B. Watson and Rayner (1920) study of an
analogue of phobia in a young boy, where a prototype of a with-

clinical
drawal design was attempted (see chapter 5). These investigators unfortu-
nately suffered the fate of many modern-day clinical researchers in that the
subject moved away before the "reversal" was complete.
Anytime that a treatment produced demonstrable effects on an observable
behavior disorder, the potential for scientific investigation was there. An
excellent example, among many, was Breuer's classic description of the treat-
ment of hysterical symptoms in Anna O. through psychoanalysis in 1895
(Breuer & Freud, 1957). In a series of treatment sessions, Breuer dealt with
one symptom at a time through hypnosis and subsequent "talking through,"
where each symptom was traced back to its hypothetical causation in circum-
stances surrounding the death of her father. One at a time, these behaviors
disappeared, but only when treatment was administered to each respective
behavior. This process of treating one behavior at a time fulfills the basic
requirement for a multiple baseline experimental design described in chapter
7, and the clearly observable success indicated that Breuer 's treatment was
effective. Of course, Breuer did not define his independent variables, in that
there were several components to his treatment (e.g., hypnosis, interpreta-
tion); but, in the manner of a good scientist as well as a good clinician, Breuer
admitted that he did not know which component or components of his
treatment were responsible for success. He noted at least two possibilities, the
suggestion inherent in the hypnosis or the interpretation. He then described
events discovered through his talking therapy as possibly having etiological
significance and wondered about the reliability of the girl's report as he
hypothesized various etiologies for the symptoms. However, he did not, at the
time, firmly link successful treatment with the necessity of discovering the
etiology of the behavior disorder. One wonders if the early development of
clinical techniques, including psychoanalysis, would have been different if
careful observers like Breuer had been cognizant of the experimental implica-
tions of their clinical work. Of course, this small leap from uncontrolled case
study to scientific investigation of the single case did not occur because of a
lack of awareness of basic scientific principles in early clinicians. The result
was an accumulation of successful individuals' case studies, with clinicians
from varying schools claiming that their techniques were indispensable to
success. In many cases their claims were grossly exaggerated. Brill noted in
1909 on psychoanalysis that "The results obtained by the treatment are
unquestionably very gratifying. They surpass those obtained by simpler
methods in two chief respects; namely, in permanence and in the prophylactic
value they have for the future" (Brill, 1909). Much later, in 1935, Kessel and
Hyman observed, "this patient was saved from an inferno and we are
convinced that this could have been achieved by no other method" (Kessel &
Hyman, 1933). From an early behavioral standpoint. Max (1935) noted the
electrical aversion therapy produced "95 percent relief" from the compulsion
of homosexuality.
These kinds of statements did little to endear the case study method to
serious applied researcherswhen they began to appear in the 1940s and 1950s.
In fact, the case study method, if anything, deteriorated somewhat over the
years in terms of the amount and nature of publicly observable data available
in these reports. Frank (1961) noted the difficulty in even collecting data from
a therapeutic hour in the 1930s due to lack of necessary equipment, reluc-
tance to take detailed notes, and concern about confidentiality. The advent of
the phonograph record at this time made it possible at least to collect raw data
from those clinicians who would cooperate, but this method did not lead to
any fruitful new ideas on research. With the advent of serious applied
research in the 1950s, investigators tended to reject reports from uncontrolled
case studies due to an inabilij)rtaiialuatjJi^fects^gfJj:eatment. Given the
extraordinary claims by clinicians after successful case studies, this attitude is
understandable. However, from the viewpoint of single-case experimental
designs, this rejection of the careful observation of behavior change in a case
report had the effect of throwing out the baby with the bathwater.
Percentage of success in treated groups
A further development in applied research was the reporting of collections

of case studies in terms of percentage of success. Many of these reports have
been cited by Eysenck (1952). However, reporting of results in this manner
probably did more harm than good to the evaluation of clinical treatment. As
Paul (1969) noted, independent and dependent variables were no better
1
defined than in most case reports, and techniques tended to be fixed and
"school" oriented. Because all procedures achieved some success, practi-
tioners within these schools concentrated on the positive results, explained
away the failures, and decided that the overall results confirmed that their
procedures, as applied, were responsible for the success. Due to the strong
and overriding theories central to each school, the successes obtained were
attributed to theoretical constructs underlying the procedure. This precluded
a careful analysis of elements in the procedure or the therapeutic intervention
that many have been responsible for certain changes in a given case and had
the effect of reinforcing the application of a global, ill-defined treatment
from whatever theoretical orientation, to global definitions of behavior disor-
ders, such as neurosis. This, in turn, led to statements such as "psy-
chotherapy works with neurotics." Although applied researchers later
rejected these efforts as unscientific, one carryover from this approach was
the notion of the average response to treatment; that is, if a global treatment
is successful on the average with a group of "neurotics," then this treatment
will probably be successful with any individual neurotic who requests treat-
ment.
Intuitively, of course, descriptions of results from 50 cases provide a more
convincing demonstration of the effectiveness of a given technique than
A modification of this approach
separate descriptions of 50 individual cases.
utilizing updated and procedures and with the focus on individual
strategies
responses has been termed clinical replication. This strategy can make a
substantial contribution to the applied research process (see chapter 10). The
major difficulty with this approach, however, particularly as itwas practiced
in early years, is that the category in which these clients are classified most
always becomes unmanageably heterogeneous. The neurotics described in
Eysenck's (1952) paper may have less in common than any group of people
one would choose randomly. When cases are described individually, however,
a clinician stands a better chance of gleaning some important information,
since specific problems and specific procedures are usually described in more
detail. When one lumps cases together in broadly defined categories, individ-
ual case descriptions are lost and the ensuing report of percentage success
becomes meaningless. This unavoidable heterogeneity in any group of pa-
tients is an important consideration that will be discussed in more detail in
this chapter and in chapter 2.
Group comparison approach in applied research
By the late 1940s, clinical psychology and, to a lesser extent, psychiatry

began to produce the type of clinician who was also aware of basic research
strategies. These scientists were quick to point out the drawbacks of both the
case study and reports of percentages of success in groups in evaluating the
effects of psychotherapy. They noted that any adequate test of psychotherapy

would have to include a more precise definition of terms, particularly out-
come criteria or dependent variables (e.g., Knight, 1941). Most of these
applied researchers were trained as psychologists, and in psychology a new
emphasis was placed on the "scientist-practitioner" model (Barlow et al.,
1983). Thus, the source of research methodology in the newly developing
areas of applied research came from experimental psychology. By this time,
the predominant methodology in experimental psychology was the between-
subjects group design.
The group design also was a logical extension of the earlier clinical reports
of percentage success in a large group of patients, because the most obvious
criticism of this endeavor is the absence of a control group of untreated
patients. The appearance of Eysenck's (1952) notorious article comparing
percentage success of psychotherapy in large groups to rates of "sponta-
neous" remission gleaned from discharge rates at state hospitals and insur-
ance company records had two effects. Firsts it reinforced the growing
conviction that the effects of psychotherapy could not be evaluated from case
reports or "percentage success groups" and sparked a new flurry of interest in
evaluating psychotherapy through the scientific method. Second, the empha-
sis on comparison between groups and quasi-control groups in Eysenck's
review strengthened the notion that the logical way to evaluate psychotherapy
was through the prevailing methodology in experimental psychology the
between-groups comparison designs.
This approach to applied research did not suddenly begin in the 1950s,
although interest certainly increased at this time. Scattered examples of
research with clinically relevant problems can be found in earlier decades.
One interesting example is a study reported by Kantorovich (1928), who
applied aversion therapy to one group of twenty alcoholics in Russia and
compared results to a control group receiving hypnosis or medication. The
success of this treatment (and the direct derivation from Pavlov's work) most
likely ensured a prominent place for aversion therapy in Russian treatment
programs for alcoholics. Some of the larger group comparison studies typical
of the 1950s also began before Eysenck's celebrated paper. One of the best
known is the Cambridge-Somerville youth study, which was reported in 1951
(Powers & Witmer, 1951) but was actually begun in 1937. Although this was
an early study, it is quite representative of the later group comparison studies
in that many of the difficuhies in execution and analysis of results were
repeated again and again as these studies accumulated.
The major difficulty, of course, was that these studies did not prove that
psychotherapy worked. In the Cambridge-Somerville study, despite the ad-
vantages of a well-designed experiment, the discouraging finding was that
"counseling" for delinquents or potential delinquents had no significant

effect when compared to a well-matched control group.
When this finding was repeated in subsequent studies (e.g., Barron &
Leary, 1955), the controversy over Eysenck's assertion on the ineffectiveness
of psychotherapy became heated. Most clinicians rejected the findings out-
right because they were convinced that psychotherapy was useful, while
scientists such as Eysenck hardened their convictions that psychotherapy was
at best ineffective and at worst some kind of great hoax perpetrated on
unsuspecting clients. This controversy, in turn, left serious applied researchers
groping for answers to difficult how to even
methodological questions on
approach the issue of evaluating effectiveness in psychotherapy. As a result,
major conferences on research in psychotherapy were called to discuss these
questions (e.g., Rubenstein & Parloff, 1959). It was not until Bergin reex-
amined these studies in a very important article (Bergin, 1966; see also Bergin
& Lambert, 1978) that some of the discrepancies between clinical evidence
from uncontrolled case studies and experimental evidence from between-
subject group comparison designs were clarified. Bergin noted that some
clients were improving in these studies, but others were getting worse. When
subjected to statistical averaging of results, these effects canceled each other
out, yielding an overall result of no effect when compared to the control
group. Furthermore, Bergin pointed out that these therapeutic effects had
been described in the original articles, but only as afterthoughts to the major
statistical findings of no effect. Reviewers such as Eysenck, approaching the
results from a methodological point of view, concentrated on the statistical
findings. These studies did not, however, prove that psychotherapy was
ineffective for a given individual. What these results demonstrated is that
people, particularly clients with emotional or behavioral disorders, are quite
different from each other. Thus attempts to apply an ill-defined and global
treatment such as psychotherapy to a heterogeneous group of clients classified
under a vague diagnostic category such as neurosis are incapable of answer-
ing the more basic question on the effectiveness of a specific treatment for a
specific individual.
The conclusion that psychotherapy was ineffective was premature, based
on this reanalysis, but the overriding conclusion from Bergin's review was
that "Is psychotherapy effective?" was the wrong question to ask in the first
place, even when appropriate between-group experimental designs were em-
ployed. During the 1960s, scientists (e.g., Paul 1967) began to realize that any
test of a global treatment such as psychotherapy would not be fruitful and
that clinical researchers must start defining the independent variables more
precisely and must ask the question: "What specific treatment is effective with
a specific type of client under what circumstances?"
1.5. LIMITATIONS OF THE GROUP COMPARISON

APPROACH
The clearer definition of variables and the call for experimental questions
that were precise enough to be answered were major advances in applied
research. The extensive review of psychotherapy research by Bergin and
Strupp (1972), however, demonstrated that even under these more favorable
conditions, the application of the group comparison design to applied prob-
lems posed many difficulties. These difficulties, or objections, which tend to
limit the usefulness of a group comparison approach in applied research, can
be classified under five headings: (1) ethical objections, (2) practical problems
in collecting large numbers of patients, (3) averaging of results over the
group, (4) generality of findings, and (5) intersubject variability.
Ethical objections
An oft-cited issue, usually voiced by clinicians, is the ethical problem

inherent in withholding treatment from a no-treatment control group. This
notion, of course, is based on the assumption that the therapeutic interven-
tion, in fact, works, in which case there would be little need to test it at all.
Despite the seeming illogic of this ethical objection, in practice many clini-
cians and other professional personnel react with distaste to withholding

some treatment, however inadequate, from a group of clients who are under-
going significant human suffering. This attitude is reinforced by scattered
examples of experiments where control groups did endure substantial harm
during the course of the research, particularly in some pharmacological
experiments.
Practical problems
On a more practical level, the collection of large numbers of clients

homogeneous for a particular behavior disorder is often a very difficult task.
In basic research in experimental psychology most subjects are animals (or
college sophomores), where matching of relevant behaviors or background
variables such as personality characteristics is feasible. When dealing with
severe behavior disorders, however, obtaining sufficient clients suitably
matched to constitute the required groups in the study is often impossible. As
Isaac Marks, who is well known for his applied research with large groups,
noted:
Having selected the technique to be studied, another difficulty arises in assem-

bling a homogeneous sample of patients. In uncommon disorders this is only
possible in centers to which large numbers of patients are regularly referred.
from these a tiny number are suitable for inclusion in the homogeneous sample
one wishes to study. Selection of the sample can be so time consuming that it
severely limits research possibilities. Consider the clinician who wishes to assem-
ble a series of obsessive-compulsive patients to be assigned at random into one of
two treatment conditions. He will need at least 20 such cases for a start, but
obsessive-compulsive neuroses (not personality) make up only 0.5-3 percent of
the psychiatric outpatients in Britain and the USA. This means the clinician will
need a starting population of about 2000 cases to sift from before he can find his
sample, and even then this assumes that all his colleagues are referring every
suitable patient to him. In practice, at a large center such as the Maudsley
Hospital, it would take up to two years to accumulate a series of obsessive
compulsives for study (Bergin & Strupp, 1972, p. 130).
To Marks 's credit, he has successfully undertaken this arduous venture on

several occasions (Marks, 1972, 1981), but the practical difficulties in execut-
enormous clinical facility at
ing this type of research in settings other than the
the Maudsley are apparent.
Even if this approach is possible in some large clinical settings, or in state
hospital settings where one might study various aspects of schizophrenia, the
related economic considerations are also inhibiting. Activities such as gather-
ing and analyzing data, following patients, paying experimental therapists,
and on and on require large commitments of research funds, which are often
unavailable.
Recognizing the practical limitations on conducting group comparison
studies in one setting, Bergin and Strupp set an initial goal in their review of
the state of psychotherapy research of exploring the feasibility of large
collaborative studies among various research centers. One advantage, at
least,was the potential to pool adequate numbers of patients to provide the
necessary matching of groups. Their reluctant conclusion was that this type
of large collaborative study was not feasible due to differing individual styles
among researchers and the extraordinary problems involved in administering
such an endeavor (Bergin & Strupp, 1972). Since that time there has been the
occasional attempt to conduct large collaborative studies, most notably the
recent National Institute of Mental Health study testing the effectiveness of
cognitive behavioral treatment of depression (NIMH, 1980). But the extreme
expense and many of the administrative problems foreseen by Bergin and
Strupp (1972) seem to ensure that these efforts will be few and far between
(Barlow et al., 1983).
Averaging of results
A third difficulty noted by many applied researchers is the obscuring of

outcome in group averages. This issue was cogently raised
individual clinical
by Sidman (1960) and Chassan (1967, 1979) and repeatedly finds its way into
the informal discussions with leading researchers conducted by Bergin and

Strupp and published in their book, Changing Frontiers in the Science of
Psychotherapy (1972). Bergin's (1966) review of large-outcome studies where
some clients improved and others worsened highlighted this problem. As
noted earlier, a move away from tests of global treatments of ill-defined
variables with the implicit question "Is psychotherapy effective?" was a step
But even when specific questions on effects of therapy
in the right direction.
in homogeneous groups are approached from the group comparison point of
view, the problem of obscuring important findings remains because of the
enormous complexities of any individual patient included in a given treat-
ment group. The fact that patients are seldom truly "homogeneous" has been
described by Kiesler (1966) in his discussion of the patient uniformity myth.
To take Marks's example, 10 patients, homogeneous for obsessive-compulsive
neurosis, may bring entirely different histories, personality variables, and
environmental situations to the treatment setting and will respond in varying
ways to treatment. That is, some improve and others will not.
patients will
The average response, however, will not represent the performance of any
individual in the group. In relation to this problem, Bergin (Bergin & Strupp,
1972) noted that he consulted a prominent statistician about a therapy
research project who dissuaded him from employing the usual inferential
statistics applied to the group as a whole and suggested instead that individual
curves or descriptive analyses of small groups of highly homogeneous pa-

tients might be more fruitful.
Generality of findings
Averaging and the complexity of individual patients also bring up some

related problems. Because resultsfrom group studies do not reflect changes in
individual patients, these findings are not readily translatable or generalizable
Chassan (1967) pointed out, the clinician
to the practicing clinician since, as
cannot determine which particular patient characteristics are correlated with
improvement. In ignorance of the responses of individual patients to treat-
know to what extent a given patient is similar to
ment, the clinician does not
patients who improved or perhaps deteriorated within the context of an
overall group improvement. Furthermore, as groups become more homoge-
neous, which most researchers agree is a necessary condition to answer
specific questions about effects of therapy, one loses the ability to make
inferential statements to the population of patients with a particular disorder
because the individual complexities in the population will not have been
adequately sampled. Thus it becomes difficult to generalize findings at all
beyond the specific group of patients in the experiment. These issues of

averaging and generality of findings will be discussed in greater detail in
chapter 2.
Intersubject variability
A final issue bothersome to clinicians and applied researchers is variability.
Between-subject group comparison designs consider only variability between

subjects as a method of dealing with the enormous differences among indi-
viduals in a group. Progress is usually assessed only once (in a posttest). This
large intersubject variability is often responsible for the "weak" effect ob-
tained in these studies, where some
clients show considerable improvement
and others deteriorate, and the average improvement is statistically significant
but clinically weak. Ignored in these studies is within-subject variability or the
clinical course of a specific patient during treatment, which is of great
practical interest to clinicians. This issue will also be discussed more fully in
chapter 2.
1.6. ALTERNATIVES TO THE GROUP COMPARISON

APPROACH
Many of these practical and methodological difficulties seemed overwhelm-
ing to clinicians and applied researchers. Some investigators wondered if
serious, meaningful research on evaluation of psychotherapy was even possi-
ble (e.g., Hyman & Berger 1966), and the gap between clinician and scientist
widened. One difficulty here was the restriction placed on the type of method-
ology and experimental design applicable to applied research. For many
scientists, a group comparison design was the only methodology capable of
yielding important information in psychotherapy studies. In view of the

dearth of alternatives available and against the background of case study and
"percentage success" efforts, these high standards were understandable and
correct. Since there were no clearly acceptable scientific alternatives, however,
applied researchers failed to distinguish between those situations where group
comparison designs were practical, desirable, and necessary (see section 2.9)
and situations where the development of alternative methodology was re-
quired. During the 1950s and 1960s, several alternatives were tested.
Many applied researchers reacted to the difficulties of the group compari-
son approach with a "flight into process" where components of the thera-
peutic process, such as relationship variables, were carefully studied (Hoch &
Zubin, 1964). A second approach, favored by many clinicians, was the
"naturalistic study," which was very close to actual clinical practice but had
dubious scientific underpinnings. As Kiesler (1971) noted, these approaches
are quite closely related because both are based on correlational methods,
where dependent variables are correlated with therapist or patient variables
either within therapy or atsome point after therapy. This is distinguished
from the experimental approach, where independent variables are systemati-
cally manipulated.
Naturalistic studies
The advantage of the naturalistic study for most clinicians was that it did
little to disrupt the typical activities engaged in
by clinicians in day-to-day
practice. Unlike with the experimental group comparison design, clinicians
were not restricted by precise definitions of an independent variable (treat-
ment, time limitation, or random assignment of patients to groups). Kiesler
(1971) noted that naturalistic studies involve "... live, unaltered, minimally
controlled, unmanipulated ^natural' psychotherapy sequences so-called ex-
periments of nature" (p. 54). Naturally this approach had great appeal to
clinicians for it dealt directly with their activities and, in doing so, promised
to consider the complexities inherent in treatment. Typically, measures of
multiple therapist and patient behaviors are taken, so that all relevant vari-
ables (based on a given clinician's conceptualization of which variables are
relevant) may be examined for interrelationships with every other variable.
Perhaps the best known example of this type of study is the project at the
Menninger Foundation (Kernberg, 1973). Begun in 1954, this was truly a
mammoth undertaking involving 38 investigators, 10 consultants, three dif-
ferent project leaders, and 18 years of planning and data collection. Forty-
two patients were studied in this project. Thisgroup was broadly defined,
although overtly psychotic patients were excluded. Assignment of patient to
therapist and to differing modes of psychoanalytic treatment was not random
but based on clinical judgments of which therapist or mode of treatment was
most suitable for the patient. In other words, the procedures were those
normally in effect in a clinical setting. In addition, other treatments, such as
pharmacological or organic interventions, were administered to certain pa-
tients as needed. Against this background, the investigators measured multi-
ple patient characteristics (such as various components of ego strength) and
correlated these variables, measured periodically throughout treatment by
referring to detailed records of treatment sessions, with multiple therapeutic
activities and modes of treatment. As one would expect, the results are
enormously complex and contain many seemingly contradictory findings. At
least one observer (Malan, 1973) noted that the most important finding is that
purely supportive treatment is ineffective with borderline psychotics, but
working through of the transference relationship under hospitalization with
this group is effective. Notwithstanding the global definition of treatment and
the broad diagnostic categories (borderline psychotic) also present in early
group comparison studies, this report was generally hailed as an extremely
important breakthrough in psychotherapy research. Methodologists, how-
ever, were not so sure. While admitting the benefits of a clearer definition of
psychoanalytic terms emanating from the project. May (1973) wondered

about the power and significance of the conclusions. Most of this criticism

concerns the purported strength of the naturalistic study that is, the lack of
control over factors in the naturalistic setting. If subjects are assigned to

treatments based on certain characteristics, were these characteristics respon-
sible for improvement rather than the treatment? What is the contribution of
additional treatments received by certain patients? Did nurses and other
therapists possibly react differently to patients in one group or another?
What was the contribution of "spontaneous remission"?
In its pure state, the naturalistic study does not advance much beyond the
uncontrolled case study in the power to isolate the effectiveness of a given
treatment, as severe critics of the procedure point out (e.g., Bergin & Strupp,
1972), but this process an improvement over case studies or reports of
is
"percentage success" in groups because measures of relevant variables are

constructed and administered, sometimes repeatedly. However, to increase
confidence in any correlational findings from naturalistic studies, it would
seem necessary to undermine the stated strengths of the study that is, the
"unaltered, minimally controlled, unmanipulated" condition prevaiHng in
the typical naturalistic project
by randomly assigning patients, limiting
access to additional confounding modes of treatment, and observing devia-
tion of therapists from prescribed treatment forms. But if this were done, the
study would no longer be naturalistic.
A further problem is obvious from the example of the Menninger project.
The practical difficulties in executing this type of study seem very little less
than those inherent in the large group comparison approach. The one excep-
tion is that the naturalistic study, in retaining close ties to the actual function-
numbers of
ing of the clinic, requires less structuring or manipulating of large
patients and therapists. The fact that this project took 18 years to complete
makes one consider the significant administrative problem inherent in main-
taining a research effort for this length of time. This factor is most likely
responsible for the admission from one prominent member of the Menninger
team, Robert S. Wallerstein, that he would not undertake such a project
again (Bergin & Strupp, 1972). Most seem to have heeded his advice because
few, if any, naturalistic studies have appeared in recent years.
Correlational studies, of course, do not have to be quite so "naturalistic"
as the Menninger study (Kazdin, 1980a; Kendall & Butcher, 1982). Kiesler
(1971) reviewed a number of studies without experimental manipulation that
contain adequate definitions of variables and experimental attempts to rule
out obvious confounding factors. Under such conditions, and if practically
feasible, correlational studies may expose heretofore unrecognized relation-
ships among variables in the psychotherapeutic process. But the fact remains
that correlational studies by their nature are incapable of determining causal
relationships on the effects of treatment. As Kiesler pointed out, the most
common error in these studies is the tendency to conclude that a relationship
between two variables indicates that one variable is causing the other. For
instance, the conclusion in the Menninger study that working through trans-
ference relationships is an effective treatment for borderline psychotics (as-
suming other confounding factors were controlled or randomized) is open to

several different interpretations. One might alternatively conclude that cer-
tain behaviors subsumed under the classification borderline psychotic caused
the therapist to behave in such a way that transference variables changed or
that a third variable, such as increased therapeutic attention during this more
directive approach, was responsible for changes.
Process research
The second alternative to between-group comparison research was the

process approach so often referred to in the APA conferences on psy-
chotherapy research (e.g., Strupp & Luborsky, 1962). Hoch and Zubin*s
(1964) popular phrase "flight into process" was an accurate description of the
reaction of many clinical investigators to the practical and methodological
difficulties of the large group studies. Typically, process research has con-
cerned itself with what goes on during therapy between an individual patient
and therapist instead of the final outcome of any therapeutic effort. In the
late 1950s and early 1960s, a large number of studies appeared on such topics
as relation of therapist behavior to certain patient behaviors in a given
interview situation (e.g., Rogers, Gendlin, Kiesler, & Truax, 1967). As such,
process research held much appeal for clinicians and scientists alike. CHni-
cians were pleased by the focus on the individual and the resulting ability to
study actual clinical processes. In some studies repeated measures during
therapy gave clinicians an idea of the patient's course during treatment.
Scientists were intrigued by the potential of defining variables more precisely
within one interview without concerning themselves with the complexities
involved before or after the point of study. The increased interest in process
research, however, led to an unfortunate distinction between process and
outcome was well stated by Lu-
studies (see Kiesler, 1966). This distinction
borsky (1959), who noted that process research was concerned with how
changes took place in a given interchange between patient and therapist,
whereas outcome research was concerned with what change took place as a
result of treatment. As Paul (1969) and Kiesler (1966) pointed out, the
dichotomization of process and outcome led to an unnecessary polarity in the
manner in which measures of behavior change were taken. Process research
collected data on patient changes at one or more points during the course of
therapy, usually without regard for outcome, while outcome research was
concerned only with pre-post measures outside of the therapeutic situation.
Kiesler noted that this was unnecessary because measures of change within
treatment can be continued throughout treatment until an "outcome" point is
reached. He Chassan (1962) on the desirability of determining

also quoted
what transpired between the beginning and end of therapy in addition to
outcome. Thus the major concern of the process researchers, perhaps as a

result of this imposed distinction, continued to be changes in patient behavior
at points within the therapeutic endeavor. The discovery of meaningful
clinical changes as a result of these processes was left to the prevailing
experimental strategy of the group comparison approach. This reluctance to
relate process variables to outcome and the resulting inability of this ap-
proach to evaluate the effects of psychotherapy led to a decline of process
research. Matarazzo noted that in the 1960s the number of people interested
in process studies of psychotherapy had declined and their students were
nowhere to be seen (Bergin & Strupp, 1972). Because process and outcome
were dichotomized manner, the notion eventually evolved that changes
in this
during treatment are not relevant or legitimate to the important question of
outcome. Largely overlooked at this time was the work of M. B. Shapiro
(e.g.,1961) at the Maudsley Hospital in London, begun in the 1950s. Shapiro
was repeatedly administering measures of change to individual cases during
therapy and also continuing these measures to an end point, thereby relating
"process" changes to "outcome" and closing the artificial gap which Kiesler
was to describe so cogently some years later.
1.7. THE SCIENTIST-PRACTITIONER SPLIT

The state of affairs of clinical practice and research in the 1960s satisfied
few people. Clinical procedures were largely judged as unproven (Bergin &
Strupp, 1972; Eysenck, 1965), and the prevailing naturalistic research was
unacceptable to most scientists concerned with precise definition of variables
and cause-effect relationships. On the other hand, the elegantly designed and
scientifically rigorous group comparison design was seen as impractical and
incapable of dealing with the complexities and idiosyncrasies of individuals
by most clinicians. Somewhere in between was process research, which dealt
mostly with individuals but was correlational rather than experimental. In
addition, the method was viewed as incapable of evaluating the clinical
effects of treatment because the focus was on changes within treatment rather
than on outcome.
These developments were a major contribution to the well-known and oft-
Commission on Mental Illness and
cited scientist-practitioner split (e.g., Joint
Health, 1961). The notion of an applied science of behavior change growing
out of the optimism of the 1950s did not meet expectations, and many
clinician-scientists stated flatly that applied research on their had no effect
clinical practice. Prominent among them was Matarazzo, who noted, "Even
after 15 years, few of my research findings affect my practice. Psychological
sciencQ per se doesn't guide me one bit. I still read avidly but this is of little
direct practical help. My clinical experience is the only thing that has helped
me in my practice to date. . .
." (Bergin & Strupp, 1972, p. 340). This opinion
was echoed by one of the most productive and best known researchers of the
1950s, Carl Rogers, who as early as the 1958 APA conference on psy-
chotherapy noted that research had no impact on his clinical practice and by
1969 advocated abandoning formal research in psychotherapy altogether
(Bergin & Strupp, 1972). Because this view prevailed among prominent
clinicians who were well acquainted with research methodology, it follows
that clinicians without research training or expertise were largely unaffected
by the promise or substance of scientific evaluation of behavior change
procedures. L. H. Cohen (1976, 1979) confirmed this state of affairs when he
summarized a series of surveys indicating that 40% of mental health profes-
sionals think that no research exists that is relevant to practice, and the
remainder believe that less than 20% of research articles have any applicabil-
ity to professional settings.
Although the methodological above were only one
difficulties outlined
contribution to the scientist-practitioner Barlow et al., 1963, for a

split (see
detailed analysis), the concern and pessimism voiced by leading researchers in

the field during Bergin and Strupp 's comprehensive series of interviews led
these commentators to reevaluate the state of the field. Voicing dissatisfaction
with the large-scale group comparison design, Bergin and Strupp concluded:
Among researchers as well as statisticians, there is a growing disaffection from

traditional experimental designs and statistical procedures which are held inap-
propriate to the subject matter under study. This judgment applies with particu-
lar force to research in the area of therapeutic change, and our emphasis on the
value of experimental case studies underscores this point. We strongly agree that
most of the standard experimental designs and statistical procedures have exerted
and are continuing to exert, a constricting effect on fruitful inquiry, and they
serve to perpetuate an unwarranted overemphasis on methodology. More accu-
rately, the exaggerated importance accorded experimental and statistical dicta
cannot be blamed on the techniques proper after all, they are merely tools
but their veneration mirrors a prevailing philosophy among behavioral scientists
which subordinates problems to methodology. The insidious effects of this trend
are tellingly illustrated by the typical graduate student who is often more in-
terested in the details of a factorial design than in the problem he sets out to
study; worse, the selection of a problem is dictated by the experimental design.
Needless to say, the student's approach faithfully reflects the convictions and
teachings of his mentors. With respect to inquiry in the area of psychotherapy,
the kinds of effects we need to demonstrate at this point in time should be
significant enough so that they are readily observable by inspection or descriptive
statistics. If this statistical and mathematical
cannot be done, no fixation upon
which obviously can come only from the
niceties will generate fruitful insights,
researcher's understanding of the subject matter and the descriptive data under
scrutiny (1972, p. 440)
1.8. A RETURN TO THE INDIVIDUAL
Bergin and Strupp were harsh in their comments on group comparison

design and failed to specify those situations where between-group methodol-
ogy may be practical and desirable (see chapter 2). However, their conclusions
on alternative directions, outlined in a paper appropriately titled "New
Directions in Psychotherapy Research" (Bergin & Strupp, 1970), had radical
and far-reaching implications for the conduct of applied research. Essentially,
Bergin and Strupp advised against investing further effort in process and
outcome studies and proposed the experimental single-case approach for the
purpose of isolating mechanisms of change in the therapeutic process. Isola-
tion of these mechanisms of change would then be followed by construction
of new procedures based on a combination of variables whose effectiveness
was demonstrated in single-case experiments. As the authors noted, "As a
general paradigm of inquiry, the individual experimental case study and the
experimental analogue approaches appear to be the primary strategies which
will move us forward in our understanding of the mechanisms of change at
this point" (Bergin & Strupp, 1970, p. 19). The hope was also expressed that
this approach would tend to bring research and practice closer together.
With the recommendations emerging from Bergin and Strupp's compre-
hensive analysis, the philosophy underlying applied research methodology
had come full circle in a little over 1(X) years. The disillusionment with large-
scale between-group comparisons observed by Bergin and Strupp and their
subsequent advocacy of the intensive study of the individual is an historical
At that
repetition of a similar position taken in the middle of the last century.
time, the noted physiologist, Claude Bernard, in An Introduction to the
Study of Experimental Medicine (1957), attempted to dissuade colleagues
who believed that physiological processes were too complex for experimental
inquiry within a single organism. In support of this argument, he noted that
the site of processes of change is in the individual organism, and group
averages and variance might be misleading. In one of the more famous
anecdotes in science, Bernard castigated a colleague interested in studying the
properties of urine in 1865. This colleague had proposed collecting specimens
from urinals in a centrally located train station to determine properties of the
average European urine. Bernard pointed out that this would yield little
information about the urine of any one individual. Following Bernard's
persuasive reasoning, the intensive scientific study of the individual in physi-
ology flourished.
But methodology in physiology and experimental psychology is not directly
applicable to the complexities present in applied research. Although the
splendid isolation of Pavlov's laboratories allowed discovery of important
psychological processes without recourse to sophisticated experimental de-
sign, it is unlikely that the same results would have obtained with a household
pet in its natural environment. Yet these are precisely the conditions under
which most applied researchers must work.
The plea of applied researchers for appropriate methodology grounded in
the scientific method to investigate complex problems in individuals is never
more evident than in the writings of Gordon Allport. Allport argued most
eloquently that the science of psychology should attend to the uniqueness of
the individual (e.g., Allport, 1961, 1962). In terms commonly used in the
1950s, Allport became the champion of the idiographic (individual) ap-
proach, which he considered superior to the nomothetic (general or group)
approach.
Why should we not start with individual behavior as a source of hunches (as we
have and then seek our generalization (also as we have in the past) but
in the past)
finally come back to the individual not for the mechanical application of laws (as
we do now) but for a fuller and more accurate assessment then we are now able
to give? I suspect that the reason our present assessments are now so often feeble
and sometimes even ridiculous, is because we do not take this final step. We stop
with our wobbly laws of generality and seldom confront them with the concrete
person. (Allport, 1962, p. 407)
Due to the lack of a practical, applied methodology with which to study the
individual, however, most of Allport 's own research was nomothetic. The
increase in the intensive study of the individual in applied research led to a
search for appropriate methodology, and several individuals or groups began
developing ideas during the 1950s and 1960s.
The role of the case study
One result of the search for appropriate methodology was a reexamination

of the role of the uncontrolled case study so strongly rejected by scientists in
the 1950s. Recognizing its inherent limitations as an evaluation tool, many
clinical investigators (e.g.. Barlow, 1980; Kazdin, 1981; Lazarus & Davison,
make important contributions to an
1971) suggested that the case study could
experimental effort.One of the more important functions of the case study is
the generation of new hypotheses, which later may be subjected to more
rigorous experimental scrutiny. As Dukes (1965) observed, the case study can
occasionally be used to shed some light on extremely rare phenomena or cast
doubt on well-established theoretical assumptions. Carefully analyzing
threats to internal validity when drawing causal inferences from case studies,
Kazdin (1981) concluded that under certain very specific conditions data from
case studies can approach data from single-case experimental manipulations.
Case studies may also make other important contributions to science
(Barlow et al., 1983; see also chapter 10). Nevertheless, the case study
generally is not capable of isolating therapeutic mechanisms of change (Her-
sen & Barlow, 1976; Kazdin, 1981; Leitenberg, 1973), and the inability of
many scientists and clinicians to discriminate the critical difference between
the uncontrolled case study and the experimental study of an individual case
has most likely retarded the implementation of single-case experimental
designs (see chapter 5).
The representative case
During this period, other theorists and methodologists were attempting to

formulate viable approaches to the experimental study of single cases. Shontz
(1965) proposed the study of the representative case as an alternative to
traditional approaches in experimental personality research. Essentially,
Shontz was concerned with validating previously established personality con-
structs or measurement instruments on individuals who appear to possess the
necessary behavior appropriate for the research problem. Shontz 's favorite
example was a study of the contribution of psychodynamic factors to epilepsy
described by Bowdlear (1955). After reviewing the literature on the presumed
psychodynamics in epilepsy, Bowdlear chose a patient who closely approxi-
mated the diagnostic and descriptive characteristics of epilepsy presented in
the literature (i.e., the representative case). Through a series of questions,
Bowdlear then correlated seizures with a certain psychodynamic concept in
this patient acting out dependency. Since this case was "representative,"
Bowdlear assumed some generalization to other similar cases.
Shontz 's contribution was not methodological, because the experiments he
cites were largely correlational and in the tradition of process research.
Shontz also failed to recognize the value of the single-case study in isolating
effective therapeutic variables or building new procedures, as suggested later
by Bergin and Strupp (1972). Rather, he proposed the use of a single-case in a
deductive manner to test previously established hypotheses and measurement
instruments in an individual who is known to be so stable in certain personal-
ity characteristics that he or she is "representative" of these characteristics.
Conceptually, Shontz moved beyond Allport, however, in noting that this
approach was not truly idiographic in that he was not proposing to investigate
a subject as a self-contained universe with its own laws. To overcome this
objectionable aspect of single-case research, he proposed replication on sub-
jects who differed in some significant way from the first subject. If the general
hypothesis were repeatedly confirmed, this would begin to establish a gener-
ally applicable law of behavior. If the hypothesis were sometimes confirmed
and sometimes rejected, he noted that "... the investigator will be in a
position either to modify his thinking or to state more clearly the conditions
under which the hypothesis does and does not provide a useful model of
psychological events" (Shontz, 1965, p. 258). With this statement, Shontz
anticipated the applied application of the methodology of direct and system-

atic replication in basic research (see chapter 10) suggested by Sidman (1960).
Shapiro's methodology in the clinic
One of the most important contributions to the search for a methodology

came from work of M. B. Shapiro in London. As early as
the pioneering
was advocating a scientific approach to the study of individual
1951, Shapiro
phenomena, an advocacy that continued through the 1960s (e.g., M. B.
Shapiro, 1961, 1966, 1970).
Unlike Allport, however, Shapiro went beyond the point of noting the
advantages of applied research with single-cases and began the difficult task
of constructing an adequate methodology. One important contribution by
Shapiro was the utilization of carefully constructed measures of clinically
relevant responses administered repeatedly over time in an individual. Typi-
cally,Shapiro would examine fluctuations in these measures and hypothesize
on the controlling effects of therapeutic or environmental influences. As
such, Shapiro was one of the first to formally investigate questions more
relevant to psychopathology than behavior change or psychotherapy per se
using the individual case. Questions concerning classification and the identi-
fication of factors maintaining the disorder and even speculations regarding
etiology were all addressed by Shapiro. Many of these studies were correla-
tional in nature, or what Shapiro refers to as simple or complex descriptive
studies (1966). As such, these efforts bear a striking resemblance to process
studies mentioned above, in that the effect of a therapeutic or potential-
maintaining variable was correlated with a target response. Shapiro at-
tempted to go beyond this correlational approach, however, by defining and
manipulating independent variables within single-cases. One good example in
the area of behavior change is the systematic alteration of two therapeutic
approaches in a case of paranoid delusions (M. B. Shapiro & Ravenette,
1959). In a prototype of what was later to be called the A-B-A design, the
authors measured paranoid delusions by asking the patient to rate the "inten-
sity" of a number of paranoid ideas on a scale of 1 to 5. The sum of the score
across 18 different delusions then represented the patient's paranoid "score."
Treatments consisted of "control" discussion concerning guilt feelings about
situations in the patient's life, unrelated to any paranoid ideation, and ra-
tional discussion aimed at exposing the falseness of the patient's paranoid
beliefs. The experimental sequence consisted of 4 days of "guilt" discussion
followed by 8 days of rational discussion and a return to 4 days of "guilt"
discussion. The authors observed an overall decline in paranoid scores during
this experiment, which they rightly noted as correlational and thus potentially
due to a variety of causes. Close examination of the data revealed, however,
thaton weekends when no discussions were held, the patient worsened during
the guilt control phase and improved during the rational discussion phase.
These fluctuations around the regression line were statistically significant.
This effect, of course, is weak and of dubious importance because overall
improvement in paranoid scores was not functionally related to treatment.
Furthermore, several guidelines for a true experimental analysis of the treat-
ment were violated. Examples of experimental error include the absence of
baseline measurement to determine the pretreatment course of the paranoid
beliefs and the simultaneous withdrawal of one treatment and introduction of
a second treatment (see chapter 3). The importance of the case and other
early work from M. B. Shapiro, however, is not the knowledge gained from
any one experiment, but the beginnings of the development of a scientifically
based methodology for evaluating effects of treatment within a single-case.
To the extent that Shapiro's correlational studies were similar to process
research, he broke the semantic barrier which held that process criteria were
unrelated to outcome. He demonstrated clearly that repeated measures within
an individual could be extended to a logical end point and that this end point
was the outcome of treatment. His more important contribution from our
point of view, however, was the demonstration that independent variables in
applied research could be defined and systematically manipulated within a
single-case, thereby fulfilling the requirements of a "true" experimental ap-
proach to the evaluation of therapeutic technique (Underwood, 1957). In
addition, his demonstration of the applicability of the study of the individual
case to the discovery of issues relevant to psychopathology was extremely
important. This approach is only now enjoying more systematic application
by some of our creative clinical scientists (e.g., Turkat & Maisto, in press).
Quasi-experimental designs
In the area of research dealing with broad-based educational or social

change, most often termed evaluation research, Campbell and Stanley (1963)
and Cook and Campbell (1979) proposed a series of important methodologi-
cal innovations that they termed quasi-experimental designs. Education re-
search, of course, is more often concerned with broad-based effects of
programs rather than individual behavioral change. But these designs, many
of which are applicable to either groups or individuals, are also directly
relevant in our context. The two designs most appropriate for analysis of
change in the individual are termed the term series design and the equivalent
term series design. From the perspective of applied clinical research, the time
series design is similar to M.
B. Shapiro's effort to extend process observation
throughout the course of a given treatment to a logical end point or outcome.
This design goes beyond observations within treatment, however, to include
observations from repeated measures in a period preceding and following a
given intervention. Thus one can observe changes from a baseline as a result
of a given intervention. While the inclusion of a baseline is a distinct method-
ological improvement, this design is basically correlational in nature and is
unable to isolate effects of therapeutic mechanisms or establish cause-effect

relationships. Basically, this design is the A-B design described in chapter 5.
The equivalent time series design, however, involves experimental manipula-

tion of independent variables through alteration of treatments, as in the M.
B. Shapiro and Ravenette study (1959), or introduction and withdrawal of
one treatment in an A-B- A fashion. Approaching the study of the individual
from a different perspective than Shapiro, Campbell and Stanley arrived at
similar conclusions on the possibility of manipulation of independent vari-
ables and establishment of cause-effect relationships in the study of a single-
case.
What was perhaps the more important contribution of these methodolo-
gists, however, was the description of various limitations of these designs in
their ability to rule out alternative plausible hypotheses (internal validity) or
the extent to which one can generalize conclusions obtained from the designs
(external validity) (see chapter 2).
Chassan and intensive designs
It remained for Chassan (1967, 1979) to pull together many of the method-
ological advances in single-case research to that point in a book that made
clear distinctionsbetween the advantages and disadvantages of what he
termed extensive (group) design and intensive (single-case) design. Drawing
on long experience in applied research, Chassan outlined the desirability and
applicability of single-case designs evolving out of applied research in the
1950s and early 1960s. While most of his own experience in single-case design
concerned the evaluation of pharmacologic agents for behavior disorders,
Chassan also illustrated the uses of single-case designs in psychotherapy
research, particularly psychoanalysis. As a statistician rather than a practic-
ing clinician, he emphasized the various statistical procedures capable of
establishing relationships between therapeutic intervention and dependent
variables within the single-case. He concentrated on the correlation type of
made occasional use of a prototype of the A-
design using trend analysis but
B-A &
Chassan, 1964), which, in this case, extended the
design (e.g., Bellak
work of M. B. Shapiro to evaluation of drug effects but, in retrospect,
contained some of the same methodological faults. Nevertheless, the sophisti-
cated theorizing in the book on thorny issues in single-case research, such as
generality of findings from a single-case, provided the most comprehensive
treatment of these issues to this time. Many of Chassan 's ideas on this subject
will appear repeatedly in later sections of this book.
1.9. THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

While innovative applied researchers such as Chassan and M. B. Shapiro
made methodological advances in the experimental study of the single-case,
their advances did not have a major impact on the conduct of applied
research outside of their own settings. As late as 1965, Shapiro noted in an
invited address to the Eastern Psychological Association that a large majority
of research in prominent clinical psychology journals involved between-group
comparisons with little and, in some cases, no reference to the individual
approach that he advocated. He hoped that his address might presage the
beginning of a new emphasis on this method. In retrospect, there are several
possible reasons for the lack of impact. First, as Leitenberg (1973) was later
to point out, many of the measures used by M. B. Shapiro in applied research
were indirect and subjective (e.g., questionnaires), precluding the observation
of direct behavioral effects that gained importance with the rise of behavior
therapy (see chapter 4). Second, Shapiro and Chassan, in studies of psy-
chotherapy, did not produce the strong, clinically relevant changes that would
impress clinicians, perhaps due to inadequate or weak independent variables
or treatments, such as instructions within interview procedures. Finally, the
advent of the work of Shapiro and Chassan was associated with the general
disillusionment during this period concerning the possibilities of research in
psychotherapy. Nevertheless, Chassan and Shapiro demonstrated that mean-
ingful applied research was possible and even desirable in the area of psy-
chotherapy. These investigators, along with several of Shapiro*s students
(e.g.,Davidson & Costello, 1969; Inglis, 1966; Yates, 1970), had an important
influenceon the development and acceptance of more sophisticated method-
ology, which was beginning to appear in the 1960s.
It is signi ficant that it w as the rediscovery of th e study of the sing le-case in
basic research, coupled_withIagewâgBJ^gEo-BI ^
in the applied area ,
that marked the beginnings qf_a_new^mphasis on the experi mental study of

the single-caseîn applied research. One indication of the broad influence of
thisômbmation of events was the emergence of a journal in 1968 {Journal of
Applied Behavior Analysis) devoted to single-case methodology in applied
research and the appearance of this experimental approach in increasing
numbers in the major psychological and psychiatric journals. The methodol-
ogy in basic research was termed the experimental analysis of behavior, the
new approach to applied problems became known as behavior modification
or behavior therapy.
Some observers have gone so far as to define behavior therapy in terms of
single-case methodology (Yates, 1970; 1975) but, as Leitenberg (1973) pointed
out, this definition is without empirical support because behavior therapy is a
clinical approach employing a number of methodological strategies (see
Kazdin, 1978, and Krasner, 1971a, for a history of behavior therapy). The
relevance of the experimental analysis of behavior to applied research is the
development of sophisticatedjnethodplogy^nabling intensive study of indi-
vidual_suB]ects. In rejecting a between-subject approaciraslEe~only" useful
scientific methodology. Skinner (1938, 1953) reflected the thoughts of the
early physiologists such as Claude Bernard and emphasized repeated objec-

tive measurement in a single subject over a long period of time under highly
controlled conditions. As Skinner noted (1966b), "... instead of studying a
thousand rats for one hour each, or a hundred rats for ten hours each, the
investigator Hkely to study one rat for a thousand hours" (p. 21), a
is
procedure that clearly recognizes the individuality of an organism. Thus,

Skinner and his colleagues in the animal laboratories developed and refined
the single-case methodology that became the foundation of a new applied
science. Culminating in the definitive methodological treatise by Sidman
(1960), entitled Tactics of Scientific Research, the assumption and conditions

of a true experimental analysis of behavior were outlined. Examples of fine-
grain analyses of behavior and the use of withdrawal, reversal, and multi-
element experimental designs in the experimental laboratories began to
appear in more applied journals in the 1960s, as researchers adapted these
strategies to the investigation of applied problems.
It is unlikely, however, that this approach would have had a significant
impact on applied clinical research without the growing popularity of behav-
ior therapy.The fact that M. B. Shapiro and Chassan were employing
rudimentary prototypes of withdrawal designs (independent of influences
from the laboratories of operant conditioning) without marked effect on
applied research would seem to support this contention. In fact, even earlier,
F. C. Thorne (1947) described clearly the principle of single-case research,

including A-B-A withdrawal designs, and recommended that clinical research
proceed in manner, without apparent effect (Barlow et al., 1983). The^
this
growth_af_thg,_behav ior therapy approach j OLapplied problems, however,
provided a vehicle for the introduction of thejiiethodology on^scal^^
attracted_attention_ftpm^^ Behavior therapyras"
the application of the principles of general-experimental and social psychol-
ogy to the clinic, also emphasized direct measurement of clinically relevant
target behaviors and experimental evaluation of independent variables or
"treatments." Since many of these "principles of learning" utilized in behav-
ior therapy originally emanated from operant conditioning, it was a small
step for behavior therapists to also borrow the operant methodology to
validate the effectiveness of these same principles in applied settings. The
initial success of this approach (e.g., Ullmann & Krasner, 1965) led to similar
evaluations of additional behavior therapy techniques that did not derive

directly from the operant laboratories (e.g., Agras et al., 1971; Barlow,
Leitenberg, & Agras, 1969). During this period, methodology originally
intended for the animal laboratory was adapted more fully to the investiga-
tion of applied problems and "applied behavior analysis" became an impor-
tant supplementary and, in some cases, alternative methodological approach
to between-subjects experimental designs.
The early pleas to return to the individual as the cornerstone of an applied
science of behavior have been heeded. The last several years have witnessed
the crumbling of barriers that precluded publication of single-case research in
any leading journal devoted to the study of behavioral problems. Since the
first edition of this book, a proHferation of important books has appeared
devoted, for example, to strategies for evaluating data from single-case

designs (Kratochwill, 1978b), to the application of these methods in social
work (Jayaratne & Levy, 1979), or to the philosophy underlying this approach
to applied research (J. M. Johnston & Pennypacker, 1980). Other excellent
books have appeared concentrating specifically on descriptions of design
alternatives (Kazdin, 1982b), and major handbooks on research are not
complete without a description of this approach (e.g., Kendall & Butcher,
1982).
More importantly, the field has not stood still. From their more recent
origins in evaluating the application of operant principles to behavior disor-
ders, single-case designs are now fully incorporated into the armamentarium
of applied researchers generally interested in behavior change beyond the
subject matter of the core mental health professions or education. Profes-
approach
sions such as rehabilitation medicine are turning increasingly to this
as appropriate to the subject matter at hand (e.g., and the
Schindele, 1981),
field is progressing. New design alternatives have appeared only recently, and
strategies involved in more traditional approaches have been clarified and

refined. We believe that the recent methodological developments and the
demonstrated effectiveness of this methodology provide a base for the estab-
lishment of a true science of human behavior with a focus on the paramount
importance of the individual. A description of this methodology is the
purpose of this book.
CHAPTER 2
General Issues in a Single-Case Approach
2.1. INTRODUCTION
TXvo issues basic to any science are variability and generality of findings.
These issues are handled somewhat differently from one area of science to
another, depending on the subject matter. The first section of this chapter
concerns variability.
In applied research, where individual behavior is the primary concern, it is
our contention that the search for sources of variability in individuals must
occur if we are to develop a truly effective clinical science of human behavior
change. After a brief discussion of basic assumptions concerning sources of
variability in behavior, specific techniques and procedures for dealing with
behavioral variability in individuals are outlined. Chief among these are
repeated measurement procedures that allow careful monitoring of day-to-
day variability in individual behavior, and rapidly changing, improvised
experimental designs that facilitate an immediate search for sources of va-
riability in an individual. Several examples of the use of this procedure to
track down sources of intersubject or intrasubject variability are presented.
The second section of this chapter deals with generality of findings. Histori-
cally, this has been a thorny issue in applied research. The seeming limitations
in establishing wide generality from results in a single-case are obvious, yet

establishment of generality from results in large groups has also proved
elusive. After a discussion of important types of generality of findings, the
shortcomings of attempting to generalize from group results in applied
research are discussed. Traditionally, the major problems have been an inabil-
ity to draw a truly random sample from human behavior disorders and the
difficulty of generalizing from groups to an individual. Applied researchers

attempted to solve the problem by making groups as homogeneous as possi-
32
General Issues in A Single-case Approach 33
ble so that results would be applicable to an individual who showed the

characteristics of the homogeneous group. An alternative method of estab-
lishing generality of findings is the replication of single-case experiments. The
relative merits of establishing generality of findings from homogeneous
groups and replication of single-case experiments are discussed at the end of
this section.
Finally, some research questions that cannot be answered through experi-
mentation on single-cases are listed, and strategies for combining some
strengths of single-case and between-subject research approaches are sug-
gested.
2.2. VARIABILITY
The notion that behavior is a function of a multiplicity of factors finds
wide agreement among and professional investigators. Most scien-
scientists
tists also agree that as one moves up the phylogenetic scale, the sources of
variability in behavior become greater. In response to this, many scientists
choose to work with lower life hope that laws of behavior will
forms in the
emerge more readily and be generalizable to the infinitely more complex area
of human behavior. Applied researchers do not have this luxury. The task of
the investigator in the area of human behavior disorders is to discover
functional relations among treatments and specific behavior disorders over
and above the welter of environmental and biological variables impinging on
the patient at any given time. Given these complexities, it is small wonder that
most treatments, when tested, produce small effects or, in Bergin and Strupp's
terms, weak results (Bergin & Strupp, 1972).
Variability in basic research
Even in basic research, behavioral variability is enormous. Injittempting to

deal with this probl em, man y pvppnmpntal pcyrhnlpgists asQiimpH thqt va-
riability was intrinsic to the org anis m rather than imposed bv experime n t al or
en vironmental factors j Sidman 1960). If variability were an intrinsic compo-
,
nent of behavior, then procedures had to be found to deal with this issue
before meaningful research could be conducted. The solution involved ex-
perimental designs and confidence level would elucidate func-
statistics that
tional relations among independent and dependent variables over and above
the intrinsic variability. Sidman (1960) noted that this is not the case in some
other sciences, such as physics. Physics assumes that variability is imposed by
error of measurement or other identifiable factors. Experimental efforts are
then directed to discovering and eliminating as many sources of variability as
possible so that functional relations can be determined with more precision.
Sidman proposed that basic researchers in psychology also adopt this strat-
egy. Rather than assuming that variability is intrinsic to the organism, one
should make every effort to discover sources of behavioral variability among
organisms such that laws of behavior could be studied with the precision and
specificity found in physics. This precision, of course, would require close
attention to the behavior of the individual organism. If one rat behaves
differently from three other rats in an experimental condition, the proper
tactic is to find out why. If the experimenter succeeds, the factors that produce
that variability can be eliminated and a "cleaner" test of the effects of the
original independent variable can be made. Sidman recognized that behav-
ioral variability may never be entirely eliminated, but that isolation of as
many sources of variability as possible would enable an investigator to
estimate how much variability actually is intrinsic.
Variability in applied research
Applied researchers, by and large, have not been concerned with this
argument. Every practitioner is aware of multiple social or biological factors
that are imposed on his or her data. If asked, many investigators might also
assume some intrinsic variability in clients attributable to capriciousness in
nature; but most are more concerned with the effect of uncontrollable but
potentially observable events in the environment. For example, the sudden
appearance of a significant relative or the loss of a job during treatment of
depression may affect the course of depression to a far greater degree than the
particular intervention procedure. Menstruation may cause marked changes
in behavioral measures of anxiety. Even more disturbing are the multiple
unidentifiable sources of variability that cause broad fluctuation in a patient's
clinical course. Most applied researchers assume this variability is imposed
rather than intrinsic, but they may not know where to begin to factor out the
sources.
The solution, as in basic research, has been to accept broad variability as an
unavoidable evil, to employ experimental design and statistics that hopefully
control variability, and to look for functional relations that supersede the
"error."
As Sidman observed when discussing these tactics in basic research:
The rationale for statistical immobilization of unwanted variables is based on the

assumed random nature of such variables. In a largegroup of subjects, the
reasoning goes, the uncontrolled factor will change the behavior of some subjects
in one direction and will affect the remaining subjects in the opposite way. When
the data are averaged over all the subjects, the effects of the uncontrolled
variables are presumed to add algebraically to zero. The composite data are then
regarded as though they were representative of one ideal subject who had never
been exposed to the uncontrolled variables at all (1960, p. 162).
Although one may question this strategy in basic research, as Sidman has, the
amount of control an experimenter has over the behavioral history and
current environmental variables impinging on the laboratory animal makes
this strategy at least feasible. In applied research, when control over behav-
ioral histories or even current environmental events is limited or nonexistent,
there is far less probability of discovering a treatment that is effective over
and above these uncontrolled variables. This, of course, was the major cause
of the inability of early group comparison studies to demonstrate that the
treatment under consideration was effective. As noted in chapter 1, some
clients were improving while others were worsening, despite the presence of
the treatment. Presumably, this variability was not intrinsic but due to current
life circumstances of the clients.
Clinical vs. statistical significance
The experimental designs and statistics gleaned from the laboratories of

experimental psychology have an added disadvantage in applied research.
The purpose of research in any basic science is to discover functional relations
among dependent and independent variables. Once discovered, these func-
tional relationships become principles that add to our knowledge of behavior.
In applied research, however, the discovery of functional relations is not
sufficient. The purpose of applied research is to effect meaningful clinical or
socially relevant behavioral changes. For example, if depression were reliably
measurable on a 0-100 scale, with 100 representing severe depression, a
treatment that improved each patient in a group of depressives from 80 to 75
would be statistically significant if all depressives in the control group re-
mained at 80. This statistical significance, however, would be of little use to
the practicing clinician because a score of 75 could still be in the suicidal
range. An improvement of 40 or 50 points might be necessary before the
clinician would consider the change clinically important. Elsewhere, we have
referred to the issue as statistical versus clinical significance (Barlow & Her-
sen, 1973), and this issue has been raised repeatedly during the last decade
(e.g., Garfield & Bergin, 1978). In this simplified example, statisticians might
observe that this issue is easily correctable by setting a different criterion level
for "effectiveness." In the jungle of applied research, however, when any
effect superseding the enormous "error" or variance in a group of heteroge-
neous clients is remarkable, the clinician and even the researcher will often
overlook this issue and consider a treatment that is statistically significant to

also be clinically effective.
As Chassan (1960, 1979) pointed out, statistical significance can underesti-
mate clinical effectiveness as well as overestimate it. This unfortunate circum-
stance occurs when a treatment is quite effective with a few members of the
experimental group while the remaining members do not improve or dete-

riorate somewhat. Statistically, then, the experimental group does not differ
from the control group, whose members are relatively unchanged. When
broad divergence such as this occurs among clients in response to an interven-
tion, statistical treatments will average out the clinical effects along with
changes due to unwanted sources of variability. In fact, this type of intersub-
ject variability is the rule rather than the exception. Bergin (1966) clearly
illustrated the years that were lost to applied research because clinical investi-
gators overlooked the marked effectiveness of these treatments on some
clients (see also, Bergin & Lambert, 1978; Strupp & Hadley, 1979). The issue
of clinical versus statistical significance is, of course, not restricted to be-
tween-group comparisons but is something applied researchers must consider
whenever statistical tests are applied to clinical data (see chapter 9).
Nevertheless, the advantages of attempting to eliminate the enormous
intersubject variability in applied research through statistical methods have
intuitive appeal for both researchers and clinicians who want quick answers
to pressing clinical or social questions. In fact, to the clinician who might
observe one severely depressive patient inexplicably get better while another
equally depressed patient commits suicide, this variability may well seem to
be intrinsic to the nature of the disorder rather than imposed by definable
social or biological factors.
Highlighting variability in the individual
In any case, whether variability in applied research is intrinsic to some

degree or not, the alternative to the treatment of intersubject variability by
statistical means is to highlight variability and begin the arduous task of
determining sources of variability in the individual. To the applied researcher,
this task is staggering. In realistic terms he or she must look at each individual
who differs from other clients in terms of response to treatment and attempt
to determine why. Since the complexities of human environments, both
external and internal, are enormous, the possible causes of these differences
number in the millions.
With the complexities involved in this search, one may legitimately ques-
tion where to begin. Since intersubject variability begins with one client
differing in response from some other clients, a logical starting point is the
individual. If one is to concentrate on individual variability, however, the
manner in which one observes this variability must also change. If one
depressed patient deteriorates during treatment while others improve or
remain stable, it is difficult to speculate on reasons for this deterioration if the
only data available are observations before and after treatment. It would be
much to the advantage of the clinical researcher to have followed this one
patient's course during treatment so that the beginning of deterioration could
be pinpointed. In this hypothetical case the patient may have begun to

improve until a point midway in treatment, when deterioration began.
Perhaps a disruption in family life occurred or the patient missed a treatment
session, while other patients whose improvement continued did not expe-
rience these events. It would then be possible to speculate on these or other
factors that were correlated with such change. In single-case research the
investigator could adjust to the variability with immediate alteration in
experimental design to test out hypothesized sources of these changes.*
Repeated measures
The basis of this searc h for sources of variability is repeated measurement

of the dependent variable or p robleni behavjoi^Tf this, tactic has a familiar
ring to practitioners, no accident, for this is precisely the strategy every
it is
practitioner uses daily. It is no secret to clinicians or other behavior change

agents in applied settings that behavioral improvement from an initial obser-
vation to some end point sandwiches marked variabiHty in the behavior
between these points. A major activity of clinicians is observing this variabil-
ity and making appropriate changes in treatment strategies or environmental
circumstances, where possible, to eliminate these fluctuations from a general

improving trend. Because measures in the clinic seldom go beyond gross
observation, and treatment consists of a combination of factors, it is difficult
for clinicians to pinpoint potential sources of variability, but they speculate;
with increased clinical experience, effective clinicians may guess rightly more
often than wrongly. In some cases, weekly observation may go on for years.
As Chassan (1967) pointed out:
The existence of variability as a basic phenomenon in the study of individual

psychopathology implies that a single observation of a patient state, in general,
can offer only a minimum of information about the patient state. While such
information is literally better than no information, it provides no more data than
does any other statistical sample of one (1967, p. 182)
He then quoted Wolstein (1954) from a psychoanalytic point of view, who

comments on diagnostic categories:
These terms are "ad hoc" definitions which move the focus of inquiry away from
repetitive patterns with observable frequencies to fixed momentary states. But
this notion of the momentary present is specious and deceptive; it is neither fixed
nor momentary nor immediately present, but an inferred condition (p. 39).
For an excellent discussion of the concept of variability and the relationship of

measurement to variability see J. M. Johnston and Pennypacker (1981).
The relation of this strategy to process research, described in chapter 1, is
obvious. But the search for sources of individual variability cannot be re-
stricted to repeated measures of one small segment of a client's course
somewhere between the beginning and the end of treatment, as in process
research. With the multitude of events impinging on the organism, significant
behavior fluctuation may occur at any time from the beginning of an
intervention until well after completion of treatment. The necessity of re-
peated, frequent measures to begin the search for sources of individual
variability is apparent. Procedures for repeated measures of a variety of
behavior problems are described in chapter 4.
Rapidly changing designs
If one is committed to determining sources of variability in individuals,

repeated measurement alone is insufficient. In a typical case, no one event is
clearly associated with behavioral fluctuation, and repeated observation will
permit only a temporal correlation of several events with the behavioral
fluctuation. In the clinic this temporal correlation provides differing degrees
of evidence on an intuitive level concerning causality. For instance, if a
claustrophobic became trapped in an elevator on the way to the therapist's
office and suddenly worsened, the clinician could make a reasonable in-
ference that this event caused the fluctuation. Usually, of course, sources of
and the applied researcher must guess from among
variability are not so clear,
However, it would add little to science if an investi-
several correlated events.
gator merely reported at the end of an experiment that fluctuation in behav-
iors were observed and were correlated with several events. The task
confronting the applied researcher at this point is to devise experimental
designs to isolate the cause of the change or the lack of change. One
advantage of single-case experimental designs is that the investigator can
begin an immediate search for the cause of an experimental behavior trend by
altering the experimental design on the spot. This feature, when properly
employed, can provide immediate information on hypothesized sources of
variability. In Skinner's words:
A prior design in which variables are distributed, for example, in a Latin square,
may be a severe handicap. When on behavior can be immediately ob-
effects
served, it is more efficient to explore relevant variables by manipulating them in
an improvised and rapidly changing design. Similar practices have been responsi-
ble for the greater part of modern science (Honig, 1966, p. 21).
More recently, this feature of single-case designs has been termed response
guided experimentation (Edgington, 1983, 1984).
2.3. EXPERIMENTAL ANALYSIS OF SOURCES OF

VARL\BILITY THROUGH IMPROVISED DESIGNS
In single-case designs there are at least three patterns of variability high-
lighted by repeated measurement. In the first pattern a subject may not
respond to a treatment previously demonstrated as effective with other sub-
jects. In a second pattern a subject may improve when no treatment is in
effect, as in a baseline phase. This "spontaneous" improvement is often
considered to be the result of "placebo" effects. These two patterns of
intersubject variability are quite common in applied research. In a third
pattern the variability is intrasubject in that marked cyclical patterns emerge
in themeasures that supersede the effect of any independent variable. Using
improvised and rapidly changing designs, it is possible to follow Skinner's
suggestion and begin an immediate search for sources of this variability.
Examples of these efforts are provided next.
Subject fails to improve
One experiment from our laboratories illustrates the use of an "improvised

and rapidly changing design" to determine why one subject did not improve
with a treatment that had been successful with other subjects. The purpose of
this experiment was to explore the effects of a classical conditioning proce-
dure on increasing heterosexual arousal in homosexuals desiring this addi-
tional arousal pattern (Herman, Barlow, & Agras, 1974a). In this study,
heterosexual arousal as measured by penile circumference change to slides of
nude females was the major dependent variable. Measures of homosexual
arousal and reports of heterosexual urges and fantasies were also recorded.
The design is a basic A-B-A-B with a baseline procedure, making it technically
an A-B-C-B-C, where A is baseline; B is a control phase, backward condition-
ing; and C is the treatment phase, or classical conditioning. In classical
conditioning the client viewed two slides for one minute each. One slide
depicted a female, which became the CS. A male slide, to which the client
became aroused routinely, became the UCS. During classical conditioning,
the client viewed the CS (female slide) for one minute, followed hnmediately
by the UCS (male slide) for 1 minute in the typical classical conditioning
paradigm. During the B, or control phase, however, the order of presentation
was reversed (UCS-CS), resulting in a backward conditioning paradigm
which, of course, should not produce any learning.
During Experiment 1 (see Figure 2-1), no increases in heterosexual arousal
were noted during baseline or backward conditioning. A sharp rise occurred,
however, during classical conditioning. This was followed by a downward
trend in heterosexual arousal during a return to the backward conditioning
o o Heterosexual urges & fantasies

- Circumference change to females
Circumference ctiange to males
Numt)er of Reported Masturbations witti Female Fantasies
100
000 0000 005 4100 000?5 rl
o
S.I 80
If
si
C M
-< c
3 . n
n
20-
:;
1 2 3 4 5 6 7 8 9 10 U 1? 13 14 13 16 17 18 19
I I
Baseline Backward Classical Backward Classical
I
Presentation I Cond. Present. Conditioning
Blocks of Two Sessions

( Circumferencecfiange to males averaged over each phase I
FIGURE 2-1. Mean and female slides expressed as a

penile circumference change to male
percentage of full and total heterosexual urges and fantasies collected from 4 days
erection
surrounding each session. Data are presented in blocks of two sessions (circumference change to
males averaged over each phase). Reported incidence of masturbation accompanied by female
fantasy is indicated for each blocked point. (Figure 1, p. 36, from: Herman, S. H., Barlow, D.
H., and Agras, W. W. [1974]. An experimental analysis of classical conditioning as a method of
increasing heterosexual arousal in homosexuals. Behavior Therapy, 5, 33-47. Copyright 1974 by
Association for the Advancement of Behavior Therapy Reproduced by permission.)
control phase, and further increases in arousal during a second classical

conditioning phase, suggesting that the classical conditioning procedure was
producing the observed increase.
In attempting to replicate this finding on a second client (see Figure 2-2),
some variation in responding was noted. Again, no increase in heterosexual
arousal occurred during baseline or backward conditioning phases; but none
occurred during the first classical conditioning phase either, even though the
number of UCS slides was increased from one to three. At this point, it was
noted that his response latency to the male slide was approximately 30
seconds. Thus the classical conditioning procedure was adjusted slightly, such
Mean UCR percentage per treatment session
54433624324 7 4 4 4124 323346342940 322015^'

1
M> Circumference change to males '
Circumference change to females !
>o Heterosexual urges and fantasies I
,S 60.
v7 ?5?
V.
Backward Classcal Simultaneous
Presentation Conditioning Presentation
Individual Sessions
(Circumference change to males averaged over each phase)
FIGURE 2-2. Mean penile circumference change to male and female slides expressed as a
percentage of full erection and total heterosexual urges and fantasies collected from 4 days
surrounding each session. Data are presented for individual sessions with circumference change to
males averaged over each phase. Mean UCR percentage is indicated for each treatment session.
(Figure 2, p. 40, from: Herman, S. H., Barlow, D. H., and Agras, W. S. [1974]. An experimental
analysis of classical conditioning as a method of increasing heterosexual arousal in homosexuals.
Behavior Therapy, 5, 33-47. Copyright 1974 by Association for the Advancement of Behavior
Therapy. Reproduced by permission.)
that 30 seconds of viewing the female slide alone was followed by 30 seconds
of viewing both the male and female slides simultaneously (side by side),
followed by 30 seconds of the male slide alone. This adjustment (labeled
simultaneous presentation) produced increases in heterosexual arousal in the
separate measurement sessions, which reversed during a return to the original
classical conditioning procedure and increased once again during the second
phase, in which the slides were presented simultaneously. The experiment
suggested that classical conditioning was also effective with this cHent but
only after a sensitive temporal adjustment was made.
Merely observing the "outcome" of the 2 subjects at the end of a fixed
point in time would have produced the type of intersubject variability so
common in outcome studies of therapeutic techniques. That is, one subject
would have improved with the initial classical conditioning procedure
whereas one subject would have remained unchanged. If this pattern contin-
ued over additional subjects, the result would be the typical weak effect
(Bergin & Strupp, 1972) with large intersubject variability. Highlighting the
variability through repeated measurement in the individual and improvising a
new experimental design as soon as a variation in response was noted (in this
case no response) allowed an immediate search for the cause of this unrespon-
siveness. It should also be noted that this research tactic resulted in immediate
clinical benefit to the patient, providing a practical illustration of the merging

of scientist and practitioner roles in the applied researcher.
Subject improves ''spontaneously''
A second source of variability quite common in single-case research is the

presence of "spontaneous" improvement in the absence of the therapeutic
variable to be tested. This effect is illustrated in a second experiment on
increasing heterosexual arousal in homosexuals (Herman, Barlow, & Agras,
1974b).
In this study, the original purpose was to determine the effectiveness of
orgasmic reconditioning, or pairing masturbation with heterosexual cues, in
producing heterosexual arousal. The heterosexual cues chosen were movies of
a female assuming provocative sexual positions. The initial phase consisted of
measurements of arousal patterns without any "treatment," which served as
a baseline of sexual arousal. Before pairing masturbation with this movie, a
control phase was administered where all elements of the treatment were
present with the exception of masturbation. That is, the subject was in-
was "treatment" and that looking at movies would help him

structed that this
learn heterosexual arousal. Although no increase in heterosexual arousal was
expected during this phase, this procedure was experimentally necessary to
isolate the pairing of masturbation with the cues in the next phase as the
effective treatment. The effects of masturbation were never tested in this
experiment, however, since the first subject demonstrated unexpected but
substantial increases in heterosexual arousal during the "control" phase, in
which he simply viewed the erotic movie (see Figure 2-3). Once again it
became necessary to improvise a new experimental design at the endôf this

control phase, in an attempt to determine the cause of this unexpected
increase. On the hunch that the erotic heterosexual movie was responsible for
these gains rather than other therapeutic variables such as expectancy, a
second erotic movie without heterosexual content was introduced, in this case
a homosexual movie. Heterosexual arousal dropped and
in this condition
increased once again when the heterosexual movie was introduced. This
experiment, and subsequent replication, demonstrated that the erotic hetero-
sexual movie was responsible for improvement. Determination of the effects
of masturbation was delayed for future experimentation.
Subject displays cyclical variability
A third pattern of variability, highlighted by repeated measurement in

individual cases, is observed when behavior varies in a cyclical pattern. The
behavior may follow a regular pattern (i.e., weekly) or may be irregular. A
MALE
BASELINE , FEMALE EXPOSURE EXPOSURE FEMALE EXPOSURE
i
62.5-
Circumference change to:
Females
Males
50-
37.5
25
12 3 4 5 6 7 8 9 10 11 12 13 14 15
BLOCKS OF THREE SESSIONS

( Circumference Change to Males Averaged Over Each Phase
FIGURE 2-3. Mean penile circumference change expressed as a percentage of full erection to
nude female (averaged over blocks of three sessions) and nude male (averaged over each phase)
slides. (Figure 1, p. 338, from: Herman, S. H., Barlow, D. H., and Agras, W. S. [1974]. An
experimental analysis of exposure to "explicit" heterosexual stimuli as an effective variable in

changing arousal patterns of homosexuals. Behaviour Research and Therapy, 12, 335-345.
Copyright 1974 by Pergamon. Reproduced by permission.)
common temporal pattern, of course, is the behavioral or emotional fluctua-

tion noted during menstruation. Of more concern to the clinician is the
marked fluctuation occurring in most behavioral disorders over a period of
time. In most instances the fluctuation cannot be readily correlated with
specific, observable environmental or psychological events, due to the extent
of the behavioral or emotional fluctuation and the number of potential
variables that may be affecting the behavior. As noted in the beginning of this
chapter, experimental clinicians can often make educated guesses, but the
technique of repeated measurement can illustrate relationships that might not
be readily observable.
A good example of this method is found in an early case of severe, daily
asthmatic attacks reported by Metcalfe (1956). In the course of assessment,
Metcalfe had the patient record in diary form asthmatic attacks as well as all
activities during the day, such as games, shopping expeditions, meetings with
her mother, and other social visits. These daily recordings revealed that
asthmatic attacks most often followed meetings with the patient's mother,
particularly if these meetings occurred in the home of the mother. After this
relationshipwas demonstrated, the patient experienced a change in her Hfe
circumstances which resulted in moving some distance away from her mother.
During the ensuing 20 months, only nine attacks were recorded despite the
fact that these attacks had occurred daily for a period of 2 years prior to
intervention. What is more remarkable is that eight of the attacks followed
her now infrequent visits to her mother.
Once again, the procedure of repeated measurement highlighted individual
fluctuation, allowing a search for correlated events that bore potential causal
It should be noted that no experimen-
relationships to the behavior disorder.
tal was undertaken in this case to isolate the mother as the cause of
analysis
asthmatic attacks. However, the dramatic reduction of high-frequency at-
tacks after decreased contact with the mother provided reasonably strong
evidence about the contributory effects of visits to the mother, in an A-B
fashion. What is more convincing, however, is the reoccurrence of the attacks
at widely spaced intervals after visits to the mother during the 20-month
follow-up. This series of naturally occurring events approximates a contrived
A-B- A-B. . . design and effectively isolates the mother's role in the patient's
asthmatic attacks (see chapter 5).
Searching for ** hidden" sources of variability
In the preceding case functional relations become obvious without experi-

mental investigation, due to the overriding effects of one variable on the
behavior in question and a series of fortuitous events (from an experimental
point of view) during follow-up. Seldom in applied research is one variable so
predominant. The more usual case one where marked fluctuations in

is
behavior occur that cannot be correlated with any one variable. In these
cases, close examination of repeated measures of the target behavior and
correlated internal or external events does not produce an obvious relation-
ship. Most likely, many events may be correlated at one time or another with
deterioration or improvement in a client. At this point, it becomes necessary
to employ sophisticated experimental designs if one is to search for the source
of variability. The experienced applied researcher must first choose the most
likely variables for investigation from among the many impinging on the
client at any one time. In the case described above, not only visits to the
mother but visits to other relatives as well as stressful situations at work might
all have contributed to the variance. The task of the clinical investigator is to
tease out the relevant variables by manipulating one variable, such as visits to
mother, while holding other variables constant. Once the contribution of
visits to mother to behavioral fluctuation has been determined, the investiga-
tor must go on to the next variable, and so on.

In many cases, behavior is a function of ah interaction of events. These

events may be naturally occurring environmental variables or perhaps a
combination of treatment variables which, when combined, affect behavior
differently from each variable in isolation. For example, when testing out a
variety of treatments for anorexia nervosa (Agras, Barlow, Chapin, Abel, &
Leitenberg, 1974), it was discovered that size of meals served to the patients
seemed related to caloric intake. An improvised design at this point in the
experiment demonstrated that size of meals was related to caloric intake only
if feedback and reinforcement were present. This discovery led to inclusion of
this procedure in a recommended treatment package for anorexia nervosa.
Experimental designs to determine the effects of combinations of variables
will be discussed in section 6.6 of chapter 6.
2.4. BEHAVIOR TRENDS

AND INTRASUBJECT AVERAGING
When testing the effects of specific interventions on behavior disorders, the
investigator is less interested in small day-to-day fluctuations that are a part
of so much behavior. In these cases the investigator must make a judgment on
how much behavioral variability to ignore when looking for functional
relations among overall trends in behavior and treatment in question. To the
investigator interested in determining all sources of variability in individual
behavior, this is a very difficult choice. For applied researchers, the choice is
often determined by the practical considerations of discovering a therapeutic

variable that "works" for a specific behavior problem in an individual. The
necessity of determining the effects of a given treatment may constrain the
applied researcher from improvising designs in midexperiment to search for a
source of each and every fluctuation that appears.
In correlational designs, where one simply introduces a variable and ob-
serves the "trend," statistics have been devised to determine the significance
of the trend over and above the behavioral fluctuation (Campbell & Stanley,
1966; & Campbell, 1979; see also chapter 9). In experimental designs
Cook
such as A-B-A-B, where one is looking for cause-effect relationships, investi-
gators will occasionally resort to averaging two or more data points within
phases. This intrasubject averaging, which sometimes called blockings will
is
usually make trends in behavior more can judge

visible, so that the clinician
the magnitude and clinical relevance of the effect. This procedure is danger-
ous, however, if the investigator is under some illusion that the variability has
somehow disappeared or is unimportant to an understanding of the control-
ling effects of the behavior in question. This method is simply a procedure to
make large and changes resulting from introduction and
clinically significant
withdrawal of treatment more apparent. To illustrate the procedure, the
Reinforcement Reinforcement Reinforcement

Base Line Reinforcement & Feedback | | Feedback
Weight m - 4,000

Caloric
Intake o>--o
3,000 o
S
2,000
- 1,000
30 40 50
Days
FIGURE 2-4, Data from an experiment examining the effect of feedback on the eating behavior
of a patient withanorexia nervosa (Patient 4). (Figure 3, p. 283, from: Agras, W. S., Barlow, D.
H., Chapin, H. N., Abel, G. G., and Leitenberg, H. [1974]. Behavior modification of anorexia
nervosa. Archives of General Psychiatry, 30, 279-286. Copyright 1974 by American Medical
Association. Reproduced by permission.)
original data on caloric intake in a subject with anorexia nervosa will be

presented for comparison with published data (Agras et al., 1974). The data
as published are presented in Figure 2-4. After the baseline phase, material
reinforcers such as cigarettes were administered contingent on weight gain in
a phase labeled reinforcement. In the next phase, informsiiiomil feedback was
added to reinforcement. Feedback consisted of presenting the subject with
meal and counts of number of
daily weight counts of caloric intake after each
mouthfuls eaten. The data indicate that caloric intake was relatively stable
during the reinforcement phase but increased sharply when feedback was
added to reinforcement. Six data points are presented in each of the rein-
forcement and reinforcement-feedback phases. Each data point represents
the mean of 2 days. With this method of data presentation, caloric intake
during reinforcement looks quite stable.
In fact, there was a good deal of day-to-day variability in caloric intake
during this phase. If one examines the day-to-day data, caloric intake ranged
from 1,450 to 3,150 over the 12-day phase (see Figure 2-5). Since the variabil-
CALORIES CONSUMED PER DAY
REINFORCEMENT

FEEDBACK
I I I I I I I I I I I I 'l I I I I I I I I I I !
1 4 8 12 16 20 24 DAYS
FIGURE 2-5. Caloric intake presented on a daily basis during reinforcement and reinforcement
and feedback phases for the patient whose data is presented in Figure 2-4. (Replotted from Figure
3, p. 283, from: Agras, W. S., Barlow, D. H., Chapin, H. N., Abel, G. G., and Leitenberg, H.
[1974]. Behavior modification of anorexia nervosa. Archives of General Psychiatry, 30, 279-286.
Copyright 1974 by American Medical Association. Reproduced by permission.)
ityassumed a pattern of roughly one day of high caloric intake followed by a

day of low intake, the average of 2 days presents a stable pattern. When
feedback was added during the next 12-day phase, the day-to-day variability
remained, but the range was displaced upward, from 2,150 to 3,800 calories
per day. Once again, this pattern of variability was approximately one day of
high caloric intake followed by a low value. In fact, this pattern obtained
throughout the experiment.
In this experiment, feedback was clearly a potent therapeutic procedure
over and above the variability, whether one examines the data day-by-day or
in blocks of 2 days. The averaged data, however, present a clear picture cf the
effect of the variable over time. Since the major purpose of the experiment
was to demonstrate the effects of various therapeutic variables with anorex-
ics, we chose to present the data in this way. It was not our intention,
however, to ignore the daily variability. The fairly regular pattern of change
suggests several environmental or metabolic factors that may account for
these changes. If one were interested in more basic research on eating patterns
in anorexics, one would have to explore possible sources of this variability in
a finer analysis than we chose to undertake here.
It is possible, of course, that feedback might not have produced the clear
and clinically relevant increase noted in these data. If feedback resulted in a
small increase in caloric intake that was clearly visible only when data were
averaged, one would have to resort to statistical tests to determine if the
increase could be attributed to the therapeutic variable over and above the
day-to-day variability (see chapter 9). Once again, however, one may question
the clinical relevance of the therapeutic procedure if the improvement in
behavior is so small that the investigator must use statistics to determine if
change actually occurred. If this situation obtained, the preferred strategy
might be to improvise on the experimental design and augment the thera-
peutic procedure such that more relevant and substantial changes were pro-
duced. The issue of clincial versus statistical significance, which was discussed
in some detail above, is a recurring one in single-case research. In the last
analysis, however, this is always reduced to judgments by therapists, educa-
tors, etc. on the magnitude of change that is relevant to the setting. In most
cases, these magnitudes are greater than changes that are merely statistically
significant.
The above example notwithstanding, the conservative and preferred ap-
proach of data presentation in single-case research is to present all of the data
so that other investigators may examine the intrasubject variability firsthand
and draw their own conclusions on the relevance of this variability to the
problem.
Large intrasubject variability is a common feature during repeated mea-
surements of target behaviors in a single-case, particularly in the beginning of
an experiment, when the subject may be accommodating to intrusive mea-
sures. How much variability the researcher is willing to tolerate before
introducing an independent variable (therapeutic procedure) is largely a

question of judgment on the part of the investigator. Similar procedural
problems arise when introduction of the independent variable itself results in
increased variability. Here the experimenter must consider alteration in length

of phases to determine if variability will decrease over time (as it often does),
clarifying the effects of the independent variable. These procedural questions
will be discussed in some detail in chapter 3.
2.5. RELATION OF VARIABILITY

TO GENERALITY OF FINDINGS
The search for sources of variabiHty within individuals and the use of
improvised and fast-changing experimental designs appear to be contrary to

one of the most cherished goals of any science the establishment of general-
ity of findings. Studying the idiosyncrasies of one subject would seem, on the
surface, to confirm Underwood's (1957) observation that intensive study of

individuals will lead to discovery of laws that are applicable only to that
individual. In fact, the identification of sources of variability in this manner
leads to increases in generality of findings.
If one assumes that behavior is lawful, then identifying sources of variabil-
ity inone subject should give us important leads in sources of variability in
other similar subjects undergoing the same treatments. As Sidman (1960)
pointed out.
Tracking down sources of variability is then a primary technique for establishing

generality. Generality and variability are basically antithetical concepts. If there
are major undiscovered sources of variability in a given set of data, any attempt
to achieve subject or principle generality is likely to fail. Every time we discover
and achieve control of a factor that contributes to variability, we increase the
likelihood that our data will be reproducible with new subjects and in different
situations. Experience has taught us that precision of control leads to more
extensive generalization of data (p. 152).
And again,
It is unrealistic to expect that a given variable will have the same effects upon all
subjects under all conditions. As we identify and control a greater number of the
conditions that determine the effects of a given experimental operation, in effect
we decrease the variability that may be expected as a consequence of the opera-
tion. It then becomes possible to produce the same results in a greater number of
subjects. Such generality could never be achieved if we simply accepted inter-
subject variability and gave equal status to all deviant subjects in an investigation
(p. 190).
In other words, the more we learn about the effects of a treatment on

different individuals, in different settings, and so on, the easier it will be to
determine if that treatment will be effective with the next individual walking
into the office. But if we ignore differences among individuals and simply
average them into a group mean, it will be more difficult to estimate the
effects on the next individual, or "generalize" the results. In applied research,
when intersubject and intrasubject variability are enormous, and putative

sources of the variability are difficult to control, the establishment of general-
ity is a difficult task indeed. But the establishment of a science of human
behavior change depends heavily on procedures to establish generality of
findings. This important issue will be discussed in the next section.
2.6. GENERALITY OF FINDINGS

Types of generality
Generalization means many things. In applied research, generalization

usually refers to the process in which behavioral or attitudinal changes in the
treatment setting "generalize" to other aspects of the client's life. In educa-
tional research this can mean generalization of behavioral changes from the
classroom to the home. Generalization of this type can be determined by
observing behavioral changes outside of the treatment setting.
There are at least three additional types of generality in behavior change
research, however, that are more relevant to the present discussion. The first
is generality of findings across subjects or clients; that is, if a treatment effects

certain behavior changes in one subject, will the same treatment also work in
other subjects with similar characteristics? As we shall see below, this is a
large question because subjects can be "similar" in many different ways. For
instance, subjects may be similar in that they have the same diagnostic labels
or behavioral disorders (e.g., schizophrenia or phobia). In addition, subjects
may be of similar age (e.g., between 14 and 16) or come from similar
socioeconomic backgrounds.
Generality across behavior change agents is a second type. For instance,
will a therapeutic technique that is effective when applied by one behavior
change agent also be effective when applied to the same problem by different
agents? A common example is the classroom. If a young, attractive, female
teacher successfully uses reinforcement principles to control disruptive behav-
ior in her classroom, will an older female teacher who is more stern also be
able to apply successfully the same principles to similar problems in her class?
Will an experienced therapist be able to treat a middle-aged claustrophobic
more effectively than a naive therapist who uses exactly the same procedure?
A third type of generality concerns the variety of settings in which clients
are found. The question here is will a given treatment or intervention applied
by the same or similar therapist, to similar clients, work as well in one setting
as another? For example, would reinforcement principles that work in the
classroom also work in a summer camp setting, or would desensitization of
an agoraphobic in an urban office building be more difficult than in a rural
setting?
These questions are very important to clinicians who are concerned with
which treatments are most effective with a given client in a given setting.
have looked to the applied researcher to answer these
Typically, clinicians
questions.
Problems in generalizing from a single-case
The most obvious limitation in studying a single-case is that one does not
know if the results from this case would be relevant to other cases. Even if
one isolates the active therapeutic variable in a given client through a rigorous
single-case experimental design, critics note that there is little basis for infer-
ring that this therapeutic procedure would be equally effective when applied
to clients with similar behavior disorders (client generality) or that different
therapists using this technique would achieve the same results (therapist
generality). Finally, one does not know if the technique would work in a
different setting (setting generality). This issue, more than any other, has
retarded the development of single-case methodology in applied research and
has caused many authorities on research to deny the utility of studying a
single-case for any other purpose than the generation of hypotheses (e.g.,
Kiesler, 1971). Conversely, in the search for generality of applied research
findings, the group comparison approach appeared to be the logical answer
(Underwood, 1957).
In the specific area of individual human behavior, however, there are issues
that limit the usefulness of a group approach in establishing generality of
findings. On the other hand, the newly developing procedures of direct,
systematic, and clinical replication offer an alternative, in some instances, for
establishing generality of findings relevant to individuals. The purpose of this
section is to outline the major issues, assumptions, and goals of generality of
findings as related to behavior change in an individual and to describe the
advantages and disadvantages of the various procedures to establishing
generality of findings.
2.7. LIMITATIONS OF GROUP DESIGNS IN

ESTABLISHING GENERALITY OF FINDINGS
In chapter 1, section 1.5, several limitations of group designs in applied
research noted by Bergin and Strupp (1972) were outlined. One of the
limitations referred to difficulties in generalizing results from a group to an
two problems stand out. The first is inferring that
individual. In this category,
resultsfrom a relatively homogeneous group are representative of a given
population. The second is generalizing from the average response of a hetero-
geneous group to a particular individual. These two problems will be dis-
cussed in turn.
Random sampling and inference in applied research
After the brilliant work of R. A. Fisher, early applied researchers were

most concerned with drawing a truly random sample of a given population,
so that results would be generalizable to this population. For instance, if one
wished to draw some conclusion on the effects of a given treatment for
schizophrenia, one would have to draw a random sample of all schizophrenics.
In reference to the three types of generality mentioned above, this means
that the clients under study (e.g., schizophrenics) must be a random sample of
all schizophrenics, not only for behavioral components of the disorder, such
as loose associations or withdrawn behavior, but also for other patient

characteristics such as age, sex, and socioeconomic status. These conditions
must be fulfilled before one can infer that a treatment that demonstrates a
statistically significant effect would also be effective for other schizophrenics
outside of the study. As Edgington (1967) pointed out, "In the absence of
random samples hypothesis testing is still possible, but the significance state-
ments are restricted to the effect of the experimental treatments on the
subjects actually used in the experiment, generalization to other individuals
being based on logical nonstatistical considerations" (p. 195). If one wishes to
make statements about effectiveness of a treatment across therapists or
settings, random samples of therapists and settings must also be included in
the study.
Random sampling of characteristics in the animal laboratories of experi-
mental psychology is most relevant
feasible, at least across subjects, since
characteristics such as genetic and environmental determinants of individual
behavior can be controlled. In clinical or educational research, however, it is
extremely difficult to sample adequately the population of a particular syn-

drome. One reason for this is the vagueness of many diagnostic categories
(e.g., schizophrenia). In order to sample the population of schizophrenics one
must be able to pinpoint the various behavioral characteristics that make up
this diagnosis and ensure that any sample adequately represents these behav-
iors. But the relative unreliability of this diagnostic category, despite improve-
ments in recent years (Spitzer, Forman, & Nee, 1979), makes it very difficult
to determine the adequacy of a given sample. In addition, the therapeutic
emphasis may differ from setting to setting. In one center, bizarre behavior
and hallucinations may be emphasized. In another center, a thought disorder
may be the primary target of assessment (Neale & Oltmanns, 1980; Wallace,
Boone, Donahoe, & Foy, in press).
A second problem that arises when one is attempting an adequate sample
of a population is the availability of clients who have the needed behavior or
characteristics to fill out the sample (see chapter 1, section 1.5). In laboratory
animal research this is not a problem because subjects with specified charac-
teristics or genetic backgrounds can be ordered or produced in the laborator-
ies. In applied research, however, one must study what is available,and this
may result in a heavy weighting on certain client characteristics and inade-
quate sampling of other characteristics. Results of a treatment applied to this
sample cannot be generalized to the population. For example, techniques to
control disruptive behavior in the classroom will be less than generalizable if
they are tested in a class where students are from predominantly middle-class
suburbs and inner-city students are underrepresented.
Even in the great snake phobic epidemic of the 1960s, where the behavior
in question was circumscribed and clearly defined, the clients to whom
various treatments were applied were almost uniformly female college sopho-
mores whose fear was neither too great (they could not finish the experiment
on time) nor too little (they would finish it too quickly). Most investigators
admitted that the purpose of these experiments was not to generalize treat-
ment results to clinical populations, but to test theoretical assumptions and
generate hypotheses. The fact remains, however, that these results cannot even
be generalized beyond female college sophomores to the population of snake
fearers, where age, sex, and amount of fear would all be relevant.
It should be noted that all examples above refer to generality of findings
across clients with simalar behavior and background characteristics. Most
studies at least consider the importance of generality of findings along this
dimension, although few have been successful. What is perhaps more impor-
tant is the failure of most studies to consider the generality problem in the
other two dimensions namely, setting generality and behavior change agent
(therapist) generality. Several investigators (e.g., Kazdin, 1973b, 1980b;
McNamara & MacDonough, 1972) have suggested that this information may
be more important than client generality. For example, Paul (1969) noted
after a survey of group studies that the results of systematic desensitization
seemed to be a function of the qualifications of the therapist rather than
differences among clients. Furthermore, in regard to setting generality,
Brunswick (1956) suggested that, "In fact, proper sampling of situations and
problems may be in the end more important than proper sampling of subjects
considering the fact that individuals are probably on the whole much more
alike than are situations among one another" (p. 39). Because of these
problems, many sophisticated investigators specializing in research methodol-
ogy have accepted the impracticability of random sampling in this context
and have sought other methods for establishing generality (e.g., Kraemer,
1981).
The failure to be able to make statistically inferential statements, even
about populations of clients based on most clinical research studies, does not
mean no statements about generality can be made. As Edgington (1966)
that
pointed out, one can make statements at least on generality of findings to
similar clients based on logical non-statistical considerations. Edgington re-
ferred to this as logical generalization, and this issue, along with generality to
settings and therapists, will be discussed below in relation to the establishment

of generality of findings from a single-case.
Problems in generalizing from the group to the individual
The above discussion might be construed as a plea for more adequate

sampling procedures involving larger numbers of clients seen in many dif-
ferent settings
by a variety of therapists in other words, the notion of the
"grand collaborative study," which emerged from the conferences on re-
search in psychotherapy in the 1960s (e.g., Bergin & Strupp, 1972; Strupp &
Luborsky, 1962). On the contrary, one of the pitfalls of a truly random
sample in applied research is that the more adequate the sample, in that all
relevant population characteristics are represented, the less relevance will this
finding have for a specific individual. The major issue here is that the better
the sample, the more heterogeneous the group. The average response of this
group, then, will be less likely to represent a given individual in the group.
Thus, ifone were establishing a random sample of severe depressives, one
should include clients of various ages, and racial, and socioeconomic back-
grounds. In addition, cHents with various combinations of the behavior and
thinking or perceptual disorder associated with severe depression must be
included. It would be desirable to include some patients with severe agitation,
others demonstrating psychomotor retardation, still others with varying de-
grees and types of depressive delusions, and those with somatic correlates
such as terminal sleep disturbance. As this sample becomes truly more
random and representative, the group becomes more heterogeneous. The
specific effects of a given treatment on an individual with a certain combina-
tion of problems becomes lost in the group average. For instance, a certain
treatment might alleviate severe agitation and terminal sleep disturbance but
have a deleterious effect on psychomotor retardation and depressive delu-
sions. If one were to analyze the results, one could infer that the treatment,
on the average, is better than no treatment for the population of patients with
severe depression. For the individual clinician, this finding is not very helpful
and could actually be dangerous if the clinician's patient had psychomotor
retardation and depressive delusions.
Most studies, however, do not pretend to draw a truly random sample of
patients with a given diagnosis or behavior disorder. Even the most recent,
excellent, example of a general collaborative study on treatments for depres-
sion where random sampling was perhaps feasible did not attempt random
sampling (NIMH, 1980). Most studies choose clients or patients on the basis
of availability after deciding on inclusion and exclusion criteria and then
randomly assign these subjects into two or more groups that are matched on
relevant characteristics. Typically, the treatment is administered to one group
while the other group becomes the no-treatment control. This arrangement,
which has characterized much clinical and educational research, suffers for
two reasons; (1) To the extent that the "available" clients are not a random
sample, one cannot generalize to the population; and (2) to the extent that the
group is heterogeneous on any of a number of characteristics, one cannot
make statements about the individual. The only statement that can be made
concerns the average response of a group with that particular makeup which,
unfortunately, is unlikely to be duplicated again. As Bergin (1966) noted, it
was even difficult to say anything important about individuals within the
group based on the average response because his analysis demonstrated that
some were improving and some deteriorating (see Strupp & Hadley, 1979).
The result, as Chassan (1967, 1979) eloquently pointed out, was that the
behavior change agent did not know which treatment or aspect of treatment
was effective that was statistically better than no treatment but that actually
might make a particular patient worse.
Improving generality of findings to the individual

through homogeneous groups: Logical generalization
What Bergin and Strupp (1972) and others (e.g., Kiesler, 1971; Paul, 1967)
recognized was that if anything important was going to be said about the
group would have to be
individual, after experimenting with a group, then the
homogeneous For example, in a study of a
for relevant client characteristics.
group of agoraphobics, they should all be in one age-group with a relatively
homogeneous amount of fear and approximately equal background (per-
sonality) variables. Naturally, clients in the control group must also be
homogeneous for these characteristics.
Although this approach sacrifices random sampling and the ability to make
about the population of agoraphobics, one can begin to
inferential statements
say something about agoraphobics with the same or similar characteristics as
those in the study through the process of logical generalization (Edgington,
1967, 1980a). That is, if a study shows that a given treatment is successful
with a homogeneous group of 20- to 30-year-old female agoraphobics with
certain personality characteristics, then a clinician can be relatively confident
that a 25-year-old female agoraphobic with those personality characteristics
will respond well to that same treatment. (Recently some experts have sug-
gested that one should not assemble groups that are too homogeneous, for
even the ability to generalize on more logical grounds might be greatly
restricted [Kraemer, 19811.)
The process of logical generalization depends on similarities between the
patients in the homogeneous group and the individual in question in the
clinician's office. Which features of a case are important for extending logical
generalization and which features can be ignored (e.g., hair color) will depend
on the judgment of the clinician and the state of knowledge at the time. But if
one can generalize in logical fashion from a patient whose results or charac-
teristics are well specified as part of a homogeneous group, then one can also
logically generalize from a single individual whose response and biographical

characteristics are specified. In fact, the rationale has enabled applied re-
searchers to generalize the results of single-case experiments for years (Dukes,
1965; Shontz, 1965). To increase the base for generalization from a single-
case experiment, one simply repeats the same experiment several times on
similar patients, thereby providing the clinician with results from a number of
patients.
2.8. HOMOGENEOUS GROUPS VERSUS

REPLICATION OF A SINGLECASE EXPERIMENT
Because the issue of generalization from single-case experiments in applied
research is a major source of controversy (Agras, Kazdin, & Wilson, 1979;
Kazdin, 1980b, 1982b; Underwood, 1957), the sections to follow will describe
our views of the relative merits of replication studies versus generalization
from homogeneous groups.
As a basis for comparison, it is useful to compare the single-case approach
with PauFs (1967, 1969) incisive analysis of the power of various experimental
designs using groups of clients. Within the context of the power of these
various designs to establish cause-effect relationships, Paul reviewed the
several procedures commonly used in applied research. These procedures
range from case studies with and without measurement, from which cause-
effect relationships can seldom if ever be extracted, through series of cases
typically reporting percentage of success with no control group. Finally, Paul
cited the two major between-group experimental designs capable of establish-
ing functional relationships between treatments and the average response of
clients in the group. The first is what Paul referred to as the nonfactorial
design with no-treatment control, in other words the comparison of an
experimental (treatment) group with a no-treatment control group. The sec-
ond design is the powerful factorial design, which not only establishes cause-
effect relations between treatments and clients but also specifies what type of
clients under what conditions improve with a given treatment; in other words,
client-treatment interactions. The single-case replication strategy paralleling
the nonfactorial design with no-treatment control is direct replication. The
replication strategy paralleling the factorial design is called systematic replica-
tion.
Direct replication and treatment/no-treatment control group design
When Paul's article was written employing single-

(1967), applied research
case designs, usually of the A-B-A variety, was
appear (e.g.,
just beginning to
Ullmann & Krasner, 1965). Paul quickly recognized the validity or power of
this design, noting that "The level of product for this design approaches that
of the nonfactorial group design with no-treatment controls" (p. 117). When
Paul spoke of level of product here he was referring, in Campbell and
Stanley's (1963) terms, to internal validity, that is, the power of the design to
isolate the independent variable (treatment) as responsible for experimental
effects and to external validity or the ability to generalize findings across
relevant domains such as client, therapist, and setting. We would agree with
Paul's notions that the level of product of a single-case experimental design
only "approaches" that of treatment/no-treatment group designs, but for
somewhat different reasons. It is our contention that the single-case A-B-A
design approaches rather than equals the nonfactorial group design with no-
treatment controls only because the number of clients is considerably less in a
single-case design(N = I) than in a group design, where 8, 10, or more clients
are not uncommon. It is our further contention that, in terms of external
validity or generality of findings, a series of single-case designs in similar
clients in which the original experiment is directly replicated three or four
times can far surpass the experimental group/no-treatment control group
design. Some of the reasons for this assertion are outlined next.
Results generated from an experimental group/no-treatment control group
study as well as a direct replication series of single-case experimental designs
yield some information on generality of findings across clients but cannot
address the question of generality across different therapists or settings.
Typically, the group study employs one therapist in one setting who applies a
given treatment to a group of clients. Measures are taken on a pre-post basis.
Premeasures and postmeasures are also taken from a matched group of
clients in the control group who do not receive the intervening treatment. For
example, 10 depressive patients homogeneous on behavioral and emotional
aspects of their depression, as well as personality characteristics, would be
compared to a matched group of patients who did not receive treatment.
Logical generalization to other patients (but not to other therapists or set-
tings) would depend on the degree of homogeneity among the depressives in
both groups. As noted above, the less homogeneous the depression in the
experiment, the greater the difficulty for the practicing clinician in determin-
ing if that treatment is effective for his or her particular patient. A solution to
this problem would be to specify in some detail the characteristics of each
patient in the treatment group and present individual data on each patient.
The clinician could then observe those patients that are most like his or her
particular client and determine if these experimental patients improved more

than the average response in the control group. For example, after describing
in detail the case history and presenting symptomatology of 10 depressives,
one could administer a pretest measuring severity of depression to the 10
depressives and a matched control group of 10 depressives. After treatment
of the 10 depressives in the experimental group, the posttest would be
administered. When results are presented, the improvement (or lack of
improvement) of each patient group could be presented
in the treatment
either graphically or in numerical form along with
the means and standard
deviations for the control group. After the usual procedure to determine
statistical significance, the clinician could examine the amount of improve-
ment of each patient in the experimental group to determine (1) if the

improvement were clinically relevant, and (2) if the improvement exceeded
any drift toward improvement in the control group. To the extent that some
patients in the treatment group were similar to the clinician's patient, the
clinician could begin to determine, through logical generalization, whether
the treatment might be effective with his or her patient.
However, a series of single-case designs where the original experiment is
replicated on a number of patients also enables one to determine generality of
findings across patients (but not across therapists or settings). For example, in
the same hypothetical group of depressives, the treatment could be adminis-
tered in an A-B-A-B design, where A
represents baseline measurement and B
represents the treatment. The comparison here is still between treatment and
no treatment. As results accumulate across patients, generality of findings is
estabHshed, and the results are readily translatable to the practicing clinician,
since he or she can quickly determine which patient with which characteristics
improved and which patient did not improve. To the extent that therapist and
treatment are alike across patients, this is the clinical prototype of a direct
replication series (Sidman, 1960), and it represents the most common replica-
tion tactic in the experimental single-case approach to date.
Given these results, other attributes of the single-case design provide added
strength in generalizing results to other clients. The first attribute is flexibility
(noted in section 2.3). If a particular procedure works well in one case but
works less well or fails when attempts are made to replicate this in a second or
third case, slight alterations in the procedure can be made immediately. In
many cases, reasons for the inability to replicate the findings can be ascer-
tained immediately, assuming that procedural deficiencies were, in fact, re-
sponsible for the lack of generality. An was outlined in
example of this result
one patient
section 2.3, describing intersubject variability. In this example,
improved with treatment, but a second did not. Use of an improvised
experimental design at this point allowed identification of the reason for
failure. This finding should increase generality of findings by enabling imme-
diate application of the altered procedure to another patient with a similar
response pattern. This is an example of Sidman's (1960) assertion that "t-ack-

ing down sources of variability is then a primary technique of establishing
generality" (see also Kazdin, 1973b; Leitenberg, 1973; Skinner, 1966b). If

alterations in the procedure do not produce clinical improvement, either
differences in background, personality characteristics, or differences within
the behavior disorder itself can be noted, suggesting further hypotheses on
procedural changes that can be tested on this type of client at a later date.
Finally, using the client as his or her own control in successive replications
provides an added degree of strength in generalizing the effect of treatment
across differing clients. In group or single-case designs employing no-treat-
ment controls or attention-placebo controls, it is possible and even quite
likely that certain environmental events in a no-treatment control group or
phase will produce considerable improvements (e.g., placebo effects). In a
nonfactorial group design, where treated clients show more improvement
than clients in a no-treatment control, one can conclude that the treatment is
effective and then proceed in generalizing results to other clients in clinical

situations. However, the degree of the contribution of nonspecific environ-
mental factors to the improvement of each individual client is difficult to
judge. In a single-case design (for example, the A-B-A-B or true withdrawal
design), the influence of environmental factors on each individual client can
be estimated by observing the degree of deterioration when treatment is
withdrawn. If environmental or other factors are operating during treatment,
improvement will continue during the withdrawal phase, perhaps at a slower
rate, necessitating further experimental inquiry. Even in a nonfactorial group
design with powerful effects, the contribution of this factor to individual
clients is difficult to ascertain.
Systematic and clinical replication and factorial designs
Direct replication series and nonfactorial designs with no-treatment con-

trols come to grips with only one aspect of generality of findings generality
across clients. These designs are not capable of simultaneously answering
questions on generality of findings across therapists, settings, or clients that
some substantial degree from the original homogeneous group. For
differ in
example, one might ask, if the treatment works for 25-year-old female
agoraphobics with certain personality characteristics, will it also work for a
40-year-old female agoraphobic with different personality characteristics?
In the therapist domain, the obvious question concerns the effectiveness of
treatment as related to that particular therapist. If the therapist in the
hypothetical study were an older, more experienced therapist, would the
treatment work as well with a young therapist? Finally, even if several
therapists in one setting were successful, could therapists in another setting
and geographical area attain similar results?
To answer all of these questions would require literally hundreds of experi-

mental group/no-treatment control group studies where each of the factors
relevant to generalization was varied one at a time (e.g., type of therapist,
type of client). Even if this were feasible, however, the results could not
always be attributed to the factor in question as replication after replication
ensued, because other sources of variance due to faulty random assignment
of clients to the group could appear.
In reviewing the status and goals of psychotherapy research, many clinical
investigators (e.g., Kazdin, 1980b, 1982b; Kiesler, 1971; Paul, 1967) proposed
most sophisticated experimental designs
the application of one of the in the

armamentarium of the psychological researcher the factorial design as an
answer to the above problem. In this design, relevant factors in all three areas
of generality of concern to the clinician can be examined. The power of this
design is in the specificity of the conclusion.

For example, the effects of two antidepressant pharmacological agents and
a placebo might be evaluated in two different settings (the inpatient ward of a
general hospital and an outpatient community mental health center) on two
groups of depressives (one group with moderate to severe depression and a
second group with mild depression). A therapist in the psychiatric ward
setting would administer each treatment to one half of each group of depres-
sives the moderate to severe group and the mild group. All depressives
would be matched as closely as possible on background variables such as age,
sex, and personality characteristics. The same therapist could then travel to
the community mental health center and carry out the same procedure. Thus
we have a 2 x 2 x 2 factorial design. Possible conclusions from this study
are numerous, but results might be so specific as to indicate that antidepres-
sants do work but only with moderate to severe depressives and only if
hospitalized in a psychiatric ward. It would not be possible to draw conclu-
sions on the importance of a particular type of therapist because this factor
was not systematically varied. Of course, the usual shortcomings of group
designs are also present here because results would be presented in terms of
group averages and intersubject variability. However, to the extent that
subjects in each experimental cell were homogeneous and to the extent that
improvement was large and clinically important rather than merely statisti-
cally significant, then results would certainly be a valuable contribution. The
clinical practitioner would be able to examine the characteristics of those
subjects in the improved group and conclude that under similar conditions
(i.e., an inpatient psychiatric unit) his or her moderate to severe depressive
patient would be likely to improve, assuming, of course, that this patient

resembled those in the study. Here again, the process of logical generalization
rather than statistical inference from a sample to a population is the active
mechanism.
Thus, while the factorial design can be effective in specifying generality of
findings across all important domains in applied research (within the limits
discussed above), one major problem remains: Applied researchers seldom do

this kind of study.As noted in chapter 1, section 1.5, the major reasons for
this are practical. The enormous investment of money and time necessary to
collect large numbers of homogeneous patients has severely inhibited this

type of endeavor. And often, even in several different settings, the necessary
number of patients to complete a study is just not available unless one is
willing to wait years. Added to this are procedural difficuhies in recruiting

and paying therapists, ensuring adequate experimental controls such as dou-
ble-blind procedures within a large setting, and overcoming resistance to
assigning a large number of patients to placebo or control conditions, as well
as coping with the laborious task of recording and analyzing large amounts of
data (Barlow & Hersen, 1973; Bergin & Strupp, 1972).
In addition, the arguments raised in the last section on inflexibility of the
group design are also applicable here. If one patient does not improve or
reacts in an unusual way to the therapeutic procedure, administration of the
procedure must continue for the specified number of sessions. The unsuccess-
ful or aberrant results are then, of course, averaged into the group results
from that experimental cell, thus precluding an immediate analysis of the
intersubject variability, which will lead to increased generality.
Systematic and clinical replication procedures involve exploring the effects
of different settings, therapists, or clients on a procedure previously demon-
strated as successful in a direct replication series. In other words, to borrow
the example from the factorial design, a single-case design may demonstrate
that a treatment for severe depression works on an inpatient unit. Several
direct replications then establish generality among homogeneous patients.
The next task is to replicate the procedure once again, in different settings
with different therapists or with patients with different background charac-
teristics. Thus the goals of systematic and clinical replication in terms of
generality of findings are similar to those of the factorial study.

At first glance, it does not appear as if replication techniques within single-
case methodology would prove any more practical in answering questions
concerning generality of findings across therapists, settings, and types of
behavior disorder. While direct replication can begin to provide answers to
questions on generality of findings across similar clients, the large questions
of setting and therapist generality would also seem to require significant
collaboration among diverse investigators, long-range planning, and a large
investment of money and time the very factors that were noted by Bergin
and Strupp (1972) to preclude these important replication effects. The sur-
prising fact concerning this particular method of replication, however, is that
these issues are not interfering with the establishment of generality of find-
ings, since systematic and clinical replication is in progress in a number of
areas of applied research. In view of the fact that systematic and clinical
replication has the same advantages of logical generalization as direct replica-

tion, the information yielded by the procedure has direct applicability to the
Examples from these ongoing systematic replication and clinical series
clinic.
and procedures and guidelines for replication will be described in chapter 10.
2.9. APPLIED RESEARCH QUESTIONS

REQUIRING ALTERNATIVE DESIGNS
It was observed in chapter 1 that applied researchers during the 1950s and
1960s often considered single-case versus between-group comparison research
as an either-or proposition. Most investigators in this period chose one
methodology or the other and eschewed the alternative. Much of this polemic
characterized the idiographic-nomothetic dichotomy in the 1950s (Allport,
1961). This type of argument, of course, prevented many investigators from
asking the obvious question: Under what condition is one type of design more
appropriate than another? As single-case designs have become more sophisti-
cated, the number of questions answered by this strategy has increased. But
there are many instances in which single-case designs either cannot answer the
relevant applied jesearch question or are less applicable. The purpose of this
book, of course, is to make a case for the relevance of single-case experimen-
tal designs and to cover those issues, areas, and examples where a single-case
approach is appropriate and important. We would be remiss, however, in
ignoring those areas where alternative experimental designs offer a better
answer.
Actuarial questions
There are several related questions or issues that require experimental

strategies involving groups. Baer (1971) referred to one as actuarial, although
he might have said political. The fact is, after a treatment has been found
effective, society wants to know the magnitude of its effects. This informa-
tion is often best conveyed in terms of percentage of people who improved
compared to an untreated group. If one can say that a treatment works in 75
out of 100 cases where only 15 out of 100 would improve without treatment,
this is the kind of information that is readily understood by society. In a
systematic replication series, the results would be stated differently. Here the
investigator would say that under certain conditions the treatment works,
while under other conditions it does not work, and other therapeutic variables
must be added. While this statement might be adequate for the practicing
clinician or educator, little information on the magnitude of effect is con-
veyed. Because society supports research and, ultimately, benefits from it, this
actuarial approach is not As Baer (1971) pointed out, this problem

trivial.
"... is similar to that of any insurance company, we merely need to know

how often a behavioral analysis changes the relevant behavior of society
toward the behavior, just as the insurance company needs to know how often
age predicts death rates" (p. 366). It should be noted, however, that a study
such as this cannot answer why a treatment works; it is simply capable of
communicating the size of the effect. But if the treatment package is the result
of a series of single-case designs, then one should already know why it works,
and demonstration of the magnitude of effect is all that is needed.
Several cautions should be noted when proceeding in this manner. First, the
cost and practical limitation of running a large-group study do not allow
unlimited replication of this effort, if it can be done at all. Thus one should
have a well-developed treatment package that has been thoroughly tested in
single-case experimental designs and replications before embarking on this
effort. Preferably, the investigator should be well into a systematic replicaton
series in order to have some idea of the client, setting, or therapeutic variables
that predict success. Groups can then be constructed in a homogeneous
fashion. Premature application of the group comparison design, where a
treatment or the conditions under which it is effective have not been ade-
quately worked out, can only produce the characteristic weak effect with
large intersubject variability that is so prevalent in group comparison studies
to date (Bergin & Strupp, 1972). Of course, well-developed clinical replication
series,where a comprehensive treatment package is replicated across many
individuals with a given problem, can also specify size or effect and the
percentage of clinical success. But the information from the comparison
group would be missing.
Modification of group behavior
A related issue on the appropriateness of group design arises when the

applied researcher is not concerned with the fate of the individual but rather
with the effectiveness of a given procedure on a well-defined group. A
particularly good example is the classroom. If the problem is a mild but
annoying one, such as disruptive behavior in the classroom, the researcher
and school administrator may be more interested in quickly determining what
procedure is remedying this problem for the classroom as a whole.
effective in
The goal changing behavior of a well-defined group rather than
in this case is
individuals within that group. It may not be important that two or three
children remain somewhat out of order if the classroom is substantially more
quiet. A good example is an experiment on the modification of
particularly
classroom noise reported in chapter 7, Figure 7-5 (C. W. Wilson & Hopkins,
1973). A similar approach might be desirable with any coexisting group of
people, such as a ward in a state hospital where the control of disruptive
behavior would allow more efficient execution of individual therapeutic

programs (see chapter 5, Figure 5-17) (Ayllon & Azrin, 1965). This stands in
obvious contrast to a series of patients with severe clinical problems who do
not coexist in some geographical location but are seen sequentially and
assigned to a group only for experimental consideration. In this case, the
applied researcher would be ill-disposed to ignore the significant human
suffering of those individuals who did not improve or perhaps deteriorated.
When group behavior is the target, however, and a comparison of treated
and untreated classrooms, for example, is desirable, one is not limited to
between-subject designs in these instances because within-subject designs are
also feasible. There are many examples where A-B-A or multiple baseline
designs have been used in classroom research with repeated measures of the
average behavior of the group (e.g., Wolf & Risley, 1971; see also chapters 5
and 6).
Once again, it is a good idea to have a treatment that has been adequately
worked out on individuals before attempting to modify behavior of a group.
If not, the investigator will encounter intolerable intersubject variability that
will weaken the effects of the intervention.
2.10. BLURRING THE DISTINCTION

BETWEEN DESIGN OPTIONS
The purpose of this book in general and this chapter in particular is to
illustrate the underlying rationale for single-case experimental designs. To
achieve this goal, the strategies and underlying rationale of more traditional
between-group designs have been placed in sharp relief relative to single-case
designs, to highlight the differences. This need not be the case. As described
throughout group designs could be carried out with close atten-
this chapter,
tion to individual change and repeated measures across time.

If one were comparing treatment and no treatment, for example, 10
depressed patients could be individually described and repeated measures
could be taken of their progress. Amount of change could then be reported in
clinically relevant terms. These data could be contrasted with the same
reporting of individual data for a no-treatment group. Of course, statistical
inferences could be made concerning group differences, based on group
averages and intersubject variability within groups, but one would still have
the individual data to back on. This would be important for purposes of
fall
logical generalization, which forms the only rational basis for generalizing
results from one group of individual subjects to another individual subject. In
our experience as editors of major journals, data from group studies are
being reported increasingly in this manner, as investigators alter their underly-
ing rationale for generality of findings from inferential to logical. With

individuals carefully described and closely tracked during treatment, the
investigator is in a position to speculate on sources of intersubject variability.
That is, one subject improves dramatically while another improves only
if
marginally or perhaps deteriorates during treatment, the investigator can

immediately analyze, at least in dipost hoc fashion, differences between these
clients. The investigator would be greatly assisted in making these judgments
by repeated measurement within these group studies because the investigator

could determine if a specific client was making good progress and then
faltered, or simply did not respond at all from the beginning of treatment.
Events correlated with a sudden change in the direction of progress could be
noted for future reference. All that the investigator would be lacking would
be the flexibility inherent in single-case design which would allow a quick
change in experimental strategy or an experimental strategy based on the
responses of the individual client (Edgington, 1983) to immediately track
down the sources of this intersubject variability. Of course, many other
factors must be considered when choosing appropriate designs, particularly
practical considerations such as time, expense, and availability of subjects.
Once again we would suggest that if one is going to generalize from group
studies to the variety of individuals entering a practitioner's office, then it is
essential that data from individual clients be described so that the process of
logical generalization can be applied in its most powerful form. In view of the
inapplicability of making statistical inferences to hypothetical populations,
based on random sampling, logical generalization is the only method avail-
able to us, and we must maximize its strength with thorough description of
individuals in the study.
With these cautions in place, and with a full understanding of the rationale
and strengths of single-case designs, the investigator can then make a
reasoned choice on design options. For example, for comparing two treat-
ments with no treatment, where each treatment should be effective but the
relative effectiveness is unknown, one might choose an alternating-treatments
design (see chapter 8) or a more traditional between-group comparison design
with close attention to individual change. The strengths and advantages of
alternating-treatments designs are fully discussed in chapter 8, but if one has
a large number of subjects available and a fixed treatment protocol that for
one reason or another cannot be altered during treatment, regardless of
progress, then one may wish to use a between-group strategy with appropriate
attention to individual data. Subsequent experimental strategies could be
employed using single-case experimental designs during follow-up to deal
with minimal responders or those who do not respond at all or perhaps
deteriorate. But sources of intersubject variability must be tracked down
eventually if we are to advance our science and ensure the generality of our
results. Treatment in between-group designs could also be applied in a rela-
lively "pure" form, much as it would be in a clinical setting. Occasionally we

will refer to these options in the context of describing the various single-case
design options throughout this book.
A further blurring of the distinction occurs when single-case designs are
applied to groups of subjects. Section 5.6 and Figure 5-17 describe the
application of an A-B-A withdrawal design to a large group of subjects.
group is discussed in
Similarly, a multiple baseline design applied to a large
section 7.2.Data are described in terms of group averages in both experi-
ments. These experimental designs, then, approach the tradition of within-
subject designs (Edwards, 1968), where the same group of subjects
experiences repeated experimental conditions. Appropriate statistical analy-
ses have long been available for these design options (e.g., Edwards, 1968).
Despite the blurring of experimental traditions that is increasingly taking
place, the overriding strength of single-case designs and their replications lies
in the use of procedures that are appropriate to studying the subject matter at
hand the individual. It is to a description of these procedures that we now
turn.
CHAPTER 3
General Procedures in
Single-case Research
3.1. INTRODUCTION
Advantages of the experimental single-case design and general issues involved
in this type of research were briefly outlined in chapter 2. In the present
chapter a more detailed analysis of general procedures characteristic of all
experimental single-case research will be undertaken. Although previous
discussion of these procedures has appeared periodically in the psychological
and psychiatric literatures (Barlow & Hersen, 1973; Hersen, 1982; Kazdin,
1982b; Kratchowill, 1978b; Levy & Olson, 1979), a more comprehensive
analysis, from both a theoretical and an applied framework, is very much
needed.
A review of the literature on applied clinical research since the 1960s shows
that there is a substantial increase in the number of articles reporting the use
of the experimental single-case design strategy. These papers have appeared in
a wide variety of educational, psychological, and psychiatric journals. How-
ever, many researchers have proceeded without the benefit of carefully
thought-out guidelines, and, as a consequence, needless errors in design and
practice have resulted. Even in the Journal of Applied Behavior Analysis,
which is primarily devoted to the experimental analysis model of research,
errors in procedure and practice are not uncommon in reported investiga-
tions.
and practical applica-

In the succeeding sections of this chapter, theoretical
tions of repeated measurement, methods for choosing an appropriate base-
line, changing one independent variable at a time, reversals and withdrawals,
length of phases, and techniques for evaluating effects of "irreversible"
67
procedures will be considered. For heuristic purposes, both correct and

incorrect applications of the aforementioned will be examined. Illustrations
of actual and hypothetical cases will be provided. In addition, discussions of
strategies to assess response maintenance following successful treatment is

provided.
3.2. REPEATED MEASUREMENT

Aspects of repeated measurement techniques have already been discussed
in chapter 2. However, in this section we
examine some of the issues in
will
greater detail. In the typical psychotherapy outcome study (e.g., Bellack,
Hersen, & Himmelhock, 1981), in which the randomly assigned or matched-
group design is used, dependent measures C^.g., Beck Depression Inventory
scores) usually are obtained only on a pretherapy, posttherapy, and follow-up
basis. Occasionally, however, a midtherapy assessment is carried out. Thus
possible fluctuations, including upward and downward trends and curvilinear
relationships, occurring throughout the course of therapy are omitted from
the analysis. However, whether espousing a behavioral, client-centered, exis-
tential, or psychoanalytic position, the experienced clinician is undoubtedly
cognizant that changes unfortunately do not follow a smooth linear function
from the beginning of treatment to its ultimate conclusion.
Practical implications and limitations
There are a number of important practical implications and limitations in

applying repeated measurement techniques when conducting experimental
single-case research (see chapter 2 for general discussion). F[rst of all, the
operations involved in o btaining such measurements (whether thev be mo -
toric, physiological, or attitudinal) must be clearly specified, observable .
public, and replicable in all respec ts. When measurement techniques require
the use of human observers, independent reliability checks must be es-
tablished (see chapter 4 for specific details). Secondly, rrif^^l'^^'^^^^S tt^V^P
r epeatedly, esp eciall y over extended periods of time, must be done under
exacting and totally standardized conditions with resp ect to measurement
devices use d,_:p ersonnel involved, time or times of day measurements are
recorded^ instructions ^^ ^^^ g"bjf ot, and specifi c pnvirnpmpntal mnditions
(e.g., location)where the mpavmrpjj^^pt SCSSionS OCCUr..
Deviations from any of the aforementioned conditions may well lead to
spurious effects in the data and might result in erroneous conclusions. This is
General Procedures in Single-case Research 69
of particular import at the point where the prevailing condition is experimen-

change from baseline to reinforcement conditions). In the
tally altered (e.g.,
event that an adventitious change in measurement conditions were to coincide

with a modification in experimental procedure, resulting differences in the
data could not be scientifically attributed to the experimental manipulation,
inasmuch as a correlative change may have taken place. Under these circum-
would either have to renew efforts or
stances, the conscientious experimenter
experimentally manipulate and evaluate the change in measurement tech-
nique.
The importance of maintaining standard measurement conditions bears
some illustration. Elkin, Hersen, Eisler, and Williams (1973) examined the
separate and combined effects of feedback, reinforcement, and increased
food presentation in a male anorexia nervosa patient. With regard to mea-
surement, two dependent variables
caloric intake and weight were ex-
amined daily. Caloric intake was monitored throughout the 42-day study
without the subject's knowledge. Three daily meals (each at a specified time)
were served to the subject while he dined alone in his room for a 30-minute
period. At the conclusion of each of the three daily meals, unknown to the
subject, the caloric value of the food remaining on his tray was subtracted
from the standard amount presented. Also, the subject was weighed daily at
approximately 2:00 p.m., in the same room, on the same scale, with his back
turned toward the dial, and, for the most part, by the same experimenter. In
this study, consistency of the experimenter was not considered crucial to
maintaining accuracy and freedom from bias in measurement. However,

maintaining consistency of the time of day weighed was absolutely essential,
particularly in terms of the number of meals (two) consumed until that point.
There are certain instances when a change in the experimenter will seriously
affect the subject's responses over time. Indeed, this was empirically evalu-
ated by Agras, Leitenberg, Barlow, and Thomson an alternating
(1969), in
treatment design (see chapter 8). However, in most single-case research,
unless explicitly planned, such change may mar the results obtained. For
example, when employing the Behavioral Assertiveness Test (Eisler, Miller, &
Hersen, 1973) over time repeatedly as a standard behavioral measure of
assertiveness, it is clear that the use of different role models to promote
responding might result in unexpected interaction with the experimental
condition (e.g., feedback or instructions) being manipulated. Even when
using more measurement tecniques, such as the mechanical strain
objective
gauge for recording penile circumference change (Barlow, Becker, Leiten-
berg, & Agras, 1970) in sexual deviates, extreme care should be exercised with
respect to instructions given and to the role of the examiner (male research
assistant) involved in the measurement session (cf. Wincze, 1982; Wincze &
Lange, 1981). A substitute for the original male experimenter, particularly in
the case of a homosexual pedophile in the early stages of his experimental

treatment, could conceivably result in spurious correlated changes in penile
circumference data.
There are several other important issues to be considered when using
repeated measurement techniques in applied clinical research. 5oi_xaniEle,
frequency of measurements obtained per unit of time should be given jnore
c areful attent ion. The experimenter oby jon^'y rruia<-<>iîr^ t^^t ^ yvffiri^t
rmmhprjj mpagiirpmpr|tg ar e recorded SO that a representative sample is
^
obtained .^ n^the^ ther hand th eLexperimenter must exercise caution to avoid

.
takin g toomany mfasnrpm^pt*; in a piyen ppnnH of time, as fatigue oa-tbe

part of the^suiajecUnav result This
is of paramount importance when taking
.
measurements that require an active response on the subject's part (e.g.,

number of erections to sexual stimuli over a specific time period, or repeated
modeling of responses during the course of a session in assertive training).
A uniqiip prnhlpm rîat^^d to measurement traditionally farpH hy mvpgtig}^-
tors working in institutional settings (state hospitals, training sfj
^honk f^r the
r etarded, etc.) involved the major environmental changes that tal^ p plare ^t
night and on weekends The . astute observer who has worked in these settings
is quite familiar with the distinction that is made between the "day" and
"night" hospital and the "work week" and the "weekend" hospital. Unless
the investigator is in the favored position to exert considerable control over
the environment (as were Ayllon and Azrin, 1968, in their studies on token
economy), careful attention should be paid to such differences. One possible
solution would be to restrict the taking of measurements across similar
conditions (e.g., measurements taken only during the day). A second solution
would involve plotting separate data for day and night measurements.
A totally different measurement problem is faced by the experimenter who
is intent on using self-report data on a repetitive basis (Herson, 1978). When
using thjsJYPe of assessment tecnique, the possibility always exists, evenjn

clinical subjects, that the subject's natural responsivitv will not be tapped, but
t hat data in confor mity to "ex perimental demand" (Orne, 1962) are being
reeorded. The use of torms and the correlation of self-report
alternate
(attitudinal) measures with motoric and physiological indexes of behavior are
some of the methods to ensure validity of responses. This is of particular
utility when measures obtained from the different response systems correlate
both highly and positively. Discrepancies in verbal and motoric indexes of

behavior have been a subject of considerable speculation and study in the
behavioral literature, and the reader is referred to the following for a more
complete discussion of those issues: Barlow, Mavissakalian, and Schofield
(1980); D. C. Cohen (1977); and Hersen (1973).
A gnal^jggilt^, rf-\ate(^ to rppâtpH Tr>pa<;iirement, involves the problem of
extiepie^daLly variability of a target behavior under study.. For example,
repetitive time sampling on a random basis within specified time limits is a
most useful technique for a variable subject to extreme fluctuations and

responsivity to environmental events (see Hersen, Eisler, Alford, & Agras,
1973; J. G. Williams, Barlow, & Agras, problems in measure-
1972). Similar
ment include the area of cyclic variation, an excellent example being the effect
of the female's estrus cycle on behavior. Issues related to cyclic variation in
terms of extended measurement sessions will be discussed more specifically in
section 3.6 of this chapter.
3.3 CHOOSING A BASELINE

In most experimental single-case designs (the exception is the B-A-B de-
sign), the initial period of observation involves the repeated measurement of
the natural frequency of occurrence of the target behaviors under study. This
initial period is defined as the baseline, and
most frequently designated as
it is
the A-phase of study (Barlow, Blanchard, Hayes, & Epstein, 1977; Barlow &
Hersen, 1973; Hersen, 1982; Risley & Wolf, 1972; Van Hasselt & Hersen,
1981). It should be noted that this phase was earlier labeled 0,020304 by
Campbell and Stanley (1966) in their analysis of quasi-experimental designs
for research (time series analysis).
The primary purpose of baseline measurement is to have a standard by
which the subsequent efficacy of an experimental intervention can be evalu-
ated. In addition, Risley and Wolf (1972) pointed out that, from a statistical
framework, the baseline period functions as a predictor for the level of the
target behavior attained in the future. A number of statistical techniques for
analyzing time series data have appeared in the literature (Edgington, 1982;
Wallace & Elder, 1980); the use of these methods will be discussed in chapter
9.
Baseline stability
When selecting a baseline, its stability and range Of variabilitymust be

carefully examined. McNamara and MacDonough (1972) have raised an issue
that is continuously faced by all of those involved in applied clinical research.
They specifically posed the following question: "How
long enough for
long is
a baseline?" (p. 364). Unfortunately, there no simple response or formula

is
that can be applied to this question, but a number of suggestions have been
made. Baer, Wolf, and Risley (1968) recommended that baseline measure-
ment be continued over time "until its stability is clear" (p. 94). McNamara
and MacDonough concurred with Wolf and Risley 's (1971) recommendation
that repeated measurement be applied until a stable pattern emerges. How-
ever, there are some practical and ethical limitations to extending initial
measurement beyond certain Hmits. The first involved a problem of logistics.
For the experimenter working in an institutional setting (unless in an ex-

tended-care facility), the subject under study will have to be discharged within
a designated period of time, whether upon self-demand, familial pressure, or
exhaustion of insurance company compensation. Secondly, even in a facility
giving extended care to its patients, there is an obvious ethical question as to
how long the applied clinical researcher can withhold a treatment application.
This assumes even greater magnitude when the target behavior under study
results in serious discomfort either to the subject or to others in the environ-
ment (see J. M. Johnston, 1972, p. 1036). Finally, although McNamara and
MacDonough (1972) argued that "The use of an extended baseline is a most
easily implemented procedure which may help to identify regularities in the
behavior under study" (p. 361), unexpected effects on behavior may be found
as a result of extended measurement through self-recording procedures (Hol-
lon & Bemis, Such effects have been found when subjects were asked
1981).
to record their behaviors under repeated measurement conditions. For exam-
ple, McFall (1970) found that when he asked smokers to monitor their rate of
smoking, increases in their actual smoking behavior occurred. By contrast,
smokers asked to monitor rate of resistance to smoking did not show parallel
changes in their behavior. The problem of self-recorded and self-reported
data will be discussed in more detail in chapter 4.
In the context of basic animal research, where the behavioral history of the
organism can be determined and controlled, Sidman (1960) has recom-
mended that, for stability, rates of behavior should be within a 5 percent
range of variability. Indeed, the "basic science" research is in a position to
create baseline data through a variety of interval and ratio scheduling effects.
However, even in animal resarch, where scheduling effects are programmed
to ensure stability of baseline conditions, there are instances where unex-
pected variations take place as a consequence of extrinsic variables. When
such variability is presumed to be extrinsic rather than intrinsic, Sidman
(1960) has encouraged the researcher to first examine the source of variability
through the method of experimental analysis. Then extrinsic sources of
variation can be systematically eliminated and controlled.
Sidman acknowledged, however, that the applied clinical researcher, by
virtue of his or her subject matter, when control over the behavioral history is
nearly impossible, is at a distinct disadvantage. He noted that "The behav-
ioral engineer must continuously take variability as he finds it, and deal with
it as an unavoidable fact of life" (Sidman, 1960, p. 192). He also acknowl-
edged that "The behavioral engineeer seldom has the facilities or the time that
would be required to eliminate variability he encounters in a given problem"
(p. 193). When variability in baseline measurements is extensive in applied
clinical research, it might be useful to apply statistical techniques for purposes
of comparing one phase to the next. This would certainly appear to be the
case when such variability exceeds a 50 percent level. The use of statistics
under these circumstances would then meet the kind of criticism that has been
who uses single-case methodology.
leveled at the applied clinical researcher
For example, Bandura (1969) argued that there is no difficulty in interpreting
performance changes when differences between phases are large (e.g., the
absence of overlapping distributions) and when such differences can be
replicated across subjects (see chapter 10). However, he underscored the
difficulties in reaching valid conclusions when there is "considerable variabil-
ity during baseline conditions" (p. 243).
Examples of baselines
With the exception of a brief discussion in Hersen (1982) and in Barlow

and Hersen 's (1973) paper, which was primarily directed toward a psychiatric
readership, the different varieties of baselines commonly encountered in
applied clinical research have neither been examined nor presented in logical
sequence in the experimental literature. Thus the primary function of this
section is to provide and familiarize the interested applied researcher with
examples of baseline patterns. For the sake of convenience, hypothetical
examples, based on actual patterns reported in the literature, will be illus-
tratedand described. Methods for dealing with each pattern will be outlined,
and an attempt to formulate some specific rules (a la cookbook style) will be
undertaken.
The issue concerning the ultimate length of the basehne measurement
phase was previously discussed in some detail. However, it should be pointed
out here that "A minimum of three separate observation points, plotted on
the graph, during this baseline phase are required to establish a trend in the
data" (Barlow & Hersen, 1973, p. 320). Thus three successively increasing or
decreasing points would constitute establishment of either an upward or
downward trend in the data. Obviously, in two sets of data in which the same
trend is exhibited, differences in the slope of the line will indicate the extent or
power of the trend. By contrast, a pattern in which only minor variation is
seen would indicate the recording of a stable baseline pattern. An example of
such a stable baseline pattern is depicted in Figure 3-1 Mean number of facial
.
tics averaged over three daily 15-minute videotaped sessions are presented for
a 6-day period. Visual inspection of these data reveal no apparent upward or
downward trend. Indeed, data points are essentially parallel to the abscissa,
minimum. This kind of baseline pattern, which
while variability remains at a
shows a constant rate of behavior, represents the most desirable trend, as it
permits an unequivocal departure for analyzing the subsequent efficacy of a
treatment intervention. Thus the beneficial or detrimental effects of the
following intervention should be clear. In addition, should there be an ab-
sence of effects following introduction of a treatment, it will also be ap-
parent. Absence of such effects, then, would graphically appear as a
S250 .
P
^ 200 -
o
<
"-
u.
O
150

^"~^*"~~"*~
~-_ ^
'-t^^"'^

*
g 100

2
LU
-
g, 50
UJ
ec
"-
1 1 1 1
3 4
DAYS
FIGURE 3-1. The stable baseline. Hypothetical data for mean number of facial tics averaged
over three daily 15-minute videotaped sessions.
continuation of the steady trend first established during the baseline measure-
ment phase.
A second type of baseline trend that frequently is encountered in applied
clinical research is such that the subject's condition under study appears to be
worsening (known as the deteriorating baseline Barlow & Hersen, 1973).
Once again, using our hypothetical data on an example of this kind
facial tics,
of baseline trend is presented in Figure 3-2. Examination of this figure shows
a steadily increasing linear function, with the number of tics observed aug-
menting over days. The deteriorating baseline is an acceptable pattern inas-
much as the subsequent application of a successful treatment intervention
should lead to a reversed trend in the data (i.e., a decreasing linear function
over days). However, should the treatment be ineffective, no change in the
slope of the curve would be noted. If, on the other hand, the treatment
application leads to further deterioration (i.e., if the treatment is actually
detrimental to the patient see Bergin, 1966), it would be most difficult to
assess its effects using the deteriorating baseline. In other words, a differen-
tial analysis as to whether a trend in the data was simply a continuation of the
baseline pattern or whether application of a detrimental treatment specifically
led to its continuation could not be made. Only if there appeared to be a
pronounced change of the curve following introduction of a
in the slope
detrimental treatment could some kind of valid conclusion be reached on the
basis of visual inspection. Even then, the withdrawal and reintroduction of
the treatment would be required to establish its controlling effects. But from
both clinical and ethical considerations, this procedure would be clearly
unwarranted.
A baseline pattern that provides difficulty for the applied clinical researcher

250
s
^
^ 200
o
1^
1^
150 ""^

> 100
o
SB ,,,__^ ^^
Ul
9
e>
50
UJ

1 2 3 4 5 6
DAYS
FIGURE 3-2. The increasing baseline (target behavior deteriorating). Hypothetical data for
mean number of facial tics averaged over three daily 15-minute videotaped sessions.
is one that reflects steady improvement in the subject's condition during the
course of initial observation. An example of this kind of pattern appears in
Figure 3-3. Inspection of this figure shows a linear decrease in tic frequency
over a 6-day period. The major problem posed by this pattern, from a
research standpoint, is that application of a treatment strategy while improve-
ment is already taking place will not allow for an adequate assessment of the
intervention. Secondly, shouldimprovement be maintained following initia-
tion of the treatment intervention, the experimenter would be unable to
attribute such continued improvement to the treatment unless a marked
change in the slope of the curve were to occur. Moreover, removal of the
treatment and its subsequent reinstatement would be required to show any
controlling effects.
An alternative (and possibly a more desirable) strategy involves the contin-
uation of baseline measurement with the expectation that a plateau will be
reached. At emerge and the effects of
that point, a steady pattern will
treatment can then be easily evaluated. improvement
It is also possible that
seen during baseline assessment is merely a function of some extrinsic vari-
able (Sidman, 1960) of which the experimenter is currently unaware. Follow-
ing Sidman's recommendations, it then behooves the methodical
experimenter, assuming that time limitations and clinical and ethical consider-
ations permit, to evaluate empirically, through experimental analysis, the
possible source (e.g., "placebo" effects) of covariation. The results of this
kind of analysis could indeed lead to some interesting hunches, which then
might be subjected to further verification through the experimental analysis
method (see chapter 2, section 2.3).
The extremely variable baseline presents yet another problem for the
^ 200
^ 150
Ik
e
100
z
UJ
g
UJ
50
"
3 4
DAYS
FIGURE 3-3. The decreasing baseline (target behavior improving). Hypothetical data for
mean number of facial tics averaged over three daily 15-minute videotaped sessions.
.
.250
o -
5200
A
^
il50
fck
kk
.
/^
/
A A
\\V
/
/
/
/
\
\
\
\
/
/
/
1
/
/
100 \
\
/
\ /
>- /
\ /
o \ / \ /
50
a
.
Y
o>
UJ
ec .
u. "
1 , ,
1 2 3 4 5 6
DAYS
FIGURE 3-4, The variable baseline. Hypothetical data for mean number of facial tics
averaged over three 15-minute videotaped sessions.
clinical researcher. Unfortunately, this kind of baseline pattern is frequently

obtained during the course of applied clinical research, and various strategies
for dealing with it are required. An example of the variable baseline is
presented in Figure 3-4. An examination of these data indicate a tic frequency
of about 24 to 255 tics per day, with no discernible upward or downward
trend clearly in evidence. However, a distinct pattern of alternating low and
high trends is present. One possibility (previously discarded in dealing with
extreme initial variability) is to simply extend the baseline observation until
some semblance of stability is attained, an example of which appears in

Figure 3-5.
A second strategy involves the use of inferential statistics when comparing
baseline and treatment phases, particularly where there is considerable over-
lap between succeeding distributions. However, if overlap is that extensive, the
statistical model will be equally ineffective in finding differences, as appropri-
ate probability levels will not be reached. Further details regarding graphic
presentation and statistical analyses of data will appear in chapter 9.
A final strategy for dealing with the variable baseline is to assess systemati-
cally the sources of variability. However, as pointed out by Sidman (1960), the
amount of work and time involved in such an analysis is better suited to the
"basic scientist" than the applied clinical researcher. There are times when the
clinical researcher will have to learn to live with such variability or to select
measures that fluctuate to a lesser degree.
Another possible baseline pattern is one in which there is an initial period of
deterioration, which is then followed by a trend toward improvement (see
Figure 3-6). This type of baseline (increasing-decreasing) poses a number of
problems for the experimenter. First, when time and conditions permit, an
empirical examination of the covariants leading to reversed trends would be
of heuristic value. Second, while the trend toward improvement is continued
in the latter half, of the baseline period of observation, application of a
treatment will lead to the same difficulties in interpretation that are present in
the improving baseline, previously discussed. Therefore, the most useful
course of action to pursue involves continuation of measurement procedures
until a stable and steady pattern emerges.
S 250
:i 200
2 150
100
50
23456789 DAYS
10
FIGURE 3-5. The variable-stable baseline. Hypothetical data for mean number of facial tics
averaged over three daily 15-minute videotaped sessions.
2 250
^ 200
o
1 150
100
50
3 4
DAYS
FIGURE 3-6. The increasing-decreasing baseline. Hypothetical data for mean number of
facial tics averaged over three daily 15-minute videotaped sessions.
o *
^ 200
2 150
100
S
z
Ui
5. 50
12 3
DAYS
4 5 6
FIGURE 3-7. The decreasing-increasing baseline. Hypothetical data for mean number of
facial tics averaged over three daily 15-minute videotaped sessions.
Very similar to the increasing-decreasing pattern is its reciprocal, the de-

creasing-increasing type of baseline (see Figure 3-7). This kind of baseline
pattern often reflects the placebo effects of initially being part of an experi-
ment or being monitored (either self or observed). Although placebo effects
are always of interest to the clinical researcher, when he or she is faced with
time pressures, the preferred course of action is to continue measurement
procedures until a steady pattern in the data is clear. If extended baseline
measurement is not feasible, introduction of the treatment, following the
worsening of the target behavior under study, is an acceptable procedure,
particularly if the controlling effects of the procedure are subsequently dem-

onstrated via its withdrawal and reinstatement.
A final baseline trend, the unstable baseline, also causes difficulty for the
applied clinical researcher. A hypothetical example of this type of baseline,
obtained under extended measurement conditions, appears in Figure 3-8.
Examination of these data reveals not only extreme variability but also the
absence of a particular pattern. Therefore, the problems found in the variable
baseUne are further compounded here by the lack of any trend in the data.
This, of course, heightens the difficulty in evaluating these data through the
method of experimental analysis. Even the procedure of blocking data
usually fails to eliminate all instability on the basis of visual analysis. To date,
no completely satisfactory strategy for dealing with the variable baseline has
appeared; at best, the kinds of strategies for dealing with the variable baseline
are also recommended here.
3.4 CHANGING ONE VARIABLE AT A TIME

A cardinal rule of experimental single-case research is to change one
variable at a time when proceeding from one phase to the next (Barlow &
Hersen, 1973). Barlow and Hersen pointed out that when two variables are
simultaneously manipulated, the experimental analysis does not permit con-
clusions as to which of the two components (or how much of each) contrib-
utes to improvements in the target behavior. It should be underscored that the
o/ie-variable rule holds, regardless of the particular phase (beginning, middle,
or end) that is being evaluated. These strictures are most important when
CA 250
o
p
^ 200
^
C9
150
ik
e
>- 100
o
3 50
O*
UJ
fie
7 9 11 13 15
DAYS
FIGURE 3-8. The unstable baseline. Hypothetical data for mean number of facial tics
averaged over three daily 15-minute videotaped sessions.
examining the interactive effects of treatment variables (Barlow & Hersen,

1973; Elkin et al., 1973; Leitenberg, Agras, Thomson, & Wright, 1968). A
more complete discussion of interaction designs appears in chapter 6, section
6.5.
Correct and incorrect applications
A frequently committed error during the course of experimental single-case

research involves the simultaneous manipulation of two variables so as to
assess their presumed interactive effects. A review of the literature suggests
that this type of error is often made in the latter phases of experimentation. In
order to clarify the issues involved, selected examples of correct and incorrect
applications will be presented.
For illustrative purposes, let us asume that baseline measurement in a study
consists of the number of social responses (operationally defined) emitted by
a chronic schizophrenic during a specific period of observation. Let us further
assume that subsequent introduction of a single treatment variable involves
application of contingent (token) reinforcement following each social re-
sponse that is observed on the ward. At this point in our hypothetical
example, only one variable (token reinforcement) has been added across the
two experimental phases (baseline to the first treatment phase). In accordance
with design principles followed in the A-B-A-B design, the third phase would
consist of a return to baseline conditions, again changing (removing) only one
variable across the second and third phases. Finally, in the fourth phase,
token reinforcement would be reinstated (addition of one variable from
Phase 3 to 4). Thus, we have a procedurally correct example of the A-B-A-B
design (see chapter 5) in wnich only one variable is altered at a time from
phase to phase.
In the following example we will present an inaccurate application of
single-case methodology. Using our previously described measurement situa-
tion, let us assume that baseline assessment is now followed by a treatment
combination comprised of token reinforcement and social reinforcement. At
this point, the experiment is labeled A-BC. Phase 3 is a return to baseline
conditions (A), while Phase 4 consists of socal reinforcement alone (C). Here
we have an example of an A-BC-A-C design, with A= baseline, BC = token
and social reinforcement, A = baseline, and C = social reinforcement. In
this experiment the researcher is hopeful of teasing the relative effects of
token and social reinforcement. However, this a totally erroneous assumption
on his or her part. From the A-BC-A portion of this experiment, it is feasible
only to assess the combined BC assuming that the
effect over baseline (A),
appropriate trends in the data appear. Evaluation of the individual effects of
the two variables (social and token reinforcement) comprising the treatment
package is not possible. Moreover, application of the C condition (social
reinforcement alone) following the second baseline also does not permit firm
conclusions, either with respect to the effects of social reinforcement alone or
in contrast to the combined treatment of token and social reinforcement. The
experimenter is not in a position to examine the interactive effects of the BC
and C phases, as they are not adjacent to one another.
If our experimenter were interested in accurately evaluating the interactive
effects of token and social reinforcement, the following extended design
would be considered appropriate: A-B-A-B-BC-B-BC. When this experimen-
tal strategy is used, the interactive effects of social and token reinforcement
can be examined systematically by comparing differences in trends between

the adjacent B (token reinforcement) and BC (token and social reinforce-
ment) phases. The subsequent return to B and reintroduction of the com-
bined BC would allow for analysis of the additive and controlling effects of
social reinforcement, assuming expected trends in the data occur.
A published example of the correct manipulation of variables across
phases appears in Figure 3-9. In this study, Leitenberg et al., (1968) examined
the separate and combined effects of feedback and praise on the mean
4801
400
360
320
280
240
FEEDBACK NO FEEDBACK FEEDBACK

1 2 3 4 5 6 7 8 9 1011 1 23456789 10 11 1 23456789 1011
BLOCKS OF TWO SESSIONS (10 TRIALS)
FIGURE 3-9. Time in which a knife was kept exposed by a phobic patient as a function of
feedback, feedback plus praise, and no feedback or praise conditions. (Figure 2, p. 131, from
Leitenberg, H., Agras, W. S., Thomson, L., & Wright, D. E. (1%8), Feedback in behavior
modification: Anexperimental analysis in two phobic cases. Journal of Applied Behavior
Analysis, 1, 131-137. Copyright 1968 by Society for the Experimental Analysis of Behavior, Inc.
Reproduced by permission.)
y
number of seconds a knife-phobic patient allowed himself to be exposed to a

knife. An
examination of the seven phases of study reveals the following
progression of variables: (1) feedback, (2) feedback and praise, (3) feedback,
(4) no feedback and no praise, (5) feedback, (6) feedback and praise, and (7)
feedback. A comparison of adjacent phases shows that only one variable was
manipulated (added or subtracted) at a time across phases. In a similar
design, Elkin et al., (1973) assessed additive and subtractive effects of
The following progression
therapeutic variables in a case of anorexia nervosa.
of variables was used in a six-phase experiment: caXoncs baseline
(1) 3,(X)0
(2) 3,(X)0 calonQSfeedback, (3) 3,000 calories^feedback and reinforce-

ment (4) 4,500 calonQsfeedback and reinforcement (5) 3,000 calories
y ,

feedback and reinforcement, (6) 4,500 calories feedback and
reinforcement. Again, changes from one phase to the next (italicized) never
involved more than the manipulation of a single variable.
Exceptions to the rule
In a number of experimental single-case studies (Barlow et al., 1969;

Eisler, Hersen, & Agras, 1973; Pendergrass, 1972; Ramp, Ulrich & Dulaney,
1971) legitimate exceptions to the rule of maintaining a consistent stepwise
progression (additive or subtractive) across phases have appeared. In this
section the exceptions will be discussed, and examples of published data will
be presented and analyzed. For example. Ramp et al. (1971) examined the
effects of instructions and delayed time-out in a 9-year-old male elementary
school student who proved to be a disciplinary problem. Two target behaviors
(intervals out of seat without permission and intervals talking without per-
mission) were selected for study in four separate phases. During baseline, the
number of 10-second time intervals in which the subject was out of seat or
talking were recorded for 15-minutes sessions. In Phase 2 instructions simply
involved the teacher's informing the subject that permission for being out of
seat and talking were required (raising his hand). The third phase consisted of
a delayed time-out procedure. A red light, mounted on the subject's desk, was
illuminated for a 1-3-second period immediately following an instance of
out-of-seat or talking behavior. Number of illuminations recorded were cu-
mulated each day, with each classroom violation resulting in a 5-minute
detention period in a specially constructed time-out booth while other
children participated in gym and recess activities. The results of this study
appear in Figure 3-10. Relabeling of the four experimental phases yields an A-
B-C-A design. Inspection of the figure shows that the baseline (A) and
instructions (B) phases do not differ significantly for either of the two target
behaviors under study. Thus although the independent variables differ across
these phases, the resuhing dependent measures are essentially alike. However,
INSTRUCTIONS DELAYED TIMEOUT CONTINGENCIES REMOVED
< Z
1Ô
S^mmmmtmm
"X
g
Am.-^
SESSIONS
FIGURE 3-10. Each point represents one session and indicates the number of intervals in which
the subject was out of his seat (top) or talking without permission (bottom). A total of 90 such
intervals was possible within a 15-minute session. Asterisks over points indicate sessions that
resulted in time being spent in the booth. (Figure 1, p. 237, from: Ramp, E., Ulrich, R., &
Dulaney, S. (1971). Delayed timeout as a procedure for reducing disruptive classroom behavior:
A case study. Journal of Applied Behavior Analysis, 4, 235-239. Copyright 1971 by Society for
the Experimental Analysis of Behavior, Inc. Reproduced by permission.)
institution of the delayed time-out contingency (C) yielded a marked decrease

in classroom violations. Subsequent removal of the time-out contingency in
Phase 4 (A) led to a renewed increase in classroom violations.
Since the two initial phases (A and B) yield similar data (instructions did
not appear to be effective), equivalence of the baseline and instructions
phases are assumed. If one then collapses data across these two phases, an A-
C-A design emerges, with some evidence demonstrated for the controlling
effects of delayed time-out. In A-C-A design follows the experi-
this case the
mental analysis used in the case of the A-B-A design (see chapter 5). However,
further confirmation of the controlling effects would require a return to the C
condition (delayed time-out). This new design would then be labeled as

follows: A = B-C-A-C. It should be noted that without the functional equiva-
lence of the first two phases (A = B) this would essentially be an incorrect
experimental procedure. The functional equivalence of different adjacent
experimental phases warrants further illustration. An excellent example was
provided by Pendergrass (1972), who used an A-B-A = C-B design strategy. In
her study, Pendergrass evaluated the effects of time-out and observation of
punishment being administered (time-out) to a cosubject in an 8-year-old
retarded boy. Two negative high-frequency behaviors were selected as targets
for study. They were (1) banging objects on the floor and on others (bang),
and (2) the subject's biting of his lips and hand (bite). Only one of the two
target behaviors (bang) was directly subjected to treatment effects, but gener-
alization and side effects of treatment on the second behavior (bite) were
examined concurrently. Results of the study are presented in Figure 3-11.
Time-out following baseline assessment led to a significant decrease in both
the punished and unpunished behaviors. A return to baseline conditions in
Phase 3 resulted in high levels of both target behaviors. Institution of the
"watch" condition (observation of punishment) did not lead to an apprecia-
ble decrease, hence the functional equivalence of Phases 3 (A) and 4 (C). In
Phase 5 the reinstatement of time-out led to renewed improvement in target
behaviors.
In this study the ineffectiveness of the watch condition is functionally
SUBJECT ^1
S 1.00
i 0.75
O O
a;0.50 A"- ,..^, z
f^- <
^
2
<
0.25
-W sV-
CO
U
5
O 0.50
|0.25
1.00
0.75
/ \. ^-
Mil-
UJ
CO
o
a.
o BASE T BASE WATCH T
a. 1 2 3 4 5
f\
FIGURE 3-11. Proportion of total intervals in which Bang (punished) and Bite (unpunished)
responses were recorded for SI in 47 free-play periods. (Figure 1, p. 88, from: Pendergrass, V. E.
(1972). Timeout from positive reinforcement following persistent, high-rate behavior in retar-
dates. Journal of Applied Behavior Analysis, 5, 85-91 Copyright 1972 by Society for Experimen-
.
tal Analysis of Behavior, Inc. Reproduced by permission.)

equivalent to the continuation of the baseline phase (A), despite obvious

differences in procedure. With it is most
respect to labeling of this design,
appropriately designated as follows: A-B-A = C-B between A
(the equal sign
and C represents their functional equivalence insofar as dependent measures
are concerned).
A further exception to the basic rule occurs when the experimenter is
package containing two or more

interested in the total impact of a treatment
components (e.g., instructions, feedback, and reinforcement). In this case,
more than one variable is manipulated at a time across adjacent experimental
phases. An
example of this type of design appeared in a series of analogue
studies reported by Eisler, Hersen, and Agras (1973). In one of their studies
the combined effects of videotape feedback and focused instructions were
examined in an A-BC-A-BC design, with A = baseline and BC = videotape
feedback and focused instructions. As is apparent from inspection of Figure
3-12, analysis of these data follows the A-B-A-B design pattern, with the
exception that the B phase is represented by a compound treatment variable
(BC). However, it should be pointed out that, despite the fact that improve-
ments over baseline appear for both target behaviors (looking and smiling)
LOOKING SMILING
^ jV Video Fdbk
& Foe Insir
4 5 6 7 8 9 10 11 12 1 4 5 6 7 8 9 10
, BLOCKS OF TWO MINUTES BLOCKS OF TWO MINUTES
FIGURE
A- Mean
et - A - 6t
number of looks and
3-12. smiles for three couples in 10-second intervals plotted
in blocks of 2 minutes for the Videotape Feedback Plus Focused Instructions Design. (Figure 3,
p. 556, from: Eisler, R. M., Hersen, M., & Agras, W. S. (1973). Effects of videotape and
on nonverbal marital interaction: An analog study. Behavior Therapy, 4,
instructional feedback
551-558. Copyright 1973 by Association for the Advancement of Behavior Therapy. Reproduced
by permission.)
^
during videotape feedback and focused instructions conditions, this type of

design will obviously allow for no conclusions as to the relative contribution
of each treatment component.
A final exception to the one-variable rule appears in a study by Barlow,
Leitenberg, and Agras (1969), in which the controlling effects of the noxious
scene in covert sensitization were examined in 2 patients (a case of pedophilia
and one of homosexuality). In each case an A-BC-B-BC experimental design
was used (Barlow & Hersen, 1973). In both cases the four experimental
phases were as follows: (1) A = baseline, (2) BC = covert sensitization
treatment (verbal description of variant sexual activity and introduction of
the nauseous scene), (3) B = verbal description of deviant sexual activity but
no introduction to the nauseous scene, and (4) BC = covert sensitization
(verbal description of sexual activity and introduction of the nauseous scene).
For purposes of illustration, data from the pedophilic case appears in Figure
3-13. Examination of the design strategy reveals that covert sensitization
treatment (BC) required instigation of both components. Thus initial dif-
ferences between baseline (A) and acquisition (BC) only suggest efficacy of
the total treatment package. When the nauseous scene is removed during
extinction (B), the resulting increase in deviant urges and card sort scores
similarly suggests the controlling effects of the nauseous scene. In reacquisi-
tion (BC), where the nauseous scene is reinstated, renewed decreases in the
30
CO
HI
Q. ACQUISITION EXTINCTION REACQUISITION
fe
Total urges O
o>-
oo< Card sort > #
oaO
<^ 6
r-c
Om
^^
Zlu OZ
5 (/><
Qo
-J
<
3
P
c
12 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
lg^ EXPERIMENTAUpAYS
EXPERIMENTAL
f\ Qp
FIGURE 3-13. Total score on card sort per experimental day and total frequency of pedophilic
sexual urges in blocks of 4 days surrounding each experimental day. (Lower scores indicate less
sexual arousal.). (Figure 1, p. 599, from: Barlow, D. H., Leitenberg, H., & Agras, W. S. (1969).
Experimental control of sexual deviation through manipulation of the noxious scene in covert
sensitization. Journal of Abnormal Psychology, 74, 5%-601. Copyright 1969 by the American
Psychological Association. Reproduced by permission.)
data confirm its controlling effects. Therefore, despite an initial exception to

changing one variable at a time across adjacent phases, a stepwise subtractive
and additive progression is maintained in the last two phases, with valid
conclusions derived from the ensuing experimental analysis.
Issues in drug evaluation
Issues discussed in the previous section that pertain to changing of vari-

ables across adjacent experimental phases and the functional equivalence in
data following procedurally different operations are identical when analyzing
the effects of drugs on behavior. It is of some interest that experimenters with
both a behavior modification bias (e.g., Liberman, Davis, Moon, & Moore,
1973) and those adhering to the psychoanalytic tradition (e.g., Bellak &
Chassan, 1964) have used remarkably similar design strategies when investi-
gating drug effects on behavior, either alone or in combination with psy-
chotherapeutic procedures.
Keeping in mind that one-variable rule, the following sequence of experi-
mental phases has appeared in a number of studies: (1) no drug, (2) placebo,
(3) activedrug, (4) placebo, and (5) active drug. This kind of design, in which
a stepwise application of variables appears, permits conclusions with respect
to possible placebo effects (no-drug to placebo phase) and those with respect
to the controlling influences of active drugs (placebo, active drug, placebo,
active drug). Within the experimental analysis framework, Liberman et al.
(1973) have labeled this sequence the A-A,-B-A,-B design. More specifically,
they examined the effects of stelazine on a number of asocial responses

emitted by a withdrawn schizophrenic patient. The particular sequence used
was as follows: (A) no drug, (Ai) placebo, and (B)
(B) stelazine, (A,) placebo,
stelazine. Similarly, within the psychoanalytic framework, Bellak and Chas-
san (1964) assessed the effects of chlordiazepoxide on variables (primary
process, anxiety, confusion, hostility, "sexual flooding," depersonalization,
ability tocommunicate) rated by a therapist during the course of 10 weekly
interviews. A double-blind procedure was used in which neither the patient
nor the therapist was informed about changes in placebo and active medica-
tion conditions. In this study, an A-A,-B-A,-B design was employed with the
following sequential pattern: (A) no drug, (A,) placebo, (B) chlordiazepox-
ide, (Al) placebo, and (B) chlordiazepoxide.
Once again, pursuing the one variable rule, Liberman et al., (1973) have
shown how the combined effects of drugs and behavioral manipulations can
be evaluated. Maintaining a constant level of medication (600 mg of
chlorpromazine per day), the controlling effects of time-out on delusional
behavior (operationally defined) were examined as follows: (1) baseline plus
600 mg of clorpromazine, (2) time-out plus 600 mg of chlorpromazine, and
(3) removal of time-out plus 600 mg of chlorpromazine. In this study (AB-
CB-AB) the only variable manipulated across phases was the time con-
tingency.
There are several other important issues related to the investigation of drug
effects in single-case experimental designs that merit careful analysis. They
include the double-blind evaluation of results, long-term carryover effects of
phenothiazines, and length of phases. These will be discussed in some detail
in section 3.6 of this chapter and in chapter 7.
3.5. REVERSAL AND WITHDRAWAL

In their survey of the methodological aspects of applied behavior analysis,
Baer et al. (1968) stated that there are two types of experimental designs that
can be used to show the controlling effects of treatment variables in individu-
als. These two basic types are commonly referred to as the reversal and
multiple-baseline design strategies. In this section we will concern ourselves
only with the reversal design. The prototypic A-B-A design and all of its
numerous extensions and permutations (see chapter 5 for details) are usually
placed in this category (Barlow et al., 1977; Barlow & Hersen, 1973; Hersen,
1982; Kazdin, 1982b; Van Hasselt & Hersen, 1981).
When speaking of a reversal, one typically refers to the removal (with-
drawal) of the treatment variable that is applied after baseline measurement
has been concluded. In practice, the reversal involves a withdrawal of the B
phase (in the A-B-A design) after behavioral change has been successfully
demonstrated. If the treatment (B phase) indeed exerts control over the
targeted behavior under study, a decreased or increased trend (depending on
which direction indicates deterioration) in the data should follow its removal.
In describing their experimental efforts when using A-B-A designs, applied
clinical researchers frequently have referred to both their procedures and
resuhing data as reversals. This, then, represents a terminological confusion
between the independent variable and the dependent variable. However, from
either a semantic, logical, or scientific standpoint, it is untenable that both a
cause and an effect should be given an identical label. A careful analysis
reveals that a reversal involves a specific technical operation, and that its
result (changes in the target behavior[s]) is simply examined in terms of rates

of the data (increased, decreased, or no change) in relation to patterns seen in
the previous experimental phase. To summarize, a reversal is an active proce-
dure; the obtained data may or may not reflect a particular trend.
The reversal design
A still finer distinction regarding reversals was made by Leitenberg (1973)

in his examination of experimental single-case design strategies. He con-
tended that the reversal design (e.g., A-B-A-B design) is inappropriately

labeled, and that the term withdrawal (i.e., withdrawal of treatment in the
second A phase) is a more accurate description of the actual technical
operation. Indeed, a distinction between a withdrawal and a reversal was
made, and Leitenberg showed how the latter refers to a specific kind of
experimental strategy. It should be underscored that, although "... this
distinction ... is typically not made in the behavior modification literature"
(Leitenberg, 1973), the point is well taken and should be considered by
applied clinical researchers.
To illustrate and clarify this distinction, an excellent example of the reversal
design, selected from the child behavior modification literature, will be pre-
sented. Allen, Hart, Buell, Harris, and Wolf (1964) were concerned with the
contingent effects of reinforcement on the play behavior of a 4y2-year-old girl
who evidenced social withdrawal with peers in a preschool nursery setting.

Two target behaviors were selected for study: (1) percentage of interaction
with adults, and (2) percentage of interaction with children. Observations
were recorded daily during 2-hour morning sessions. As can be seen in Figure
3-14, baseline data show that about 15 percent of the child's time was spent
interacting with children, whereas approximately 45 percent of the time was
spent in interactions with adults. The remaining 40 percent involved "isolate"
play. Inasmuch as the authors hypothesized that teacher attention fostered
interactions with adults, in the second phase of experimentation an effort was
made to demonstrate that the same teacher attention, when presented con-
tingently in the form of praise following the child's interaction with other
children, would lead to an increase in such interactions. Conversely, isolate
play and approaches to adults were ignored. Inspection of Figure 3-14 reveals
that contingent reinforcement (praise) increased the percentage of interaction
with children and led to a concomitant decrease in interactions with adults. In
the third phase a "true" reversal of contingencies was put into effect. That is
to say, contingent reinforcement (praise) was now administered when the
childapproached adults, but interaction with other children was ignored.
Examination of Phase 3 data reflects the reversal in contingencies. Percentage
of time spent with children decreased substantially while percentage of time
spent with adults showed a marked increase. Phase 2 contingencies were then
reinstated in Phase 4, and the remaining points on the graph are concerned
with follow-up measures.
Reversal and withdrawal designs compared
A major difference between the reversal and withdrawal designs is that in

the third phase of the reversal design, following instigation of the therapeutic
procedure, the same procedure is now applied to an alternative but incom-
patible behavior. By contrast, in the withdrawal design, the A phase following
Single-case Experimental Designs
12 3 4 9 10 II 12 13 14 IS \h\ 17 It 19 20 21 22 23 24 2S 31 31 40 SI
Basalin* Rinf. Inttract. with Rvrsl Rinf. IntMâct. with Children Post
Children Clitckt
D y t
FIGURE 3-14. Daily percentages of time spent in social interaction with adults and with children
during approximately 2 hours of each morning session. (Figure 2, p. 515, from: Allen K. E.,
Hart, B. M., Buell, J. S., Harris, R R., & Wolf, M. M. (1964). Effects of social reinforcement on
isolate behavior of a nursery school child. Child Development, 35 511-518. Copyright 1964.
Reproduced by permission of The Society for Research in Child Development, Inc.)
introduction of the treatment variable (e.g., token reinforcement) simply

involves removal and a return to baseline conditions. Leitenberg (1973)
its
it can be quite dramatic is

argued that "Actually, the reversal design although
somewhat more cumbersome ."
(pp. 90-91) than the more frequently
. .
employed withdrawal design. Moreover, the withdrawal design is much better

suited for investigations that do not emanate from the operant (reinforce-
ment) framework (e.g., the investigation of drugs and examination of nonbe-
havioral therapies).
Withdrawal of treatment
The specific point at which the experimenter removes the treatment vari-
able (second A phase in the A-B-A design) in the withdrawal design is
multidetermined. Among the factors to be considered are time Hmitations
imposed by the treatment setting, staff cooperation when working in institu-
tions (J. M. Johnston, 1972), and ethical considerations when removal of
treatment can possibly lead to some harm to the subject (e.g., head banging
in a retardate) or others in the environment (e.g., physical assaults toward
wardmates in disturbed inpatients). Assuming that these important environ-

mental considerations can be dealt with adequately and judiciously, a variety
of parametric issues must be taken into account before instituting withdrawal
of the treatment variable. One of these issues involved the overall length of
adjacent treatment phases; this will be examined in section 3.6 of this chapter.
In this section we will consider the implementation of treatment withdrawal
in relation to data trends appearing in the first two phases (A and B) of study.
We will illustrate both correct and incorrect applications using hypothetical
data. Let us consider an example in which A refers to baseline measurement
of the frequency of social responses emitted by a withdrawn schizophrenic.
The subsequent treatment phase (B) involves contingent reinforcement in the
form of praise, while the third phase (A) represents the withdrawal of
treatment and a return to original baseline conditions. For purposes of
illustration, we will assume stability of "initial" baseline conditions for each
of the following examples.

In our first example (see Figure 3-15) data during contingent reinforcement
show a clear upward trend. Therefore, institution of withdrawal procedures
at the conclusion of this phase will allow for analysis of the controlling effects
of reinforcement, particularly if the return to baseline results in a downward
trend in the data. Equally acceptable is a baseline pattern (second A phase) in
which there is an immediate loss of treatment effectiveness, which is then
maintained at a low-level stable rate (this pattern is the same as the initial
baseline phase).
In our second example (see Figure 3-16) data during contingent reinforce-
ment show the immediate effects of treatment and are maintained throughout
the phase. After these initial effects, there is no evidence of an increased rate
of responding. However, the withdrawal of contingent reinforcement at the
conclusion of the phase does permit analysis of its controlling effects. Data in
the second baseline show no overlap with contingent reinforcement, as there
is a return to the stable but low rate of responding seen in the first baseline (as
15
BASEUNE CONT. BASELINE

I REINF.
I 12
9 11 13 15 17
DAYS
FIGURE 3-15. Increasing treatment phase followed by decreasing baseline. Hypothetical data
for frequency of social responses in a schizophrenic patient per 2-hour period of observation.
in Figure 3-16). Equally acceptable would be a downward trend in the data as

depicted in the second baseline in Figure 3-14.
In our third example of a correct withdrawal procedure, examination of
Figure 3-17 indicates that contingent reinforcement resulted in an immediate
increase in rate, followed by a linear decrease, and then a renewed increase in
rate which then Although it would be advisable to analyze contrib-
stabilized.
uting factors to the decrease and subsequent increase (Sidman, 1960), institu-
tion of the withdrawal procedure at the conclusion of the contingent
reinforcement phase allows for an analysis of its controlling effects, particu-
larly as a decreased rate was observed in the second baseline.
An example of the incorrect application of treatment withdrawal appears
, 15
UJ CONT.
BASELINE BASELINE
REINF.
i 12
u
e
y. 6
1 3 5 7 9 H 13 15 17
DAYS
FIGURE 3-16. High-level treatment phase followed by low-level baseline. Hypothetical data
for frequency of social responses in a schizophrenic patient per 2-hour period of observation.
CONT
BASELINE BASELINE
REINF.
i
bU
ce
12
\\ /
A^ y\ \
< 9
\y \
i
^-V
^
2 3
1

[V^
7 9 11 13 15 17 19 21 23
FIGURE 3-17. Decreasing-increasing-stable treatment phase followed by decreasing baseline.

Hypothetical data for frequency of social responses in a schizophrenic patient per 2-hour
period of observation.
in Figure 3-18. Inspection of the figure reveals that after a stable pattern is
obtained in baseline, introduction of contingent reinforcement leads to an

immediate and dramatic improvement, which is then followed by a marked
decreasing linear function. This trend is in evidence despite the fact that the
last data point in contingent reinforcement is clearly above the highest point
achieved in baseline. Removal of treatment and a return to baseHne condi-
tionson Day 13 similarly result in a decreasing trend in the data. Therefore,
no conclusions as to the controlling effects of contingent reinforcement are
possible, as it is not clear whether the decreasing trend in the second baseline
is a function of the treatment's withdrawal or mere continuation of the trend
begun during treatment. Even if withdrawal of treatment were to lead to the
stable low-level pattern seen in the first baseline period, the same problems in
interpretation would be posed.
When the aforementioned trend appears during the course of experimental
treatment, it is recommended that the phase be continued until a more
consistent pattern emerges. However, if this strategy is pursued, the equiva-
lent length of adjacent phases is altered (see section 3.6). A second strategy,
although admittedly somewhat weak, is to reintroduce treatment in Phase 4
(thus, we have an A-B-A-B design), with the expectation that a reversed trend
in the data will reflect improvement. There would then be limited evidence for
the treatment's controlling effects.
A similar problem ensues when treatment is withdrawn in the example that
appears in Figure 3-19. In spite of an initial upward trend in the data when
contingent reinforcement is first introduced (B), the decreasing trend in the
latter half of the phase, which is then followed by a similar decline during the
second baseline (A), prevents an analysis of the treatment's controlling ef-
CONT.
<o 15f BASELINE BASELINE
LU REINF.
/>
g
t/i
12
kU
oc
^ 9
5
o
C/3
l^ 6-
O
'>
u \-\
z
UJ 5-
3
o>
LU
QC
i
-I I l_l k_JL I I I '
1 5 5 7 9 11 13 15 17
DAYS
FIGURE 3-18. High-level decreasing treatment phase followed by decreasing baseline.

Hypothetical data for frequency of social responses in a schizophrenic patient per 2-hour
r, , period of observation.
c 15 CONT.
BASELINE REINF. BASELINE
12
< 9
1 3 5 9 11 13 15 17
DAYS
FIGURE 3-19. Increasing-decreasing treatment phase followed by decreasing behavior. Hy-

pothetical data for frequency of social responses in a schizophrenic patient per 2-hour period of
observation.
fects. Therefore, the same recommendations made in the case of Figure 3-18
apply here.
Limitations and problems
As mentioned earlier, the applied clinical researcher faces some unique

problems when intent on pursuing experimental analysis by withdrawing a
particular treatment technique. These problems are heightened in settings
where one exerts relatively little control, either with respect to staff coopera-
tion or in terms of other important environmental contingencies (e.g., when
dealing with individual problems in the classroom situation, responses of
other children throughout the varying stages of experimentation may spu-
riously affect the results). Although these concerns have been articulated
elsewhere in the behavioral literature (Baer et al., 1968; Bijou, Peterson,
Harris, Allen, & Johnston, 1969; Hersen, 1982; Kazdin & Bootzin, 1972;
Leitenberg, 1973), a brief summary of the issues at stake might be useful at
this point.
A frequent criticism leveled at researchers using single-case methodology is
that removal of the treatment will lead to the subject's irreversible deteriora-
tion (at least in terms of the behavior under study). However, as Leitenberg
(1973) pointed out, this is a weak argument with no supporting evidence to be
found in the experimental literature. If the technique shows initial beneficial
effects and it exerts control over the targeted behavior being examined, then,
when reinstated, its controlling effects will be established. To the contrary,
Krasner (1971b) reported that recovery of low levels of baseline
initially
performance often fails to occur in extended applications of the A-B-A design

where multiple withdrawals and reinstatements of the treatment technique are
instituted (e.g., A-B-A-B-A-B-A-B). Indeed, the possible carryover effects

across phases and concomitant environmental events leading to improved
conditions contribute to the researcher's difficulties in carrying out scienti-
fically acceptable studies.
A problem encountered is one of staff resistance. Usually, the
less subtle
researcher working in an applied setting (be it at school, state institution for

the retarded, or psychiatric hospital) is consulting with house staff on difficult
problems. In efforts to remediate the problem, the experimenter encourages
staff to apply treatment strategies that are likely to achieve beneficial results.
When staff members are subsequently asked to temporarily withdraw treat-
ment procedures, some may openly rebel. "What teacher, seeing Johnny for
the first time quietly seated for most of the day, would like to experience
another week or two of bedlam just to satisfy the perverted whim of a
psychologist?" (J. M. Johnston, 1972, p. 1035). In other cases the staff
member or parent (when establishing parental retraining programs) may be
unable to revert to his or her original manner of functioning (i.e., his or her
way of previously responding to certain classes of behavior). Indeed, this
happened by Hawkins, Peterson, Schweid, and Bijou
in a study reported
(1966). Leitenberg (1973) argued that "In such cases, where the therapeutic
procedure cannot be introduced and withdrawn at will, sequential ABA
designs are obviated" (p. 98). Under these circumstances, the use of alterna-
tive experimental strategies such as multiple baseline (Hersen, 1982) or al-
ternating-treatment designs (Barlow & Hayes, 1979) obviously are better

suited (see chapters 7 and 8).
To summarize, the researcher using the withdrawal design must ensure that
(1) there is full staff or parental cooperation on an a priori basis; (2) the
withdrawal of treatment will lead to minimal environmental disruptions (i.e.,
no injury to subject or others in the environment will result) (see R. F.
Peterson & Peterson, 1968); (3) the withdrawal period will be relatively brief;
(4) outside environmental influences will be minimized throughout baseline,
treatment, and withdrawal phases; and (5) final reinstatement of treatment to
its logical conclusion will be accomplished as soon as it is technically feasible.
3.6. LENGTH OF PHASES

Although there has been some intermittent discussion in the literature with
regard to the length of phases when carrying out single-case experimental
research (Barlow & Hersen, 1973; Bijou et al., 1969; Chassan, 1967; J. M.
Johnston, 1972; Kazdin, 1982b), a complete examination of the problems
faced and the decision to be made by the researcher has yet to appear.
Therefore, in this section the major issues involved will be considered includ-
ing individual and relative length of phases, carryover effects and cyclic
variations. In addition, these considerations will be examined as they apply to
the study of drugs on behavior.
Individual and relative length
When considering the individual length of phases independently of other

factors (e.g., time limitations, ethical considerations, relative length of
phases), most experimenters would agree that baseline and experimental
some semblance of stability in the data is
conditions should be continued until
apparent. J. M. Johnston (1972) has examined these issues with regard to the
study of punishment. He stated that:
It is necessary that each phase be sufficiently long to demonstrate stability (lack

of trend and a constant range of variability) and to dispel any doubts of the
reader that the data shown are sensitive to and representative of what was
happening under the described condition (p. 1036).
He notes further:
That if there is indication of an increasing or decreasing trend in the data or

widely variable rates from day to day (even with no trend) then the present
condition should be maintained until the instability disappears or is shown to be
representative of the current conditions (p. 1036).
The aforementioned recommendations reflect the ideal and apply best

when each experimental phase is considered individually and independently
of adjacent phases. If one were to fully carry out these recommendations, the
possibiHty exists that widely disparate lengths in phases would result. The
strategic difficulties inherent in unequal phases has been noted elsewhere by
Barlow and Hersen (1973). Indeed, they cited the advantages of obtaining a
relatively equal number of data points for each phase.
Let us illustrate the importance of their suggestions by considering the
following hypothetical example, in which the effects of time-out on frequency
of hitting other children during a free-play situation are assessed in a 3-year-
old child. Examination of Figure 3-20 shows a stable baseline pattern, with a
high frequency of hitting behavior exhibited. Data for Days 5-7, when
treatment (time-out) is first instigated, show no effects, but on Day 8 a slight
decline in frequency appears. If the experimenter were to terminate treatment
at this point, it is obvious that few statements about its efficacy could be
made. Thus the treatment is continued for an additional 4 days (9-12), and an
appreciable decrease in hitting is obtained. However, by extending (doubling)
the length of the treatment phase, the experimenter cannot be certain whether
additional treatment in itself leads to changes, whether some correlated
General ProcecU:es^mSmgle-case Research 97
BASELINE TIME-OUT BASELINE
7 9 11 13 15
DAYS
FIGURE 3-20. Extension of the treatment phase in an attempt to show its effects. Hypothetical
data in which the effects of time-out on daily frequency of hitting other children (based on a 2-
hour free-play situation) in a 3-year-old male child are examined.
variable (e.g., increased teacher attention to incompatible positive behaviors

emitted by the child) results in changes, or whether the mere passage of time
(maturational changes) accounts for the decelerated trend. Of course, the
withdrawal of treatment on Days 13-16 (second baseline) leads to a marked
incrased in hitting behavior, thus suggesting the controlling effects of the
time-out contingency. However, the careful investigator would reinstate time-
out procedures, to dispel any doubts as to its possible controlling effects over
the target behavior of hitting. Additionally, once the treatment (time-out)
phase has been extended to 8 days, it would be appropriate to maintain
equivalence in subsequent baseline and treatment phases by also collecting
approximately 8 days of data on each condition. Then, questions as to
whether treatment effects are due to maturational or other controllable
influences will be satisfactorily answered.
As previously noted, the actual length of phases (as opposed to the ideal
length) is often determined by factors aside from design considerations.
However, where possible, the relative equivalence of phase lengths is desir-

able. If exceptions are to be made, either the initial baseline phase should be
lengthened to achieve stability in measurement, or the last phase (e.g., second
B phase in the A-B-A-B design) should be extended to insure permanence of
the treatment effects. In fact, with respect to this latter point, investigators
should make an effort to follow their experimental treatments with a full
clinical application of the most successful techniques available.
An example of the ideal length of alternating behavior and treatment
phases appears in Miller's (1973) analysis of the use of Retention Control
Training (RCT) in a "secondary enuretic" child (see Figure 3-21). Two larget
behaviors, number of enuretic episodes and mean frequency of daily urina-
tion, were selected for study in an A-B-A-B experimental design. During
baseline, the child recorded the natural frequency of target behaviors and
received counselingfrom the experimenter on general issues relating to home
and school. Following baseline, the first week of RCT involved teaching the
child topostpone urination for a 10-minute period after experiencing each
urge. Delay of urinationwas increased to 20 and 30 minutes in the next 2
weeks. During Weeks 7-9 RCT was withdrawn, but was reinstated in Weeks
10-14.
Examination of Figure 3-21 indicates that each of the first three phases
consisted of 3 weeks, with data reflecting the controlling effects of RCT on
both target behaviors. Reinstatement of RCT in the final phase led to re-
newed control, and the treatment was extended to 5 weeks to ensure main-
tenance of gains.
It might be noted that phase and data patterns do not often follow the ideal
sequence depicted in the Miller (1973) study. And, as a consequence, experi-
menters frequently are required to make accommodations for ethical, proce-
DAILY URHUTION
ENURETK EPISODES
Rtttntion Ntontlon
lattline Control . iaMlIno Control
1 TralnlRc! Trainlni
,
j
I
' \ ,
! i'
i ^ - '\ ^ .'^
*
y
\/
i\
1/"
\
1 \_l \
\
12 3 4 5 7 8 9 K) 11 12 13 M 15 16
CONSECUTIVE DAYS
FIGURE 3-21. Number of enuretic episodes per week and mean number of daily urinations per
week for Subject 1. (Figure 1, p. 291, from: Miller, P. M. (1973). An experimental analysis of
retention control training in the treatment of nocturnal enuresis in two institutionalized adoles-
cents. Behavior Therapy, 4, 288-294. Copyright 1973 by Association for the Advancement of
Behavior Therapy. Reproduced by permission.)
dural, or parametric reasons. Moreover, when working in an unexplored area

where the issues are of social significance, deviations from some of our
proposed rules during the earlier stages of investigation are acceptable.
However, once technical procedures and major parametric concerns have
been dealt with satisfactorily, a more vigorous pursuit of scientific rigor would
be expected. In short, as in any scientific endeavor, as knowledge accrues, the
level of experimental sophistication should reflect its concurrent growth.
Carryover effects
A parametric issue that is very much related to the comparative lengths of

adjacent baseline and treatment phases one of pverlappijQg (earryQyer)^
is
effects. Carryover effects in behavioral from drug) studies usually

(as distinct
appear in the second baseline phase of the A-B-A-B type design andâre
characterized by t he experimenter^Jiiability_toj^^^
baseline respondingTNot only is the original baseline rate not recoverable in
somecaseslelgTûlt, Peterson, & Bijou, 1968; Hawkins et al., 1966), but on
occasion (e.g., Zeilberger, Sampen, & Sloane, 1968) the behavior under study
undergoes more rapid modification the second time the treatment variable is
introduced.
Presence of carryover effects has been attributed to a variety of factors
including changes in instructions across experimental conditions (Kazdin,
1973b), the estabhshment of new conditioned reinforcers (Bijou et al., 1969),
the maintenance of new behavior through naturally occurring environmental
contingencies (Krasner, 1971b), and the differences in stimulus conditions
across phases (Kazdin & Bootzin, 1972). Carryover effects in behavioral
research are an obvious clinical advantage, but pose a problem experimen-
tally,as the controlling effects of procedures are then obfuscated.
Proponents of the group comparison approach (e.g., Bandura, 1969)
contend that the presence of carryover effects in single-case research is one of
its major shortcomings as an experimental strategy. Both in terms of drug
evaluation (Chassan, 1967) and with respect to behavioral research (Bijou et

al., 1969), short^geriods of experimentation (appHcation of the treatment
recommended to counteract th ese difficulties. Examining the
variable) were
problem from the operanfTramework, BijoiTerarrmêd that "In studies
involving stimuli with reinforcing properties, relatively short experimental
periods are advocated, since long ones might allow enough time for the
establishment of new conditioned reinforcers" (p. 202). Carryover effects are
also an important consideration in alternating treatment designs but are more
easily handled through counterbalancing procedures (see chapter 8).
A major difficulty in carrying out meaningful evaluations of drugs on
behavior using single-case methodology involves their carryover effects from
one phase to the next. This is most problematic when withdrawing active drug
treatment (B phase) and returning to the placebo (A, phase) condition in the
A-A,-B-A,-B design. With respect to such effects, Chassan (1967) pointed out
that "This, for instance, is thought likely to be the case in the use of
monoaminoxidase inhibitors for the treatment of depression" (p. 204). Simi-
larly, when using phenothiazine derivatives, the experimenter must exercise
caution inasmuch as residuals of the drugs have been found to remain in body
tissues for extended periods of time (as long as 6 months in some cases)
following their discontinuance (Ban, 1969).
However, it is possible to examine the short-term effects of phenothiazines
on designated target behaviors (Liberman et al., 1973), but it behooves the
experimenter to demonstrate, via blood and urine laboratory studies, that
controlling effects of the drug are truly being demonstrated. That is to say,
and graphic data patterns) between behavioral
correlations (statistical
changes and drug levels in body tissues should be demonstrated across
experimental phases.
Despite the carryover difficulties encountered with the major tranquilizers
and antidepressants, the possibility of conducting extended studies in long-
term facilities should be explored, assuming that high ethical and experimen-
tal standards prevail. In addition, study of the short-term efficacy of the
minor tranquilizers and amphetamines on selected target behaviors is quite
feasible.
Cyclic variations
A most neglected issue in experimental single-case research is that of cyclic

and 2.3, for a more general discussion
variations (see chapter 2, sections 2.2
of Although the importance of cyclic variations was given atten-
variability).
tion by Sidman (1960) with respect to basic animal research, and J. M.
Johnston & Pennypacker (1981) in a more applied context, the virtual ab-
sence of serious consideration of this issue in the applied literature is striking.
This issue is of paramount concern when using adult female subjects as their
own controls in short-term (one month or less) investigations. Despite the
fact that the effects of the estrus cycle on behavior are given some consider-
ation by Chassan (1967), he argued that ". . .a 4- week period (with random
phasing) would tend to distribute menstrual weeks evenly between treat-
ments" (p. 204). However, he did recognize that "The identification of such
weeks in studies involving such patients would provide an added refinement
for the statistical analysis of the data" (p. 204).
Whether one is examining drug effects or behavioral interventions, the
implications of cyclic variation for single-case methodology are enormous.
Indeed, the psychiatric literature is replete with examples of the deleterious
effects (leading to increased incidence of psychopathology) of the premen-
strual and menstrual phases of the estrus cycle on a wide variety of target
behaviors in pathological and nonpathological populations (e.g., Dalton,

1959, 1960a, 1960b, 1961; G. S. Glass, Heninger, Lansky, & Talan, 1971;
Mandell & Mandell, 1967; Rees, 1953).
To illustrate, we will consider the following possibility. Let us assume that
alternating placebo and active drug conditions are being evaluated (one week
each per phase) on the number of physical complaints issued daily by a young
hospitalized female. Let us further assume that the first placebo condition
coincides with the premenstrual and early part of the subject's menstrual
cycle. Instigation of the active drug would then be confounded with cessation
of the subject's menstrual phase. Assuming that resulting data suggest a
decrease in somatic complaints, it is entirely possible that such change is
primarily due to correlated factors (e.g., effects of the different portions of

the subject's menstrual cycle). Of two phases
course, completion of the last
(A and B) of this A-B-A-B design mightno change in data patterns
result in
across phases. However, interpretation of data would be complicated unless
the experimenter were aware of the role played by cyclic variation (i.e., the
subject's menstrual cycle).
The use of extended measurement phases under these circumstances in
addition to direct and systematic replications (see chapter 10) across subjects
is absolutely necessary in order to derive meaningful conclusions from the
data.
3.7. EVALUATION OF IRREVERSIBLE PROCEDURES

There are certain kinds of procedures (e.g., surgical lesions, therapeutic
instructions) that obviously cannot be withdrawn once they have been ap-
plied. Thus, in assessment of these procedures in single-case research, the use
of reversal and withdrawal designs is generally precluded. The problem of
irreversibility of behavior has attracted some attention and
is viewed as a
major limitation of single-case design by some (e.g., Bandura, 1969). The

notiori jiere is that^ôme^hetapeutic procedures proûcej:gsuJtsiii"jeamlng^
thatwiUjiaLjieversejwh^ Thus, one isjiinabie to
iso late that proced ureas^ effective In response to this, some have advocated
.
withdrawing the procedure early in the treatment phase to effect a reversal.

This strategy is based on the hypothesis that behavioral improvements may
begin as a result of the therapeutic technique but are maintained at a later
point by factors in the environment that the investigators cannot remove (see
Kazdin, 1973; Leitenberg, 1973, also see chapter 5). The most extreme cases
of irreversibility may on
involve a study of the effects of surgical lesions
behavior, or psychosurgery. Here the effect is clearly irreversible. This prob-
lem is easily solved, however, by turning tp ^ irmltiple baseline design In fact.
the multiple baseline strategy is ideally suited for studying such variables, in
that withdrawals of treatment are not required to show the controlling effects
of particular techniques (Baer et al., 1968; Barlow & Hersen, 1973; Hersen,
1982; Kazdin, 1982b). A complete discussion of issues related to the varieties
of multiple baseline designs currently being employed by applied researchers
appears in chapter 7.
In this section, however, the limited use and evaluation of therapeutic

instructions in withdrawal designs will be examined and illustrated. Let us
consider the problems involved in "withdrawing" therapeutic instructions. In
contrast to a typical reinforcement procedure, which can be introduced,
removed, and reintroduced at will, an instructional set, after it has been
given, technically cannot be withdrawn. Certainly, it can be stopped (e.g.,
Eisler, Hersen, & Agras, 1973) or changed (Agras et al., 1969; Barlow,
Agras, Leitenberg, Callahan, & Moore, 1972), but it is not possible to remove
it in the same sense as one does in the case of reinforcement. Therefore, in
light of these issues, when examining the interacting effects of instructions
and other therapeutic variables (e.g., social reinforcement), instructions are
typically maintained constant across treatment phases while the therapeutic
variable is introduced, withdrawn, and reintroduced in sequence (Hersen,
Gullick, Matherne, & Harbert, 1972).
Exceptions
There are some exceptions to the above that periodically have appeared in
the psychological literature. In two separate studies the short-term effects of
instructions (Eisler, Hersen, & Agras, 1973) and the therapeutic value of
instructional sets (Barlow et al., 1972) were examined in withdrawal designs.
In one of a series of analogue studies, Eisler, Hersen and Agras investigated
the effects of focused instructions ("We would you to pay attention as to
like
how much you are looking at each other") on two nonverbal behaviors
(looking and smiling) during the course of 24 minutes of free interaction in
three married couples. An A-B-A-B design was used, with A consisting of 6
minutes of interaction videotaped between a husband and wife in a small
television studio. The B phase also involved 6 minutes of videotaped interac-
tion, but focused instructions on looking were administered three times at 2-
minute intervals over a two-way intercom system by the experimenter from
the adjoining control room. During the second A phase, instructions were
discontinued, while in the second B they were renewed, thus completing 24
minutes of taped interaction.
Retrospective ratings of looking and smiling for husbands and wives (mean
data for the three couples were used, as trends were similar in all cases)
appear in Figure 3-22. Looking duration in baseline for both spouses was
moderate in frequency. In the next phase, focused instructions resulted in a

substantial increase followed by a slightly decreasing trend. When instruc-

tions were discontinued in the second baseline, the downward trend was
maintained. But reintroduction of instructions in the final phase led to an
upward trend in looking. Thus, there was some evidence for the controlling
of introducing, discontinuing, and reintroducing the instructional set.
effects
However, data for a second but "untreated" target behavior smiling
showed almost no parallel effects.
Barlow et al. (1972) examined the effects of negative and positive instruc-
tional sets administered during the course of covert sensitization therapy for
homosexual subjects. In a previous study (Barlow, Leitenberg, & Agras,
1969), pairing of the nauseous scene with undesired sexual imagery proved to
be the controlling ingredient in covert sensitization. However, as the possibil-
ity was raised that therapeutic instructions or positive expectancy of subjects
may have contributed to the treatment's overall efficacy, an additional study

was conducted (Barlow et al., 1972).
The dependent measure in the study by Barlow and his associates was mean
percentage of penile circumference change to selected slides of nude males.
LOOKING SMILING
*
8
VA
Baseline Foe. Instr. Baseline Foe. Instr.
2 3 4 5 ( 7 8 9 10 11 12 23 456 789 10
BLOCKS O TWO MINUTES BLOCKS OF TWO MINUTES
FIGURE 3-22. Mean number of looks and smiles for three couples in 10-second intervals plotted
in blocks of 2 minutes for the Focused Instructions Alone Design. (Figure 4, p. 556, from: Eisler,
R. M., Hersen, M., & Agras, W. S. (1973). Effects of videotape and instructional feedback on
An analog study. Behavior Therapy, 4, 551-558. Copyright 1973
nonverbal marital interactions:
by Association for the Advancement of Behavior Therapy. Reproduced by permission.)
2
Four homosexuals served as subjects in A-BC-A-BD single-case designs.

During A (baseline placebo), a positive instructional set was administered, in
that subjects were told that descriptions of homosexual scenes along with
deep muscle relaxation would lead to improvement. In the BC phase, stan-
dard covert sensitization treatment was paired with a negative instructional
set (subjects were informed that increased sexual arousal would occur). In the
next phase a return to baseline placebo conditions was instituted (A). In the
final phase (BD) standard covert sensitization treatment was paired with a
positive instructional set (subjects were informed that pairing of the nauseous
scene with homosexual imagery, based on a review of their data, would lead
to greatest improvement).
Mean data for the four subjects presented in blocks of two sessions appear
in Figure 3-23. Baseline data suggest that the positive set failed to effect a
decreased trend. In the next phase (BC), a marked improvement was noted as
a function of covert sensitization despite the instigation of a negative set. In
the third phase (A), some deterioration was apparent although a positive set
had been instituted. Finally, in the last phase (BD), covert sensitization
coupled with positive expectation of treatment resulted in renewed improve-
ment.
Baseline lAcquisition with Extinction with |

Reocquisition
plocebo ' negative in str. ther. instr. .with ther. instr.
I I
50
E >
40 \.
c
^
>
30-
2L^ i /
20-
\.
10
J I I I
12 3 4 5 6 7 8 9 10 11 1
Blocks of two sessions
FIGURE 3-23. Mean penile circumference changes to male slides for 4 Ss, expressed as a
percentage of full erection. In each phase, data from the first, middle, and last pair of sessions are
shown. (Figure 1, p. 413, from: Barlow, D. H., Agras, W. S., Leitenberg, H., Callahan, E. J., &
Moore, R. C, (1972). The contribution of therapeutic instruction to covert sensitization. Beha-
viour Research and Therapy, 10, 411-415. Copyright 1972 by Pergamon. Reproduced by
permission.)
In summary, datafrom this study show that covert sensitization treatment

is and that therapeutic expectancy is definitely not the
the effective procedure
primary ingredient leading to success. To the contrary, a positive set paired
with a placebo-relaxation condition in baseline did not yield improvement in
the target behavior.
Although the design in this study permits conclusions as to the efficacy of
positiveand negative sets, a more direct method of assessing the problem
could have been accomplished in the following design: (1) baseline placebo,
(2) acquisition with positive instructions, (3) acquisition with negative instruc-
tions, and (4) acquisiton with postive instructions. When labeled alphabeti-
cally, it provides an A-BC-BD-BC design. In the event that negative
instructions were to exert a negative effect in the BD phase, a reversed trend
in the data would appear. On the other hand, should negative instructions
have no effect or a negligible effect, then a continued downward linear trend
would appear across phases BC, BD and the return to BC.
3.8. ASSESSING RESPONSE MAINTENANCE

work on single-case strategies, it is
In reviewing the theoretical and applied
clear that most of the attention has been directed to determining the func-
tional relationship between treatment intervention and behavioral change.
That is, the emphasis is on response acquisition. (Indeed, this has been the
case in behavior therapy in general.) More recently, greater emphasis has been
accorded to evaluating and ensuring response maintenance following success-
ful treatment (see Hersen, 1981). Specifically with respect to single-case
experimental designs, Rusch and Kazdin (1981) described a methodology for
assessing such response maintenance. Techniques outlined are applicable to
multiple baseline designs (see chapter 7) but also in some instances to the
basic and more complicated withdrawal designs (see chapters 5 and 6).
As noted by Rusch and Kazdin (1981):
In acquisition studies investigators are interested in demonstrating, unequivo-

cally, that a functional relationship exists between treatment and behavioral
change. In maintenance studies, on the other hand, investigators depend on the
ability of the subject to discern and respond to changes in the environment when
the environment is altered; the latter group relies upon subject's failure to
discriminate between those very same stimuli or, possibly, upon the subject's
failure to discriminate among functionally similar stimulus [sic] . . .
(pp.
131-132)
Rusch and Kazdin referred to three types of response maintenance evalua-

tion strategies: (1) sequential-withdrawal, (2) partial-withdrawal, and (3)
partial-sequential withdrawal. In each instance, however, a compound treat-

ment one comprised of several elements or strategies) was being evalu-
(i.e.,
ated. Let us consider the three response maintenance evaluation strategies in

turn.
In sequential-withdrawal, one element of treatment is withdrawn subse-
quent to response acquisition (e.g., reinforcement). In the next phase a
second element of the treatment (e.g., feedback) may be withdrawn, and then
a third (e.g., prompting). This, then, allows the investigator to determine
which, if any, of the treatment elements is required to ensure response
maintenance postacquisition. Examples of this strategy appear in Sowers,
Rusch, Connis, and Cummings (1980) in a multiple baseline design and in
O'Brien, Bugle, and Azrin (1972) in a withdrawal design.
The partial-withdrawal strategy requires use of a multiple baseline design.
Here a component of treatment from one of the baselines or the entire
treatment for one of the baselines is removed (see Russo & Koegel, 1977).
This, of course, allows a comparison between untreated and treated baselines
following response acquisiton. Thus if removal of a part or all of treatment
leads to decremental performance, it would be clear that response main-
tenance following acquisition requires direct and specific programming.
Treatment, then, could be reimplement ed or altered altogether. It should be
noted, however, that, "The possibility exists that the information obtained
from partially withdrawing treatment or withdrawing a component of treat-
ment may not represent the characteristic data pattern for all subjects,
behaviors, or situations included in the design" (Rusch & Kazdin, 1981, p.
136).
Finally, in the partial-sequential withdrawal strategy, a component of treat-
ment from one of the baselines or the entire treatment for one of the baselines
is removed. (To this point, the approach followed is identical to the proce-
dures used in iht partial-withdrawal strategy.) But, this is followed in turn by

subsequent removal of treatment in succeeding baselines. Irrespective of
whether treatment loss appears across the baselines, Rusch and Kazdin (1981)
argued that, "By combining the partial- and sequential-withdrawal design
strategies, investigators can predict, with increasing probability, the extent to
which they are controlling the treatment environment as the progression of
withdrawals is extended to other behaviors, subjects, or settings" (p. 136).
CHAPTER 4
Assessment Strategies
by Donald P. Hartmann
4.1. INTRODUCTION
Assessment strategies that best complement single-case experimental designs
are direct, ongoing or repeated, and intraindividual or ideographic rather
than interindividual or normative. The search is for the determinants of
behavior through examination of the individual's transactions with the social
and physical environment. Thus behavior is a sample, rather than a sign of
the individual's repertoire in the specific assessment setting. This approach,
with its various strategies and philosophical underpinnings, has burgeoned of
late within the general area of behavioral assessment (Hartmann, Roper, &
Bradford, 1979). However, as noted throughout the book, the implementa-
tion of these strategiesis not in any way limited to behavioral approaches to
therapy. The treatment-related functions of assessment are to aid in the

choice of target behavior(s), selection and refinement of intervention tactics,
and evaluation of treatment effectiveness (e.g., Hawkins, 1979; Mash &
Terdal, 1981).
The relative emphasis on these treatment-related functions differs depend-
ingon whether assessment is serving single-case research or between-group
comparison. In the particularly those involving
latter case, selection goals
subjects or target behaviors assume greater importance. In the former case,
treatment refinement, or calibration, assumes greater importance. The imple-
Thanks to Lynne Zarbatany for her critical reading of an earlier draft of this
chapter and to Andrea Stavros for her typing and editorial assistance.
107
mentation of treatment-related functions also varies as a function of single-

subject versus group design. For example, nicthods of evaluating treatment
effectiveness in single case designs (see chapter 2) place much greater empha-
sis on repeated measurement (e.g., Bijou, Peterson, & Ault, 1968). Indeed, as
described in chapter 3, repeated measurement of the target behavior is a
common, critical feature of all single-case experimental designs.
Just as assessment serves diverse functions, it also varies in its focus.
Assessment can be used to evaluate overt motor behaviors such as approach
responses to feared objects, physiological-emotional reactions such as ecto-
dermal reactions and heart-rate acceleration, or cognitive-verbal responses
such as hallucinations and subjective feelings of pain (Nelson & Hayes,
1979).' Assessors may be interested in some or all of these components of the
triple response system, as well as in their covariation (Lang, 1968; also see
Cone, 1979). While assessment can accommodate most any potential focus,
the most common (and perhaps the most desirable) focus in individual
subject research is overt motor behavior.
Because the content focus of assessment may vary widely, a variety of
assessment techniques or methods have been developed. These techniques
include direct observation, self-reports including self-monitoring, question-
naires, structured interviews, and various types of instrumentation, particu-
larly for the measurement of psychophysiological responding (e.g., Haynes,
1978). Though any technique conceivably could be paired with any content
domain, current practices favor certain associations between content and
method: motor acts with direct observations, cognitive responses with self-
report, and physiological responses with instrumentation.
Just as individual subjects researchers prefer to target motor acts, most
also prefer the assessment technique associated with that domain, direct
observation. Indeed, direct observation has been referred to as the
"hallmark," the ''sine qua non,'' and the "greatest contribution" not only of
behavioral assessment but of behavior analysis and modification (see Hart-
mann & Wood, 1982). Though direct observation is indeed overwhelmingly
the most popular assessment technique in published work in the area of
behavior modification (P. H. Bornstein, Bridgwater, Hickey, & Sweeney,
1980), it is noteworthy that the assessment practices of therapists, even
behavior therapists, are considerably more varied (e.g.. Wade, Backer, &
Hartmann, 1979).
This chapter will address issues of particular importance in using assess-
ment techniques for choosing target behaviors and subsequently tracking
them for the purposes of refining and evaluating treatment using repeated
measurement strategies. In keeping with their importance in applied behav-
ioral research, these issues will be addressed in the context of the assessment
of motor behavior using direct observations. Issues featured include defining
target behaviors, selecting response dimensions and the conditions of obser-
Assessment Strategies 109
vation, developing observational procedures, reactivity and other observer

effects, selecting and training observers, and assessing reliability and validity.
Finally, brief mention will be made of other assessment devices used in the
assessment of common target behaviors.
4.2. SELECTING TARGET BEHAVIORS

The phases have been
in assessment, particularly behavioral assessment,
likened to a funnel (e.g.. Cone &
Hawkins, 1977). At its inception, assess-
ment is concerned with such general and broad issues as "Does this individual
have a problem?", and, if so, "What is the nature and extent of the prob-
lem?" Interviews, questionnaires, and other self-report measures often pro-
vide initial answers to such questions, with direct observations in contrived
and norm- or criterion-referenced tests pinpointing the behavioral
settings
components requiring remediation and indicating the degree of disturbance
(Hawkins, 1979). However, the utility of assessment devices for these pur-
poses has not been established (e.g.. Mash & Terdal, 1981). In fact, there is
some evidence by behavioral

that the use of behavioral assessment techniques
assessors produces inconsistent target behavior selection (see Evans &
Wilson, 1983).^
Disagreements in target behavior selection might be limited if behaviors
identified as targets for intervention met one or more of the following criteria
(Kazdin, 1982b; Mash & Terdal, 1981; Wittlieb, Eifert, Wilson, & Evans,
1979): (1) The behavior is considered important to the client or to people who
are close to the client such as spouse or parent; (2) the activity is dangerous to
the client or others; (3) the response is socially repugnant; (4) the actions
seriously interfere with the client's functioning; (5) the behavior represents a
clear departure from normal functioning. Even if an individual's behavior
meets one or more of these criteria, the problem's severity or future course
may be unknown or the specific intervention target may be unclear. This
continued ambiguity might be due to the problem's being poorly defined, or
to its some unknown component of a chain such as long divi-
representing
sion, a symptom complex such as depression, or a construct such as social
skills. A number of empirical methods may help to clarify the problem in
such circumstances.
One method involves comparing the individual's behavior to a standard or
norm to determine the nature and extent of the problem (e.g., Hartmann et
al., comparison procedure was used by Minkin et al. (1976)
1979). This social
improving the conversational skills of predelin-
to identify potential targets to
quent girls. Normative conversational samples provided by effectively func-
tioning youth were examined to determine their distinguishing features. These
features, including asking questions and providing feedback, were then tar-
geted for the predelinquent girls.
no Single-case Experimental Designs
In a second method, subjective evaluation, ratings of response adequacy or

importance are solicited from qualified judges (see Goldfried & D'Zurilla,
1969). For example, Werner et al. (1975) asked police to identify the behav-
iors of suspected delinquents that were important in police-adolescent in-
teractions.These behaviors, including responding politely and cooperatively,
served as target behaviors in a subsequent training program. Subjective
evaluation and social-comparison methods are often referred to as social
validation procedures (Ksizdin, 1977; Wolf, 1978). Methodological appraisals
of social validation procedures have been provided (Forehand, 1983).
In a third method, a careful empirical-logical analysis is conducted of the
problematic behavior to determine which component or components are
performed inadequately (Hawkins, 1975). Task analyses have been conducted
on diverse behaviors, including dart throwing (Schleien, Weyman, & Kiernan,
1981) and janitorial skills (Cuvo, Leaf, & Borakove, 1978). This approach
bears strong similarity to criterion-referencing testing as used to identify
academic deficiencies (e.g., Carver, 1974). Other less-common approaches for
clarifying problem behaviors, including those based on component analysis

and regression techniques, were reviewed by Nelson and Hayes (1981).
If multiple problem behaviors have been targeted following this winnowing
and clarifying procedure, a final decision concerns the order of treating target
behaviors. While the existing (and scant) data on this issue suggest that the
order of treatment of target behaviors may have no effect on outcome
(Eyberg & Johnson, 1974), a number of suggestions have been offered for
choosing the first behavior to be treated (Mash & Terdal, 1981; Nelson &
Hayes, 1981). Behaviors recommended for initial treatment include those that
are (1) dangerous to the client or others; (2) most irritating to individuals in
the client's immediate social environment such as spouse or parent; (3) easiest
to modify; (4) most produce generalized positive effects; (5) earliest
likely to
in a chain or prerequisite to other important behaviors; or (6) most difficult to
modify. Of course this decision, as well as many others faced by therapists,
may have to be based on more mundane considerations, such as skill level of
the therapist or demands of the referral source.
4.3. TRACKING THE TARGET BEHAVIOR USING

REPEATED MEASURES
The stem of the assessment funnel represents the baseline, treatment, and
follow-up phases of an intervention study. Measurement during these phases
requires a more narrow focus on the target behavior for purposes of refining,
and in some cases, extensively modifying, the intervention and subsequently
evaluating its impact.' Assessment during these phases typically employs
direct observation of the target behavior(s) in either contrived or natural
1
Assessment Strategies 1 1
settings (e.g., M. B. Kelly, 1977). A first step in developing or utilizing an

existing observational or other assessment procedure is to operationally
define the target behavior and select the response dimension or property best
suited for the purpose of the study.
Defining the target behavior
After pilot observations have roughly mapped the target behavior by

providing a narrative record of the how, what, when, and where of respond-
ing (e.g., Hawkins, 1982), the investigator will be ready to develop an
operational definition for the behavior. In defining responses, one can either
emphasize topography or function (e.g., J. M. Johnston & Pennypacker,
1980). Topographically based definitions emphasize the movements compris-
ing the response, whereas functionally based definitions emphasize the conse-
quences of the behavior (Hutt & Hutt, 1970; Rosenblum, 1978).
Thumb-sucking might be defined topographically as "the child having his
thumb or any other finger touching or between his lips or fully inserted into
his mouth between his teeth" (Gelfand & Hartmann, 1984). On the other
hand, aggression might be defined functionally as "an act whose goal re-
sponse is injury to an organism" (Dollard, Dobb, Miller, Mowrer, & Sears,
1939, p. 11). According to Hawkins (1982), functional units provide more
valuable information than do topographical units, but they also tend to entail
more assumptions on the part of the instrument developer and more in-
ferences on the part of the observer.
Whether the topographical or functional approach is followed, the defini-
tion should provide meaningful and replicable data. Meaningful, as used
here, is similar in meaning to the term convergent validity (e.g., Campbell &
Fiske, 1959). The definition of the target behavior should agree or converge
with the common uses of the label given the target behavior, and with the
definition used by the referral source and in related behavior change studies
(e.g., Gelfand & Hartmann, 1984)." Replicable refers to the extent to which
similar results would be obtained if the measurement were obtained either in

another laboratory or by two independent observers in the same laboratory
(interobserver agreement).
Interobserver disagreements and other definitional problems can be reme-
died by making definitions objective, clear, and complete (Hawkins & Dobes,
1977). Objective definitions refer only to observable characteristics of the
target behavior; they avoid references to intent, internal states, and other
private events. Clear definitions are unambiguous, easily understood, and
readily paraphrased. A complete definition includes the boundaries of the
behavior, so that an observer can discriminate it from other, related behav-
iors. Complete definitions include the following components (Hawkins,
1982): a descriptive name; a general definition, as in a dictionary; an elabora-
tion that describes the critical parts of the behavior; typical examples of the
TABLE 4-1. Sample Definition of Peer Interaction
Target Behavior: Peer interaction.
Definition: Peer interaction refers to a social relationship between agemates

such that they mutually influence each other (Chaplin, 1975).
Elaboration: Peer interaction is scored when the child is (a) within three feet
of a peer and either (b) engaged in conversation or physical
activity with the peer or (c) jointly using a toy or other play
object.
Example: "Gimme a cookie" directed at a tablemate.

Hitting another child.
Sharing a jar of paint.
Questionable Instances: Waiting for a turn in a group play activity (scored).

Not interacting while standing in line (not scored).
Two children independently but concurrently talking to a
teacher (not scored).
Note. From Gelfand, D. M. & Hartmann, D. P. Child behavior: Analysis and therapy (2nd ed.).
Elmsford, NY: Pergamon Press. Copyright 1984. Reproduced by permission.
behavior; and questionable instances borderline or difficult examples of

both occurrences and nonoccurrences of the behavior. An illustrative defini-
tion of peer interaction meeting these requirements is given in Table 4-1.
Selecting observation settings
The settings used for conducting behavioral investigations have been lim-
ited only by the creativity of investigators and the location of subjects.
Because the occurrences of many behaviors are dependent upon specific
environmental stimuli, behavior rates may well vary across settings contain-
ing different stimuli (e.g., Kazdin, 1979). Thus, for example, drinking as-
sessed in a laboratory bar may not represent the rate of the behavior observed
in more natural contexts (Nathan, Titler, Lowenstein, Solomon, & Rossi,
1970), and cooperative behavior modified in the home may not generalize to
the school setting (R. G. Wahler, 1969b). Even within the home, desirable and
undesirable child behaviors may vary with temporal and climatic variables
(Russell & Bernal, 1977). Thus unless the purpose of an investigation is
limited to modifying a behavior in a narrowly defined treatment context,
observations need to be extended beyond the setting in which treatment
occurs. Observations conducted in multiple settings are required (1) if gener-
alization of treatment effects is to be demonstrated; (2) if a representative
portrayal of the target behavior is to be obtained; and (3) if important
contextual variables that control responding and that may be used to generate
effective interactions are to be identified (e.g., Gelfand & Hartmann, 1984;
Hutt & Hutt, 1970). Given the infrequency with which settings are typically
3
sampled (P. H. Bornstein et al., 1980), these issues either have not captured
the interests of behavior change researchers, or the cost of conducting obser-
vations in multiple settings has exceeded available resources.
While most investigators would prefer to observe behavior as it naturally
number of factors may require that observa-
occurs (e.g., Kazdin, 1982b), a
tions be conducted elsewhere. The reasons for employing contrived or ana-
logue settings include convenience to observers and clients; the need for
standardization or measurement sensitivity; or the fact that the target behav-
ior naturally occurs as a low rate, and observations in natural settings would
involve excessive dross. All of these factors may have determined R. T. Jones,
Kazdin and Haney's (1981b) choice of a contrived setting to assess the
effectiveness of a program to improve children's skill in escaping from home
emergency fires.
The correspondence between behavior observed in contrived observational
settings and in naturalistic settings varies as a function of (1) similarities in
persons present, and (3) the control
their physical characteristics, (2) the
exerted by the observation process (Nay, 1979). Even if assessments are
conducted in naturalistic settings, the observations may produce variations in
the cues that are normally present in these settings. For example, setting cues
may change when structure is imposed on observation settings. Structuring
may range from presumably minor restrictions in the movement and activities
of family members during home observations to the use of highly contrived
situations, as in some assessments of fears and social skills. Haynes (1978),
McFall (1977), and Nay (1977, 1979) provided examples of representative
studies that employed various levels and types of structuring in observation
settings; they also discussed the potential advantages and limitations of
structuring relative to cost, measurement sensitivity, and generalizability.
Cues in observation settings may also be affected by the type of observers
used and their relationship to the persons observed. Observers can vary in
their level of participation with the observed. At the one extreme are nonpar-
ticipant (independent) observers whose only role is to gather data. At the
other extreme are self-observations conducted by the subject or client. In-
termediate levels of participant-observation are represented by significant
others, such as parents, peers, siblings, teachers, aides, and nurses, who are
normally present in the setting where observations take place (e.g., Bickman,
1976). The major advantages of participant-observers is that they may be
present at times that might otherwise be inconvenient for independent obser-
vers, and their presence may be less obtrusive. On the other hand, they may
be less dependable, more subject to biases, and more difficult to train and
evaluate than are independent observers (Nay, 1979).
When observation settings vary from natural life settings either because of
the presence of possibly obtrusive external observers or the imposition of
structure, the ecological validity of the observations is open to question (e.g.,
Barker & Wright, 1955; Rogers-Warren & Warren, 1977). Methods of limiting
these threats to ecological validity are discussed in the section on observer
effects.
Though selection of observation settings is an important issue, investiga-
tors must also determine how best to sample behaviors within these settings.
Sampling of behavior is influenced by how observations are scheduled.
Behavior cannot be continuously observed and recorded except by partici-
pant-observers and when the targets are low-frequency events (see, for exam-
ple, the Clinical Frequency Recording System employed by Paul & Lentz,
1977), or when self-observation procedures are employed (see Nelson, 1977).

Otherwise, the times in which observations are conducted must be sampled,
and decisions must be made about the number of observation sessions to be
scheduled and the basis for scheduling. More samples are required when
behavior rates are low, variable, and changing (either increasing or decreas-
ing); when events controlling the target behaviors vary substantially; and
when observers are asked to employ complex coding procedures (Haynes,
1978).
Once a choice has been made about how frequently to schedule sessions, a
session duration must be chosen. In general, briefer sessions are necessary to
limit observer fatigue when a complex coding system is used, when coded
behaviors occur at high rates, and when more than one subject must be
observed simultaneously. Ultimately, however, session duration, as well as the
number of observation sessions, should be chosen to minimize costs and to
maximize the representativeness, sensitivity,and reliability of data and the
output of information per unit of time. For an extended discussion of these
issues as they apply to scheduling, see Arrington (1943). If observations are to
be conducted on more must be made concerning
than one subject, decisions
the length of time and the order in which each subject will be observed.
Sequential methods, in which subjects are observed for brief periods in a
previously randomized, rotating order, are superior to fewer but longer
observations or to haphazard sampling (e.g., Thomson, Holmberg, & Baer,
1974).
Selecting a response dimension
Behaviors vary in frequency, duration, and quality. The choice of response

dimension(s) ordinarily is based on the nature of the response, the availability
of suitable measurement devices, and the purpose of the study (e.g., Bake-
man, 1978; Sackett, 1978).
Response frequency is assessed when the target behavior occurs in discrete
units that are equal in other important respects, such as duration. Frequency
measures have been taken (1) of a variety of freely occurring responses such
as conversations initiated and headbangs; (2) with discrete-trial or discrete-
5
category responses such as pitches or instructions complied with; and (3)

hit,
when measurement units, such as the number

individuals are themselves the
of individuals who litter, overeat, commit murder, or are in their seats at the
end of recess (Kazdin, 1982b). Behaviors such as crying, for which individual
incidents vary in temporal or in other important respects or which may be
difficult to classify into discrete events, are better evaluated using another
response dimension such as duration.
When response occurrences are easily discriminated, and occur at moder-
ate tolow rates, frequencies can be tallied conveniently by moving an object,
such as a paper clip, from one pocket to another; by placing a check mark on
a sheet of paper; or by depressing the knob on a wrist counter. When
responses occur at very low rates, even a busy participant can record a wide
range of behavior for a large number of individuals (e.g.. Wood, Callahan,
Alevizos, & Teigen, 1979). More complex observational settings require the
use of a complicated recording apparatus or of multiple observers; sampling
of behaviors, individual or both; or making repeated passes through either
video or audio recordings of the target behaviors (e.g.. Holm, 1978; Simpson,
1979).
Response duration, or one of its derivatives such as percentage of time
spent in an activity, is assessed when a temporal characteristic of a response is
targeted such as the length of time required to perform the response, the
response latency, or the interresponse time (Cone & Foster, 1982). While
duration is less commonly observed than is frequency (e.g., M. B. Kelly,
1977), duration has been measured for a variety of target responses including
the length of time that a claustrophobic, patient sat in a small room (Leiten-
berg et al., 1968) and latency to comply with classroom instructions
(Fjellstedt & Sulzer-Azaroff, 1973).
Duration measures require the availability of a suitable timing device and a
target response with clearly discernible onsets and offsets. In single-variable
studies, the general availability and convenience of digital wristwatches with
real time and stopwatch functions may enable even a participant observer to
serve as the primary source of data. In the case of multiple-target behaviors, a
complex timing device such as a multiple-channel event recorder such as a
Datamyte is required.
Response quality is typically assessed when target behaviors vary either in
(1) intensity or amplitude, such as noise level and penile erection; (2) ac-
curacy, such as descriptions of place and time used to test general orientation;
or (3) acceptability, such as the appropriateness of assertion and the intelligi-
bility of speech (Cone & Foster, 1982). These qualitative dimensions may be
evaluated on continuous or discrete scales, and the discrete scales can them-
selves be dichotomous or multi-categorical. For example, assessment of the
amount of food spilled by a child could be made by weighing the child and
the food on his or her plate before and after each meal (quantitative,
continuous), by counting the number of spots on the tablecloth (quantitative,

discrete), or by determining for each meal whether or not spilling had
occurred (dichotomous, discrete). The selection of a particular measurement
scale determined by the discriminatory capabilities of observers, the
is
precision of information required by the study, cost factors, and the

availability of suitable rating devices (e.g., Gelfand & Hartmann, 1984).
To avoid the problems of bias associated with qualitative ratings, particu-
larly of global ratings (e.g. Shuller & McNamara, 1976), scale values should
be anchored or identified in terms of critical incidents or graded behavioral
examples. For example, the anchor associated with a value of five on a seven-
point scale for rating spelling accuracy might be "two errors, including
and excessive letters." P. C. Smith
substitutions, omissions, letter reversals,
and Kendall (1963) described how to develop behavioral rating scales with
empirically formulated anchors, and additional suggestions are given by
Cronbach (1970, chapter 17). Examples of how complex qualitative judg-
ments can be made reliably can be found in Goetz and Baer (1973) and in
Hopkins, Schutte, and Gar ton (1971). Because all qualitative scales can be
conceived of as either frequency or duration measures, they must conform to
the requirements previously described for measurement of these response
dimensions.
Selecting observation procedures
Ahmann's (1974) description of observation procedures (traditionally

called sampling procedures) contained at least five techniques of general use
for applied behavioral researchers. Selection of one of these procedures will
be determined in part by which response characteristics are recorded, and in

turn will determine how the behavioral stream is segregated or divided.
Real-time observations involve recording both event frequency and dura-
tion on the basis of their occurrence in the noninterrupted, natural time flow
(Sanson-Fisher, Poole, Small, & Fleming, 1979). Data from real-time record-
ing are powerful, rigorous, and flexible, but these advantages may come at
the cost of expensive recording devices (e.g., Hartmann & Wood, 1982). The
real-time
method and event recording the technique discussed next are the
only two procedures commonly employed to obtain unbiased estimates of
response frequency, to determine rate of responses, and to calculate condi-
tional probabilities (e.g., Bakeman, 1978).
Event recording, sometimes called frequency recordings, the tally method,
or trial scoring when applied to discrete trial behavior, is used when frequency
is the response dimension of interest. With event recording, initiations of the
target behavior are scored for each occurrence in an observation session or
during brief intervals within a session (H. F. Wright, 1960). Event recording
has the overwhelming advantage of simplicity. Its disadvantages include (1)
7
the fragmentary picture it gives of the stream of behavior; (2) the difficulty of
identifying sources of disagreements between observers, unless the observa-

tions are locked into real time; (3) the unrehability of observations when
response onset or offset are difficult to discriminate; and (4) the tendency of
observers to nod off when coded events occur infrequently (Nay, 1979; Reid,
1978; Sulzer-Azaroff & Mayer, 197). Despite these disadvantages, event re-
cording is a commonly used method in behavior change research (M. B. Kelly,
(1977).
Duration recording is used when one of the previously discussed temporal
aspects of responding is targeted. According to M.B. Kelly (1977), duration
recording is the least used of the common recording techniques, perhaps in
part because of the belief that frequency is a more basic response characteris-
tic (e.g., Bijou 1969),

et al.,and perhaps in part because of the apparent ease
of estimating duration by either of the two methods described next.
Scan sampling, also referred to as instantaneous time sampling, momen-
tary time sampling, and discontinuous probe time sampling, is particularly
useful with behaviors for which duration (percentage of time occurrence) is a

more meaningful dimension than is frequency. With scan sampling, the
observer periodically scans the subject or client and notes whether or not the
behavior is occurring at the instant of the observation. The brief observation
periods that give this technique its name can be signaled by the beep of a
digital watch, an oven timer, or an audiotape played through an earplug, on
either a fixed or random schedule. Impressive applications of scan sampling
with chronic mental patients were described by Paul and his associates (Paul
& Lentz, 1977; Power, 1979).
The final procedure, interval recording, is also referred to as time sampling,
one-zero recording, and the same time one of the
Hansen system. It is at the
most popular recording methods (M. B. Kelly, 1977) and one of the most
troublesome (e.g., Altman, 1974; Kraemer, 1979). With this technique, an
observation session is divided into brief observe-record intervals, and each
interval is scored if the target behavior occurs either throughout the interval,
or, more commonly, during any part of the interval (Powell, Martindale,
& Kulp, 1975). The observation and recording intervals can be signaled
efficiently and unobtrusively by means of an earpiece speaker used in con-
junction with a portable cassette audio recorder. The observers listen to an
audiotape on which is recorded the number of each observation and record-
ing interval, separatedby the actual length of these intervals. If data sheets
are similarly numbered, the likelihood of observers getting lost is substan-
tially reduced in comparison to the use of other common signaling devices.
While interval recording procedures have been recommended for their

ability to measure both response frequency and response duration, recent
research indicates that this method may provide seriously distorted estimates
of both of these response characteristics (see Hartmann & Wood, 1982). As a
measure of frequency, the rate of interval-recorded data will vary depending

upon the duration of the observation interval. With long intervals, more than
one occurrence of a response may be observed, yet only one response would
be scored. With short intervals, a single response may extend beyond an
interval and thus would be scored in more than one interval. As a measure of
response duration, interval-recorded data also present problems. For exam-
ple, duration will be overestimated whenever responses are scored, yet occur
for only a portion of any observation interval. The interval method will only
provide a good estimate of duration when observation intervals are very short
in comparison with the mean duration of the target behavior. Under these
conditions the interval method becomes procedurally similar to scan sam-
pling.
Despite these and other limitations (see Sackett, 1978; Sanson-Fisher et al.,
1979), interval recording continues to enjoy the favor of applied behavioral

researchers (Hawkins, 1982). This popularity is due, no doubt, to the tech-
nique's ease of application to multiple-behavior coding systems, particularly
when some of the behaviors included in the system cannot readily be divided
into discrete units, and its convenience for detecting sources of interobserver
unreliability (Cone & Foster, 1982). Nonetheless, if accurate estimates of
frequency and duration are required, investigators would be well advised to
consider alternatives to interval recording. If real-time sampling is not re-
quired or is prohibitively expensive, adequate measures of response duration
and frequency can result from combining the scan and event recording
techniques. However, data produced by combining these two methods do not
have the same range of applications as data obtained by the real-time proce-
dure.
More
detailed guidelines for selecting an observation procedure were given
inGelfand and Hartmann (1975), in Nay (1979), and in Sulzer-Azaroff and
Mayer (1977). Table 4-2 summarizes the most important of these guidelines.
Additional suggestions for dealing with special recording problems, such as
those involved in observing more than one subject, are available in Bijou et
al. (1968), in Boer (1968), and in Paul (1979).
Observer effects
Observer effects represent a conglomerate of systematic or directional

errors in behavior observations that may result from using human observers.
The most widely recognized and potentially hazardous of these effects include
reactivity, bias, drift, and cheating (e.g., Johnson & Bolstad, 1973; Kent &
Foster, 1977; Wildman & Erickson, 1977).
Reactivity refers to the fact that subjects may respond atypically as a result
of being aware that their behavior is being observed (Weick, 1968). The
factors that contribute to reactivity (e.g., Arrington, 1939; Kazdin, 1982a)
TABLE 4-2. Factors to Consider in Selecting an Appropriate Recording Technique
METHOD ADVANTAGES AND DISADVANTAGES
Real-Time Recording Advantages:

Provides unbiased estimates of frequency and duration.
Data capable of complex analyses such as conditional
probability analysis.
Data susceptible to sophisticated reliability analysis.
Disadvantages:
Demanding task for observers.
May require costly equipment.
Requires responses to have clearly distinguishable
beginnings and ends.
Event or Duration Recording Advantages:

Measures are of a fundamental response characteristic
frequency or duration).
(i.e.,
Can be used by participant-observers (e.g., parents or

teachers) with low rate responses.
Disadvantages:
Requires responses to have clearly distinguishable
Unless responses are located in real time (e.g., by dividing
a session into brief recording intervals), some forms of
reliability assessment may be impossible.
May be difficult with multiple behaviors unless mechanical
aids are available.
Momentary Time Samples Advantages:

Response duration of primary interest.
Time-saving and convenient.
Useful with multiple behaviors and/or children.
Applicable to responses without clear beginnings or ends.
Disadvantages:
Unless samples are taken frequently, continuity of
behavior may be lost.
May miss most occurrences of brief, rare responses.
Interval Recording Advantages:
Sensitive to both response frequency and duration.
Applicable to wide range of responses.
Facilitates observer training and reliability assessments.
Applicable to responses without clearly distinguishable
Disadvantages:
Confounds frequency and duration.
May under- or overestimate response frequency and
duration.
Note. Adapted from Gelfand, D. M. & Hartmann, D. P. (1984). Child behavior: Analysis and
therapy (2nd ed,). Elmsford, NY: Pergamon Press. Copyright 1984. Reproduced by permission.
include the following: (1) Socially desirable or appropriate behaviors may be

facilitated while socially undesirable or "private" behaviors may be sup-
pressed when subjects are aware of being observed (e.g., Baum, Forehand, &
Zegiob, 1979); (2) the more conspicuous or obvious the assessment proce-
dure, the more likely it is to evoke reactive effects; however, numerous
contrary findings have been obtained, and such factors as observer proximity
to subjects and instructions that alert subjects to observations do not guaran-
tee reactive responding (see Hartmann & Wood, 1982); (3) observer attributes
such as sex, activity level/responsiveness, and age appear to influence reactiv-
whereas adults are influenced by observers' appearance, tact,
ity in children,
and public-relations skills (e.g., Haynes, 1978; also see Johnson & Bolstad,
1973); (4) young children under the age of six and subjects who are open and
confident or perhaps merely insensitive may react less to direct observation
than subjects who do not share these characteristics; and (5) the rationale for
observation may affect the degree to which subjects respond in an atypical
manner (see discussion by Weick, 1968). Johnson and Bolstad (1973) recom-
mended providing a thorough rationale for observation procedures in order
to reduce subject concerns and potential reactive effects due to the observa-
tion process. Other methods for reducing reactivity also may prove useful
(Kazdin, 1979; 1982a).
1. Use unobtrusive observational procedures (see Sechrest, 1979; Webb et al.,

1981). For example, Hollands worth, Glazeski, and Dressel (1978) evalu-
ated the effects of training on the social-communicative behavior of an
anxious, verbally deficient clerk by observing him unobtrusively at work
while he interacted with customers.
2. Reduce the degree of obtrusiveness by hiding observers behind one-way
mirrors or making them less conspicuous, that is, by having them avoid
eye contact with the observee. Table 4-3 lists suggestions for classroom
observers that are intended to decrease their obtrusiveness and hence the
reactivity of their observations.
3. Increase reliance on reports from informants who are a natural part of the
client's social environment.
4. Obtain assessment data from multiple sources differing in method arti-
fact.
5. Allow subjects to adapt to obervations before formal data collection

begins. Unfortunately, the length of time or number of observation ses-
sions required for habituation is unclear, and recommended adaptation
periods range as high as six hours for observations conducted in homes
(see Haynes, 1978).
Observer bias is a systematic error in assessment usually associated with

observers' expectancies and prejudices as well as their information-processing
.
TABLE 4-3. Suggestions for School Observers
1 Obtain the caretaker's permission to observe the child in the classroom or other
school environment.
2. Consult the classroom teacher prior to making observations and agree upon an
acceptable introduction and explanation for your presence in the classroom. Also
arrange for mutually agreeable observation times, location, etc.
3. Insofar as possible, coordinate your entry and exit from the classroom with normal
breaks in the daily routine.
4. Be inconspicuous in your personal appearance and conduct.
5. Do not strike up conversations with the children.
6. Sit in an inconspicuous location from which you can see but cannot easily be seen.
7. Disguise your interest in the target child by varying the apparent object of your
glances.
8. Do not begin systematic behavioral observations until the children have become
accustomed to your presence.
9. Minimize disruptions by taking your observations at the same time each day.
10. Thank the teacher for allowing you to visit the classroom.
Note. Adapted from Gelfand, D. M. & Hartmann, D. P. (1984). Child behavior: Analysis and
therapy (2nd ed.). Elmsford, NY: Pergamon Press. Copyright 1984. Reproduced by permission.
limitations. Observers may, for example, impose patterns of regularity and

orderliness on otherwise complex and unruly behavioral data (Hollenbeck,
1978; Mash & Makohoniuk, 1975). Other systematic errors are due to obser-
vers* expectancies including explicit or implicit hypotheses about the purposes
of an investigation, how subjects should behave, or perhaps even what might
constitute appropriate data (e.g., Haynes, 1978; Kazdin, 1977; Nay, 1979).
Observers may also develop biases on the basis of overt expectations resulting
from knowledge of experimental hypotheses, subject characteristics, and
prejudices conveyed exphcitly or implicitly by the investigator (e.g., O'Leary,
Kent, & Kanowitz, 1975).
Methods of controlling biases include using professional observers; using
videotape recording with subsequent rating of randomly ordered sessions;
maintaining experimental naivete among observers; cautioning observers
about the potential lethal effects of bias; employing stringent training criteria;
and using precise, low-inference operational definitions (Haynes, 1978; Kaz-
din, 1977; Redfield & Paul, 1976; Rosenthal, 1976; also see Weick, 1968). If
there any reason to doubt the effectiveness with which observer bias is
is
being controlled, investigators should assess the nature and extent of bias by
systematically probing their observers (Hartmann, Roper, & Gelfand, 1977;
Johnson & Bolstad, 1973).
Observer drift, or instrument decay (Cook & Campbell, 1979; Johnson &
Bolstad, 1973), occurs when observer consistency or accuracy decreases, for
example, from the end of training to the beginning of formal data collection
(e.g., Taplin & Reid, 1973).^ Drift occurs when a recording-interpretation bias
has gradually evolved over time (Arrington, 1939, 1943) or when response
definitions or measurement procedures are informally altered to suit novel
changes in the topography of some target behavior (Doke, 1976). Drift can
also result from observer satiation or boredom (Weick, 1968). Observer drift
can cause inflated estimates of interobserver reliability when these estimates
are based on data obtained (1) during training sessions, (2) from overt
reliability assessment no matter when scheduled, or (3) from a long-standing,
familiar team of observers during the course of a lengthy investigation (see
Hartmann & Wood, 1982).
Drift can be limited or its effects reduced by providing continuing training
throughout a project, by training and recalibrating all observers at the same
time, and by inserting random and covert reliability probes throughout the
course of the investigation. Alternatively, investigators can take steps to
evaluate the presence of observer drift by having observers periodically rate
prescored videotapes (sometimes referred to as criterion videotapes), by
conducting reliability assessment across rotating members of observation
teams, and by using independent reliability assessors (see reviews by Cone &
Foster, 1982; Hartmann & Wood, 1982; Haynes, 1978).
Observer cheating has been reported only rarely (e.g., Azrin, Holz, Ulrich,
& Goldiamond, 1961). More commonly, observers have been known to
though these calculation mistakes are
calculate inflated reliability coefficients,
not necessarily the result of intentional fabrication (e.g., Rusch, Walker, &
Greenwood, 1975). Precautions against observer cheating include random,
unannounced reliabihty spot checks; collection of data forms immediately
after an observation session ends; restriction of data analysis and rehability
calculations to individuals who did not collect the data; provision of pens
rather than pencils to raters (obvious corrections might then be evaluated as
an indirect measure of cheating); and reminders to observers about the
canons of science and the dire consequences of cheating (Hartmann & Wood,
1982). See the section on staging reliability assessments (p. 124) for further
suggestions regarding limiting observer drift and observer cheating.
Selecting and training observers
Unsystematic or random observer errors as well as many of the systematic

sources of error in observational data just described may be partially con-
trolled by properly selecting observers and training them well.
Behavioral researchers seem unaware of the substantial amount of research
on individual differences in observational skills (see Boice, 1983). In general,
observational skills increase with age and are better developed in women than
in men. There is also some evidence components of social
to suggest that the
skills, such as the ability to perceive nonverbally communicated affect, may
be related to observer accuracy, and that the perceptual-motor skills of

observers may prove directly relevant to training efficiency and to the main-
tenance of desired levels of observer performance (e.g., Nay, 1979). Addi-
tional observer attributes that may be important include morale, intelligence,
motivation, and attention to detail (e.g., Boice, 1983; Hartmann & Wood,
1982; Yarrow & Waxier, 1979).
Once potential observers are selected, they require systematic training in
order to perform adequately. Recent reviews of the observer-training litera-
ture (e.g., Hartmann & Wood, 1982; Reid, 1982) suggest that observers
should progress through a sequence of training experiences that includes
general orientation, learning the observation manual, conducting analogue
observations, in situ practice, retraining-recalibration, and debriefing. Train-
ing should begin with a suitable rationale and introduction that explains to
the observers the need for tunnel vision for remaining naive regarding the
purpose of the study and its experimental hypotheses. They should be warned
against attempts to generate their own hypotheses and instructed to avoid
private discussions of coding procedures and problems. Observers should
also become Conduct of
familiar with the APA's Ethical Principles in the
Research with Human Participants (1973); particular emphasis should be
placed upon issues confidentiality, the canons of science, and observer
etiquette.
Next, observer trainees should memorize verbatim the operational defini-
tions, scoring procedures, and examples of the observation system as pre-
sented in a formal observation training manual (Paul & Lentz, 1977).
(Suggestions for constructing observation manuals are given by Nay, 1979, p.
237.) Oral drills, pencil-and-paper tests, and scoring of written descriptions of
behavioral vignettes can be employed for training and evaluation at this
stage. Investigators should utilize appropriate instructional principles such as
successive approximations and ample positive reinforcement in teaching their
observer trainees appropriate observation, recording, and interpersonal
skills.Having passed the written test, observers should next be trained to
criterion accuracy and consistency on a series of analogue assessment samples
portrayed via film clips or role playing. Training should begin with exposure
to simple or artificially simplified behavioral sequences; later material should
present rather complex interactional sequences containing unpredictable and
variable patterns of responding. The observers should be overtrained on these
materials in order to minimize later decrements in performance. Immediately
after observers complete each training segment, their protocols should be
reviewed, and both correct and incorrect entries should be discussed (Reid,
1982). During this phase, observers should recode training segments until
100*^0agreement with criterion protocols is achieved (Paul & Lentz, 1977).
Discussion of procedural problems and confusions should be encouraged
throughout this training phase, and all scoring decisions and clarifications
should be posted in an observer log or noted in the observation manual that
each observer carries.
Practice in the observation setting follows. Practice observations can serve
the dual purpose of desensitizing observers to fears about the setting (i.e.,
inpatient psychiatric unit) and allowing subjects or clients to habituate to the

observation procedures. Training considerations outlined in the previous step
are also relevant here. Particular attention should be given to observer
motivation. Reid (1982) suggests that observer motivation and morale may be
strengthened by providing observers with (1) varied forms of scientific stimu-
lation such as directed readings on topics related to the project, and (2)
incentives for obtaining reliableand accurate data.
During the course of the investigation, periodic retraining and recalibration
sessions should be conducted with all observers: recalibration could include
spot tests on the observation manual, coding of prescored videotapes, and
covert reliability assessments. If data quality declines, extra retraining ses-
sions should be held. At the end of the investigation, observers should be
interviewed to ascertain any biases or other potential confounds that may
have influenced their observations. Observers should be informed about the
nature and results of the investigation and should receive acknowledgment in
technical reports or publications.
Reliability
Observational instruments require periodic assessments to ensure that they

promote correct decisions regarding treatment effectiveness. Such evaluations
are particularly critical for relatively untried observational instruments, for
those that attempt to obtain scores on multiple-response dimensions, and for
those that are applied in uncontrolled, naturalistic settings by unprofessional
personnel. Traditionally, these evaluations have fallen under the domain of
one of the various theories of reliability (or more recently of generalizability)
and its associated methods (Cronbach et al., 1972; Nunnally, 1978).
Any reliability analysis requires a series of decisions. These decisions
involve selecting the dimensions of observation that require formal assess-
ment; deciding on the conditions under which reliability data will be
gathered; choosing a unit of analysis; selecting a summary reliability statistic;
interpreting the values of reliability statistics; modifying, if necessary, the
data collection plan; and reporting reliability information.
The first step in assessing data quality is to decide the dimensions (or facets)
of the data that are important to the research question. Potentially relevant
dimensions can include observers, coding categories, occasions, and settings
(e.g.. Cone, 1977). With the exception of interobserver reliability,* these
dimensions have not engaged the systematic attention of researchers using
observations (Hartmann & Wood, 1982; Mitchell, 1979). This is unfortunate

because sessions or occasions clearly deserve as much attention as observers
have already received (Mitchell, 1979) and are particularly important in single-
case research. Without observation sessions of adequate number and dura-

tion, the resulting data will be unstable. Data that are unstable, either because
of variability or because of trends in the changeworthy direction, may pro-
duce inconclusive tests of treatment effects (see chapter 9). Because of the
pivotal importance of observers and sessions to the use of observational
codes, the remainder of this section will refer to these two aspects of observa-
tional reliability.
Conditions of observation can affect the performance of both subjects and
observers and, hence, estimates of data quality or dependability (e.g., Hart-
mann & Wood, For example, observer performance improves, some-
1982).
times substantially, under overt, in comparison to disguised, reliability
assessment conditions. Because most reliability assessments are conducted
under overt conditions, much of our observational data are substantially less
adequate than our interobserver reliability analyses suggest. The performance
by observers also can deteriorate substantially from training to the later

phases of an investigation, and in response to increases in the complexity of
the behavior displayed by subjects (e.g.. Cone & Foster, 1982). The quality of
data recorded by observers can also vary as a function of their expectations
and biases and as a result of calculation errors and fabrication, as previously
discussed.
To counter the distortions that these conditions can produce, (1) subjects
and observers should be given time to acclimate to the observational setting
before reliability data are collected; (2) observers should be separated and, if
possible, kept unaware of both when reliability assessment sessions are sched-
uled and the purpose of the study; (3) observers should be reminded of the
importance of accurate data and regularly retrained with observational stim-
ulivarying in complexity; (4) reliability assessments should be conducted
throughout the investigation, particularly in each part of multiphase behav-
ior-change investigations; and (5) the task of calculating reliability should be
undertaken by the investigator, not by the observers (Hartmann, 1982).
Before a reliabihty analysis can be completed, the investigator must deter-
mine the appropriate behavioral units (or the levels of data) on which the
analysis will be conducted (Johnson & Bolstad, 1973). A common, molar unit
is obtained by combining the scores of either empirically or logically related
molecular variables. For example, scores on tease can be added to scores on

cry, humiliate, and the like to generate a total aversive behavior score (R. R.
Jones, Reid, & Patterson, 1975). Still other composite units can be based on
aggregation of scores over time. For example, students* daily question asking
can be combined over a 5-day period to generate weekly question-asking
scores.
SCED E*
Because the reliability of composites differs from the reliability of their

components (e.g., Hartmann, 1976), investigators should be careful not to
make inferences about the reliability of composites based upon the reliability
of their components, and vice versa. To ensure that reliability is neither
overestimated nor underestimated, reliability calculations should be per-
formed on the level of data or units of behavior that will be subjected to
substantive analysis. Thus if weekly behavior rate is the focus of analysis, the
reliability of the rate measure should be assessed at the level of data summed
over the seven days of a week. However, in some situations, it may be useful
to assess reliability at a finer level of data than that at which substantive
analyses are conducted. For example, even if data are analyzed at the level of
daily session totals, assessment of reliability on individual trial scores can be
useful in identifying specific disagreements that indicate the need for more
observer training, for revision of the observer code, or for modification of
recording procedures (Hartmann, 1977).
Investigators have a surfeit of statistical indexes to use in summarizing their
reliability data. Berk (1979) described 22 different summary reliability statis-
tics, and both Fleiss (1975) and House, House, and Campbell (1981) dis-
cussed 20 partially overlapping sets of procedures for summarizing the
reliability of categorical ratings provided by two judges. Still other summary
statistics were described by Prick and Semmel (1978), Tinsley and Weiss
(1975), and Wallace and Elder (1980). These statistics differ in their
appropriateness for various forms of data, their inclusion of a correction
for chance agreement, the factors that lower their numerical value (con-
tribute to error), their underlying measurement scale, their capacity for
summarizing scores for the entire observational system with a single index,
and their degree of computational complexity and abstractness (Hartmann,
1982).
Observation data are typically obtained in one or both of two forms: (1)
categorical data such as occur-nonoccur, correct-incorrect, or yes-no that
might be observed in brief time intervals or scored in response to discrete
trials; and (2) quantitative data such as response frequency, rate, or duration.
Somewhat different summary statistics have been developed for the two
kinds of data.
Table 4-4 includes a two-by-two table for summarizing categorical data and
the statistics commonly used
or recommended for these data. These statistics
all raw agreement (referred to as percent agreement in its
are progeny of
common form), the most common index for summarizing the interobserver
consistency of categorical judgments (M. B. Kelly, 1977). Raw agreement has
been repeatedly criticized, largely because the value of this statistic may be
inflated when the target behavior occurs at extreme rates (e.g., Mitchell,
1979). A variety of techniques have been suggested to remedy this problem.
Some procedures differentially weight occurrence and nonoccurrence agree-
TABLE 4-4. TWo-by-TVo Summary Table of Relative Proportion of Occurrence

of a Behavior as Recorded by T\vo Observers,
with Selected Statistical Procedures Applicable to These Data
SUMMARY TABLE
Occurrence Nonoccurrence Total
Occurrence .60 = a .05 = b .65 = p^

0\ Nonoccurrence .10 = c .25 = d .35 = Qi
Total .70 = p2 .30 = Q2 1.00
Raw Agreement = a + d = .85
Occurrence Agreement = a /(a + b + c) = .SO
Nonoccurrence Agreement = d/ib + c + d) = .63
Kappa = (a + d - p.p^ - q^QiVil - PiP2 - q^gi) - .66
Note. Some of the summary statistics described here commonly employ a percentage scale (for
example, raw agreement). For convenience, these statistics are defined in terms of a proportion
scale. (Adapted from Hartmann, D. P. (1982). Assessing the dependability of observational data.
In D. Hartmann, (Ed.), Using observers to study behavior: New directions for methodology of
P.
social and behavioral science. San Francisco: Jossey-Bass. Copyright 1982 by D. P. Hartmann.
ments (e.g., Cone & Foster, 1982; Hawkins & Dotson, 1975), whereas other
procedures provide formal correction for chance agreements. The most pop-
ular of these corrected statistics is Cohen's kappa (J. Cohen, 1960). Kappa
has been discussed and illustrated by Hartmann (1977) and HoUenbeck
(1978), and a useful technical bibliography on kappa appears in Hubert
(1977). Kappa may be used for summarizing observer agreement as well as
accuracy (Light, 1971), for determining consistency among many raters
(A. J. Conger, 1980), and for evaluating scaled (partial) consistency among
observers (J. Cohen, 1968).
Table 4-5 includes qualitative data from a subject scores from six sessions
for
two observers and analyses of these data. The percentage agreement for
these data, sometimes called marginal agreement (Frick & Semmel, 1978), is
the ratio of the smaller value (frequency or duration) to the larger value
obtained by two observers, multiplied by 100. This form of percentage
agreement also has been criticized for potentially inflating reliability estimates
(Hartmann, 1977). Berk (1979) advocated use of generalizability coefficients,
as these statistics provide more information and permit more options than do
either percentage agreement or simple correlation coefficients (also see Hart-
mann, 1977; Mitchell, 1979; and Shrout & Fleiss, 1979). Despite these advan-
tages, some researchers argue that generalizability and related correlational
approaches should be avoided because their mathematical properties may
TABLE 4-5. Days-by-Observers Data and Analysis of These Data
OBSERVERS
Sessions 0, O2 "Percentage Agreement"
1 11 9 82%
2 8 6 75<7o
3 9 7 78<7o
4 10 9 90Vo
5 12 11 92<ô
6 8 8 100<?7o
ANALYSIS OF VARIANCE SUMMARY

Sources Mean Squares (MS)
Between Sessions (BS) 5.40

Within Sessions (WS) 1.16
Observers (0) 5.33
S X .33
GENERALIZABILITY OR INTERCLASS COEFFICIENTS (ICQ
ICC (1,1) = (MSbs - MSyysVWSss + (k- \)MS^s] =

(5.40 - 1.16)/[5.40 + 5(1.16)] = .38
=
ICC (3.1) = (MSss - MSsôVIMSbs + {k -DMSsxol =

(5.40 - .33)/[5.40 + 5(.33)] = .72
Note. Adapted from Hartmann, D.P. (1982). Assessing the dependability of observational data.
In D. P. Hartmann (Ed.), Using observers to study behavior: New directions for methodology of
social and behavioral science. San Francisco: Jossey-Bass. Copyright 1982 by D. P Hartmann.
inhibit applied behavior analysis from becoming a "people's science" (Baer,

1977a; Hawkins & Fabry, 1979).
Disagreement about procedures for summarizing observer reliability are
also related to differing recommendations for "acceptable values" of obser-

ver reliability estimates. Given the variety of available statistics
with various
statistics based on different metrics and employing different conceptions of
error a common standard for satisfactory reliability seems unlikely. Never-

theless, recommendations have ranged from .70 to .90 for raw agreement,
and from .60 to .75 for kappa-like statistics (see Hartmann, 1982). While
these recommendations will be adequate for many, even most, research
purposes, the overriding basis for judging the adequacy of data is whether
they provide a powerful means of detecting experimentally produced or
naturally occurring response covariation.
Power depends not only on data quality, but also on the magnitude of
number of available investigative units (for
covariation to be detected, the
example, sessions), and the experimental design. Thus, data quality must be
evaluated in the context of these factors (Hartmann & Gardner, 1979). If
consideration of these factors indicates that the data are of adequate quality,
further modification of the observational system is not required. However,
if one or more forms of reliability prove unacceptable, revision of the
research plan is in order.
If the quality judged unsatisfactory, a number of options are
of data is
available to the investigator.For example, if consistency across observers is

inadequate, the investigator can train observers more extensively, improve
observation and recording conditions, clarify definitions, use more than one
observer to gather data and analyze the average of the observers' scores, or
employ some combination of the options just described (Hartmann, 1982).
If the performance of observer is adequate, but the target behavior varies
substantially across occasions, the researcher may modify the observational
setting by removing distracting stimuli or by adding a brief habituation
period to each observational session (e.g., Sidman, 1960), increase the length
of each observation period until a session duration is discovered which will
provide consistent data, or increase the number of sessions and then average
scores over the number of sessions required to achieve stable performance.
The option that is selected will depend upon the purpose of the study and
on practical considerations, such as the investigator's ability to identify and
control undesirable sources of variability and the feasibility of increasing the
number or length of observation sessions (Hartmann & Gardner, 1981).
Recommendations for reporting reliability information have ranged from
the suggestion that investigators embellish their primary data displays with
disagreement ranges and chance agreement levels (Birkimer & Brown, 1979)
to advocacy of what appear to be cumbersome tests of statistical significance
(Yelton, Wildman, & Erickson, 1977). The recommendations that follow
were proposed by Hartmann and Wood (1982): (1) Reliability estimates
should be reported on interobserver accuracy, consistency, or both, as well as
on session reliability; (2) in the case of interobserver consistency or accuracy
assessed with agreement statistics, either a chance-corrected index or the
chance level of agreements for the index used should be reported; (3) reliabil-
ity should be reported for covert reliability assessments scheduled periodically
throughout the course of the study, for different subjects (if relevant), and
across experimental conditions; and (4) reliability should be reported for each
variable that is the focus of substantive analysis.
VaUdity
Validity, or the extent to which a score measures what it is intended to

measure, has not received much attention in observation research (e.g.,
Johnson & Bolstad, 1973; O'Leary, 1979). In fact, observations have been
considered inherently valid insofar as they are based on direct sampling of

behavior and they require minimal inferences on the part of observers
(Goldfried & Linehan, 1977). According to Haynes (1978) the assumption of
inherent validity in observations involves a serious epistemological error. The
data obtained by human observers may not be veridical descriptions of
behavior. As previously discussed, accuracy of observations can be attenu-
ated by various sources of unreliability and contaminated by reactivity effects
and other sources of measurement bias. The occurrence of such measure-
ment-specific sources of variation provides convincing evidence for the need
to validate observation scores. Validation is further indicated when observa-
tions are combined to measure some higher-level construct such as deviant
behavior or when observation scores are used to predict other important
behaviors (e.g., Hartmann et al., 1979; Hawkins, 1979). Validation may take
the form of content, criterion-related (concurrent and predictive), or con-
struct validity.
Although each of the traditional types of validity is relevant to observation
systems (e.g., Hartmann et al., 1979), is especially important
content validity
development of a behavior coding schema. Content validity is
in the initial
assessed by determining the adequacy with which an observation instrument
samples the behavioral domain of interest (Cronbach, 1971). According to
Linehan (1980), three requirements must be met to establish content validity.
Firsty the universe of interest (i.e., domain of relevant events) must be
completely and unambiguously defined. Depending upon the nature and
purposes of an observation system, this requirement may apply to the behav-
iors of the target subject, to antecedent and consequent events provided by
other persons, or to settings and temporal factors. Next, these relevant
factors should be representatively sampled for inclusion in the observation
system. Finally, the method and combining observations to
for evaluating
form scores should be specified.
The criterion-related validity of assessment scores refers primarily to the
degree to which one source of behavioral assessment data can be substituted
for by another. Though the literature on the consistency between alternative
sources of assessment data is small and inconclusive, there is evidence of poor
correspondence between observation data obtained in structured (analogue)
and in naturaHstic settings (e.g.. Cone & Foster, 1982; Nay, 1979).
settings
Poor correspondence has also been shown when contrasting observation data
with less reactive assessment data (Kazdin, 1979). These results suggest that
behavioral outcome data might have restricted generalizability and under-
score the desirability of criterion-related validity studies when observational
and alternative data sources are used to assess treatment outcome.
Construct validityis indexed by the degree to which observations accu-
rately measure some psychological construct. The need for construct validity
is most apparent when observation scores are combined to yield a measure of
some molar behavior category or construct such as "assertion." G. R. Patter-

(e.g., Johnson & Bolstad, 1973; R. R. Jones, Reid, &
son and his colleagues
Patterson, 1975; Weinrott, Jones, & Boler, 1981) have illustrated construct
validation procedures with their composite. Total Deviancy. Their investiga-
tions have demonstrated, for example, that the Total Deviancy score discrimi-
nates between clinical and nonclinical groups of children and is sensitive to
the social-learning intervention strategies for which it was initially developed.
Despite the impressive work done by Patterson and his associates, as well as
by other behavioral investigators (e.g., Paul, 1979), the validation of an
instrument is an ongoing process. Observations may have impressive validity
for one purpose, such as for evaluating the effectiveness of behavioral
interventions (see Nelson & Hayes, 1979), but they may be only moderately
valid or even invalid measures for subsequent assessment purposes. The
validity of observation data for each assessment function must be indepen-
dently verified (e.g.. Mash & Terdal, 1981).
4.4 OTHER ASSESSMENT TECHNIQUES

Target behaviors may be identified for which direct observations are im-
practical, impossible, or unethical (e.g.. Cone & Foster,
1982). In such cases,
one or more alternative assessment techniques are required. These techniques
may include products of behavior, self-report measures, or physiological
procedures. Measurement of behavioral products, such as number of emptied
Hquor containers, may be particularly useful when the target behavior is
relatively inaccessible to direct observation because of its infrequency, sub-
tlety, or private nature; when either the behavior or its observation causes
embarrassment to the client; or when observation by others would otherwise
disrupt or seriously distort the form, incidence, or duration of the response.
Self-report measures also may be useful in such circumstances, though they
are prey to a number of distorting influences. At other times, physiological
measures may be required, because either the response is ordinarily inaccessi-
ble to unaided human observers or observers cannot provide measures of
sufficient precision. It is to these classes of measures that we briefly turn next.
Behavioral products
Many target behaviors have relatively enduring effects on the environment.

Measuring these behavioral effects or products allows the investigator to
make inferences about the target behaviors associated with the products. This
indirect approach to assessment has several advantages including conven-
ience, nonreactivity, and economy. Because the products remain accessible for
some length of time, they can be accurately and precisely measured at a time,

and perhaps a location, convenient to the investigator (Nay, 1979). Further-

more, because behavioral products do not require the immediate presence of
an observer, they can be measured unobtrusively (and hence nonreactively)
and with relatively little cost.
Behavioral products have been used by a large number of behavioral
investigators (Kazdin, 1982c). For example, Stuart (1971) used client weight
as a measure of eating, and Hawkins, Axelrod, and Hall (1976) assessed
various academic behaviors using task-related behavioral products such as
number of solved math problems. Webb, Campbell, Schwartz, and Sechrest
(1966) lent some order to the array of possible behavioral products by
organizing them into three classes: (1) erosion measures such as shortened
fingernails used to index nail biting (McNamara, 1972); (2) trace measures
such as clothes-on-the-floor to assess "cabin-cleaning" (Lyman, Richard, &
Elder, 1975); (3) and archival records such as number of irregular hospital
discharges to indicate discontent with the hospital (P. J. Martin & Lindsey,
1976). Both Sechrest (1979) and Webb et al. (1981) presented impressive
catalogs of these indirect measures of behavior.
Behavioral by-products, as well as any other indirect or proxy measures,
require validation before they can be used with confidence. Until such valida-
tion is undertaken, questions remain regarding how accurately the product
measure corresponds to the behavior it presumably indexes (J. M. Johnston
& Pennypacker, 1981). For example, weight loss, a common index of eating
reduction, also may reflect increased exercise and the use of diuretics or
stimulants (Haynes, 1978). The distance of behavioral products from their
target behaviors also may be troublesome (Nay, 1979). As a result of working
with the product, rather than the behavior itself, information on controlling
variables may be lost, and changes produced in the target behavior may not
be indicated quickly enough. Furthermore, if behavioral products are conse-
quated, the temporal delay of reinforcement may be too great to strengthen
appropriate target responding.
Self-report measures
In the tripartite classification of responses (motor, cognitive, and physiolo-

gical), self-repoft measures are associated with the assessment of the cognitive
domain thoughts, beliefs, preferences, and other subjective dimensions
because of the inaccessibility of this domain to more direct assessment
approaches. However, self-report techniques also can be used to measure
motor and physiological responses that potentially could be assessed objec-
tively (e.g.. Barrios, Hartmann, & Shigetomi, 1981). The latter use of self-
reports is common when cost is a critical concern or when the client is not
part of an "observable social system" (Haynes, 1978).
Like other assessment devices, self-report measures can be used to generate
information at any part of the assessment funnel, from initial screening

decisions to evaluation of treatment outcome. However, they are most pop-
ular as an economical means of getting started during the initial phases of
assessment (Nay, 1979). The use of self-report procedures in treatment evalu-
ation traditionally has been frowned on by investigators, in large part because
of these reports' susceptibility to various forms of bias and distortion, their
lack of specificity, and their mediocre correspondence with objective measures
(e.g., Bellack & Hersen, 1977). However, more recent behavioral self-report
procedures have gained in acceptance for the evaluation of behavioral inter-

vention, particularly in pre-post group treatment investigations (e.g.,
Haynes, 1978) and when used to assess client satisfaction (e.g., Bornstein &
Rychtarik, 1983; McMahon & Forehand, 1983).
Self-report measures come in a variety of forms including paper-and-pencil
self-rating inventories, surveys and questionnaires, checklists, and self-moni-
toring procedures. Discussion of these measures will largely be limited to
paper-and-pencil questionnaries and self-monitoring techniques, as they have
been most widely utilized by behavioral assessors (e.g.. Swan & McDonald,
1978).
Numerous pencil-and-paper self-report questionnaires are available on
which clients are asked to indicate, in response to a series of items (e.g.,
situations or behaviors) their likelihood of engaging in a response (McFall &

Lillisand, 1971), their degree of emotional arousal (e.g., Geer, 1965), or the
frequency with which they engage in particular behaviors(e.g., Lewinsohn &
These inventories or questionnaires provide assessment data on

Libet, 1972).
a broad range of target responses including assertive and other forms of
social behavior, fears, appetitive or ingestive behaviors such as smoking and
drinking, psychophysical responses such as pain, depression, and marital
interactions, to name but a few. In fact, if a behavior has been studied by two
investigators, the chances are very good that at least two different self-report
questionnaires are available for assessing the behavior.' For extensive surveys
of existing behavioral questionnaires, see Haynes (1978), Haynes and Wilson
(1979), and recent reviews of specific content domains published in mono-
graphs devoted to behavioral assessment (e.g.. Barlow, 1981; Hersen &
Bellack, 1981; Mash & Terdal, 1981) and in behavioral assessment journals.
Because self-report inventories vary so substantially in quality and are
potentially prey to a variety of distortions, promising inventories should be
checked against the following evaluative criteria before a final selection is
made (Bellack & Hersen, 1977; Haynes, 1978; Haynes & Wilson, 1979)."
1. Can the inventory be administered repeatedly to clients? If the inventory's

form or content precludes repeated application, or if the scores change
systematically with repeated administration, the self-report procedure is
not suitable for tracking the target response in an individual-subject

investigation.However, even if the inventory does not meet this criterion,

it may be an aid to selecting subjects, target behaviors, or
suitable as
treatments (e.g., Hawkins, 1979).
2. Does the questionnaire provide the required degree of specific information
regarding the target behavior? Many traditional self-report techniques
were based on trait assumptions of temporal, situational, and behavioral
(item) homogeneity or consistency that have proven to be incorrect (e.g.,
Mischel, 1968). Although the increased response and situational specificity
of behavioral self-report measures improve their correspondence with
objective measures (e.g.. Lick, Sushinsky, & Malow, 1977), the term
behavior in an instrument's title does not guarantee the requisite degree of
specificity.
3. Is the inventory sensitive enough to detect changes in performance as a

result of treatment? Although most questionnaires evaluated for sensitiv-
ity have passed this validity hurdle, not all have done so successfully (e.g.,
Wolfe & Fodor, 1977).

4. Does the questionnaire guard against the biases common to the self-report
genre? Self-report measures are susceptible to a variety of test -related and
subject-related distortions. As regards test-related biases, the wording of
items may be so ambiguous that idiosyncratic interpretations by respon-
dents are common Cronbach, 1970). Furthermore, items may re-
(e.g.,
quest information that beyond subjects' discrimination, storage, or

is
recall capabilities, or they may be arranged so as to effect scores (response

bias). Scores may also be effected by clients' attempts at impression
management. Clients may, for example, endorse socially valued responses
(social desirability), agree with strongly worded alternatives (acquies-
cence), endorse responses that they expect to be positively regarded by the
investigator (demand effects), or engage in outright faking or lying. Biases
due to impression management are particularly troublesome in the assess-
ment of subjective experiences, as independent verification of the accuracy
of responding may be difficult or impossible. Unfortunately, few question-
naires include scales designed to detect biased responding or guard against
its occurrence (Evans, 1983).
5. Finally, does the inventory meet expected reliability and validity require-
ments and possess appropriate norms for the population of interest in the
present investigation? Self-report questionnaires may be adequate for one
group, but not for another, so an instrument's technical information must
be examined with care.
Self-monitoring, the second popular type of self-report among behavioral

chnicians, is one major exception: The
similar to direct observation, but with
client is the observer. Data from self-monitoring have been used for target
behavior and treatment selection, as well as for treatment evaluation. How-
ever, in the latter case, objective assessments typically play a more important
role, except when the target is itself a subjective response.
Self-monitoring has proven particularly useful for assessing rare and sensi-
tivebehaviors and responses that are only accessible to the client such as pain
due to migraine headaches (Feuerstein & Adams, 1977) and obsessive rumina-
tions (Emmelkamp & Kwee, 1977). Other responses assessed via self-moni-
toring include appetitive urges, hallucinations, hurt and depressed feelings,
sexual behaviors, and waking time (for insomniacs). An array of behaviors
more susceptible to direct observations also has been monitored by the client,
including weight gain or loss, caloric intake, nail biting, exercise, academic
behaviors, alcohol consumption, and whining. Haynes (1978), Haynes and
Wilson (1979), Nay (1979), and Nelson (1977) surveyed applications of target
behaviors and recording procedures used in self-monitoring.
Self-monitoring procedures share a number of method-related problems.
Foremost among these is reactivity (Haynes & Wilson, 1979; Nelson, 1977).
Reactivity effects vary as a function of the social desirability of the behavior
recorded, with the frequency of positively valued responses likely to increase
and negatively valued acts likely to decrease during the course of self-
monitoring. The obtrusiveness, the timing, and the frequency of self-moni-

toring also may influence the level of subject reactivity. Indeed, because of
these reactive effects, self-monitoring has been included in a number of
treatment packages as an intervention technique (e.g.. Nay, 1979).
A more serious, problem is the variable accuracy of
second, and perhaps
Haynes & Wilson, 1979; Nelson, 1977). Inaccurate self-
self-monitoring (e.g.,
monitoring can be improved by many of the same stratagems used to improve
the accuracy of direct observation: arrange recording procedures that are
convenient, habitual, and generally nonaversive; provide prior training in
self-monitoring; and encourage and dispense contingencies for accuracy. Self-
monitoring accuracy also can be enhanced by means of various social-
influence procedures such as a public commitment to self-monitor (P. H.
Bornstein, Hamilton, Carmody, Rychtarik, & Veraldi, 1977). Despite the fact
that accuracy can be increased through use of these manipulations, there are
numerous factors adversely affecting the validity of self-monitoring; hence
this approach should be used with caution when it is the only method
available for monitoring the progress or outcome of treatment (Haynes,

1978).
Psychophysiological measures
Psychophysiological measures involve the surface recording of physiologi-

cal events, most of which are controlled by the autonomic nervous system
(Haynes, 1978). The assessment of psychophysiological responses has become
increasingly important to behavioral clinicians as a result of the (perhaps

premature) popularity of biofeedback training (Bradley & Prokop, 1982) and

of the application of behavioral intervention techniques to a variety of
physiological responses that can be assessed only imprecisely with self-report
measures.
Because of the expense of psychophysiological assessments, their use has
been limited largely to the intermediate and lower levels of the behavioral
assessment funnel. Their objectivity and precision have made them particu-
larly useful in identifying psychophysiological and psychophysiologically me-
diated problem behaviors and their etiologies. For example, strain gauges
have been used to assess the sexual preferences of males based on their
responsiveness to erotic stimuU (e.g., see Freund & Blanchard, 1981), and
muscular reactivity (EMG) and temperature measures have been used to
distinguish muscular tension from vascular headaches
(e.g., see Blanchard,
1981). Other problems assessed with psychophysiological techniques include

insomnia, ulcers, hypertension, pain, asthma, inadequate circulation (Ray-
naud's disease), a variety of sexual dysfunctions (e.g., Haynes, 1978; Haynes
& Wilson, 1979) and a variety of anxiety disorders (Mavissakalian & Barlow,
1981c; Taylor & Agras, 1981; Vermilyea, Boice, & Barlow, in press).
Perhaps even more common is the role performed by psychophysiological
assessments in monitoring the effects of interventions intended to modify
physiological responding. For example, heart rate and blood pressure often
have been included in the evaluation of tension reduction techniques like
relaxation training (e.g., see Nietzel & Bernstein, 1981), and brain wave
patterns (EEG) have been considered the criterion for assessing experimental
interventions to improve the sleep of insomniacs (e.g., Coates & Thoresen,
1981).
The most common physiological responses recorded by behavioral investi-
gators include muscular activity (EMG), and ectodermal respond-
heart rate,
ing such as GSR (Haynes & Wilson, However, other responses such as
1979).
pupil size, temperature, respiration rate, blood pressure and flow, and EEG
also are recorded by behavioral investigators (e.g., Haynes, 1978). EMG
recording is used to assess muscle tension, in large part because of the widely
held belief that muscle tension mediates anxiety and that muscular relaxation
training decreases levels of autonomic arousal. Recordings of muscle tension
are particularly common of tension headaches and of fears
in the assessment
and anxiety (see, for example, Blanchard, 1981; Nietzel & Bernstein, 1981).
The popularity of recording heart from the ease with which this
rate stems
response can be measured and analyzed, and from the apparent relationship
of heart rate to stress and anxiety. Despite the utility of this recording to
behavioral assessors (see Haynes & Wilson, 1979), caution is required because
heart rate is also related to the individual's "... evaluation of the situation,
his prior experience, and his previously established reaction pattern" (Nay,
1979, p. 262).
The final common physiological measure is of ectodermal activity (EDR)
usually skin conductance or its reciprocal, skin resistance. EDRs have been
viewed as a measure of activation or autonomic arousal; thus, they often are
used to monitor changes in response to fear stimuli as a result of behavioral
interventions (e.g.. Barlow, Leitenberg, Agras, & Wincze, 1969). However,
the use of ectodermal responding as a measure of arousal also must be done
cautiously,as scores vary depending on the EDR response component
measured (conductance, fluctuations, latency, and wave form), the time-
sampling parameters utilized, and the specific measurement site and proce-
dures used (e.g., Edelberg, 1972; Venables & Christie, 1973).
Sophisticated uses of physiological measures have been made primarily by
laboratory investigators rather than practicing clinicians, due to the expense
of the equipment, the inconvenience associated with its use, and the need for
extensive knowledge of physiology and electronics (Nietzel & Bernstein,

1981).' Equipment for measuring psychophysiological responses includes (1)
a sensing device, such as electrodes or some form of transducer for detecting
relevant input, (2) a central processor that may include amplifiers for
strengthening the incoming signal and filters for removing "noise;" and (3) an
output for displaying the electronic signals, such as a pen-tracing or a
digitized printout. Because malfunctioning of these components may result in
missing data (a particularly serious problem in individual subject investiga-
tions), special precautions should be followed in conducting physiological
assessments. For example, laboratory assistants should be thoroughly famil-
iar with the equipment, including its maintenance and calibration, and would
be well advised to practice with nonclinical subjects before actually moni-
toring physiological responding during experimental interventions (Hersen &
Barlow, 1976).
In conducting any physiological measurement, investigators should be
aware of the range of variables that may invalidate their records (e.g., Haynes
& Wilson, 1979; Ray & Raczynski, 1981). Aspects of the physical environ-
ment, including temperature, lighting, humidity, ambient noise, and un-
shielded electrical sources, may affect the client's or subject's responding.
Control of these variables is and subjects should be habituated or
necessary,
adapted to the laboratory setting before recording occurs. Similarly, record-
ing techniques, such as the preparation of the recording site, nature of the
conductive medium, and type, location, and attachment of electrodes or
transducers also can affect the resulting physiological record. Investigators
should consult standard references in this area (e.g., Greenfield & Sternbach,
1972; Stern, Ray, & Davis, 1980; Venables & Martin, 1967) in order to avoid
problems due to unstandardized recording procedures. Procedural variables
also can interact with measurement procedures to determine the nature of
clients' responses. Thus aspects of the procedure such as the presence and
characteristics of the examiner should be held constant throughout an investi-
gation.
Not surprisingly, the characteristics of the response assessed will determine
the nature of the resulting record. For example, some responses display
substantial habituation or adaptation effects; that is, the same stimulus
evokes lowered levels of responding following repeated stimulation, both

within and across sessions (cf. Barlow, Leitenburg, & Agras, 1969; Montague
& Coles, 1966). Responsivity to stimulation also will vary inversely with the
prestimulus level of that response. According to this "law of initial values," a
change in heart rate from 120 to 125 is different from, and probably greater
than, a change from 70 to 75. Thus some form of data transformation may
be necessary to equate response changes at various ranges of the response
dimension (e.g., Ray & Raczynski, 1981). Individuals also may show response
specificity, or a particular pattern of responding across related stimuli (e.g.,
Lacey, 1959). Because individuals vary in the response system that is most
reactive, investigators should assess their clients' reactivity before selecting a
measure that will be sensitive to the changes resulting from treatment. Some
physiological systems also may be responsive to circadian rhythms, and to
diurnal as well as layer cyclic effects (Haynes & Wilson, 1979); again,
familarity with standard technique references is critical to the judicious
selection of measurement procedures.
NOTES
1. The by-products, or traces (e.g., Webb, Campbell, Schwartz, Sechrest, & Grove,
1981), of behaviors such as pounds gained and cigarettes smoked also are consid-
ered grist for the assessment mill.
2. The inconsistency in target behavior selection is due in part to variations in

individual assessors' notions of what is socially important (Baer et al., 1968), their
personal values regarding the relative desirability of alternative behaviors, their

conceptions of deviancy, and their familiarity with the immediate and long-term
consequences of various forms of problem behavior. The operation of these factors
can be seen in the recent controversies centering on modifying feminine sex-role
behaviors among boys and annoying, but only mildly disruptive, classroom behav-
iors (e.g., Winett & Winkler, 1972; Winkler, 1977).
3. Not infrequently, additional behaviors will be monitored during one or more of the
aforementioned phases. For example, measurements may be regularly or periodi-
cally obtained on the independent, or treatment, variable to ensure that it is
manipulated in the intended manner. L. Peterson, Homer, and Wonder lich (1982)
argued that the infrequent use of independent variable checks seriously threatens
the reliability and validity of applied behavior studies. Along with J. M. Johnston
and Pennypacker (1980), they suggested a variety of methods of assessing the
integrity of independent variable manipulations. Similar recommendations are
given in related treatment literatures (e.g., Hartmann, Roper, & Gelfand, 1977;
Paul & Lentz, 1977).
At other times the investigator may choose to measure environmental events
such as the opportunities to perform the target response (Hawkins, 1982). For
example, when the target is "instruction following," assessing the client's perfor-
mance may require measurement of the occurrence of each instruction or request.
Without such an assessment, it may be impossible to distinguish changes in

compliance by the client from changes in requesting by the client's environment.
More complicated sets of environmental events also may be monitored regularly
when patterns of responding rather than single events are targeted, as illustrated in
the work by Patterson (1982) and by Gottman (1979).
Other client behaviors also may be monitored, including behaviors that might be
expected to reflect collateral effects of treatment either beneficial generalized
effects or undesirable side effects (Drabman, Hammer, & Rosenbaum, 1979;
Kazdin, 1982c; Stokes & Baer, 1977).
4. A very important, but often overlooked, practical advantage of defining target

behaviors consistently with the definitions employed in earlier studies is that the
observational systems used in these studies may be readily adapted to current
needs. See Haynes (1978, pp. 119-120) and Haynes and Wilson (1979, pp. 49-52)
for a sample listing of observational systems; Simon and Boyer (1974) for an
anthology; and Barlow (1981), Ciminero, Calhoun, and Adams (1977), Hersen and
Bellack (1981); and Mash and Terdal (1981) for surveys of topic-area reviews.
5. When observers perform consistently, yet inaccurately, the phenomenon is labeled

consensual observer drift (Johnson & Bolstad, 1973).
6. Reliability sometimes refers to consistency between standard scores from observers

(or settings or occasions), whereas agreement refers to consistency between their
raw scores (Tinsley & Weiss, 1975). A related term, observer accuracy, refers to
comparisons between an observer and an established criterion. Various investiga-
tors have argued that observer accuracy assessments should be preferred to interob-
server reliability or agreement assessments (e.g.. Cone, 1982). Possible accuracy
criteria include audio- or video-recorded behaviors orchestrated by a predeter-
mined script, mechanically generated responses, and mechanical measurements of
behavior (Boy kin & Nelson, 1981). However, the development of criterion ratings is
infeasible in many situations. Even when it is feasible, agreement with criterion

ratings can provide unrepresentative estimates of accuracy if observers can dis-
criminate between accuracy assessments and more typical observations. In such a
case, users of observational systems are left with interobserver reliability as an
indirect measure of accuracy.
7. Self-report measures have proliferated at such a rapid rate that at least one well-
known behavioral assessor suggested that journal editors limit these devices by not
considering for publication those studies employing new instruments that are not
demonstrably superior to existing ones (see comments by blue-ribbon panelists in
Hartmann, 1983).
8. Criteria for selecting or constructing measures of consumer satisfaction with treat-

ment, an increasingly popular complement to objective assessment of treatment
outcome, were described in a Behavior Therapy miniseries (Forehand, 1983).
9. Though physiological measurement typically occurs in an environmentally con-

trolled context (a laboratory), advances in telemetry have permitted in situ re-
cordings of various physiological responses (Rugh & Schwitzgebel, 1977; Vermill-
yea et al., in press).
CHAPTER 5
Basic A-B-A Withdrawal Designs
5.1. INTRODUCTION
In this chapter we will examine the prototype of experimental single-case
research the A-B-A design and its many variants. The primary objective
is to inform and familiarize the reader as to the advantages and limitations of
each design strategy while illustrating from the clinical, child, and behavior
modification literatures. The development of the A-B-A design will be traced,
beginning with its roots in the clinical case study and in the application of
"quasi-experimental designs" (Campbell & Stanley, 1966). Procedural issues
discussed at length in chapter 3 will also be evaluated here for each of the
specific design options as they apply. Both "ideal" and "problematic" exam-
ples, selected from the applied research area, will be used for illustrative
purposes.
Since the publication of the first edition of this book (Hersen & Barlow,
1976) the literature has become replete with examples of A-B-A designs.
However, there has been very little change with respect to basic procedural
Therefore, we have retained most of the original design illustrations
issues.
but have added some more recent examples from the applied behavioral
literature.
Limitations of the case study approach
For many years, descriptions of uncontrolled case histories have pre-

dominated in the psychoanalytic, psychotherapeutic, and psychiatric litera-
tures (see chapter 1). Despite the development of applied behavioral
methodology (presumably based on sound theoretical underpinnings) in the
late 1950s and early to mid-1960s, the case study approach was still the
primary method for demonstrating the efficacy of innovative treatment tech-
140
Basic A-B-A Withdrawal Designs 141
niques (cf. Ashem, 1963; Barlow, 1980; Barlow et al., 1983; Lazarus, 1963;
UUmann & Krasner, 1965; Wolpe, 1958, 1976).
Although there can be no doubt that the case history method yields
interesting (albeit uncontrolled) data, that it is a rich source for clinical
speculation, and that ingenious technical developments derive from its appli-
cation, the multitude of uncontrolled factors present in each study do not
permit sound cause-and-effect conclusions. Even when the case study method
is applied at its best (e.g., Lazarus, 1973), the absence of experimental control
and the lack of precise measures for target behaviors under evaluation remain
mitigating factors. Of course, proponents of the case study method (e.g.,
Lazarus & Davison, 1971) are well aware of its inherent limitations as an
evaluative tool, but they show how it can be used to advantage to generate
hypotheses that later may be subjected to more rigorous experimental
scrutiny. Among their advantages, the case study method can be used to (1)
foster clinical innovation, (2) cast doubt on theoretic assumptions, (3) permit
study of rare phenomena (e.g., Gilles de la Tourette's Syndrome), (4) develop
new technical skills, (5) buttress theoretical views, (6) result in refinement of
techniques, and (7) provide clinical data to be used as a departure point for
subsequent controlled investigations.
With respect to the last point, Lazarus and Davison (1971) referred to the
use of "objectified single case studies." Included are the A-B-A experimental
designs that allow for an analysis of the controlling effects of variables, thus
permitting scientifically valid conclusions. However, in the more typical case
study approach, a subjective description of treatment interventions and re-
sulting behavioral changes is made by the therapist. Most frequently, several
techniques are administered simultaneously, precluding an analysis of the
relative merits of each procedure. Moreover, evidence for improvement is
usually based on the therapist's "global" clinical impressions. Not only is
there the strong possibility of bias in these evaluations, but controls for the
treatment's placebo value are unavailable. Finally, the effects of time (ma-
turational factors) are confounded with application of the treatment(s), and
the specific contribution of each of the factors is obviously not distinguished.
More Kazdin (1981) has pointed out how "... the scientific yield
recently,
from case reports might be improved in clinical practice where methodologi-
cal alternatives are unavailable" (p. 183). In ascending order of rigor, three
types are described: (1) cases with preassessment and postassessment, (2)
cases with repeated assessment and marked changes, and (3) multiple cases
with continuous assessment and stability information (e.g., no change in a
patient's condition over extended periods of time despite prior therapeutic
efforts). However, notwithstanding improvements inherent in the aforemen-
tioned case approaches, threats to internal validity are still present to one
degree or another.
A very modest improvement over the uncontrolled case study method
elsewhere (Browning & Stover, 1971) has been labeled the "B Design." In this
"design," baseline measurement is omitted, but the investigator monitors one
of a number of target measures throughout the course of treatment. One
might also categorize this procedure as the simplest of the time series analyses
(see G. V. Glass, Willson, & Gottman, 1973). Although this strategy ob-
viously yields a more objective appraisal of the patient's progress, the con-
founds that typify the case study method apply equally here. In that sense the
B Design is essentially an uncontrolled case study with objective measures
taken repeatedly. This, of course, is the same as Kazdin's (1981) description of
cases with repeated assessment and marked changes.
5.2. A-B DESIGN
The A-B design, although the simplest of the experimental strategies,

corrects for some of the deficiencies of the case study method and those of the
B Design. In this design the target behavior is and repeated
clearly specified,
measurement is taken throughout the A and B phases of experimentation. As
in all single-case experimental research, the A phase involves a series of
baseline observations of the natural frequency of the target behavior(s) under
study. In the B phase the treatment variable is introduced, and changes in the
dependent measure are noted. Thus, with some major reservations, changes
in the dependent variable are attributed to the effects of treatment (Barlow &
Hersen, 1973; Campbell, 1969; Campbell & Stanley, 1966; Cook & Campbell,
1979; Hersen, 1982; Kazdin, 1982b; Kratochwill, 1978b).
Let us now examine some of the important reservations. In their evaluation
of the A-B strategy. Wolf and Risley (1971) argued that "The analysis^pro-
vided no information about^ât the natural course of the behavior would
"have been had we not intervened with our treatment condition" (pp.
314-315). That is to say, it is very possible that changes in the B phase might
have occurred regardless of the introduction of treatment or that changes in B
might have resulted as a function of correlation with some fortuitous (but
uncontrolled) event. When considered in this light, the A-B strategy does not
permit a full experimental analysis of the controlling effects of the treatment
inasmuch as its correlative properties are quite apparent. Indeed, Campbell
and Stanley (1966) referred to this strategy as a "quasi-experimental design."
Risley and Wolf (1972) presented an interesting discussion of the limita-
tions of theA-B design with respect to predicting, or "forecasting," the B
phase on the basis of data obtained in A. T\vo hypothetical examples of the
A-B design were depicted, with both showing a mean increase in the amount
of behavior in B over A. However, in the first example, a steady and stable
trend in baseline is followed by an abrupt increase in B, which is then
maintained. In the second case, the upward trend in A is continued in B.

Therefore, despite the equivalence of means and variances in the two cases,
the importance of the trend in evaluating the data is underscored. Some
tentative conclusions can be reached on the basis of the first example, but in
the second example the continued linear trend in A permits no conclusions as
to the controlling effects of the B treatment variable.
In further analyzing the difficulties inherent in the A-B strategy, Risley and
Wolf (1972) contended that:
The weakness in this design is that the data in the experimental condition is
compared with a forecast from the prior baseline data. The accuracy of an
assessment of the role of the experimental procedure in producing the change
rests upon the accuracy of that forecast. A strong statement of causality there-
fore requires that the forecast be supported. This support is accomplished by
elaborating the A-B design, (p. 5)
Such elaboration is found in the A-B-A design discussed and illustrated in

section 5.3 of this chapter.
Despite these aforementioned limitations, it is shown how in some settings
(where control-group analysis or repeated introduction and withdrawals of
treatment variables are not feasible) the A-B design can be of some utility
(Campbell & Stanley, 1966; Cook & Campbell,

For example, the use of
1979.
the A-B strategy in the private-practice setting has previously been recom-
mended in section 3.2 of chapter 3 (see also Barlow et al., 1983).
Campbell (1969) presented a comprehensive analysis of the use of the A-B
strategy in field experiments where more traditional forms of experimentation
are not at all possible (e.g., the effects of modifying traffic laws on the
documented frequency of accidents). However one uses the quasi-experimen-
tal design, Campbell cautioned the investigator as to the numerous threats to
internal validity (history, maturation, instabihty, testing, instrumentation,

regression artifacts, selection, experimental mortality, and selection-matura-
tion interaction) and external validity (interaction effects of testing, interac-
tion of selection and experimental treatment, reactive effects of experimental
arrangements, multiple-treatment interference, irrelevant responsiveness of
measures, and irrelevant replicability of treatments) that may be encountered.
The interested reader is referred to Campbell's (1969) excellent article for a
full discussion of the issues involved in large-scale retrospective or prospective
field studies.
In summary, it should be apparent that the use of a quasi-experimental

design such as the A-B weak conclusions. This design
strategy results in rather
is and is best
subject to the influence of a host of confounding variables
applied as a last-resort measure when circumstances do not allow for more
extensive experimentation. Examples of such cases will now be illustrated.
A-B with single target measure an djollow-uy
Epstein and Hersen (1974) used an A-B design with a follow-up procedure
to assess the effects of reinforcement on frequency of gagging in a 26-year-old
psychiatric inpatient. The patient's symptomatology had persisted for ap-
proximately 2 years despite repeated attempts at medical intervention. During
baseline (A phase), the patient was instructed to record time and frequency of
each gagging episode on an index card, collected by the experimenter the
following morning at ward rounds. Treatment (B phase) consisted of present-
ing the patient with $2.00 in canteen books (exchangeable at the hospital store
for goods) for a decrease (N -
1) from the previous daily frequency. In
addition, zero rates of gagging were similarly reinforced. In order to facilitate

maintenance of gains after treatment, no instructions were given as to how
the patient might control his gagging. Thus emphasis was placed on self-
management of the disorder. At the conclusion of his hospital stay, the patient
was requested to continue recording data at home for a period of 12 weeks. In
this case, treatment conditions were not withdrawn during the patient's
hospitalization because of clinical considerations.
Results of this study are plotted in Figure 5-1. Baseline frequency of
gagging fluctuated between 8 and 17 episodes per. day but stabilized to some
extent in the last 4 days. Institution of reinforcement procedures in the B
phase resulted in a decline to zero within 6 days. However, on Day 15,
frequency of gagging rose again to seven daily episodes. At this point, the
criterion for obtaining reinforcement was reset to that originally planned for
ow-up
2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12
DAYS WEEKS
FIGURE 5-1. Frequency of gagging during baseline, treatment, and follow-up. (Figure 1, p. 103,
from: Epstein, L. H., & Hersen, M. (1974). Behavioral control of hysterical gagging. Journal of
Clinical Psychology, 30, 102-104. Copyright 1974 by American Psychological Association.
Day 13. Renewed improvement was then noted between Days 15-18, and
treatment was continued through Day 24. Thus the B phase was twice as long
as baseline, butit was extended for very obvious clinical considerations.
The 12-week follow-up period reveals a zero level of gagging, with the
exception of Week 9, when three gagging episodes were recorded. Follow-up
data were corroborated by the patient's wife, thus precluding the possibility
that treatment only affected the patient's verbal report rather than diminution
of actual symptomatology.
Although treatment appeared to be the effective ingredient of change in
this study, particularly in light of the longevity of the patient's disorder, it is
conceivable that some unidentified variable coincided with the application of

reinforcement procedures and actually accounted for observed changes.
However, the A-B design does not permit a definitive answer to this question.
It might also be noted that the specific use of this design (baseline, treatment,
and follow-up) could readily have been carried out in an outpatient facility
(clinic or private-practice setting) with a minimum of difficulty and with no
deleterious effects to the patient.
Lawson (1983) also used an A-B design with a single target behavior
(alcohol consumption) and obtained a follow-up assessment. His case in-
volved a divorced 35-year-old male with a history of problem drinking
beginning at age 16. He periodically would experience blackouts as a function
of his drinking. But despite the chronicity of his problem, with the exception
of a few AA meetings, the subject had not obtained any form of treatment
for his alcoholism. Baseline data (based on the subject's self-report) indicated
that he consumed an average of 65 drinks per week (see Figure 5-2). This was
confirmed by his girlfriend.
Treatment (B phase) began in the third week, and, on the basis of the
behavioral analyses performed, three goals were identified: (1) to decrease
alcohol consumption, (2) to improve social relationships, and (3) to diminish
frequency of anxiety and depression episodes. Thus the comprehensive
therapy program involved goal setting with regard to number of drinks
consumed, rate-reduction strategies, stimulus-control strategies, development
of new social relationships and recreational activities, assertion training, and
self-management of depression.
Examination of data in Figure 5-2 indicates that there were substantial
improvements in rate of drinking during the course of therapy (to about 10
drinks per week) that appeared to be maintained at the 3-month follow-up
(also confirmed by the girlfriend). Indeed, an informal communication re-
ceived by the therapist 1 Vi years subsequent to treatment further confirmed
that the subject still was drinking in a socially acceptable manner.
Treatment did appear to be responsible for change in Lawson's (1983)
alcoholic, particularly given the 19-year history of excessive drinking. This,
then, from a design standpoint, fits in nicly with Kazdin's notion of repeated
3 MONTH
BASELINE TREATMENT FOLLOW
70 UP
g50
00
Z 40
a
O30
\A-
4 6
WEEKS
FIGURE 5-2. Weekly self-monitored alcohol consumption during baseline, treatment, and at 3-
month follow-up. (Figure 6-1, p. 165, from: Lawson, D. M. Alcoholism. In M. Hersen (Ed.).
(1983). Outpatient behavior therapy: A clinical guide. New York: Grune & Stratton. Copyright
1983 by M. Hersen. Reproduced by permission.)
assessment with marked changes and stability information improving the

quality of case study. But, in spite of this, the A-B design does not allow for a
clear demonstration of the controlling effects of the treatment. For that we
require an A-B-A or A-B-A-B strategy.
A-B with multiple-target measures
In our next example we will examine the use of an A-B design in which a
numherôf target behaviors were monitored simultaneously^(Eisler & Hersen,
1973). The effectroffolcen economy on points earned, behavioral ratings of
depression (WiUiams et al., 1972), and self-ratings of depression (Beck
Depressive Inventory A. T. Beck, Ward, Mendelsohn, Mock, & Erbaugh,
1961) were assessed in a 61 -year-old reactively depressed male patient. In this
study the treatment variable was not withdrawn due to time limitations.
During baseline (A), the patient was able to earn points for a variety of
specified target behaviors (designated under general rubrics of work, personal
hygiene, and responsibility), but these earned points were exchangeable for
ward privileges and material goods in the hospital canteen. During each
phase, the patient filled out a Beck Depressive Inventory (three alternate
forms were used to prevent possible response bias) at daily morning "Bank-
ing Hours," at which time points previously earned on the token economy
were tabulated. In addition, behavioral ratings (talking, smiling, motor activ-
ity) of depression (high ratings indicate low depression) were obtained sur-
reptitiously on the average of one per hour between the hours of 8:00 A.M.
and 10:00 P.M. during non-work-related activities.
The results of this study appear in Figure 5-3. Inspection of these data
indicates that number of points earned in baseline increased slightly but then
stabilized. Baseline ratings of depression show stability, with evidence of
greater daytime activity. Beck scores ranged from 19-28. Institution of token
economy on Day 5 resulted in a marked linear increase in points earned, a
substantial increase in day and evening behavioral ratings of depression, and
a linear descrease in self-reported Beck Inventory scores.
Thus it appears that token economy effected improvement in this patient's
depression as based on both objective and subjective indexes. However, as
was previously pointed out, this design does not permit a direct analysis of
the controlling effects of the therapeutic variable introduced (token
economy), as does our example of an A-B-A design seen in Figure 5-7
(Hersen, Eisler, Alford, & Agras, 1973). Nonetheless, the use of an A-B
design in this case proved to be useful for two reasons. First ^ from a clinical
standpoint, it was possible to obtain some objective estimate of the treat-
ment's success during the patient's abbreviated hospital stay. Second, the
results of this study prompted the further investigation of the effects of token
economic procedures in three additional reactively depressed subjects (Her-
sen, Eisler, Alford, & Agras, 1973). In that investigation more sophisticated
experimental strategies confirmed the controlling effects of token economy in
neurotic depression.
A-B with multiple-target measures and follow-up
A more recentand more complicated example of an A-B design with

multiple-target measures and follow-up was described by St. Lawrence,
Bradlyn, and Kelly (1983). The subject was a 35-year-old male with a 20-year
history of homosexual functioning, but whose interpersonal adjustment was
unsatisfactory. Treatment, therefore, was directed to enhancing several com-
ponents of social skill. Five components requiring modification were iden-
tified during two baseline assessments: (1) percentage of eye contact, (2)
smiles, (3) extraneous movements, (4) appropriate verbal content, and (5)
overall social skill. Assessment involved the patient and a male confederate
role-playing 16 scenes (8 commendatory; 8 refusal) that were videotaped.
Social skills training was conducted twice a week for nine weeks and
consisted of modeling, instructions, behavior rehearsal, cognitive modifica-
tion,and in vivo practice. Training was carried out with half of the commen-
datory and refusal scenes; the other half served as a measure of
generalization. In addition, follow-up sessions were conducted at 1 and 6
months after conclusion of treatment.
The results of this A-B analysis appear in Figure 5-4, with the left half
LJJ
2
QC
& 20-
u. 10
O
QC
UJ
00
5 BASELINE TOKEN ECONOMY
-8AM 4PM
- 4PM 10PM
0-
BASELINE TOKEN ECONOMY
t I I
30
\ i
20
V^ ^
i
1
\
i
10 -
i \.
!
1
BASELINE 1
TOKEN ECONOMY
1 !
4 5
DAYS
FIGURE 5-3. Number of points earned, mean behavioral ratings, and Beck Depression Scale
scores during baseline and token economy in a reactively depressed patient. (Figure 1, from:
Eisler, R. M., Hersen, M. (1973). The A-B design: Effects of token economy on behavioral and
subjective measures in neurotic depression. Paper presented at the meeting of the American
Psychological Association, Montreal, August 29.)
F : T r
COMMENDATORY SCENES REFUSAL SCENES
BASCIINC TMINlie WtlWHUP ,^ BASCIINC TUININC OiiSmjf

A A A A A_^ 100,
<!
is
5ju I I I 1-n T r-Tp
^1 "V
^
li
I
î 11^
ii
=s^ ^ I
VI I I
t^
A X
sis
I I I I 1 ir-n
TMINCD
aGENEMIIZATION
^<4 !t=
111
8j
1*
^ l^
I I t I
I I I I I
FIGURE 5-4. Mean frequency of targeted behaviors in refused and commendatory role-play
situations. (Figure 1, p. 50, from: St. Lawrence, J. S., Bradlyn, A. S., & Kelly, J. A. (1983).
Interpersonal adjustment of a homosexual adult: Enhancement via social skills training. Behavior
Modification, 7, 41-55. Copyright 1983 by Sage Publications. Reproduced by permission.)

SCED
portraying commendatory scenes and the right half refusal scenes. In general,
improvements during training suggest that the treatment was effective for
both categories (commendatory and refusal) and that there was transfer of
gains from trained to generalization scenes. Moreover, gains appeared to
remain in follow-up, with the exception of smiles (commendatory). However,
a closer examination does reveal a number of problems with these data. First,
for the commendatory scenes there are only one- or two-point baselines.
Therefore, complete establishment of baseline trends was not possible. Also,
for two of the behaviors (smiles, appropriate verbal content), improvements
in training similarly appear to be the continuation of baseline trends. Second,
this also seemed to be the case with regard to refusal scenes for the following
components: eye contact, extraneous movements, appropriate verbal con-
tent, and overall social skill. Thus, although the subject was obviously
clinically improved, these data do not clearly reflect experimental confirma-
tion of such improvement, given the limited confidence one can ever have
with the A-B strategy.
A-B with follow-up and boos

TiTournext an A-B design, clinical considerations necessi-
illustration of
tated a short baseline period and also contraindicated the withdrawal of
treatment procedures (Harbert, Barlow, Hersen, & Austin, 1974). However,
during the course of extended follow-up assessment, the patient's condition
deteriorated and required the reinstatfimenLoftreatmerU in booster^s^
Renewed improvement immediately followed, thus lending additional sup-
port for the treatment's efficacy. When examined from a design standpoint,
the conditions of the more complete A-B-A-B strategy are approximated in
this experimental case study. ^ ^- ^ ^ -
More specifically, Harbert et al. (1974) examined the effects of covert

sensitization therapy on and physiological
self-report (card sort technique)
(mean penile circumference changes) indices in a 52-year-old male inpatient
who complained of a long history of incestuous episodes with his adolescent
daughter. The card sort technique consisted of 10 scenes (typed on cards)
depicting the patient and his daughter. Five of these scenes were concerned
with normal father-daughter relations; the remaining five involved descrip-
tions of incestuous activity between father and daughter. The patient was
asked to rate the 10 scenes, presented in random sequence, on a 0-4 basis,
with representing no desire and 4 representing much desire. Thus measures
of both deviant and nondeviant aspects of the relationship were obtained
throughout all phases of study. In addition, penile circumference changes
scored as a percentage of full erection were obtained in response to
audiotaped descriptions of incestuous activity and in reaction to slides of the
daughter. Three days of self-report data and 4 days of physiological measure-
ments were taken during baseline (A phase).
Covert sensitization treatment (B phase) consisted of approximately 3

weeks of daily sessions in which descriptions of incestuous activity were
paired with the nauseous scene as used by Barlow, Leitenberg, and Agras
(1969). However, as nausea proved to be a weak aversive stimulus for this
patient, a "guilt" scene in which the patient is discovered engaging in sexual
activity with the daughter by his current wife and a respected priest was
substituted during the second week of treatment. The flexibility of the single-
case approach is exempHfied here inasmuch as a "therapeutic shift of gears"
follows from a close monitoring of the data.

Follow-up assessment sessions were conducted after termination of the
patient's hospitalization at 2-week, 1-, 2-, 3-, and 6-month intervals. After,
each fpllpw-up session, brief booster covert sensitization was administered.
The results of this study appear in Figure 5-5 and 5-6. Inspection of Figure
5-5 indicates that mean penile circumference changes to audiotapes in base-
line ranged from 18<ô 35% (mean = 22-8%). Penile circumference
to
changes to slides ranged from \S% to 15% (mean = 43-5%). Examination of
Figure 5-6 shows that nondeviant scores remained at a maximum of 20 for all
three baseline probes; deviant scores achieved a level of 17 throughout.
Introduction of standard covert sensitization, followed by use of the guilt
imagery resulted in decreased penile responding to audiotapes and slides (see
Figure 5-5) and a substantial decrease in the patient's self-reports of deviant
COVERT
BASELINE SENSITIZATION FOLLOW-UP
80
Slides
2 60
o
S 20
z
10
12 3 4 5 6 7 8 9 101112 13
|
CNJ ' C\J CO CO
PROBE DAYS
FIGURE 5-5.Mean penile circumference change to audiotapes and slides during baseline, covert
and follow-up. (Figure 1, p. 83, from: Harbert, T. L., Barlow, D. H., Hersen, M.,
sensitization,
& Austin, J. B. (1974). Measurement and modification of incestuous behavior: A case study.
Psychological Reports, 34, 79-86. Copyright 1974 by Psychological Reports. Reproduced by
permission.)
interests in his daughter (see Figure 5-6). Nondeviant interests, however,

remained at a high level.
Follow-up data in Figure 5-5 reveal that penile circumference changes
remained at zero during the first three probes but increased slightly at the 3-
month assessment. Similarly, Figure 5-6 data show a considerable increase in
deviant interests at the 3-month follow-up. This coincides with the patient's
reports of marital disharmony. In addition, nondeviant interests diminished
during follow-up (at that point the patient was angry at his daughter for
rejecting his positive efforts at being a father).
As there appeared to be some deterioration at the 3-month follow-up, an
additional course of outpatient covert sensitization therapy was carried out in
three weekly sessions. The final assessment period at 6 months appears to
reflect the effects of additional treatment in that (1) penile responding was
negligible, and (2) deviant interests had returned to a zero level.
5.3. A-B-A DESIGN
The A-B-A design is the simplest of the experimental analysis strategies in

which the treatment v a ri able is introduced and then withdrawn. For this
reason, this strategy as well as thosetHatTollow, are most often^êrred to as
^wUkdra^valdesigns^ Whereas the A-B design permits only tentative conclu-
sions as to a treatment's influence, the A-B-A design allows for an analysis of
the controlling effects of its introduction and subsequent removal. If after
. Deviant
oNon- Deviant
BASELiNE 1 COVERT SENSITIZATION FOLLOW-UP

00
UJ 20 ^--' I ^ r^
15 -i\/ 1 ,j
\^^N^--
o
a
10
\
> hy
5
/ \
*"*'--. . ^-, ^, ^
1 2 3 4 5 6 7 8 9 1011121314 g HI i
CM 1 CM CO CO
PROBE DAYS
FIGURE 5-6. Card sort scores on probe days during baseline, covert sensitization, and follow-
up. (Figure 2, p. 84, from: Harbert, T. L., Barlow, D. H., Hersen, M., & Austin, J. B. (1974).
Measurement and modification of incestuous behavior: A case study. Psychological Reports, 34,
79-86. Copyright 1974 by Psychological Reports. Reproduced by permission.)
measurement (A) the application of a treatment (B) leads to improve-

baseline
ment and coversely results in deterioration after it is withdrawn (A), one can
conclude with a high degree of certainty that the treatment variable is the
agent responsible for observed changes in the target behavior. Unless the
natural history of the behavior under study were to follow identical fluctua-
tions in trends, most improbable that observed changes would be due to
it is
any influence (e.g., some correlated or uncontrolled variable) other than the
treatment variable that is systematically changed. Also, replication of the A-
B-A design in different subjects strengthens conclusions as to power and
controlling forces of the treatment (see chapter 10).
Although the A-B-A strategy is acceptable from an experimental stand-
point, it has one major undesirable feature when considered from the clinical
context. Unfortunately for t he patient or subi ecUJhis paradigm ends on the

A or baseline phas eof study, therefore denying him orTTeTthe full Beiiefitrof
experimental treatmentT Along these lines, Bartow and Hersen~(1973) have^
argued^hirr^"^"^^^"^-
On an ethical and moral basis it certainly behooves the experimenter-clinician to

continue some form of treatment to its ultimate conclusion subsequent to
completion of the research aspects of the case. A further design, known as the A-
B-A-B design, meets this criticism as study ends on the B or treatment phase, (p.
321).
However, despite this limitation, the A-B-A design is a useful research tool
when time factors (e.g., premature discharge of a patient) or clinical aspects
of a case (e.g., necessity of changing the level of medication in addition to
reintroducing a treatment variable after the second A phase) interfere with
the correct application of the more comprehensive A-B-A-B strategy.
A second problem with the A-B-A strategy concerns the issues of multiple-
treatment interference, particularly sequential confounding (Bandura, 1969;
Cook & Campbell, 1979). The problem of sequential confounding in an A-B-
A design and its variants also somewhat limits generalization to the clinic. As
Bandura (1969) and Kazdin (1973b) have noted, the effectiveness of a thera-
peutic variable in the final phase of an A-B-A design can only be interpreted
in the context of the previous phases. Change occurring in this last phase may
not be comparable to changes that would have occurred if the treatment had
been introduced initially. For instance, in an A-B-BC-B design, when A is
baseline and B and C are two therapeutic variables, the effects of the BC
phase may be more or less powerful than if they had been introduced initially.
This point has been demonstrated in studies by O'Leary and his associates
(O'Leary & Becker, 1967; O'Leary, Becker, Evans, & Saudargas, 1969), who
noted that the simultaneous introduction of two variables produced greater
change than the sequential introduction of the same two variables.
Similarly, the second introduction of variable A in a withdrawal A-B-A

design may affect behavior differently than the first introduction. (Generally
our experience is that behavior improves more rapidly with a second intro-
duction of the therapeutic variable.) In any case, the reintroduction of
therapeutic phases is a feature of A-B-A designs that differs from the typical
applied clinical situation, when the variable is introduced only once. Thus,
appropriate cautions must be exercised in generalizing results from phases
occurring late in an experiment to the clinical situation.
In dealing with this problem, the clinical researcher should keep in mind
that the purpose of subsequent phases in an A-B-A design is to confirm the
effects of the independent variable (internal validity) rather than to generalize
to the clinical situation. The results that are most generalizable, of course, are
data from the first introduction of the treatment. When two or more variables
are introduced in sequence, the purpose again is to test the separate effects of
each variable. Subsequently, order effects and effects of combining the
was the case with the
variable can be tested in systematic replication series, as
O'leary, Becker, Evans, and Saudergas (1969) study.
T\vo examples of the A-B-A design, one selected from the clinical literature
and one from the child development area, will be used for illustration. At-
tention will be focused on some of the procedural issues outlined in chapter 3.
A-B-A from clinical literature
In pursuing their study of the effects of token economy on neurotic

depression, Hersen and his colleagues (Hersen, Eisler, Alford, & Agras, 1973)
used A-B-A The results for
strategies with three reactively depressed subjects.
one of these subjects (52-year-old, white, married farmer who became de-
pressed after the sale of his farm) appear in Figure 5-7. As in the Eisler and
Hersen (1973) study, described in detail in section 5.2 of this chapter, points
earned in baseline (A) had no exchange value, but during the token reinforce-
ment phase (B) they were exchangeable for privileges and material goods.
Unlike the Eisler and Hersen study, however, token reinforcement procedures
were withdrawn, and a return to baseline conditions (A) took place during
Days 9-12. The effects of introducing and removing token economy were

examined on two target behaviors points earned and behavioral ratings
(higher ratings indicate lowered depression).
A careful examination of baseline data reveals a slightly decreased trend in
behavioral ratings, thus indicating some very minor deterioration in the
patient's condition. As was noted in section 3.3 of chapter 3, the deteriorating
baseline is considered to be an acceptable trend. However, there appeared to
be a concomitant but slight increase in points earned during baseline. It will
be recalled that an improved trend in baseline is not the most desirable trend.
^
Peintt Irn4
havlaral Rating*
Token
It alnf orcamant
i i I 10 11 n
DAYS
FIGURE 5-7. Number of points earned and mean behavioral ratings for Subject 1. (Figure 1, p.
394, from: Hersen, M., Eisler, R. M., Alford, G. S., & Agras, W. S. (1973). Effects of token
economy on neurotic depression: An experimental analysis, Behavior Therapy, 4, 392-397.
Copyright 1973 by Association for the Advancement of Behavior Therapy. Reproduced by
permission.)
However, as the slope of the curve was not extensive, and in light of the
primary focus on behavioral ratings (depression), we proceeded with our
change in conditions on Day 5. Had there been unlimited time, baseline
conditions would have been maintained until number of points earned daily
stabilized to a greater extent.
We might note parenthetically at this point that all of the ideal conditions
(procedural rules) outlined in our discussion in chapter 3 are rarely approxi-
mated when conducting single-case experimental research. Our experience
shows that procedural variations from the ideal are required, as data simply
do not conform to theoretical expectation. Moreover, experimental finesse is
sometimes sacrificed at the expense of time and clinical considerations.
Continued examination of Figure 5-7 indicates that instigation of token
economic procedures on Day 5 resulted in a marked linear increase in both
points earned and behavioral ratings. The abrupt change in slope of the
curves, particularly in points earned, strongly suggests the influence of the
token economy variable, despite the slightly upward trend initially seen in
baseline. Removal of token economy on Day 9 led to an initially large drop in
behavioral ratings, which then stabilized at a somewhat higher level. Points

earned also declined but maintained throughout the second 4-day
stability
baseline period. The obtained decrease in target behaviors in the second
baseline phase confirms the controlling effects of token economy over
neurotic depression in this paradigm. We might also point out here that an
equal number of data points appears in each phase, thus facilitating interpre-
tation of the trends.
These results were replicated in two additional reactively depressed subjects
(Hersen, Eisler, Alford, & Agras, 1973), lending further credence to the
notion that token economy exerts a controlling influence over the behavior of
neurotically depressed individuals.
A-B-A from child literature
Walker and Buckley (1968) used an A-B-A design in their functional

analysis of the effects of an individualized educational program for a 9!/2-
year-old boy whose extreme distractibility in a classroom situation interfered
with task-oriented performance (see Figure 5.8). During baseline assessment
(A), percentage of attending behavior was recorded in 10-minute observation
sessions while the subject was engaged in working on programmed learning
materials. Following baseline measurement, a reinforcement contingency (B)
was instituted whereby the subject earned points (exchangeable for a model
of his choice) for maintaining his attention (operationally defined for him) to
the learning task. During this phase, a progressively increasing time criterion
for attending behaviors over sessions was required (30 to 6(X) seconds of
attending per point). The extinction phase (A) involved a return to original
baseUne conditions.
Examination of baseline data shows a slightly decreasing trend followed by
a slightly increasing trend, but within stable limits (mean = 33%). Institution
of reinforcement procedures led to an immediate improvement, which then
increased to its asymptote in accordance with the progressively more difficult
criterion. Removal of the reinforcement contingency in extinction resulted in
a decreased percentage of attending behaviors to approximately baseline
levels. After completion of experimental study, the subject was returned to his
classroom where a variable interval reinforcement program was used to
increase and maintain attending behaviors in that setting.
With respect to experimental design issues, we might point out that Walker
and Buckley (1968) used a short baseline period (6 data points) followed by
longer B (15 data points) and A phases (14 data points). However, in view of
the fact that an immediate and large increase in attention was obtained during
reinforcement, the possible confound of time when using disparate lengths of
phases (see section 3.6, chapter 3) does not apply here. Moreover, the shape
100
40 .
20 .
Number of Ten-Min Observation Seuiom
FIGURE 5-8. Percentage of attending behavior in successive time samples during the individual
conditioning program. (Figure 2, p. 247, from: Walker, H. M., & Buckley, N. K. (1968). The use
of positive reinforcement in conditioning attending behavior. Journal of Applied Behavior
of the curve in extinction (A) and the relatively equal lengths of the B and A
phases further dispel doubts that the reader might have as to the confound of
time.
Secondly, with respect to the decreasing-increasing baseline obtained in the
first A phase, although it might be preferable to extend measurement until
full stability is achieved (see section 3.3, chapter 3), the range of variability is
very constricted here, thus delimiting the importance of the trends.
5.4. A-B-A-B DESIGN

The A-B-A-B strategy, referred to as an equivalent time-samples design by
Campbell and Stanley (1966), controls for the deficiencies presentuTtHFA'-B-
A design. Specifically, the A-B-A-B design ends on a treatment phase (B),
which then can be extended beyond the experimental requirements of study

for cHnical reasons (e.g., Miller, 1973). In addition, this design strategy
provides for two occasions (B to A and then A to B) for demonstrating the
positive effects of the treatment variable. This, then, strengthens the conclu-
sions that can be derived as to its controlling effects over target behaviors
under observation (Barlow & Hersen, 1973).
In the succeeding subsections we will provide four examples of the use of
the A-B-A-B strategy. In the first we will present examples from the child
which illustrate the ideal in procedural considerations. In the second
literature
we examine the problems encountered in interpretation when improve-
will
ment fortuitously occurrs during the second baseline period. In the third we
will illustrate the use of the A-B-A-B design when concurrent behaviors are
monitored in addition to targeted behaviors of interest. Finally, in the fourth
we will examine the advantages and disadvantages of using the A-B-A-B
strategy without the experimenter's knowledge of results throughout the
different phases of study.
A-B-A-B from child literature
An excellent example of the A-B-A-B design strategy appears in a study

conducted by R. V. Hall et al. (1971). In this study the effects of contingent
teacher attention were examined in a 10-year-old retarded boy whose "talk-
ing-out" behaviors during special education classes proved to be disruptive,
as other children then emulated his actions. Baseline observations of talk-outs
were recorded by the teacher (reliability checks indicated 84ô to 100%
agreement) duringfive daily 15-minute sessions. During these first five ses-
sions, the teacher responded naturally to talk-outs by paying attention to
them. However, in the next five sessions, the teacher was instructed to ignore
talk-outs but to provide increased attention to the child's productive behav-

The third series of five sessions involved a return to baseline conditions,
iors.
and the last series of five sessions consisted of reinstatement of contingent

attention.
The results of this study are plotted in Figure 5-9. The presence of equal
phases in this study facilitates the analysis of results. Baseline data are stable
and range from three to five talk-outs, with three of the five points at a level
of four talk-outs per session. Institution of contingent attention resulted in a
marked decrease that achieved a zero level in Sessions 9 and 10. Removal of
contingent attention led to a linear increase of talk-outs to a high of five.
However, reinstatement of contingent attention once again brought talk-outs

under experimental control. Thus application and withdrawal of contingent
attention clearly demonstrates its controlling effects on talk-out behaviors.
CONTINOINT CONTINOINf
ATTINTION, AtfllNI^ ATTINTIOM,
<
CD V
D
Z
10 15 20
SESSIONS
FIGURE 5-9. A record of talking out behavior of an educable mentally retarded student.
Baseline I
before experimental conditions. Contingent Teacher Attention, systematic ignoring
of talking out and increased teacher attention to appropriate behavior. Baselinei reinstatment
of teacher attention to talking out behavior. (Figure 2, p. 143, from: Hall, R. V., Fox, R., Willard,
D., Goldsmith, L., Emerson, M., Owen, M., Davis, T, & Porcia, E. (1971). The teacher as
observer and experimenter in the modification of disputing and talking-out behaviors. Journal of
Applied Behavior Analysis, 4, 141-149. Copyright 1971 by Society for the Experimental Analysis
of Behavior, Inc. Reproduced by permission.)
This is twice-documented, as seen in the decreasing and increasing data trends

in the second set of A and B phases.
Let us now consider a more recent example of an A-B-A-B design taken
from the child literature. In this experimental analysis, Hendrickson, Strain,
Tremblay, and Shores (1982) documented how a normally functioning pre-
school child (the peer confederate) was taught to make specific initiations
toward three "withdrawn" preschool boys (each four years of age). This peer
confederate was a 4-year-old female, with a well-developed repertoire of
expressive language and social interaction skills. Prebaseline observation
indicated no evidence of physically aggressive behavior. She interacted pri-
marily with adults, and infrequently initiated positive behavior to other
children. She did, however, respond positively and consistently when other
children initiated play to her. This child was involved in the treatment
program as a "model" youngster (p. 327).
During baseline and intervention phases the children were brought to a
playroom for two 15-minute sessions. Three behaviors were observed and
coded during these sessions: (1) initiations of play organizers (proposes a role
or activity in a game), (2) shares (offers or gives toy to another child), and (3)
assists (provides help to another child).
Examination of baseline data in Figure 5-10 indicates that the peer confe-
. Initiations of Play Organizers (1),

^ Shares (2). and Assists (3)
A Responses to and 3s
ui 15
FIGURE 5-10. Experiment 1: Frequency of confederate initiations of play organizers, shares,

and assists and subject's positive responses to these approach behaviors. (Figure 1, p. 335, from:
Hendrickson, J. M., Strain, P. S., TVemblay, A., & Shore, R. E. (1982). Interactions of beha-
viorally handicapped children: Functional effects of peer social initiations. Behavior Modifica-
tion, 6, 323-353. Copyright 1982 by Sage Publications. Reproduced by permission.)
derate neither initiated any of the three targeted behaviors nor responded to
any initiations of the three withdrawn children. However, during the first
intervention phase, when the confederate was prompted, instructed, and

reinforced for playing, there was a marked increase in the three categories of
behavior. This was noted both in terms of initiations and responses. When
intervention was removed in the second basehne, frequency of such initiating
and responding returned to the original baseline level. Finally, in the second
intervention phase, high levels of initiating and responding were easily rein-
stated. Throughout this study, mean interobserver agreement for behaviors
targeted was 89% for all subjects.
With respect to design considerations, we have here a very clear demonstra-
tion of the efficacy of the intervention on two occasions. As was the case in
our prior example (R. V. Hall et al., 1971) baselines (especially the second)
were shorter than treatment phases. However, in light of the zero level of
baseline responding and the immediate and dramatic improvements as a

result of the intervention, the possible confound of time and length of
adjacent phases does not apply in this analysis.
A-B-A-B with unexpected improvement in baseline
In our next examplewe will illustrate the difficulties that arose in interpre-
tation when unexpected improvement took place during the latter half of the
second series of baseline (A) measurements. Epstein, Hersen, and Hemphill
(1974) used an A-B-A-B design in their assessment of the effects of feedback
on frontalis muscle activity in a patient who had suffered from chronic
headaches for a 16-year period. EMG recordings were taken for 10 minutes
following 10 minutes of adaptation during each of the six basehne (A)
sessions. EMG data were obtained while the patient relaxed in a reclining
chair in the experimental laboratory. During the six feedback (B) sessions, the
patient's favorite music (prerecorded on tape) was automatically turned on
whenever EMG activity decreased below a preset criterion level. Responses
above that turned off recordings of music. Instructions to the
level conversely
patient during this phasewere to "keep the music on." In the next six sessions
baseline (A) conditions were reinstated, while the last six sessions involved a
return to feedback (B). Throughout all phases of study, the patient was asked
to keep a record of the intensity of headache activity.
Examination of Figure 5-11 indicates that EMG activity during baseline

ranged from 28 to 50 seconds (mean = 39- 18) per minute that contained
integrated responses above the criterion microvolt level. Institution of feed-
60 Baseline Feedback Baieline i Feedback
^ 50
o
t 40
^ 30
^ 20
o
.^
^ 10
2 4 10 12 14 16 18 20 22 24
SESSIONS
FIGURE 5-11. Mean seconds per minute that contained integrated responses above criterion
microvolt level during baseline and feedback phases. (Figure 1, p. 61, from: Epstein, L. H.,
Hersen, M., & Hemphill, D. P. Music feedback as a treatment for tension headache: An
(1974).
experimental case study. Journal of Behavior Therapy and Experimental Psychiatry, 5, 59-63.
back procedures resulted in decreased activity (mean = 23- 18). Removal of

feedback in the second baseline initially resulted in increased activity in
Sessions 13-15. However, an unexplained but decreased trend was noted in
the last half of that phase. This downward trend, to some extent, detracts
from the interpretation that music feedback was the responsible agent of
change during the first B phase. In addition, the importance of maintaining
equal lengths of phases is highlighted here. Had baseline measurement been
concluded on Day 15, an unequivocal interpretation (though probably erro-
neous) would have been made. However, despite the downward trend in
baseline, mean data for this phase (30-25) were higher than for the previous
feedback phase (23 18).
In the final phase, feedback resulted in a further decline that was generally
maintained at low levels (mean = 14-98). Unfortunately, it is not fully clear
whether this further decrease might have occurred naturally without the
benefits of renewed introduction of feedback. Therefore, despite the presence
of statistically significant differences between baseline and feedback phases
and confirmation of EMG
differences by self-reports of decreased headache
intensity during feedback, the downward trend in the second baseline pre-
vents a definitive interpretation of the controlling effects of the feedback
procedure.
When it is recommended, where
the aforementioned data pattern results,
improvement in baseline be ex-
possible, that variables possibly leading to
amined through additional experimental analyses. However, time limitations
and pressing clinical needs of the patient or subject under study usually
preclude such additional study. Therefore, the next best strategy involves a
replication of the procedure with the
same subject or with additional sub-
jects bearing the same kind of diagnosis (see chapter 10).
A-B-A-B with monitoring of concurrent behaviors
When using the withdrawal strategy, such as the A-B-A-B design, most
experimenters have been concerned with the effects of their treatment vari-
able on one behavior the number of
targeted behavior. However, in a
reports (Kazdin, 1973a; Kazdin, 1973b; Lovads &
Simmons, 1969; Risley,
1968; Sajwaj, Twardosz, & Burke, 1972; Twardosz & Sajwaj, 1972) the
importance of monitoring concurrent (nontargeted) behaviors was docu-
mented. This is of particular importance when side effects of treatment are
possibly negative (see Sajwaj, Twardosz, & Burke, 1972). Kazdin (1973b) has
listedsome of the potential advantages in monitoring the multiple effects of
treatment on operant paradigms.
One initial advantage is that such assessment would permit the possibility of
determining response generalization. If certain response frequencies are in-
creased or decreased, it would be expected that other related operants would be
influenced. would be a desirable addition to determine generalization of

It
beneficial response changes by looking at behavior related to the target response.

In addition, changes in the frequency of responses might also correlate with
topographical alterations, (p. 527)
We might note here that the examination of collateral effects of treatment

should not be restricted to operant paradigms when using experimental
single-case designs.
In our following example the investigators (Twardosz & Sajwaj, 1972) used
an A-B-A-B design to evaluate the efficacy of their program to increase sitting
in a 4-yearjold, hyperactive, retarded boy who was enrolled in an experimen-
tal preschool class. In addition to assessment of the target behavior of interest
(sitting), the effects of treatment procedures on a variety of concurrent
behaviors (posturing, walking, use of toys, proximity of children) were
monitored. Observations of this child were made during a free-play period
(one-half hour) in which class members were at liberty to choose their
playmates and toys. During baseline (A), the teacher gave the child instruc-
prompt him to sit or praise
tions (as she did to all others in class) but did not
him when he did. Institution program (B) involved prompting
of the sitting
the child (placing him in a chair with toys before him on the table), praising
him for remaining seated and for evidencing other positive behaviors, and
awarding him tokens (exchangeable for candy) for in-seat behavior. In the
third phase (A) the sitting program was withdrawn and a return to baseline
conditions took place. Finally, in phase four (B) the sitting program was
reinstated.
The results of this study appear in Figure 5-12. Examination of the top part
of the graph shows that the sitting program, with the exception of the last day
in the first treatment phase, effected improvement over baseline conditions on
both occasions. Continued examination of the figure reveals that posturing
decreased during the sitting program, but walking remained at a consistent
rate throughout all phases of study. Similarly, use of toys and proximity to
children increased during administrations of the sitting program. In discus-
sing their results, Twardosz and Sajwaj (1972) stated that:
This study . . . points out the desirability of measuring several child behaviors,
although a modification procedure might focus on only one. In this way the
preschool teacher can assess the efficacy of her program based upon changes in
other behaviors as well as the behavior of immediate concern, (p. 77)
However, in the event that nontargeted behaviors remain unmodified or that

deterioration occurs in others, additional behavioral techniques can then be
applied (Sajwaj, TWardosz, & Burke, 1972). Under these circumstances it
SITTING REVER- SITTING

too PROGRAM
BASELINE PROGRAM ; SAL
60
40
a/
100
eo-
60
40
20
^ /\J^ ^
too-
60-
60-
40-
20-
''V^
100
80
60-
40-
20-
OO '
Js^ v
80-
tS eo
IS
SCHOOL DAYS
^s A^
FIGURE 5-12. Percentages of Tim's sitting, posturing, walking, use of toys, and proximity to
children during freeplay as a function of the teacher's ignoring him when he did not obey a
command to sit down. (Figure 1, p. 75, from: TVardosz, S., & Sajwaj, T. (1972). Multiple effects
of a procedure to increase sitting in a hyperactive retarded boy. Journal of Applied Behavior
might be preferable to use a multiple baseline strategy (Barlow & Hersen,

1973) in which attention to each behavior can be programed in advance (see
chapter 7).
A-B-A-B with no feedback to experimenter
A major advantage of the single-case strategy (cited in section 3.2 of

chapter 3) is that the experimenter is in a position to alter therapeutic
approaches in accordance with the dictates of the case. Such flexibility is
possible because repeated monitoring of target behaviors is taking place.
Thus changes from one phase to the next are accompHshed with the experi-
menter's full knowledge of prior results. Moreover, specific techniques are
then applied with the expectation that they will be efficacious. Although these
factors are of benefit to the experimental clinician, they present certain
difficultiesfrom a purely experimental standpoint. Indeed, critics of th^
single-case approach have concerned themselves with the possibilities of bias
in evaluation and in actual application and withdrawal of specified tech-
niques. One method of preventing such "bias" is to determine lengths o
baseline and experimental phases on an a priori basis, while keeping the
experimenter uninformed as to trends in the data during their collection. A
problem with this approach, however, is that decisions regarding choice of
baselines and those concerned with appropriate timing of institution and
removal of therapeutic variables are left to change.
The above-discussed strategy was carried out in an A-B-A-B design in
which target measures were rated from video tape recordings for all phases on
a postexperimental basis. Hersen, Miller, and Eisler (1973) examined the
effects of varying conversational topics (nonalcohol and alcohol-related) on
duration of looking and duration of speech in four chronic alcoholics and
their wives in ad libitum interactions videotaped in a television studio. Fol-
lowing 3 minutes of "warm-up" interaction, each couple was instructed to
converse for 6 minutes (A phase) about any subject unrelated to the hus-
band's drinking problem. Instructions were repeated at 2-minute intervals
over a two-way intercom from an adjoining room to ensure maintenance of
the topic of conversation. In the next 6 minutes (B phase) the couple was
instructed to converse only about the husband's drinking problem (instruc-
tions were repeated at 2-minute intervals). The last 12 minutes of interaction
consisted of identical replications of the A and B phases.
Mean data for the four couples are presented in Figure 5-13. Speech
duration data show no trends across experimental phases for either husbands
or wives. Similarly, duration of looking for husbands across phases does not
vary greatly. However, duration of looking for wives was significantly greater
during alcohol- than nonalcohol-related segments of interaction. In the first
nonalcohol phase, looking duration ranged from 26 to 43 seconds, with an

upward trend in evidence. In the first alcohol phase (B), duration of looking
ranged from 57 to 70 seconds, with a continuation of the upward linear trend.
Reintroduction of the nonalcohol phase (A) resulted in a decrease of looking
(38 to 45 seconds). In the final alcohol segment (B), looking once again
increased, ranging from 62 to 70 seconds.
An analysis of these data does not allow for conclusions with respect to the
initialA and B phases inasmuch as the upward trend in A continued into B.
However, the decreasing trend in the second A phase succeeded by the
increasing trend in the second B phase suggests that topic of conversation had
a controlling influence on the wives' rates of looking. We might note here that
DURATION OF LOOKING SPEECH DURATION

80
H H --
80
if)
ieo
70
/ y ieol
70
o o
o o
to 50 i50
u. 11.
O O
DC 40 a:40
lU / ID
m m
30
z J 30
z
.?i20 1 20
10
t 10
Non-Alc. Ale. ;Non-Alc. Ale. Npn-Alc Ale. Non-Alc.i Ale.

1 2 3 78 9 101112
4 5 6 1 2 3 7 8 9 10 1112
BLOCKS OF TWO MINUTES BLOCKS OF TWO MINUTES
FIGURE 5-13. Looking and speech duration in nonalcohol- and alcohol-related interactions of
alcoholics and their wives. Plotted in blocks of 2 minutes. Closed circles husbands; open
circles wives. (Figure 1, p. 518, from: Hersen, M., Miller, P. M., & Eisler, R. M. (1973).
Interactions between alcoholics and their wives: A descriptive analysis of verbal and non-verbal
behavior. Quarterly Journal of Studies on Alcohol, 34, 516-520. Copyright 1973 by Journal of
Studies on Alcohol, Inc. New Brunswick, N.J. 08903. Reproduced by permission.)
if the experimenters were in position to monitor their results throughout all
experimental phases, the segment probably would have been extended

initial
until the wives' looking duration achieved stability in the form of a plateau.
Then the second phase would have been introduced.
5.5. B-A-B DESIGN

The B-A-B design has frequently been usedJ^^LJavestigator^valuating
effectiveness of their treatment procedures- (Agr as, Leitenberg, S^^Barlow,
1968; Ayllon &~Azrin, T965; Leitenbert et Moss, 1973; al., 1968; Mann &
Rickard & Saunders, 1971). In this experimental strategy thephase (B) first
usually involves the application of a treatment. In the second phase (A) the
treatment is withdrawn and in the final phase (B) it is reinstated. Some
investigators (e.g., Agras et al., 1968) have introduced an abbreviated base-
line session prior to the major B-A-B phases. The B-A-B design is superior to
the A-B-A design, described in section 5.3, in that the treatment variable is in
effect in the terminal phase of experimentation. However, absence of an
initial baseline measurement session precludes an analysis of the effects of

treatment over the natural frequency of occurrence of the targeted behaviors
under study (i.e., baseline). Therefore, as previously pointed out by Barlow
and Hersen (1973), the use of the more complete A-B-A-B design is preferred
for assessment of singular therapeutic variables.
We will illustrate the use of the B-A-B strategy with one example selected
from the operant literature and a second drawn from the Rogerian frame-
work. In the first, an entire group of subjects underwent introduction,
removal, and reintroduction of a treatment procedure in sequence (Ayllon &
Azrin, 1965). In the second, a variant of the B-A-B design was imployed by
proponents of client-centered therapy (Truax & Carkhuff, 1965) in an attempt
to experimentally manipulate levels of therapeutic conditions.
B-A-B with group data
Ayllon and Azrin (1965) used the B-A-B strategy on a group basis in their
evaluation of the effects of tokeneconomy on the work performance of 44
"backward" schizophrenic subjects. During the first 20 days (B phase) of the
experiment, subjects were awarded tokens (exchangeable for a large variety
of "backup" reinforcers) for engaging in hospital ward work activities. In the
next 20 days (A phase) subjects were given tokens on a noncontingent basis,
regardless of their work performance. Each subject received tokens daily,
based on the mean daily rate obtained in the initial B phase. In the last 20
days (second B phase) the contingency system was reinstated. We might note
at this point that this design could alternately be labeled B-C-B, as the middle
phase is not a true measure of the natural frequency of occurrence of the
target measure (see section 5.6).
Work performance data (total hours per day) for the three experimental
phases appear in Figure 5-14. During the first B phase, total hours per day
worked by the group averaged about 45 hours. Removal of the con-

entire
tingency in A resulted in a marked linear decrease to a level of one hour per
day on Day 36. Reinstitution of the token reinforcement program in B led to
an immediate increase in hours worked to a level approximating the first B
phase. Thus, Ayllon and Azrin (1965) presented the first experimental demon-
stration of the controlling effects of token economy over work performance
in state hospital psychiatric patients.
It should be pointed out here that when experimental single-case strategies,
such as the B-A-B design, are used on a group basis, it behooves the
experimenter to show that a majority of those subjects exposed to and then
withdrawn from treatment provide supporting evidence for its controlling
data presented for selected subjects can be quite useful,
effects. Individual
particularly if data trends differ. Otherwise, difficulties inherent in the tradi-
tional group comparison approach (e.g., averaging out of effects, effects due
REINFOICfMINT
NOT i
CONTINGENT
^50 r
*
-

!
1
!
UPON
PIRFORMANCI
|
1
1 40
RIINfORCIMINT i;
11
i REINfORCEMENT

CONTINOINT II CONTINCINT
UPON 1
1 UPON
RIRrORMANCI 1 | PERFORMANCE
d 30 1
l\
I 20
1
1

>

N=44
UJ ^
'\
\.
S 10
S \J
20 40 60
DAYS
FIGURE 5-14, Total

B
number of hours of on-ward performance by a group of 44 patients, Exp.
Ill (Figure 4, p. 373, redrawn from: Ayllon, T., & Azrin, N. H. (1965). The measurement and
reinforcement, of behavior of psychotics. Journal of the Experimental Analysis of Behavior, 8,
357-383. Copyright 1965 by Society for the Experimental Analysis of Behavior, Inc. Reproduced
by permission.)
to a small minority while the majority remains unaffected by treatment) will
be carried over to the experimental analysis procedure. In this regard, Ayllon

and Azrin (1965) showed that 36 of their 44 subjects decreased their perfor-
mance from contingent to noncontingent reinforcement. Conversely, 36 of 44
subjects increased their performance from noncontingent to contingent rein-
forcement. Eight subjects were totally unaffected by contingencies and main-
tained a zero level of performance in all phases.
B-A-B from Rogerian framework
Although the withdrawal design has been used in physiological research for
years, and has been associated with the operant paradigm, the experimental
strategies that are applied can easily be employed in the investigation of
nonoperant (both behavioral and traditional) treatment procedures. In this
connection, Truax and Carkhuff (1965) systematically examined the effects of
high and low "therapeutic conditions" on the responses of 3 psychiatric
patients during the course of initial 1-hour interviews. Each of the interviews
consisted of the three 20-minute phases. In the first phase (B) the therapist
was instructed to evidence high levels of "accurate empathy" and "uncondi-
tional positive warmth" in his interactions with the patient. In the following
A phase the therapist experimentally lowered these conditions, and in the

final phase (B) they were reinstated at a high level.
Each of the three interviews was audiotaped. From these audiotapes, five 3-
minute segments for each phase were obtained and rerecorded on separate
spools. These were then presented to raters (naive as to which phase the tape
originated in) in random order. Ratings made on the basis of the Accurate
Empathy Scale and the Unconditional Positive Regard Scale confirmed
(graphically and statistically) that the therapist followed directions as indi-
cated by the dictates of the experimental design (B-A-B).
The effects of high and low therapeutic conditions were then assessed in
terms of depth of the patient*s intrapersonal exploration. Once again, 3-
minute segments from the A and B phases were presented to "naive" raters in
randomized order. These new ratings were made on the basis of the Truax
Depth of Interpersonal Exploration Scale (reliability of raters per segment =
78). Data with respect to depth of intrapersonal exploration are plotted in
Figure 5-15. Visual inspection of these data indicates that depth of intraper-
sonal exploration, despite considerable overlapping in adjacent phases, was
somewhat lowered during the middle phase (A) for each of the three patients.
Although these data are far from perfect (i.e., overlap between phases), the
study does illustrate that the controlling effects of nonbehavioral therapeutic
variables can be investigated systematically using the experimental analysis of
behavior model. Those of nonbehavioral persuasion might be encouraged to
assess the effects of their technical operations more frequently in this fashion.
PATIENT A
1 1
lTl. leNO
7.0 llOWIIIO llOWIIIO
0OMDniOM|<ONBII>OHt
1
' 1
|S6.5
22
*
6.0
>-
r^
Sx
o -
^'^
x2e
>~
as!
Ui
5.0
4.5
^r \l fli
1 3 5 7 9 11 1315 1 3 5 7 9 11 13 15 1 3 5 7 9 11 13 15
TIME <3 MINUTE BLOCKS) TIME (3 MINUTE BLOCKS) TIME (3 MINUTE BLOCKS)
FIGURE 5-15. Depth of intrapersonal exploration. (Figure 4, p. 122, redrawn from: Thiax,
C.B., & Carkhuff, R. R. (1965). Experimental manipulation of therapeutic conditions, Journal
of Consulting Psychology, 29, 1 19-124. Copyright 1965 by the American Psychological Associa-
tion. Reproduced by permission.)
1 70 Single-case Experimental Designs
5.6. A-B-C-B DESIGN

The A-B-C-B design, a variant of the A-B-A-B design, has been ^sed to
evaluate-the-^ffects^^QfjdnfQK^menLpmcedure . Whereas in thTA-B-A^^
and treatment (e.g., contingent reinforcement) are alter-
strategy, baseline
nated in sequence, in the A-B-C-B strategy only the first two phases of
experimentation consist of baseline and contingent reinforcement. In the
third phase (C), instead of returning to baseline observation, reinforcement is
administered in proportions equal to the preceding B phase but on a totally

noncontingent basis. This phase controls for the added attention ("attention-
placebo") that a subject receives for being in a treatment condition and is
analogous to the A, phase (placebo) used in drug evaluations (see chapter 6).
In the final phase, contingent reinforcement procedures are reinstated. Thus
the last three phases of study are identical to those used by Ayllon and Azrin
(1965) in the example described in section 5.5 (however, there the study is
labeled B-A-B).
In the A-B-C-B design the A and C phases are not comparable, inasmuch
as experimental procedures differ. Therefore, the main experimental analysis
is derived from the B-C-B portion of study. However, baseline observations
are of some value, as the effects of B over A are suggested (here we have the
limitations of the A-B analysis). We will illustrate the use of the A-B-C-B
design with one example concerned with the control of drinking in a chronic
alcoholic.
A-B-C-B with a biochemical target measure
Miller, Hersen, Eisler, and Watts (1974) examined the effects of monetary
reinforcement in a 48-year-old "skid row" alcoholic. During all phases of
study, a research assistant obtained breathalyzer samples, analyzed biochemi-
cally shortly thereafter for blood alcohol concentration, from the subject
(psychiatric outpatient) in various locations in his community. To avoid
possible bias in measurement, the subject was not informed as to specific
times that probe measures were to be taken. In fact, these times were
randomized in all phases to control for measurement bias.
During baseline (A phase), eight probe measures were obtained. During
contingent reinforcement (B), the subject was awarded $3.00 in canteen
booklets (redeemable at the hospital commissary for material goods)
whenever a negative blood alcohol sample was obtained. In the noncon-
tingent reinforcement phase (C), reinforcement ($3.00 in centeen booklets)
was administered regardless of blood alcohol concentration. In the final
phase, contingent reinforcement was reinstituted.
Inspection of Figure 5-16 reveals a variable baseline pattern ranging from a
00 to -27 level of blood alcohol. In contingent reinforcement, five of the six
CONT. NON-CONt CONT.

BSLN. REINF. . REINF. REINF.
30
S 5 20
g S 10
^ e
.00 /v
1 3 5 7 9 11 13 15 17 19 21 23 25
PROBE DAYS
^
FIGURE 5-16. Biweekly blood-alcohol concentrations for each phase. (Figure 1, p. 262, from:
Miller, P. M., Hersen, M., Eisler, R. M., & Watts,G. (1974). Contingent reinforcement of
J.
lowered blood/alcohol levels in an outpatient chronic alcoholic. Behaviour Research and

Therapy, 12, 261-263. Copyright 1974 by Pergamon. Reproduced by permission.)
probe measures attained a 00 level. During noncontingent reinforcement,

blood alcohol concentration measures rose, but to lower levels than in
baseline. When contingent reinforcement was reinstated, four of the six
probe measures yielded 00 levels of blood alcohol. Therefore, it appears that
monetary reinforcement resulted in decreases in drinking in this chronic
alcoholic while the contingency was in effect.
A-B-C-B in a group application and follow-up
A most interestingappfication of the A-B-C-B design to a group of subjects

was reported by Porterfield, Blunden, and Blewitt (1980). Subjects in this
experimental analysis were "profoundly mentally handicapped" adults at-
tending a center for the retarded. The behavior targeted for modification was
participation in activities during a 1-hour period so designated during the 19
days of the study. Participation was defined by 12 separate activities and
involved some of the following: watching television, dancing, responding to a
verbal command, talking to another subject, and eating without assistance.
The baseline phase (A) lasted 3 days, with three staff members interacting
with subjects in normal fashion. No specific instructions were given at this
point. The B phase (room manager) members
lasted 5 days, with two staff
alternating for half-hour periods. Subjects in this condition were prompted
and differentially reinforced for their participation. The C phase (no distrac-
tion) lasted 6 days and involved a maximum of two prompts to engage in

activity, but subjects were not differentially reinforced. In the fourth phase
(B) the room manager condition was reinstated. Then there was a 69-day
follow-up period involving the room manager condition in the absence of the
experimenter.
Data appear in Figure 5-17 and are presented as the percentage of subjects
trainees) engaged in activity. It is clear that baseline (A) functioning was
(i.e.,
poor, ranging from 25.7<ô to 37.9% participation. Introduction of the room

manager (B) condition led to marked increases in participation (72.9*^0 to
90.9<yo).
However, when the no-distraction (C) condition was introduced, participa-
tion decreased to near baseline levels (21.5% to 48.0%). When the room
manager condition was reintroduced, in the second B phase, level of partici-
pation once again increased to 84.7% to 88.1%. This second application of
the room manager condition clearly documented the controlling effects of the
contingency. Furthermore, data in follow-up confirmed that participation
TRAINEE ENGAGEMENT
Room Room
100 Baseline
Monoger No -Distraction
i i Monoger Follow-up
80 / \
/
&
S 60
40
20
123 4 5678 9 I0III2I3I4 15 16 I7I8I9 tlGÎ? 440>4U8485
Study days
FIGURE 5-17. Percentage of trainees engaged during the activity hour for 19 days and follow-up
days. (Figure 1, p. 236 from: Porterfield, J., Blunden, R., & Blewitt, E. (1980). Improving
environments for profoundly handicapped adults: Using prompts and social attention to main-
tain high group engagement. Behavior Modification, 4, 225-241. Copyright 1980 by Sage
Publications. Reproduced by permission.)
could be maintained (71.5% to 91.1%) in the absence of experimental

prompting.
There are two noteworthy features in this particular example of the A-B-C-
B design. First, even though the A

and C phases were technically dissimilar,
they certainly were functionally alike. That is, the resulting data pattern was
the same as an A-B-A-B design. However, contrary to the A-B-A-B design,
where there are two instances of confirmation of the contingency, only the B-
C-B portion of the design truly reflected the controlling aspects of the room
manager intervention. Second, by making the dependent measure the "per-
centage of trainees engaged," the experimenters obviated the necessity of
providing individual data. However, from a single-case perspective, data as to
percentage of time active /or each trainee would be most welcome indeed.
CHAPTER 6
Extensions of the A-B-A Design,

Uses in Drug Evaluation and
Interaction Design Strategies
6.1. EXTENSIONS AND VARIATIONS

OF THE A-B-A WITHDRAWAL DESIGN
The applied behavioral literature is replete with examples of extensions and
variations of the more basic A-B-A experimental design. These designs can be
broadly classified into five major categories. The first category consists of
designs in which the A-B pattern is replicated several times. Advantages here
are that (1) repeated control of the treatment variable is demonstrated, and
(2) extended study can be conducted until full clinical treatment has been
achieved. An example of this type of strategy appears in Mann's (1972) work,
where he used an A-B-A-B-A-B design to study the effects of contingency
contracting on weight loss in overweight subjects.
In the second category separate therapeutic variables are compared with
baseline performance during the course of experimentation (e.g., R. V. Hall
et al., 1972; Pendergrass, 1972; Wincze, Leitenberg, & Agras, 1972). Sub-
sumed under this category are the A-B-A-C-A designs discussed in section 3.4
of chapter 3. There it was pointed out that comparison of differential
effectiveness of B and C variables is when both variables appear to
difficult
effect change over baseline levels. However, in the A-B- A-B-A-C-A design the
individual controlling effects of B and C variables can be determined. A
careful distinction should be made between these kinds of designs and designs
where the interactive effects of variables are investigated (e.g., A-B-A-B-BC-
B-BC). In the latter design the effects of C above those of B can be assessed
experimentally. Once again, in the A-B-A-C-A design the effects of B and C
174
Extensions of the A-B-A Design 175
over A can be evaluated. However, interpreting the relative efficacy of B and

C is problematic in this strategy.
In the third category specific variations of the treatment procedure are
examined during the course of experimentation (e.g., Bailey, Wolf, & Phillips,
1970; Coleman, 1970; Conrin, Pennypacker, Johnston, & Rast, 1982;
Hopkins et al., 1971; Kaufman & O'Leary, 1972; McLaughlin & Malaby,
1972; Wheeler & Sulzer, 1970). For example, in some operant paradigms the
treatment procedure may be faded out (e.g., Bailey, Wolf, & Phillips, 1970).
In other paradigms, differing amounts of reinforcement may be assessed
experimentally or in graduated progression (Hopkins et al., 1971) following
demonstration of the controlling effects of variables in the A-B-A-B portion of
the design. This experimental strategy is occasionally termed 2i parametric one.
In a fourth category, the interaction of additive effects of two or more
variables are examined through variations in the basic A-B-A design (e.g.,
Agras et al., 1974; Bernard, Kratochwill, & Keefauver, 1983; Hersen et al.,
1972; Leitenberg et al., 1968; TUrner, Hersen, & Alford, Such analysis
1974).
is accomplished by examining the effects of both variables alone and in
combination, to determine the interaction. This extends beyond analysis of
two therapeutic variables over baseline as represented
the separate effects of
by the A-B-A-C-A type design described in the second category. It also
extends a stop beyond merely adding a variation of a therapeutic variable on
the end of an A-B-A-B series (e.g., A-B-A-B-BC), since no experimental
analysis of the additive effects of BC is performed. Properly run, interaction
designs are complex and usually require more than one subject (see section
6.5.).
The fifth category consists of the changing-criterion design (Hartmann &
Hall, 1976) and its variant, the periodic-treatments design (cf. Hayes, 1981).
Basically, in the changing-criterion design, baseline is followed by treatment
until a preset criterion is met. This then becomes the new baseline (A'), and a
new criterion is set. Such repetition, of course, continues until eventually the
final criterion is reached (see Hersen, 1982).
The following subsections present examples of extensions and variations,
with illustrations selected from each of the five major categories.
6.2. A-B-A-B-A-B DESIGN

Mann (1972) repeatedly introduced and withdrew a treatment variable
(contingency contracting) during extended study with overweight subjects
who had agreed, prior to experimentation, to achieve a designated weight loss
within a specified time period. At the beginning of study, each subject entered
into a formal contractual arrangement with the experimenter. In each case the
subject agreed to surrender a number of his prized possessions (valuables) to
.
the experimenter. During contingency conditions, the subject was able to

regain possession of each vahiable (one at a time) by evidencing a 2-pound
weight loss over his previous low >veight. A further 2-pound weight loss over
that resulted in the return of still another valuable, and so on. Conversely, a
2-pound weight gain over the previous low weight led to the subject's per-
manently losing one of the valuables. In addition to these short-term con-
tingency arrangements, 2-week and terminal contingencies (using similar
principles) were put into effect during treatment phases. Valuables lost by
each subject were subsequently disposed of by the experimenter in equitable
fashion (i.e., he did not profit from or retain them). During baseHne and
"reversal" conditions contractual arrangements were temporarily suspended.
The results of this study for a prototypical subject are plotted in Figure 6-1
Inspection of that figure clearly shows that when contractual arrangements
310.
M\ ^
^
300*
\1 1
290*
280-
5
Vi
o
S 270-
260*
BASE- TREATMENT ; REVER- ;
LINE ;
SAL ;
250 i
A B ; A ;
^"""^^ I I I I
h r^
e>
FIGURE 6-1. A record of the weight of Subject 1 during all conditions. Each open circle
(connected by the thin solid line) represents a 2-week minimum weight loss requirement. Each
solid dot (connected by the thick solid line) represents the subject's weight on each day that he was
measured. Each triangle indicates the point at which the subject was penalized by a loss of
valuables, either for gaining weight or for not meeting a 2-week minimum weight loss require-
ment. NOTE: The subject was ordered by his physician to consume at least 2,500 calories per day
for 10 days, in preparation for medical tests. (Figure la, p. 104, from: Mann, R. A. [1972]. The
behavior-therapeutic use of contingency contracting to control an adult behavior problem:
Weight control. Journal of Applied Behavior Analysis, 5, 99-109. Copyright 1972 by Society for
the Experimental Analysis of Behavior, Inc. Reproduced by permission.)
Extensions of the A-B-A Design 1 71
were in force the subject evidenced a steady linear decrease in weight. By

contrast, during basehne conditions, weight loss ceased, as indicated by a
plateau and slightly upward trend in the data. In short, the effects of the
treatment variable were repeatedly demonstrated in the alternately increasing
and decreasing data trends.
6.3. COMPARING SEPARATE THERAPEUTIC

VARIABLES, OR TREATMENTS
A-B-A-C-A-C'-A design
Wincze et al. (1972) conducted a series of 10 experimental single-case

designs in which the effects of feedback and token reinforcement were
examined on the verbal behavior of delusional psychiatric patients. In one of
these studies an A-B-A-C-A-C'-A design was used, with B and C representing
feedback and token reinforcement phases, respectively. During all phases of
study, a delusional patient was questioned daily (15 questions selected ran-
domly from a pool of 105) by his therapist to elicit delusional material. Per-
centage of responses containing delusional verbalizations was recorded. In
addition, percentage of delusional talk on the ward (token economy unit) was
monitored by nursing staff on a randomly distributed basis 20 times per day.
During baseline (A), the patient received "free" tokens as no contingencies
were placed with respect to delusional verbalizations. During feedback (B),
the patient continued to receive tokens noncontingently, but corrective state-
ments in response to delusional verbalizations were offered by the therapist in
individual sessions. The third phase (A) consisted of a return to baseline
procedures. In Phase 4 (C) a stringent token economy system embracing all
aspects of the patient's ward life was Tokens could be earned by the
instituted.
patient for "talking correctly" (nondelusionally) both in individual sessions
and on the ward. Tokens were exchangeable for meals, luxuries, and privi-
leges. Phase 5 (A) once again involved a return to baseline. In the sixth phase
(C) token bonuses were awarded on a predetermined percentage basis for

talking correctly (e.g., speaking delusionally less than lOô of the time during
designated periods). This condition was incorporated to counteract the ten-
dency of the patient to earn tokens merely for increasing frequency of
nondelusional talk while still maintaining a high frequency of delusional
verbalizations. In the last phase of experimentation (A), baseline conditions
were reinstated for the fourth time.
Results of this experimental analysis for one subject appear in Figure 6-2.
Percentage of delusional talk in individual sessions and on the ward did not
differ substantially during the first three sessions, thus suggesting the ineffec-
tiveness of the feedback variable. Institution of token economy in Phase 4,
2 3 4 5
SESSIONS a-
S4 ;^ WARD -
\/ ^
1 I I I I t I I I > I r 1 1 r-r-Ff-r-f-i t-i I I T I 1
7 8 18 19 25 26 32 33 39 40 46 47 53
DAYS
FIGURE 6-2. Percentage of delusional talk of Subject 4 during tnerapist sessions and on ward for
each experimental day. (Figure 4, p. 256, from: Wincze, J. P., Leitenberg, H., & Agras, W. S.
[1972]. The effects of token reinforcement and feedback on the delusional verbal behavior of
chronic paranoid schizophrenics. Journal of Applied Behavior Analysis, 5, 247-262. Copyright
1972 by Society for the Experimental Analysis of Behavior, Inc. Reproduced by permission.)
however, resulted in a marked decrease of delusional talk in individual

sessions. But it change in delusional talk on the ward.
failed to effect a
Removal of token economy Phase 5 led to a return to initial levels of
in
delusional talk during individual sessions. Throughout the first five phases,
percentage of delusional talk on the ward was consistent, ranging from 0% to
30^0. Introduction of the token bonus in Phase 6 again resulted in a drop of
delusional verbalizations in individual sessions. Additionally, percentage of
delusional talk on the ward decreased to zero. In the last phase (baseline)
delusional verbalizations rose both on the ward and in individual sessions.
In this case, feedback (B) proved to be an ineffective therapeutic agent.
However, token economy (C) and token bonuses (C), respectively, controlled
percentage of delusional talk in individual sessions and on the ward. Had
feedback also effected changes in behavior, the comparative efficacy of
feedback and token economy would be difficult to ascertain using this design.
Such analysis would require the use of a group comparison design. This is
because one variable, token reinforcement, follows the other variable, feed-
back. Therefore, it is conceivable that tokens were effective only if instituted
after a feedback phase and would not be effective if introduced initially. Thus
a possible confound of order effects exists. Of course, the more usual case is
that the first treatment would be effective to an extent that it would not leave
much room for improvement in the second treatment. In other words, a
"ceiling" effect would prevent a proper comparison between treatments, due
to the order of their introduction.
Extensions of the A-B-A Design 1 79
To compare two treatments in this fashion, the investigator would have to

administer two treatments with baseline interspersed to two different individ-
uals (and their replications), with the order of treatments counterbalanced.
For example, 3 subjects could receive A-B-A-C-A, where B and C were two
distinct treatments, and 3 could receive A-C- A-B-A. In fact, Wincze et al.
(1972) carried out this necessary counterbalancing with half of their subjects
in order to analyze the effects of feedback on token reinforcement.
This design, then, approximates the group crossover design or the counter-
balanced within-subject group comparison (e.g., Edwards, 1968), with the
exception of the presence of repeated measures and individual analyses of the
data. Each design option suffers from possible multiple-treatment inter-
ference or carryover effects (see chapter 8 for a discussion of multiple-

treatment interference). In group designs, any carryover effects are averaged
into group differences and treated statistically as part of the error. In the A-B-
A-C-A single-case design,on the other hand, data are usually presented more
descriptively, with visual analysis sometimes combined with statistical descrip-
tions (rather than inferences) to estimate the effect of each treatment. Wincze
et al. (1972) did an excellent job of this in their series, which is fully described
in chapter 10. But analysis depends on comparing individuals experiencing
different orders of treatments. Thus the functional analysis cannot be carried
out within one individual with all of the experimental control that it affords.
Other alternatives to comparing two treatments include a between-groups
comparison design or an alternating-treatments design (see chapter 8).
As noted above, this direct replication series will be discussed in greater
detail in chapter 10.
6.4. PARAMETRIC VARIATIONS OF THE

BASIC THERAPEUTIC PROCEDURES
A-B-A-B'-B '-B " DESIGN
Our example from the third category of extensions of the A-B-A design is
drawn from the child classroom literature. Hopkins et al. (1971) systemati-
cally assessed the effects of access to a playroom on the rate and quality of
writing in rural elementary schoolchildren. Target measures selected for study
were most relevant in that these children came from homes where learning
was not a high priority (parents were migrant or seasonal farm workers).
Throughout all phases of study, first- and second-grade students were given
daily standard written assignments during class periods (class periods were 50
minutes long during the first four phases).
In baseline (A), after each child had completed the assignment, handed it
to the teacher, and waited for it to be scored, he or she was expected to return
to his or her seat and remain there quietly until all others in class had turned
in their papers. In the next phase (B) each child was permitted access to an
adjoining playroom, containing attractive toys, after his or her paper was
scored. The child was allowed to remain there until the 50-minute period was
terminated, unless he or she became too noisy; then he or she was required to
return to his or her seat. The next two phases (A and B) were identical to the
first two. In the last three phases each child was permitted access to the
playroom after his or her paper had been scored, but the length of class
periods was gradually decreased (45, 40, 35 minutes). A procedural exception
to the aforementioned was made in the last phase on Days 47-54 inasmuch as
the teacher noted that a concomitant of increased speed was decreased quality
(number of errors) in writing. Therefore, during the last 8 days a quality
criterion was imposed before the child gained access to the playroom. In some
cases the child was required to recopy a portion of writing.
Data for first-grade children are plotted in Figure 6-3. Examination of the
bottom half of the figure shows that access to the playroom (50-minute
period) increased the rate of letter writing over baseline levels. This was
confirmed on two occasions in the A-B-A-B portion of study. When total time
of classroom periods systematically decreased, a corresponding increase in
rate of writing resulted. However, data for the last three phases are correla-
tive, asan experimental analysis was not performed. For example, a sequen-
tial comparison of 50-, 45- and 50-minute periods was not made. Therefore,
the controlling effects of time differences were not fully documented.

Examination of the top part of the graph shows considerable fluctuation
with respect to mean number of errors per letter. However, this did not appear
to represent a systematic increase when class periods were shortened. To the
contrary, there was a general decrease in error rate from the first to the last
phase of study. Nonetheless, the effects of practice cannot be discounted
when total length of the investigation is considered.
A-B-B'-B '-A-B' design
A more recent example of a study involving variations of the basic thera-

peutic procedure appears in a study by Conrin et al. (1982), in which differen-
tial reinforcement of other behaviors (DRO) was used to treat chronic
rumination in mentally retarded individuals. In this study an A-B-B'-B"-A-
B' design was followed. The subject (Bob) was a 19-year-old male (53 in. tall,
56 lbs. at baseline) who was profoundly retarded and who ruminated (emesis
of previously chewed food, rechewing food, and reswallowing food). The
disorder had begun some 17 years earlier.
Baseline (A) observations took place one hour after the subject had con-
sumed his meal. After each meal Bob was brought to the cottage lounge and
observed. Duration of rumination (cheek swelling, chewing, and swallowing)
G
PLAYROOM -50 MINUTES PLAYROOM- 35 MINUTES

229
209
K
UJ
ui .109
-I
ae
Ui
a- t69i
i/>
ec
o
gj49
UJ
129
K
W
a
109
I
z
woes
.069i
IS
u 12
f mST GRADE PRINTINO
Ui
-i
o
Ui
a
?
z
7
/v
z
2 6
DAYS / // Jo
fv 6 /)
FIGURE 6-3. The mean number of letters printed per minute by first-grade children are shown on
the lower coordinates, and the mean proportion of letters scored as errors are on the upper
coordinates.Each data point represents the mean averaged over all children for that day. The
horizontal dashed lines are the means of the daily means averaged over all days within the
experimental conditions noted by the legends at the top of the figure. (Figure 1, p. 81, from:
Hopkins, B. L., Schutte, R. C, & Carton, K. L. [19711. The effects of access to a playroom on
the rate and quality of printing and writing of first- and second-grade students. Journal of
Applied Behavior Analysis, 4, 77-87. Copyright 1971 by Society for the Experimental Analysis of
Behavior, Inc. Reproduced by permission.)
was timed. In the second phase (B) a DRO procedure was implemented. This
consisted of giving Bob small portions of cookies or bits of peanut butter
contingent on no rumination. In the B phase reinforcement was provided if
no rumination occurred for 15 seconds or more (IRT> 15"). In the next phase
SCED
(B') this was increased to 30 seconds (IRT>30"), followed by an IRT>60"

in phase B " Then
. there was a return to baseline (A) and reintroduction of
IRT>30".
Interrater agreement for behavioral observations ranged from 94% to
100%. Examination of data in Figure 6-4 reveals a high duration of rumina-
tion (5 to 22 minutes; mean = 7 minutes) during baseline (A). Introduction
of DRO (IRT> was
15") resulted in a zero duration after 18 sessions, which
maintained during the thinning of the reinforcement schedule in B'
(IRT>30") and B" (IRT>60"). A return to baseline conditions (A) resulted
in marked increases in rumination (mean = 10 minutes per session), but was
once again reduced to zero when DRO procedures (IRT>30") were reintro-
duced in the B' phase.
In summary, this experimental analysis clearly documents the controlling
effects of DRO over duration of rumination. It also shows how it was
possible to thin the reinforcement schedule from IRT> 15" to IRT>60" and
still maintain rumination at near zero levels.
IRT>I5" IRT>30" IRT>60" BL IRT>30"
10 20 30 40 50 60 80 90 100 110 1^0 130 140
Successive meals
A 6 (?'
S" A- 6
FIGURE 6-4. Duration of ruminations after meals by Bob. (Figure 2, p. 328, from: Conrin, J.,
Pennypacker, H. S., Johnston, J. M., & Rast, J. [1982]. Differential reinforcement of other
behaviors to treat chronic rumination of mental retardates. Journal of Behavior Therapy and
Experimental Psychiatry, 13, 325-329. Copyright 1982 by Pergamon. Reproduced by permission.
6.5. DRUG EVALUATIONS

The group comparison approach generally has predominated in the ex-
amination of the effects of drugs on behavior. However, examples in which
the subjects have served as their own controls in the experimental evaluation
of pharmacological agents are now seen more frequently in the psychological
and psychiatric literatures (e.g., Agras, Bellack, & Chassan, 1964; Chassan,
1967; K. V. Davis, Sprague, & Werry, 1969; Grinspoon, Ewalt, & Shader,
1967; Hersen & Breuning, in press; Liberman et al., 1973; Lindsley, 1962;
McFarlain & Hersen, 1974; Roxburgh, 1970). Indeed, Liberman et al. (1973)
have encouraged researchers to use the within-subject withdrawal design in
assessing drug-environment interactions. In support of their position they
contend that:
Useful interactions among the drug-patient-environment system can be obtained

using this type of methodology.The approach is reliable and rigorous, efficient
and inexpensive to mount, and permits sound conclusions and generalizations to
other patients with similar behavioral repertoires when systematic replications
are performed ... (p. 433)
There is no doubt that this approach can be of value in the study of both the
major forms of psychopathology and those of more exotic origin (Hersen &
Breuning, in press). The single-case experimental strategy is especially well
suited to the latter, as control group analysis in the rarer disorders is obviously
not feasible.
Specific issues
It should be pointed out that all procedural issues discussed in chapter 3

pertain equally to drug evaluation. In addition, there are a number of
considerations specific to this area of research: (1) nomenclature, (2) car-
ryover effects, and (3) single- and double-bHnd assessments.
With respect to nomenclature, A is designated as the baseline phase. A, as
the placebo phase, B as the phase evaluating the first active drug, and C as the
phase evaluating the second active drug. The A, phase is an intermediary
phase between A (baseline) and B (active drug condition) in this schema. This
phase controls for the subject's expectancy of improvement associated with
mere ingestion of the drug rather than for its contributing pharmacological
effects.
Some of the above-mentioned considerations have already been examined
in section 3.4 of chapter 3 in relation to changing one variable at a time across
experimental phases. With regard to this one-variable rule, it becomes ap-
parent, then, that A-B, A-B-A, B-A-B, and A-B-A-B designs in drug research
involve the manipulation of two variables (expectancy and condition) at one

time across phases. However, under certain circumstances where time limita-
tions and clinical considerations prevail, this type of experimental strategy is
justified. Of course, when conditions permit, it is preferable to use strategies

in which the systematic progression of variables across phases is carefully
followed (see Table 6-1, Designs 4,6, 7, 9-13). For example, this would be the
case in the A,-B-A, design strategy, where only one variable at a time is
manipulated from phase to phase. Further discussion of these issues will
appear in the following section, in which the different design options avail-
able to drug researchers will be outlined.
The problem of carryover effects from one phase to the next has already
been discussed in section 3.6 of chapter 3. There some specific recommenda-
tions were made with respect to short-term assessments of drugs and the
concurrent monitoring of biochemical changes during different phases of
study. In this connection. Barlow and Hersen (1973) have noted that "Since
continued measurements are in effect, length of phases can be varied from
experiment to experiment to determine precisely the latency of drug effects
after beginning the dosage and the residual effects after discontinuing the
dosage" (p. 324). This may, at times, necessitate the inequality of phase
lengths and the suspension of active drug treatment until biochemical mea-
surements (based on blood and urine studies) reach an acceptable level. For
example, Roxburgh (1970) examined the effects of a placebo and thiopropa-
zate dihydrochloride on phenothiazine-induced oral dyskinesia in a double-
blind crossover in two subjects. In both cases, placebo and active drug
treatment were separated by a 1-week interruption during which time no
placebo or drug was administered.
A third issue specific to drug evaluation involves the use of single- and
double-blind assessments. The double-blind clinical trial is a standard precau-
tionary measure designed to control for possible experimenter bias and
patient expectations of improvement under drug conditions when drug and
placebo groups are being contrasted. "This is performed by an appropriate
method of assigning patients to drugs such that neither the patient nor the
investigator observing him knows which medication a patient is receiving at
any point along the course of treatment" (Chassan, 1967, pp. 80-81). In these
studies, placebos and active drugs are identical in size, shape, markings, and
color.
While the double-blind procedure is readily adaptable to group comparison
research, it is difficult to some of the single-case strategies and
engineer for
impossible for others. Moreover, in some cases (see Table 6-1, Designs 1, 2, 4,
5, 8) even the single-blind strategy (where only the subject remains unaware
of differences in drug and placebo manipulations) is not applicable. In these
designs the changes from baseline observation to either placebo or drug
conditions obviously cannot be disguised in any manner.
Extensions of the A-B-A pesign 185
TABLE 6-1. Single-Case Experimental Drug Strategies
NO. DESIGN TYPE BLIND POSSIBLE
1. A-A, Quasi-experimental None

2. A-B Quasi-experimental None
3. A,-B Quasi-experimental Single or double
4. A-A,-A Experimental None
5. ABA Experimental None
6. A.-B-A, Experimental Single or double
7. A,-A-A, Experimental Single or double
8. BAB Experimental None
9. B-A.-B Experimental Single or double
10. A-A,-A-A, Experimental Single or double
11. A-B-A-B Experimental None
12. A.-B-A.-B Experimental Single or double
13. A-A, -B-A.-B Experimental Single or double
14. A-A, -A-A, -B-A.-B Experimental Single or double
15. A.-B-A.-C-A.-C Experimental Single or double
Note: A = no drug; A. = placebo; B = drug 1; C = drug 2,
A major difficulty in obtaining a true double-blind trial in single-case

research is related to the experimenter *s monitoring of data (i.e., making
decisions as to when baseline observation is to be concluded and when various

phases are to be introduced and withdrawn) throughout the course of investi-
gation. It is possible to program phase lengths on an a priori basis, but then
one of the major advantages of the single-case strategy (i.e., its flexibility) is
lost. However, even though the experimenter is fully aware of treatment
changes, the spirit of the double-blind trial can be maintained by keeping the
observer (often a research assistant or nursing staff member) unaware of drug
and placebo changes (Barlow & Hersen, 1973). We might note here addi-
of
tionally that despite the use of the double-blind procedure, the side effects
drugs in some cases Parkinsonism following administration of large
(e.g.,
doses of phenothiazines) and the marked changes in behavior resulting from

removal of active drug therapy in other cases often betray to nursing person-
nel whether a placebo or drug condition is currently in operation. This
problem is equally troublesome for the researcher concerned with group
comparison designs (see Chassan, 1967, chap. 4).
Different design options
In some of the investigations in which the subject has served as his or her
own method of study, where the
control, the standard experimental analysis
treatment variable is introduced, withdrawn, and reintroduced following
initial measurement, has not been followed rigorously. Thus the controlling
effects of the drug under evaluation have not been fully documented. For
example, K. V. Davis et al. (1969) used the following sequence of drug and
no-drug conditions in studying rate of stereotypic and nonstereotypic behav-

ior in severe retardates: (1) methylphenidate, (2) thioridazine, (3) placebo,
and (4) no drug. Despite the fact that thioridazine significantly (at the
statistical level) decreased the rate of stereotypic responses, failure to reintro-
duce the drug in a final phase weakens the conclusions to some extent from an
experimental analysis standpoint.
A careful survey of the experimental analysis of behavior literature reveals
relatively little discussion with regard to procedural and design issues in the
assessment of drugs. Therefore, in light of the unique problems faced by the
drug researcher and in consideration of the relative newness of this area, we
will outline the basic quasi-experimental and experimental analysis design
strategies for evaluating singular application of drugs. Specific advantages
and disadvantages of each design option will be considered. Where possible,
we will illustrate with actual examples selected from the research literature.
However, to date, most of these strategies have not yet been implemented.
A number of possible single-case strategies suitable for drug evaluation are
presented in Table 6-1 . The first three strategies fall into the A-B category and
are really quasi-experimental designs, in that the controlling effects of the
treatment variable (placebo or active drug) cannot be determined. Indeed, it
was noted in section 5.2 of chapter 5 that changes observed in B might

possibly result from the action of a correlated but uncontrolled variable (e.g.,
time, maturational changes, expectancy of improvement). These quasi-ex-

perimental designs can best be applied in settings (e.g., consulting room
practice) where limited time and facilities preclude more formal experimenta-
tion. In the first design the effects of placebo over baseline conditions are
suggested; in the second the effects of active drug over baseline conditions are
suggested; in the third the effects of an active drug over placebo are sug-
gested.
Examination of Strategies 4-6 indicates that they are basically A-B-A
designs in which the controlling effects of the treatment variable can be
ascertained. In Design 4 the controlling effects of a placebo manipulation
over no treatment can be assessed experimentally. This design has great
potential in the study of disorders such as conversion reactions and histrionic
personalities, where attentional factors are presumed to play a major role.
Also, the use of this type of design in evaluating the therapeutic contribution
of placebos in a variety of psychosomatic disorders could be of considerable
importance to clinicians. In Design 5, the controlling effects of an active drug
are determined over baseline conditions. However, as previously noted, two
variables are being manipulated here at one time across phases. Design 6
corrects for this deficiency, as the active drug condition (B) is preceded and
followed by placebo (A,) conditions. In this design the one-variable rule
across phases is carefully observed.
An example of an A,-B-A, design appears in a series of single-case drug

evaluations reported by Liberman et al. (1973). In one of these studies the
effects of fluphenazine on eye contact, verbal self-stimulation (unintelligible
or jumbled speech), and motor self-stimulation were examined in a double-
blind trial for a 29-year-old regressed schizophrenic who had been continu-
ously hospitalized for 13 years. Double-blind analysis was facilitated by the
fact that fluphenazine (10 mg, b.i.d.) or the placebo could be administered
twice daily in orange juice without its being detected (breaking of the double-
blind code) by the patient or the nursing staff, as the drug cannot be
distinguished by either odor or taste. During all phases of study, 18 randomly
distributed 1 -minute observations of the patient were obtained daily with
respect to incidence of verbal and motor self-stimulation. Evidence of eye
contact with the patient *s therapist was obtained daily in six 10-minute
sessions. Each eye contact was reinforced with candy or a puff on a cigarette.
The results of this study are plotted in Figure 6-5. During the first placebo
phase (A,), stable rates were obtained for each of the target behaviors.
PLACEBO FLUPHENAZINE PLACEBO
75
eye -A ^ ?'o^. .^
contoct 5Q.
25H ,/
o-o'' 'o
'wA b-6
0-
% of
motor
self-stim
75-
50-
25H
P-o-tx
A^ b-o' \/
75-1 \ jO CX
/
<
tr oo
% of
verboi 50-
seif-stim
25H
<to
..^vVV
rr t T I r T I I 1 1 I I T T I I I I I I I t
'
2 4 6 8 10 12 14 16 18 20 22 24 26
SESSIONS
FIGURE 6-5. Interpersonal eye contact, motor, and self-stimulation in a schizophrenic young
man during placebo and fluphenazine (20 mg daily) conditions. Each session represents the
average of a 2-day block of observations. (Figure 3, p. 437, from: Liberman, R. P., Davis, J.,
Moon, W., & Moore, J. [1973]. Research design for analyzing drug-environment-behavior
interactions. Journal of Nervous and Mental Disease, 156, 432-439. Copyright 1973. Reproduced
by permission.)
Introduction of fluphenazine in the second phase (B) resulted in a very slight

increase in eye contact, and increased variability in motor self-stimulation,
and a linear increase in verbal self-stimulation. Withdrawal of fluphenazine
and a return to placebo conditions in the final phase (A,) failed to yield data
trends. On the contrary, eye contact increased slightly while verbal self-
stimulation increased dramatically. Motor self-stimulation remained rela-
tively consistent across phases. These data were interpreted by Liberman et
al. (1973) as follows: "The failure to gain a reversal suggests a drug-initiated
response facilitation which is seen most clearly in the increase of verbal self-
stimulation, and less so in rate of eye contact" (p. 437). It was also suggested
that residual phenothiazines during the placebo phase may have contributed
to the continued increase in eye contact. However, in the absence of concur-
rent monitoring of biochemical factors (phenothiazine blood and urine
levels), this hypothesis cannot be confirmed. In summary, Liberman et al.
(1973) were not able to confirm the controlling effects of fluphenazine over
any of the target behaviors selected for study in this Ai-B-A, design.
Let us now continue our examination of drug designs listed in Table 6-1.
Strategies 7-9 can be classified as B-A-B designs, and the same advantages
and limitations previously outlined in section 5.5 of chapter 5 apply here.
Strategies 10-12 fall into the general category of A-B-A-B designs and are
superior to the A-B-A and B-A-B designs for several reasons: (A) The initial
observation period involves baseline or baseline-placebo measurement; (2)
there are two occasions in which the controlling effects of the placebo or the
treatment variables can be demonstrated; and (3) the concluding phase ends
on a treatment variable.
Agras (1976) used an A-B-A-B design to assess the effects of chlorproma-
zine in a 16-year-old, black, brain-damaged, male inpatient who evidenced a
wide spectrum of disruptive behaviors on the ward. Included in his repertoire
were: temper tantrums, stealing food, eating with his fingers, exposing him-
self, hallucinations, and begging for money, cigarettes, or food. A specific
token economy system was devised for this youth, whereby positive behaviors
resulted in his earning tokens, and inappropriate behaviors resulted in his
being penalized with fines. Number of tokens earned and number of tokens
fined were the two dependent measures selected for study. The results of this
investigation appear in Figure 6-6. In the first phase (A) no thorazine was
administered. Although improvement in appropriate behaviors was noted,
the patient's disruptive behaviors continued to increase markedly, resulting in
his being fined many times. This occurred in spite of the addition of a time-
out contingency. On Day 9, thorazine (300 mg per day) was intro-
Hospital
duced (B phase) in an attempt to control the patient's impulsivity. This dosage
was subsequently decreased to 200 mg per day, as he became drowsy. Ex-
amination of Figure 6-6 reveals that fines decreased to a zero level whereas
tokens earned for appropriate behaviors remained at a stable level. In the
No
No Thorazine Thorazine Tho Tho
40r - Earned j
o-^ Fined
j
CO I
I
0) 30- I
I
I
I
I
I
f I
20- .n '
E 10-
oo^>o o o o o-
i-j. ' '

I I I
1 3 5 7 9 11 13 15 17 19 21 23
Hospital Days
FIGURE 6-6. Behavior of an adolescent as indicated by tokens earned or fined in response to

chlorpromazine, which was added to token economy. (Figure 15-3, p. 556, from: Agras, W. S.
[1976]. Behavior modification in the general hospital psychiatric unit. In H. Leitenberg [Ed.],
Handbook of behavior modification. Englewood Cliffs, NJ: Prentice-Hall. Copyright 1976 by H.
Leitenberg. Reproduced by permission.)
third phase (A) chlorpromazine was temporarily discontinued, resulting in an

increase in fines for disruptive behavior. The no-thorazine condition (A) was
only in force for 2 days, as the patient's renewal of disruptive activities caused
nursing personnel to demand reinstatement of his medication. When thora-
zine was reintroduced in the final phase (B), number of tokens fined once
again decreased to a zero level. Thus the controlling effects of thorazine over
disruptive behavior were demonstrated. But Agras (1976) raised the question
as to the possible contribution of the token economy program in controlling
this patient's behavior. Unfortunately, time considerations did not permit him
to systematically tease out the effects of that variable.
We might also note that in the A-B-A-B drug design, where the single- or
double-blind trial is not feasible, staff and patient expectations of success
during the drug condition are a possible confound with the drug's pharmaco-
logical actions. Designs listed in Table 6-1 that show control for these factors
are 12 (A,-B-A,-B) and 13 (A-A,-B-A,-B). Design 13 is particularly useful in
this instance. In the event that administration of the placebo fails to lead to

behavioral change (A, phase of experimentation) over baseline measurement

(A), the investigator is in a position to proceed with assessment of the active
drug agent inan experimental analysis whereby the drug is twice introduced
and once withdrawn (the B-A,-B portion of study). If, on the other hand, the
placebo exerts an effect over behavior, the investigator may wish to show its
controlling effects as in Design 10 (A- A, -A- A,), which then can be followed
with a sequential assessment of an active pharmacologic agent (Design 14
A-A,-A-A,-B-A,-B). This design, however, does not permit an analysis of the
interactive effects of a placebo (A,) and a drug (B), as this would require the
use of an interactive design (see section 6.5).
An example of the A-A,-B-A,-B strategy appears in the series of drug
evaluations conducted by Liberman et al. (1973). In their study, the effects of
a placebo and trifluperazine (stelazine) were examined on social interaction
and content of conversation in a 21 -year-old, withdrawn, male inpatient
whose behavior had progressively deteriorated over a 3 -year period. At the
time the experiment was begun, the patient was receiving stelazine, 20 mg per
day. Two dependent measures were selected for study: (1) willingness to
engage in 18 daily, randomly time sampled, one-half minute chats with a
member of the nursing staff, and (2) percentage of the chats that contained
"sick talk." During the first phase of experimentation (A), the patient's
medication was discontinued. In the second phase (A,) a placebo was intro-
duced, followed by application of stelazine, 60 mg per day, in the next phase
(B). Then the A, and B phases were repeated. A double-blind trial was
conducted, as the patient and nursing staff were not made aware of placebo
and drug alternations.
Results of this study with regard to the patient's willingness to partake in
brief conversations appear in Figure 6-7. In the no-drug condition (A) a
marked linear increase in number of asocial responses was observed. Institu-
tion of the placebo in phase two (A,) first led to a decrease, followed by a
renewed increase in asocial responses, suggesting the overall ineffectiveness of
the placebo condition. In Phase 3 (B), administration of stelazine (60 mg per
day) resulted in a substantial decrease in asocial responses. However, a return
to placebo conditions (A,) again led to an increase in refusals to chat. In the
final phase (B), reintroduction of stelazine effected a decrease in refusals. To
summarize, in this experimental analysis, the effects of an active pharmaco-
logical agent were documented twice, as indicated by the decreasing data
trends in the stelazine phases. Data with respect to content of conversation
were not presented graphically, but the authors indicated that under stelazine
conditions, rational speech increased. However, administration of stelazine
did not appear to modify frequency of delusional and hypochondriacal
statements in that they remained at a constant level across all phases of study.
Let us now return to and conclude our examination of drug designs in
r r
NO DRUG PLACEBO STELAZINE PLACEBO STELAZINE

14-1
5^ 12
S
5) 10 J
f
T I I
T 17
I I
11 13 15 19 21 23 25
SESSIONS
FIGURE 6-7. Average number of refusals to engage in a brief conversation. (Figure 2, p. 435,
from: Liberman, R. P., Davis, J., Moon, W., & Moore, J. [1973]. Research design for analyzing
drug-environment-behavior interactions. Journal of Nervous and Mental Disease, 156, 432-439.
Copyright 1973 Williams & Wilkins. Reproduced by permission.)
Table 6-1. In Design 15 (A,-B-A,-C-A,-C) the controlling effects of two drugs

(B and C) over placebo conditions (A,) can be assessed. However, as in the A-
B-A-C-A design, cited in section 6. 1 , the comparative efficacy of variables B
and C are not subject to direct analysis, as a group comparison design would
be required.
We should point out here that many extensions of these 15 basic drug
designs are possible, including those in which differing levels of the drug are
examined. This can be done within the structure of these 15 designs during
active drug treatment or in separate experimental analyses where dosages are
systematically varied (e.g., low-high-low-high) or where pharmacological
agents are evaluated after possible failure of behavioral strategies (or vice
versa). However, as in the A-B-A-C-A design com-
cited in section 6.1, the
parative efficacy of variables B and C is subject to a number of restrictions
and is, in general, a rather weak method for comparing two treatments.
The following A-B-C-A-D-A-D experimental analysis illustrates how, after
two behavioral strategies (flooding, response prevention) failed to yield im-
provements in ritualistic behavior, a tricyclic (imipramine) led to some behav-
ioral change, but only when administered at a high dosage (Hirner, Hersen,
Bellack, Andrasik, & Capparell, 1980).
The subject was a 25-year-old woman with a 7-year history cf hand-

washing and toothbrushing rituals. She had been hospitalized several times,
with no treatment proving successful (including ECT). Throughout the seven
phases of the study (with the exception of response prevention), mean dura-
tion of hand-washing and toothbrushing was recorded. Following a 7-day
baseline period (A), flooding (B) was initiated for 8 days, and then response
prevention (C) for 7 days. Then there was a 5-day return to baseline (A).
Imipramine (C) was subsequently administered in increasing doses (75 mg to
250 mg) over 23 days, followed by withdrawal (A) and then reinstitution (C).
In addition, 4 weeks of follow-up data were obtained.
Resulting data in Figure 6-8 are fairly clear-cut. Neither of the two behav-
ioral strategies effected any change in the two behaviors targeted for modi-
fication. Similarly, imipramine, until it reached a level of 200 mg per day was
ineffective. However, from 200-250 mg per day the drug appeared to reduce
the duration of hand- washing and toothbrushing. When imipramine was
withdrawn, hand-washing and toothbrushing increased in duration but de-
creased again when it was reinstated. Improvement was greatest at the higher
dosage levels and was maintained during the 4- week follow-up.
From a design perspective, phases 4-7 (A-C-A-C) essentially are the same
as Design 1 1 (A-B-A-B) in Table 6-1. Of course, the problem with the A-B-A-
B design is that the intervening A' or placebo phase is bypassed, resulting in
two variables being manipulated at once (i.e., ingestion and action of the
drug). Therefore, one cannot discount the possible placebo effect in the
TUrner et al. (1980) analysis, ahhough the long history of the disorder makes
this interpretation unlikely.
FIGURE 6-8. Mean duration of hand-washing and toothbrushing per day. (Figure 3, p. 654,
from: Tbrner, S. M., Hersen, M., Bellack, A. S., Andrasik, E, & Capparell, H. V. [1980].
Behavioral and pharmacological treatment of obsessive-compulsive disorders. Journal of Ner-
vous and Mental Disease, 168, 651-657. Copyright 1980 The Williams and Wilkins Co., Balti-
more. Reproduced by permission.)
6.6. STRATEGIES FOR

STUDYING INTERACTION EFFECTS
Most treatments contain a number of therapeutic components. One task of
the clinical researcher is to experimentally analyze these components to
determine which are effective and which can be discarded, resulting in a more
efficient treatment. Analyzing the separate effects of single therapeutic vari-
ables is a necessary way to begin to build therapeutic programs, but it is
obvious that these variables may have different effects when interacting with
other treatment variables. In advanced stages of the construction of complex
treatments it becomes necessary to determine the nature of these interactions.
Within the group comparison approach, statistical techniques, such as analy-
sis of variance, are quite valuable in determining the presence of interaction.
These techniques are not capable, however, of determining the nature of the
interaction or the relative contribution of a given variable to the total effect in
an individual.
To evaluate the interaction of two (or more) variables, one must analyze the
effects of both variables separately and in combination in one case, followed
by replications. However, one must be careful to adhere to the basic rule of
not changing more than one variable at a time (see chapter 3, section 3.4).
Before discussing examples of strategies for studying interaction, it will be
helpful to examine some examples of two or more vari-
designs containing
ables that are not capable of isolating interactive or additive effects. The first
example is one where variations of a treatment are added to the end of a
successful A-B-A-B (e.g., A-B-A-B'-B'-B' described above or an A-B-A-B-BC
design in which C is a different therapeutic variable). If the BC variable
produced an effect over and above the previous B phase, this would provide a
clue that an interaction existed, but the controlling effects of the BC phase
would not have been demonstrated. To do this, one would have to return to
the B phase and reintroduce the BC phase once again.
A second design, containing two or more variables where analysis of
interaction is not possible, occurs if one performs an experimental analysis of
one variable against a background of one or more variables already present in
the therapeutic situation. For example, O'Leary et al. (1969) measured the
disruptive behavior of seven children in a classroom. Three variables (rules,
educational structure, and praising appropriate behavior while ignoring dis-
ruptive behavior) were introduced sequentially. At this point, we have an A-
B-BC-BCD design, where B is rules, C is structure, and D is praise and
ignoring. With the exception of one child, these procedures had no effect on
disruptive behavior. A fourth treatment token economy was then added.
In five of six cfhildren this was effective, and withdrawal and reinstatement of
the token economy confirmed its effectiveness. The last part of the design can

be represented as BCD-BCDE-BCD-BCDE, where E is token economy.

Although this experiment demonstrated that token economy works in this
of the first three variables is not clear. It is possible that any
setting, the role
one of the variables or all three are necessary for the effectiveness of the
token program or at least to enhance its effect. On the other hand, the initial
three variables may not contribute to the therapeutic effect. Thus we know
that a token program works in this situation, against the background of these
three variables, but we cannot ascertain the nature of the interaction, if any,
because the token program was not analyzed separately.
A third example, where analysis of interaction is not possible, occurs if one
is testing the effects Two examples of this
of a composite treatment package.
strategy were presented in chapter 3, section 3.4. In one example (see Figure
3-13) the effects of covert sensitization on pedophilic interest were examined
(Barlow, Leitenberg, & Agras, 1969). Covert sensitization, where a patient is
instructed to imagine both unwanted arousing scenes in conjunction with
aversive scenes, contains a number of variables such as therapeutic instruc-
tion, muscle relaxation, and instructions to imagine each of the two scenes. In
this experiment, the whole package was introduced after baseline, followed
by withdrawal and reinstatement of one component the aversive scene. The
design can be represented as A-BC-B-BC, where BC is the treatment package
and C is the aversive scene. (Notice that more than one variable was changed
during the transition from A-BC. This is in accordance with an exception to
the guidelines outlined in chapter 3, section 3.4.)
Figure 3-13 demonstrates that pedophilic interest dropped during the treat-
ment package, rose when the aversive scene was removed, and dropped again
after reinstatement of the aversive scene. Once again, these data indicate that
the noxious scene is important against the background of the other variables
present in covert sensitization. The contribution of each of the other variables
and the nature of these interactions with the aversive scene, however, have not
been demonstrated (nor was this the purpose of the study). In this case, it
would seem that an interaction is present because it is hard to conceive of the
aversive scene alone producing these decreases in pedophilic interest. The
nature of the interaction, however, awaits further experimental inquiry.
The preceding examples outlined designs where two or more variables are
simultaneously present but analysis of interactive or additive effects is not
possible. While these designs can hint at interaction and set the stage for
further experimentation, a thorough analysis of interaction as noted above
requires an experimental analysis of two or more variables, separately and in
combination. To illustrate this complex process, two series of experiments will

be presented that analyze the same variables feedback and reinforcement
in two separate populations (phobics and anorexics). One experiment from
the first series of phobics was presented in chapter 3, section 3.4, in connec-
tion with guidelines for changing one variable at a time.
In that series (Leitenberg et al., 1968) the firstsubject was a severe knife
phobic. The target behavior selected for study was the amount of time (in
seconds) that the patient was able to remain in the presence of the phobic
object. The design can be represented as B-BC-B-A-B-BC-B, where B repre-
sents feedback, C represents praise, and A is basehne. Each session consisted
of 10 trials. Feedback consisted of informing the patient after each trial as to
the amount of time spent looking at the knife. Praise consisted of verbal
reinforcement whenever the patient exceeded a progressively increasing time
criterion. The results of the study are reproduced in Figure 6-9. During
feedback, a marked upward linear trend in time spent looking at the knife
was noted. The addition of praise did not appear to add to the therapeutic
effect. Similarly, the removal of praise in the next phase did not subtract from
the progress. At this point, it appeared that feedback was responsible for the
therapeutic gains. Withdrawal and reinstatement of feedback in the next two
205-j
PHASES: 1
3 4 5
120
100-
AV:
/I
Z
<
60-
k /
5 40-
20
NO FB
FEEDBACK (FB) FB + PRAISE FB NO FB FB FB
ALONE
15
BLOCKS OF
20
: ALONE
25
PRAISE
30 35 40

ALONE PRAISEi ALONE
75
SESSIONS (40 TRIALS)
B 6t B A ^C s
FIGURE 6-9. Time
which a Rnife was kept exposed by a phobic patient as a function of
in
feedback, feedback plus praise, and no feedback or praise conditions. (Figure 2, p. 136, from:
Leitenberg, H., Agras, W. S., Thomson, L. E., & Wright, D. E. [1%8]. Feedback in behavior
modification: An experimental analysis in two phobic cases. Journal of Applied Behavior
phases confirmed the controlHng effects of feedback. Addition and removal

of praise in the remaining two phases repHcated the beginning of the experi-
ment, in that praise did not demonstrate any additive effect.
This experiment alone does not entirely elucidate the nature of the interac-
tion. At this point, two tentative conclusions are possible. Either praise has
no effect on phobic behavior, or praise does have an effect, which was
masked or overridden by the powerful feedback effect. In other words, this
patient may have been progressing at an optimal rate, allowing no opportun-
ity for a praise effect to appear. In accordance with the general guidelines of
analyzing both variables separately as well as in combination, the next
experiment reversed the order of the introduction of variables in a second
knife phobic patient (Leitenberg, 1973).
Once again, the target behavior was the amount of time the subject was
able to remain in the presence of the knife. The design replicated the first
experiment, with the exception of the elimination of the last phase. Thus the
design can be represented as B-BC-B-A-B-BC. In this experiment, however, B
refers to praise or verbal reinforcement and C represents feedback of amount
of time looking at the knife, which is just the reverse of the last experiment.
In this subject, little progress was observed during the first verbal reinforce-
ment phase (see Figure 6-10). However, when feedback was added to praise in
the second phase, performance increased steadily. Interestingly, this rate of
improvement was maintained when feedback was removed. After a sharp
gain, performance stabilized when both feedback and praise were removed.
Once again, the introduction of praise alone did not produce any further
improvement. The addition of feedback to praise for the second time in the
experiment resulted in marked improvement in the knife phobic. Direct
replication of this experiment with 4 additional subjects, each with a different
phobia, produced similar results. That is, praise did not produce improve-
ment when initially introduced, but the addition of feedback resulted in
marked improvement. In several cases, however, progress seemed to be
maintained in praise after feedback was withdrawn from the package, as in
Figure 6-10. In fact, feedback of progress, in its various forms, has come to
be a major motivational component within exposure-based programs for
phobia (Mavissakalian & Barlow, 1981b).
The overall results of the interaction analysis indicate that feedback is the
most active component because marked improvement occurred during both
feedback alone and feedback plus praise phases. Praise alone had little or no
effect although it was capable of maintaining progress begun in a prior
feedback phase in some cases. Similarly, praise did not add to the therapeutic
effect when combined with feedback in the first subject. Accordingly, a more
efficient treatment package for phobics would emphasize the feedback or
knowledge-of-results aspect and deemphasize or possibly eliminate the social
reinforcement component. These results have implications for treatments of
FEEDBACK NO FEEDBACK PRAISE FEEDBACK & PRAISE

& PRAISE NO PRAISE
< 140
O
z
8 120
z 100
Si
80
/
FIGURE
g ec 6^6
_
10 11 12
SESSIONS (BLOCKS OF
13
FIVE)
15 16 17
et
6-10. (Figure 1, from: Leitenberg, H. [1973]. Interaction designs. Paper read at the
18 19 20 21 22 23
American Psychological Association, Montreal, August. Reproduced by permission.)
phobics by other procedures such as systematic desensitization, where knowl-

edge of results provided by self-observation of progress through a discrete
hierarchy of phobic situations is a major component.
The interaction of reinforcement and feedback was also tested in a series of
subjects with anorexia nervosa (Agras et al., 1974). From the perspective of
interaction designs, the experiment is interesting because the contribution of a
third therapeutic variable, labeled size of meals, was also analyzed. To
illustrate the interaction design strategy, several experiments from this series
will be presented. All patients were hospitalized and presented with 6,000
calories per day, divided into four meals of 1,500 calories each. 1\vo measures
of eating behavior weight and caloric intake were recorded. Patients were
also asked to record number of mouthfuls eaten at each meal. Reinforcement
consisted of granting privileges based on increases in weight. If weight gain
exceeded a certain criterion, the patient could leave her room, watch televi-
sion, play table games with the nurses, and so on. Feedback consisted of
providing precise information on weight, caloric intake, and number of
mouthfuls eaten. Specifically, the patient plotted on a graph the information
that was provided by hospital staff.
In one experiment the effect of reinforcement was examined against a

background of feedback. The design can be represented as B-BC-BC'-BC,
where B is fefedback, C is reinforcement, and C
is noncontingent reinforce-
ment. During the first feedback phase (labeled baseline on the graph), slight
gains in caloric intake and weight were noted (see Figure 6-11). When
reinforcement was added to feedback, caloric intake and weight increased
sharply. Noncontingent reinforcement produced a drop in caloric intake and
a slowing of weight gain, while reintroduction of reinforcement once again
produced sharp gains in both measures. These data contain hints of an
Noncontingent
Base Line Reinforcement Reinforcement Reinforcement
45 4,000
Weight t ,
Caloric
Intake o -o
- 3,500
o
43 - 3,000 o
2,500
- 2.000
/V^l
I
15 30
Days
6e gfi. ic
FIGURE 6-11. Data from an experiment examining the effect of positive reinforcement in the
absence of negative reinforcement (Patient 3). (Figure 2, p. 281, from: Agras, W. S., Barlow, D.
H., Chapin, H, N., Abel, G. G., & Leitenberg, H. [1974]. Behavior modification of anorexia
nervosa. Archives of General Psychiatry, 30, 279-286. Copyright 1974 American Medical Asso-
ciation. Reproduced by permission.)
interaction, in that caloric intake and weight rose slightly during the first
feedback phase, a finding that replicated two earlier experiments. The addi-
tion of reinforcement, however, produced increases over and above those for
feedback alone. The drop and subsequent rise of caloric intake and rate of
weight gain during the next two phases demonstrated that reinforcement is a
when combined with feedback.
controlling variable
These data only hint at the role of feedback in this study, in that some
improvement occurred during the initial phase when feedback alone was in
effect. Similarly, we cannot know from this experiment the independent
effects of reinforcement because was not analyzed separately. To
this aspect
accompHsh this, two experiments were conducted where feedback was intro-
duced against a background of reinforcement. Only one experiment will be
presented, although both sets of data are very similar. The design can be
represented as A-B-BC-B-BC, where A is baseline, B is reinforcement, and C
is feedback (see Figure 6-12). It should be noted that the patient continued to
be presented with 6,000 calories throughout the experiment, a point to which
we will return later. During baseline, in which no reinforcement or feedback
was present, caloric intake actually declined. The introduction of reinforce-
Reinforcement Reinforcement Reinforcement

Base Line i Reinforcement & Feedback & Feedback
40-
4,000
3,000
oP
<o
20
^ o
2.2.
HE
2,000
1.000
40 50 60
Days
FIGURE 6-12. Data from an experiment examining the effect of feedback on the eating behavior
of a patient with anorexia nervosa (Patient 5). (Figure 4, p. 283, from: Agras, W. S., Barlow, D.
H., Chapin, H. N., Abel, G. G., & Leitenberg, H. [1974]. Behavior modification of anorexia
nervosa. Archives of General Psychiatry, 30, 279-286. Copyright 1974 American Medical Asso-
ciation. Reproduced by permission.)
ment did not result in any increases; in fact, a slight decline continued.
Adding feedback to reinforcement, however, produced increases in weight
and caloric intake. Withdrawal of feedback stopped this increase, which
began once again when feedback was reintroduced in the last phase.
With this experiment (and its replications) it becomes possible to draw
conclusions about the nature of what is in this case a complex interaction.
When both variables were presented alone, as in the initial phases in the
respective experiments, reinforcement produced no increases, but feedback
produced some increase. When presented in combination, reinforcement
added to the feedback effect and, against a background of feedback, became
the controlling variable, in that caloric intake decreased when contingent
reinforcement was removed. Feedback, however, also exerted a controlling
effect when was removed and reintroduced against a background of rein-
it
forcement. Thus, it seems that feedback can maximize the effectiveness of
reinforcement to the point where it is a controlling variable. Feedback alone,

however, is capable of producing therapeutic results, which is not the case
with reinforcement. Feedback, thus, is the more important of the two vari-
ables, although both contribute to treatment outcome.
It was noted earlier that the contribution of a third variable size of
meals
was also examined within the context of this interaction. In keeping
with the guidelines of analyzing each variable separately and in combination
with other variables, phases were examined when the large amount of 6,000
calories was presented without the presence of either feedback or reinforce-
ment. The baseline phase of Figure 6-12 represents one such instance. In this
phase caloric intake declined steadily. Examination of other baseline phases in
the replications of this experiment revealed similar results. To complete the
interaction analysis size of meal was varied against a background of both
feedback and reinforcement. The design can be represented as ABC- ABC '-
ABC, where A is feedback, B is reinforcement, C is 6,000 calories per day,
and Cis 3,000 calories per day.
Under this condition, size of meal did have an effect, in that more was
eaten when 6,000 calories were served than when 3,000 calories were pre-
sented (see Figure 6-13). In terms of treatment, however, even large meals
were incapable of producing weight gain in those phases where it was the only
therapeutic variable. Thus this variable is not as strong as feedback. The
authors concluded this series by summarizing the effects of the three variables
alone and in combination across five patients:
Thus large meals and reinforcement were combined in four experimental phases
and weight was lost in each phase. On the other hand, large meals and feedback
were combined in eight phases and weight was gained in all but one. Finally, all
three variables (large meals, feedback, and reinforcement) were combined in 12
phases and weight was gained in each phase. These findings suggest that informa-
o
6.000 Calories 3.000 Calories 6.000 Caiofies

Served Served Served
3.000 J
.--'--
2.800
o- -<f
2.600
f---
2.400
2.200.
2.000
t
Days
FIGURE 6-13. The effect of varying the size of meals upon the caloric intake of a patient with
anorexia nervosa (Patient 5). (Figure 5, p. 285, from: Agras, W. S., Barlow, D. H., Chapin, H.
N., Abel, G. G., & Leitenberg, H. [1974]. Behavior modification of anorexia nervosa. Archives
of General Psychiatry, 30, 279-286. Copyright 1974 American Medical Association. Reproduced
by permission.)
tional feedback is more important in the treatment of anorexia nervosa than

positive reinforcement, while serving large meals is least important. However, the
combination of all three variables seems most effective. (Agras et al., 1974,
p. 285)
As in the phobic series, the juxtaposition of variables within the general

framework of analyzing each variable separately and in combination pro-
vided information on the interaction of these variables.
Let us now consider two more recent applications of the beginnings of an
interaction design strategy in order to illustrate why they are incomplete at
this point in time, in contrast with the experiments described above. One
example is the evaluation of cognitive strategies (M. E. Bernard et al., 1983)
and the other is concerned with the possible combined effects of drugs and
behavior therapy (Rapport, Sonis, Fialkov, Matson, & Kazdin, 1983). M. E.
Bernard et al. (1983) evaluated the effects of rational-emotive therapy (RET)
and self-instructional training (SIT) in an A-B-A-B-BC-B-BC-A design with
follow-up. The subject was a 17-year old, overweight female who suffered
from trichotillomania (i.e., chronic hair pulling), especially while studying at
home. Throughout the study the subject self-monitored time studying and
number of hairs pulled out (deposited in an envelope). The dependent vari-
able was the ratio of hairs pulled out per minute of study time.
In baseline (A) the subject simply self-monitored. During the B phase, RET
was instituted, followed by a return to baseline (A) and reintroduction of
RET (B). In the next phase, (BC), SIT, consisting of problem-solving dia-
logues, was added to RET Then, SIT was removed (B) and subsequently
reintroduced (BC). In the last phase (A) all treatment was removed, and then
follow-up was conducted.
Results of this study appear in Figure 6-14. The first four phases comprise
an A-B-A-B analysis and do appear to confirm the controlling effects of RET
in reducing hair pulling. However, at this point the subject, albeit improved,
still was engaging in the behavior a significant proportion of the time.
Numbtrof 1.8 B ! BC BC ; A
hairs pulled
Up
out per "'fi"
minute of
''*
study time
1.2
1.0 H
0.8
0.6
0.4
0.2-
n n M lA
12
iii lii ilii II
3
i|iiiiif
4
i
5
iiii|i
6
I II iliii 1^1
7 8 9 10 11 '
12 13 14 15 20 36
Weeks
Note: 'Subject did not study
FIGURE 6-14. The number of hairs pulled out per minute of study time over baseline treatment
and follow-up phases. Missing data (*) reflect times when the subject did not study. (Figure 1, p.
277, from: Bernard, M. E., Kratochwill, T. R., & Keefauver, L. W. [1983]. The effects of rational-
emotive therapy and self-instructional training on chronic hair pulling. Cognitive Therapy and
Research, 7, 273-280. Copyright 1983 Plenum Publishing Corporation. Reproduced by permission.)
Phases 4-7 represent the interaction portion of the design (B-BC-B-BC). In

Phase 5, addition of SIT to RET yielded additional improvement to near zero
levels. When SIT then was removed in B, a moderate return of hair pulling
was noted, which was again decreased to zero levels when SIT was added
(BC). These gains subsequently held up in the final A phase and follow-up.
Although these data seem to confirm the therapeutic effect of SIT above
and beyond that obtained by RET alone, the reader should be aware of two
possible problems. First, all data are self-monitored and subject to experi-
mental demand characteristics. Second, the BC phases are longer than each B
phase; thus, there may be a possible confound with time. That is, a portion of
the extra effect brought about by combining RET and SIT simply may be due
to increased time of the combined treatment. However, this is unlikely, given
the long-standing nature of the disorder.
In addition, a study of the interactional effects is not yet possible because
SIT was not analyzed in isolation, but only against a background of RET.
Thus it is possible that introducing SIT first would have a somewhat different
effect, as would adding RET to SIT rather than the other way around, as in
this experiment. While this is a noteworthy beginning, a more thorough
evaluation of the interaction of SIT and RET awaits further experimental
inquiry. Ideally, this experiment would be directly replicated at least twice,
followed by the same experiment with SIT introduced first in three additional
subjects. But we do not live in an ideal world, and trichotillomanics are few
and far between.
Our final example of an interaction design involves a BC-BC -B-BC-B-BD '
design, with two drugs (sodium valproate, carbamazepine) and one behav-
ioral technique (differential reinforcement of other behavior [DRO]) evalu-
ated (Rapport et al., 1983). The subject in this experimental analysis was a
13.7-year-old mentally retarded female who suffered from seizures and exhib-
ited aggressive behavior toward others. She had a long history of hospitaliza-
tions and had been tried on a large variety of medications, but with little
success. Aggressive behaviors included grabbing, biting, kicking, and hair
pulling. Aggression was the primary dependent measure in this study and was
recorded by inpatient staff with a high degree of interrater agreement (range
= 9207o-100<ô).
The subject received carbamazepine (4(X) mg, t.i.d.) in each phase of the
study. In the first phase (BC) she received sodium valproate (1,2(X) mg) as
well. This was gradually withdrawn in phase 2 (BC) and removed altogether
in Phase 3 (B). In Phase 4 (BD) a DRO procedure (edible reinforcements
delivered contingently for 15 -minute time periods in which no aggression
occurred; then increased to 30 and 60 minutes) was added to carbamazepine.
DRO was discontinued in Phase 5 (B) and then reinstated in Phase 6 (BD).
Examination of Figure 6-15 shows a high rate of aggressive incidents (mean
= 15 per day) in the first phase (BC), which decreased (mean = 3 per day)
CARBAMAZEPINC
SODIUM WITH-
VALHlOATE CMAWN
48-, ^ ^
NUMBER OF
INCIDENTS
DAYS
FIGURE 6-15. Data points represent the daily frequency of aggressive behavior during the child's
hospital stay. (Arrows indicate days when nocturnal enuresis was observed.) (Figure 1, p. 262,
from: Rapport, M. D., Sonis, W. A., Fialkov, M. J., Matson, J. L., & Kazdin, A. E. [1983].
Carbamazepine and behavior therapy for aggressive behavior: Treatment of a mentally retarded,
postencephalic adolescent with seizure disorder. Behavior Modification, 7, 255-264. Copyright
1983 by Sage Publication. Reproduced by permission.)
when sodium valproate was withdrawn (BC). However, when the patient was
totally withdrawn in Phase 3 (B), aggression rose to a mean of 10 a day.
Institution of DRO in Phase 4 (BD) led to a dramatic decrease (0), rose to 4-8
when DRO was withdrawn (B) on days 63 and 64, and gradually decreased to
zero againwhen DRO was reintroduced (BD) on days 65-91.
Although there was only a 2-day withdrawal of DRO procedures, this is
truly justified given the aggressive nature of the behavior being observed.
Indeed, it is quite clear that although the drug, carbamazepine had a minor
role in controlling aggression, the addition of DRO was the major controUing
force. Moreover, effectiveness of DRO allowed the subject to be discharged
to her family, with DRO procedures subsequently implemented at school in
order to ensure generalization of treatment gains.
Once again, replication on additional
and a subsequent reordering
subjects
of the experimental strategy so that DRO
was analyzed separately and then
combined with the drug would be necessary for a more complete study of
interactions. Finally, the nature of this experimental strategy deserves some
comment, particularly when compared to other strategies attempting to
answer the same questions. First, in any experiment there are more things
interacting with treatment outcome than the two or more treatments or
variables under question. Foremost among these are client variables. This, of
course, is the reason for direct replication (see chapter 10). If the experimental
operations are replicated (in this example the interaction), despite the dif-
ferent experiences clients bring with them to the experiment, then one has
increasing confidence in the generality of the interactional finding across
subjects.
Second, as pointed out in chapter 5 and discussed more fully in chapter 8,
the latter phases of these experiments are subject to multiple-treatment inter-
ference. In other words, the effect of a treatment or interaction in the latter
phases may depend to some extent on experience in the earlier phases. But if
the interaction effect is consistent across subjects, both early and late in the
experiment, and across different "orders" of introduction of the interaction,

as in the first two examples described in this section (Agras et al., 1974;
Leitenberg et al., 1968), then one has greatly increased confidence in both the
fact and the generality of the effect. As with A-B-A withdrawal designs,
however, the most easily generalizable data from the experiment to applied
situations are the early phases before multiple treatments build up. This is
because the early phase most closely resembles the applied situation, where
the treatment would also be introduced and continued without a prior back-
ground of several treatments.
The other popular method of studying interactions is the between-group
factorial design. In this case, of course, one group would receive both
Treatments A and B, while two other groups would receive just A or just B.
(If the factorial were complete, another group would receive no treatment.)
Here treatments are not delivered sequentially, but the more usual problems
of intersubject variability, inflexibility in altering the design, infrequent mea-
surement, determination of results by statistical inference, and difficulties
generalizing to the individual obtain, as discussed in chapter 2. Each approach
to studying interactions obviously has its advantages and disadvantages.
6.7. CHANGING CRITERION DESIGN

The changing-criterion design, despite the fact that it has not to date
enjoyed widespread application, is a very useful strategy for assessing the
shaping of programs to accelerate or decelerate behaviors (e.g., increase
interactions in chronic schizophrenics; decrease motor behavior in overactive
children). As a specific design strategy, it incorporates A-B design features on
a repeated basis. After initial baseline measurement, treatment is carried out
until a preset criterion is met, and stability at that level is achieved. Then, a
more stringent criterion is set, with treatment applied until this new level is
met. If baseline is A and the first criterion is B, when the new criterion is set
the former B serves as the new baseline (A') with B' as the second criterion.
This continues in graduated fashion until the final target (or criterion) is
achieved at a stable level. As noted by Hartmann and Hall (1976), "Thus,

each phase of the design provides a baseline for the following phase. When
the rate of the target behavior changes with each stepwise change in the
criterion, therapeutic change is replicated and experimental control is demon-
strated" (p. 527).
This design, by its very nature, presupposes ". .a close correspondence
.
between the criterion and behavior over the course of the intervention phase"
(Kazdin, 1982b, p. 160). When such close correspondence fails to materialize,
with stability not apparent in each successive phase, unambiguous interpreta-
tions of the data are not possible. One solution, of course, is to partially
withdraw treatment by returning to a lower criterion, followed by a return to
the more stringent one (as in a B-A-B withdrawal design). This adds experi-
mental confidence to the treatment by clearly documenting its controlling
effects. Or, on a more extended basis, one can reverse the procedure and
experimentally demonstrate successive increases in a targeted behavior fol-
lowing initial demonstration of successive decreases. This is referred to as bi-
directionality. Finally, Kazdin (1982b) pointed out that some experimenters
have dealt with the problem of excessive variability by showing that the mean
performance over adjacent subphases reflects the stepwise progression.
None of the aforementioned solutions to variability in the subphases is
ideal. Indeed, it behooves researchers using this design to demonstrate close
correspondence between the changing criterion and actually observed behav-
ior. Undoubtedly, as this design is employed more frequently, more elegant
solutions to this problem will be found.

Hartmann and Hall (1976) presented an excellent illustration of the chang-
ing-criterion design in which a smoking-deceleration program was evaluated.
Baseline level of smoking is depicted in panel A of Figure 6-16. In the next
phase (B treatment), the criterion rate was set at 95% of the baseline rate (i.e.,
46 cigarettes a day). An increasing response cost of $1 was established for
smoking an additional cigarette (i.e., Number 47) and $2 for Number 48, and
on and on. An escalating bonus of $0.10 a cigarette was established if the
subject smoked less than the criterion number set. Subsequently, in phases
C-G, the criterion for each succeeding phase was established at 94% of the
previous one.
Careful examination of Figure 6-16 clearly indicates the success of treat-
ment in reducing cigarette smoking by 2% or more from each preceding
phase. Further, from the experimental analysis perspective, there were six
replications of the contingencies applied. In each instance, experimental
control was documented, with the treatment phase serving as baseline with
respect to the decreasing criterion for the next phase, and so on.
Related to the changing criterion design is a strategy that Hayes (1981) has
referred to as i\iQ periodic-treatments design. This design, at our writing, has
been used most infrequently and really only has a quasi-experimental basis.
DAYS: I 8 15 22 29 36 43 50 57 64 78 85
PHASES: BASELINE TREATMENT
FIGURE 6-16. Data from a smoking-reduction program used to illustrate the stepwise criterion
change design. The solid horizontal lines indicate the criterion for each treatment phase. (Figure
2, p. 529, from: Hartmann, D. P., & Hall, R. V. [1976]. The changing criterion design. Journal of
Applied Behavior Analysis, 9, 527-532. Copyright 1976 by Soc. for the Experimental Analysis of
Behavior. Reproduced by permission.)
Indeed, it is best suited for application in the private-practice setting (Barlow

et al., 1983).
The logic of the design is quite simple. Frequently, marked improvements

in a targeted behavior are seen immediately after a given therapy session. If
this is plotted graphically, one can begin to see the relationship between the
session (loosely conceptualized as an A phase) and time between sessions
(loosely conceptualized as B phases). Thus, if steady improvement occurs, the
scalloped display seen in the changing criterion design also will be observed
here.
Hypothetical data for this design possibility are presented in Figure 6-17.
But, as Hayes (1981) noted:
These data do not show what about the treatment produced the change (any
more than an A-B-A design would). It may be therapist concern or the fact that
the client attended a session of any kind. These possibilities would then need to
be eliminated. For example, one could manipulate both the periodicity and
nature of treatment. If the periodicity of behavior change was shown only when
a particular type of treatment was in place, this would provide evidence for a
more specific effect, (p. 203)
FIGURE 6-17. The periodic treatments effect is shown on hypothetical data. (Data are graphed in
raw data form in the top graph.) Arrows on the abscissa indicate treatment sessions. This
apparent B-only graph does not reveal the periodicity of improvement and treatment as well as
the bottom graph, where each two data points are plotted in terms of the difference from the
mean of the two previous data points. Significant improvement occurs only after treatment. Both
graphs show an experimental effect; the lower is merely more obvious. (Figure 3, p. 202, from:
Hayes, S. C. [1981]. Single case experimental design and empirical clinical practice. [1981].
Journal of Consulting and Clinical Psychology, 49, 193-211. Copyright 1981 by American
Psychological Association. Reproduced by permission.)
CHAPTER 7
Multiple Baseline Designs
7.1. INTRODUCTION
The use of sequential withdrawal or reversal designs is inappropriate when
treatment variables cannot be withdrawn or reversed due to practical limita-
tions, ethical considerations, or problems in staff cooperation (Baer et al.,
1968; Barlow et al., 1977; Barlow & Hersen, 1973; Birnbauer, Peterson, &
Solnick, 1974; Hersen, 1982; Kazdin & Kopel, 1975; Van Hasselt & Hersen,
1981). Practical limitations arise when carryover effects appear across adja-
cent phases of study, particularly in the case of therapeutic instructions
(Barlow & Hersen, 1973). A similar problem may occur when drugs with
known long-lasting effects are evaluated in single-case withdrawal designs.
Despite discontinuation of medication in the withdrawal (placebo) phase,
active agents persist psychologically and, with the phenothiazines, traces have
been found in body tissues many months later (Goodman & Oilman, 1975).
Also, when multiple behaviors within an individual are targeted for change,
withdrawal designs may not provide the most elegant strategy for such
evaluation.
Ethical considerations are of paramount importance when the treatment
variable is effective in reducing self- or other-destructive behaviors in sub-
jects. Here the withdrawal of treatment is obviously unwarranted, even for
brief periods of time. Related to the problem of undesirable behavior is the
matter of environmental cooperation. Even if the behavior in question does
not have immediate destructive effects on the environment, if it is considered
to be aversive (i.e., by teachers, parents, or hospital stafO the experimenter
will not obtain sufficient cooperation to carry out withdrawal or reversal of
treatment procedures. Under these circumstances, it is clear that the applied
clinical researcher must pursue the study using different experimental strate-
gies. In still other instances, withdrawal of treatment, despite absence of
209
harm to the subject or others in his or her environment, may be undesirable

because of the severity of the disorder. Here the importance of preserving
therapeutic gains is given priority, especially when a disorder has a lengthy
history and previous efforts at remediation have failed.
Multiple baseline designs and their variants and alternating treatment
designs (see chapter 8) have been used by applied clinical researchers with
increased frequency when withdrawals and reversals have not been feasible.
Indeed, since publication of the first edition of this book in 1976, we find that
the pages of our behavioral journals are replete with the innovative use of the
multiple baseline strategy, for individuals as well as groups of subjects. A list
of some recent, published examples of this design strategy appears in Table

7-1.
In this chapter we will examine in detail the rationale and procedures for
multiple baseline designs. Examples of the three principal varieties of multiple
baseline strategies will be presented for illustrative purposes. In addition, we
will consider the more recent varieties and permutations, including the non-
concurrent multiple baseline design across subjects, the multiple-probe tech-
nique, and the changing criterion design. Finally, the application of the
multiple baseline across subjects in drug evaluations will be discussed.
7.2 MULTIPLE BASELINE DESIGNS

The rationale for the multiple baseline design first appeared in the applied
behavioral literature in 1968 (Baer et al.), although a within-subject multiple
baseline strategy had been used previously by Marks and Gelder (1967) in
their assessment of electrical aversion therapy for a sexual deviate. Baer et al.
(1968) point out that:
In the multiple-baseline technique, a number of responses are identified and

measured over time to provide baselines against which changes can be evaluated.
With these baselines established, the experimenter then applies an experimental
variable to one of the behaviors, produces a change in it, and perhaps notes little
or no change in the other baselines, (p. 94)
Subsequently, the experimenter applies the same experimental variable to a

second behavior and notes rate changes in that behavior. This procedure is
continued in sequence until the experimental variable has been applied to all
of the target behaviors under study. In each case the treatment variable is
usually not applied until baseline stability has been achieved.

Baseline and subsequent treatment interventions for each targeted behavior
can be conceptualized as separate A-B designs, with the A phase further
extended for each of the succeeding behaviors until the treatment variable is
Multiple Baseline Designs 211
finally applied. The experimenter is assured that the treatment variable is
effective when a change in rate appears after its application while the rate of
concurrent (untreated) behaviors remains relatively constant. A basic as-
sumption is that the targeted behaviors are independent from one another. If
they should happen to covary, then the controlling effects of the treatment
variable are subject to question, and limitations of the A-B analysis fully
apply (see chapter 5).
The independence of behaviors within a single subject raises some

issue of
interesting problems from an experimental standpoint, particularly if the
experimenter is involved in a new area of study where no precedents apply.
The experimenter is then placed in a position where an a priori assumption of
independence cannot be made, thus leaving an empirical test of the proposi-
tion. Leitenberg (1973) argued that:
If general effects on multiple behaviors were observed after treatment had been
applied to only one, there would be no way to clearly interpret the results. Such
results may reflect a specific therapeutic effect and subsequent response general-
ization, or they may simply reflect non-specific therapeutic effects having little to
do with the specific treatment procedure under investigation, (p. 95)
In some cases, when independence of behaviors is not found, application

of the alternating treatment design may be recommended (see chapter 8). In
other cases, application of the multiple baseline design across different sub-
jects might yield useful information. Surprisingly, however, in the available
published reports the problem of independence has not been insurmountable
(Leitenberg, 1973). Although problems of independence of behaviors ap-
parently have been infrequently reported, some of the solutions referred to
may not be viable if the experimenter is interested in targeting several behav-
iors within the same subject for sequential modification.
In attempting to prevent occurrence of the problem in interpretation when
"onset of the intervention for one behavior produces general rather than
specific changes," Kazdin and Kopel (1975) offered three specific recommen-
dations. The first, of course, is to include baselines that topographically are as
distinct as possiblefrom one another. But this may be difficult to ascertain on
an a priori basis. The second is to use four or more baselines rather than two
or three. However, there always is the statistical probability that interdepen-
dence will be enhanced with a larger number. The third (on an ex post facto
basis) is and then reintroduce treatment for the correlated
to withdraw
B-A-B design), thus demonstrating the controUing effects
baseline (as in the
ovcLthat targeted response. Even though the multiple baseline strategy was
implemented in the first place to avoid treatment withdrawal, as in the A-B-A-
B design, the rationale for such temporary (or partial) withdrawal in the
multiple baseline design across behaviors seems reasonable when indepen-
dence of baselines cannot be documented. But, as noted by Hersen (1982),

"A problem with the Kazdin and Kopel solution is that in the case of
instructions a true reversal or withdrawal is not possible. Thus their recom-
mendations apply best to the assessment of such techniques as feedback,
reinforcement, and modeling" (p. 191).
The multiple baseline design is considerably weaker than the withdrawal
design, as the controlling effects of the treatment on each of the target
behaviors are not directly demonstrated (e.g., as in the A-B-A design). As
noted earlier, the effects of the treatment variable are inferred from the
untreated behaviors. This raises an issue, then, as to how many basehnes are
needed before the experimenter is able to establish confidence in the control-
ling effects of his or her treatment. A number of interpretations have ap-
peared in the literature. Baer et al. (1968) initially considered this issue to be
an "audience variable" and were reluctant to specify the minimum number of
baselines required. Although theoretically only a minimum of two baselines is
needed to derive useful information. Barlow and Hersen (1973) argued that
". . the controlling effects of that technique over at least three target
.
behaviors would appear to be a minimum requirement" (p. 323). Similarly,

Wolf and Risley (1971) contended that "While a study involving two baselines
can be very suggestive, a set of replications across three or four basehnes may
be almost completely convincing" we would recom-
(p. 316). At this point,
mend a minimum of and experimental
three to four baselines if practical
considerations permit. As previously noted, Kazdin and Kopel (1975) recom-
mended four or more baselines.
Although demonstration of the controlling effects of a treatment variable
is obviously weaker in the multiple baseline design, a major advantage of this
strategy is that it fosters the simultaneous measurement of several concurrent

target behaviors. This is most important for at least two major reasons. Firsts
the monitoring of concurrent behaviors allows for a closer approximation to
naturalistic conditions, where a variety of responses are occurring at the same
time. Second, examination of concurrent behaviors leads to an analysis of
covariation among the targeted behaviors. Basic researchers have been con-
cerned with the measurement of concurrent behaviors for some time (Cata-
nia, 1968; Herrnstein, 1970; Honig, 1966; G. S. Reynolds, 1968; Sidman,
1960). Applied behavioral researchers also have evidenced a similar interest
(Kazdin, 1973b; Sajwaj et al., 1972; TVardosz & Sajwaj, 1972). Kazdin
(1973b) underscored the importance of measuring concurrent (untreated)
behaviors when assessing the efficacy of reinforcement paradigms in applied
settings. He stated that:
While changes in target behaviors are the raison d'etre for undertaking treatment
or training programs, concomitant changes may take place as well. If so, they
should be assessed. It is one thing to assess and evaluate changes in a target
a
behavior, but quite another to insist on excluding nontarget measures. It may be

that investigators are short-changing themselves in evaluating the programs,
(p. 527)
As mentioned earlier, there are three basic types of multiple baseline

designs. In the first the multiple baseline design across behaviors the same
treatment variable is applied sequentially to separate (independent) target
behaviors in a single subject. A possible variation of this strategy, of course,
involves the sequential application of a treatment variable to targeted behav-
iors for an entire group of subjects (see Cuvo & Riva, 1980). In this connec-
tion, R. V. Hall, Cristler, Cranston, and Tlicker (1970) note that ". . . these
multiple baseline designs apply equally well to the behavior of groups if the
behavior of the group members is summed or averaged, and the group is
treated as a single organism" (p. 253). However, in this case the experimenter
would also be expected to present data for individual subjects, demonstrating
that sequential treatment applications to independent behaviors affected
most subjects in the same direction.
In the second design
the multiple baseline design across subjects
particular treatment is applied in sequence across matched subjects presum-
ably exposed to "identical" environmental conditions. Thus, as the same
treatment variable is applied to succeeding subjects, the baseline for each
subject increases in length. In contrast to the multiple baseline design across
behaviors (the within-subject multiple baseline design), in the multiple base-
line design across subjects a single targeted behavior serves as the primary
focus of inquiry. However, there is no experimental contraindication to
monitoring concurrent (untreated) behaviors as well. Indeed, it is quite likely
that the monitoring of concurrent behaviors will lead to additional findings of
merit.
As with the multiple baseline design across behaviors, a possible variation
of the multiple baseline design across subjects involves the sequential applica-
tion of the treatment variable across entire groups of subjects (see Domash et
al., But here, too, it behooves the experimenter to show that a large
1980).
majority of individual subjects for each group evidenced the same effects of
treatment.
We might note that the multiple baseline design across subjects has also
been labeled a time-lagged control design (Gottman, 1973; Gottman, McFall,
& Barnett, 1969). In fact, this strategy was followed by Hilgard (1933) some
50 years ago in a study in which she examined the effects of early and delayed
practice on memory and motoric functions in a set of twins (method of co-
twin control).
In the third design the multiple baseline design across settings a partic-
ular treatment is applied sequentially to a single subject or a group of subjects
across independent situations. For example, in a classroom situation, one
might apply time-out contingencies for unruly behavior in sequence across

different classroom periods.The baseline period for each succeeding class-
room period, then, increases in length before application of the treatment. As
in the across-subjects design, assessment of treatment is usually based on rate
changes observed in a selected target behavior. However, once again the
monitoring of concurrent behaviors might prove to be of value and should be
encouraged where possible.
To recapitulate, in the multiple baseline design across behaviors, a treat-
ment variable is applied sequentially to independent behaviors within the
same subject. In the multiple baseline design across subjects, a treatment
variable is applied sequentially to the same behavior across different but
matched subjects sharing the same environmental conditions. Finally, in the
multiple baseline design across settings, a treatment variable is applied se-
TABLE 7-1. Recent Examples of Multiple Baseline Designs
STUDY DESIGN SUBJECTS
Alford, Webster, & Sanders (1980) Across behaviors Sexual deviate

AUison & Ayllon (1980) Across subjects Sports team members
Across behaviors
Barmann, Katz, O'Brien, & Across subjects Developmentally disabled enuretics
Beauchamp (1981)
Bates (1980) Across behaviors Retarded adults
Bellack, Hersen, & Tbrner (1976) Across behaviors Schizophrenics
Berler, Gross, & Drabman (1982) Across behaviors Learning disabled children
M. R. Bornstein, Bellack, & Across behaviors Unassertive children
Hersen (1977)
M. R. Bornstein, Bellack, & Across behaviors Aggressive child inpatients
Hersen (1980)
Breuning, O'Neill, & Ferguson Across subjects Retarded adults
(1980) (groups)
Bryant & Budd (1982) Across subjects Preschoolers
Burgio, Whitman, & Johnson Across subjects Retarded children
(1980)
Cuvo & Riva (1980) Across behaviors Retarded children
Domash et al. (1980) Across subjects Police officers
(groups)
Dunlap & Koegel (1980) Across behaviors Autistic children
(groups)
Dyer, Christian, & Luce (1982) Across subjects Autistic children
Egel, Richman, & Koegel (1981) Across subjects Autistic children
Epstein et al. (1981) Across subjects Families of dialectic children
Fairbank & Keane (1982) Across settings Vietnam veteran
C. Hall, Sheldon-Wildgen, & Across behaviors Retarded adults
Sherman (1980) (scenes)
Halle, Baer, & Spradlin (1981) Across subjects Developmentally delayed children
Hay, Nelson, & Hay (1980) Across subjects Grade-schoolers
Hundert (1982) Across subjects Deaf children
R. T. Jones, Kazdin, & Haney Across subjects Third graders
(1981a)
R. T. Jones, Kazdin, & Haney Across subjects Third graders
(1981b) (Continued)
TABLE 7-1. Recent Examples of Multiple Baseline Designs (Continued)
STUDY DESIGN SUBJECTS
J. A. Kelly, Urey, & Patterson Across behaviors Psychiatric patients

(1980)
R. E. Kirchner et al. (1980) Across settings High-rate burglary areas
(groups)
Kistner, Hammer, Wolfe, Across subjects Grade-schoolers
Rothblum, & Drabman (1982) (groups)
Matson (1981) Across subjects Phobic retarded children
Matson (1982) Across behaviors Depressed retarded adults
Melin & Gotestam (1981) Across behaviors Geriatric patients
(groups)
OUendick (1981) Across settings Children with nervous tics
Poche, Brouwer, & Swearingen Across subjects Preschoolers
(1981)
Rosen & Leitenberg (1982) Across settings Anorexia nervosa patient
(meals)
Russo & Koegel (1977) Across behaviors Autistic child
Singh, Dawson, & Gregory (1980) Across settings Retarded female
Singh, Manning, & Angell (1982) Across subjects Retarded monozygotic twins
Slavin, Wodarski, & Blackburn Across subjects College dorm residents
(1981) (groups)
Stokes & Kennedy (1980) Across subjects Grade-schoolers
Stravynski, Marks, & Yule (1982) Across behaviors Neurotic outpatients
(groups)
Sulzer-Azaroff & deSantamaria Across subjects Industrial supervisors
(1980) (groups)
Van Biervliet, Spangler, & Across settings Retarded males
Marshall (1981) (groups)
Van Hasselt, Hersen, Kazdin, Across behaviors Blind adolescents
Simon, & Mastantuono (1983)
Whang, Fletcher, &. Fawcett (1982) Across subjects Counselor trainees
Wong, Gaydos, & Fuqua (1982) Across behaviors Mildly retarded pedophile
quentially to the same behavior across different and independent settings in

the same subject. Recently published examples of the three basic types of
multiple baseline strategies are categorized in Table 7-1 with respect to design
type and subject characteristics.
In the following three subsections we will illustrate the use of basic multiple
baseline strategies in addition to presenting examples of variations selected
from the child, chnical, behavioral medicine, and applied behavioral analysis
literatures.
Multiple baseline across behaviors
M. R. Bornstein, Bellack, and Hersen (1977) used a multiple baseHne

strategy (across behaviors) to assess the effects of social skills training in the
role-played performance of an unassertive 8-year-old male third grader (Tom)
whose passivity led to derision by peers. Generally, if he experienced conflict
with a peer, he cried or reported the incident to his teacher. Three target
behaviors were selected for modification as a resuh of role-played perfor-
mance in baseline: ratio of eye contact to speech duration, number of words,
and number of requests. In addition, independent evaluations of overall
assertiveness, based on role-played performance, were obtained. As can be
seen in Figure 7-1, baseline responding for targeted behaviors was low and
stable. Following baseline evaluation, Tom received 3 weeks of social skills
training consisting of three 15-30 minute sessions per week. These were
applied sequentially and cumulatively over the 3-week period. Throughout
training, six role-played scenes were used to evaluate the effects of treatment.
In addition, three scenes (on which the subject received no training) were used
to assess generalization from trained to untrained scenes.
The resuhs for training scenes appear in Figure 7-1. Examination of the
graph indicates that institution of social skills training for ratio of eye contact
to speech duration resulted in marked changes in that behavior, but rates for
number of words and number of requests remained constant. When social
skills training was applied to number of words itself, the rate for number of
requests remained the same. Finally, when social skills training was directly
applied to number of requests, marked changes were noted. Thus it is clear
that social skills training was effective in increasing the rate of the three target
behaviors, but only when treatment was applied directly to each. Indepen-
dence of the three behaviors and absence of generalization effects from one
behavior to the next facilitate interpretation of these data. On the other hand,
had nontreated behaviors covaried following application of social skills train-
ing, unequivocal conclusions as to the controlling effects of the training could
not have been reached without resorting to Kazdin and Kopel's (1975) solu-
tion to withdraw and reinstate the treatment.
The reader should also note in Figure 7-1 that, despite the fact that overall
assertiveness was not treated directly, independent ratings evinced gradual
improvement over the 3 -week period, with treatment gains for all behaviors
maintained in follow-up.
Examination of data for the untreated generalization scenes indicates that
similar results were obtained, confirming that transfer of training occurred
from treated to untreated items. Indeed, the patterns of data for Figures 7-1
and 7-2 are remarkably alike.
Liberman and Smith (1972) also used a multiple baseline design across
behaviors in studying the effects of systematic desensitization in a 28-year-
old, multiphobic female who was attending a day treatment center. Four
specificphobias were identified (being alone, menstruation, chewing hard
foods, dental work), and baseline assessment of the patient's self-report of
each was taken for 4 weeks. Subsequently, in vivo and standard systematic
desensitization (consisting of relaxation training and hierarchical presentation
of items in imagination) were administered in sequence to the four areas of
TRAINING SCENES
Bsin Social Skills Training Follow-up
5 7 9 11 2-4-
Probe Sessions Weeks
FIGURE 7-1. Probe sessions during baseline, social skills treatment, and follow-up for training
scenes for Tom. A multiple baseline analysis of ratio of eye contact while speaking to speech
duration, number of words, number of requests, and overall assertiveness. (Figure 3, p. 190,
from: Bornstein, M. R., Bellack, A. S., Hersen, M. [1977]. Social-skills training for unassertive
children: A multiple-baseline analysis. Journal of Applied Behavior Analysis, 10, 183-195.
Copyright 1977 by Society for Experimental Analysis of Behavior. Reproduced by permission.)
phobic concern. Specifically, in vivo desensitization was administered in

relation to fears of being alone and chewing hard foods, while fears of
menstruation and dental work were treated imaginally.
Results of this study, presented in Figure 7-3, indicate that the sequential
application of desensitization affected the particular phobia being treated,
GOOAUZATDN SCENES
Bsln. Social Skills Training Follow-up
X
^\A
J j
'
\^^^^
\
:a
J
AA^ \
5
^ 3
/
1 3 5 7 9 11 2-4-
FIGURE 7-2. Probe sessions during

baseline, social skills treatment, and follow-up for general-
ization scenes forTom. A multiple baseline analysis of ratio of eye contact while speaking to
speech duration, number of words, number of requests and overall assertiveness. (Figure 4, p.
191, from: Bornstein, M. R., Bellack, A. S., & Hersen, M. (1977]. Social-skills training for
unassertive children: A multiple-baseline analysis. Journal of Applied Behavior Analysis, 10,
183-195. Copyright 1977 by Society for the Experimental Analysis of Behavior. Reproduced by
permission.)
but no evidence of generalization to untreated phobias was noted. Indepen-

dence of the four target behaviors and rate changes when desensitization was
finally applied to each support the conclusion that treatment was effective
and that it exerted control over the dependent measures (self-reports of
degrees of fear). Although the authors argued that a positive set for improve-
r
BASELINE DESENSITIZATION
12-1 Being Alone '
// T-
\Z-{ Menstruation I-
y
8
o
Chewing
U n-T r r I
- I
//
Hard Foods
IIIIUI n *T r I
- -// T
12-1 ê/7/a/ Work
6-
1 2 3 4 5
ll
6 7 8
lill..
9 10 11 12 13 14 15
-//-
23
Weeks
FIGURE 7-3. Multiple baseline evaluation of desensitization in a single case with four phobias.
(Figure 1, p. 600, from: Liberman, R. P., & Smith, V. [1972]. A multiple baseline study of
systematic desensitization in a patient with multiple phobias. Behavior Therapy^ 3, 597-603.
Copyright 1972 by Association for the Advancement of Behavior Therapy. Reproduced by
permission.)
ment was maintained throughout all phases of study, the possibility that
expectancy of improvement and actual treatment effects were confounded
cannot be discounted, especially in light of the primary reliance on self-report
data. However, casually conducted behavioral observations corroborate self-
report data.
Despite the above-mentioned limitations, Liberman and Smith's (1972)
investigation is of interest from a number of standpoints. Firsts as most
multiple baseline studies emanate from the operant framework, this study
lends credence to the notion that nonoperant procedures (e.g., systematic
can be assessed in this paradigm. Second as the particular
desensitization) y
dependent measure (ratings of subjective fear on the Target Complaint Scale)

is based on the patient's self-report, it would appear that this type of single-
case research might easily be carried out in inpatient facilities and even in
consulting room practice (see chapter 3, section 3.2). Finally, the treatment
was fully implemented by a mental health paraprofessional who had only one
year's training in psychiatry.
In our next example of a multiple baseline design across behaviors, a
psychological measure (erectile strength as assessed with a penile gauge) was
used to determine efficacy of covert sensitization in the treatment of a 21-
year-old married male, admitted for inpatient treatment of exhibitionism and
obscene phone calling (Alford, Webster, & Sanders, 1980). History of exhibi-
tionism began at age 16, and obscene phone calling had taken place over the
previous year. During baseline assessment:
Audiotapes of both deviant and nondeviant sexual scenes were used to elicit
arousal during physiological monitoring sessions. Deviant stimulus material

included three tapes depicting various obscene phone calls . . and three tapes of
.
exhibitionism. . . . TWo nondeviant tapes . . . that depicted normal heterosexual

behavior were also used. . . . They consisted of verbal descriptions designed to
closely parallel the patient's own sexual behavior and fantasy, (p. 17)
These included one taped description of intercourse with his wife and another
with different sexual partners.
Covert sensitization sessions were conducted twice daily in the hospital at
various locations. This treatment consisted of imaginally pairing the deviant
sexual approach (i.e., obscene phone calls, exhibitionism) with aversive stim-
uli such as suffocation, nausea, and arrest. Each session involved 20 pairings
of the deviant scenarios with aversive imagery. Following baseline assess-
ment, covert sensitization was first applied to obscene phone calling and then
to exhibitionism. In addition to therapist-conducted treatment sessions, the
patient was instructed to use covert imagery on his own initiative whenever he
experienced deviant sexual urges.
Data for this multiple baseline analysis are presented in Figure 7-4. During
baseline evaluation, penile tumescence in response to tapes of obscene phone
calling and exhibitionism was quite high. Similarly, tumescence was above
75^0 in response to nondeviant tapes of sexual activity with females other
than his wife, but only slightly higher than 25% in response to lovemaking
with his wife.
Institution of covert sensitization for obscene phone calling resulted in
marked diminution in penile responsivity to taped descriptions of that behav-
ior, eventually resulting in only a negligible response. However, such treat-
ment also appeared to affect changes in penile response to one of the
exhibitionism tapes (Ex. even though that behavior had not yet been
1),
specifically targeted. (We have here an instance where the baselines are not
independent from one another.) However, when treatment subsequently was
directed to exhibitionism itself, there was marked diminution in penile re-
COVERT SENSITIZATION
OPC 1
OBSCENE PHONE CALLING
o OPC 2
o OPC 3 GENERALIZATION
I 2
I
3 4| 5 6 |7
BSIN CS/OPC CS/OPC EX SACS
^
INPATIENT DISCHARGED 1 CS/OPCEX Ri
FIGURE 7-4. Percentage of full phone call (OPC) exhibitionistic (EX), and
erection to obscene
heterosexual stimuli (ND) during and follow-up phases. (Figure 1, p. 20,
baseline, treatment,
from: Alford, G. S., Webster, J. S., & Sanders, S. H. [1980]. Covert aversion of two interrelated
deviant sexual practices: Obscene phone calling and exhibitionism. A single case analysis.
Behavior Therapy, 11, 13-25. Copyright 1980 by Association for the Advancement of Behavior
sponse to tapes Ex. 2 and Ex. 3 in addition to continued decreases to tape Ex.
1. During the course of treatment, penile responsivity to nondeviant hetero-
sexual interactions remained high, increasing considerably with respect to
lovemaking with the wife.
The reader might note that "the patient was preloaded with 36 oz of beer 90
to 60 minutes prior to Assessments 10 and 11" (Alford et al., 1980, p. 19).
This was carried out inasmuch as he had claimed that alcohol had disinhibited
deviant sexuality. However, experimental data did not seem to confirm this.
One, 2-, and 10-month follow-up assessments indicated that all gains were
maintained, with the exception of decreased penile responsivity to taped
descriptions of intercourse with the wife. In addition, 10-month collateral
information from the patient's wife, parents, and attorney, as well as police,
court, and telephone company records revealed no incidents of sexual de-
viance.
Our illustration reveals a clinically successful intervention evaluated
SCED H
through the multiple baselinestrategy. However, because of some correlation

between the two baselines (obscene phone calling and exhibitionism), the
first
experimental control of the treatment over targeted behaviors is somewhat

unclear. Retrospectively, a more elegant experimental demonstration might
have ensued if the experimenters had temporarily withdrawn treatmentfrom
the second baseline and then reinstated it (in B-A-B fashion), in order to show
the specific controlling power of the aversive strategy. However, from the
clinical standpoint, given the length of the disorder, it is most likely that the
aversive intervention was responsible for ultimate change.
The study by Barton, Guess, Garcia, and Baer (1970) illustrates the use of a
multiple baseline design in which treatment was applied sequentially to sepa-
rate targeted behaviors for an entire group of subjects. Sixteen severely and
profoundly retarded males served as subjects in an experiment designed to
improve their mealtime behaviors through the use of time-out procedures.
Several undesirable mealtime behaviors were selected as targets for study
during preliminary observations. They included stealing (taking food from
another resident's tray), fingers (eating food with the fingers that should have
been eaten with utensils), messy utensils (e.g., using a utensil to push food off
the dish, spilling food), Sind pigging (eating spilled food from the floor, a tray,
etc.; placing mouth food without the use of a utensil). Observa-
directly over
tions of these behaviors were made 5 days per week during the noon and
evening meals by using a time-sampling procedure. Independent observations

were also obtained as reliability checks. The treatment time-out involved
removing the subject (cottage resident) from the dining area for the remain-
der of a meal or for a designated time period contingent upon his evidencing
undesirable mealtime behavior.
The full time-out contingency (removal from the dining area for the entire
meal) was initially applied to stealing following 6 days of baseline recording.
Time-out contingencies for fingers^ messy utensils^ and pigging were then
applied in sequence, each time maintaining the contingency in force for the
previously treated behavior. During the application of time-out for fingers^
the contingency involved time-out from the entire meal for 1 1 subjects, but
only 15 seconds time-out for 5 of the subjects. This differentiation was made
in response to nursing staff's concerns that a complete time-out contingency
for the five subjects might jeopardize their health. Time-out procedures for
messy utensils and pigging were limited to 15 seconds per infraction for all 16
subjects.
The results of this study are presented in Figure 7-5. Examination of the
graph indicates that when time-out was applied to stealing and fingers rates ^
for these behaviors decreased. However, application of time-out to fingers

also resulted in a concurrent increase in the rate for messy utensils. But
subsequent application of time-out for messy utensils effected a decrease in
> TIMEOUT FROM MEAl
l^ 'WV^\.A_J\A-Jwy\Arv a-a-z:l
TIMEOUT FROM MEAl FOR 11, FOR IS" FOR S'
^
. ffh
ÂV^'A^'V-An/^ ^=^
- MESSY .TIMEOUT FOR 15'
. . NEAT
J/'^^V'wV, #Ayri
/\,AA^
jjMfSUL
X g
at 'a, 1%
i?'
o J <s
10 20 30 40 50 60 70
Âw^ W>^J
80 90 100 110
^^
120
*
*"
SUCCESSIVE MEALS OF THE STUDY
FIGURE 7-5. Concurrent group rates of Stealing, Fingers, Utensils, and Pigging behaviors, and
the sum of Stealing, Fingers, and Pigging (Total Disgusting Behaviors) through the baseline and
experimental phases of the study. (Figure 1, p. 80, from: Barton, E. S., Guess, D., Garcia, E., &
Baer, D. M. Improvement of retardates' mealtime behaviors by time-out procedures using
[1970].
multiple baseline techniques. Journal of Applied Behavior Analysis, 3, 77-84. Copyright 1970 by
Society for Experimental Analysis of Behavior, Inc. Reproduced by permission.)
rate for that behavior. Finally, application of time-out for pigging proved
successful in reducing its rate.
Independence of the target behaviors was observed, with the exception of
messy utensils, which increased in rate when the time-out contingency was
applied to fingers. Although group data for the 16 subjects were presented, it
would have been desirable if the authors had presented data for individual
subjects. Unfortunately, the time-sampling procedure used by Barton et al.
(1970) precluded obtaining such information. However, this factor should not
overshadow the clinical and social significance of this study, in that (1)
mealtime behaviors improved significantly; (2) a result of improved mealtime
behaviors was a concomitant improvement in staff morale, facilitating more
favorable interactions with the subjects; and (3) staff in other cottages were
sufficiently impressed with the results of this study to begin to implement
similar mealtime programs for their own retarded residents.
A more recent example of a multiple baseline design across behaviors
(carried out in group format) was presented by Bates (1980). This study is of
particular interest inasmuch as he contrasted the effects of interpersonal skills
training (i.e., social skills training) for an experimental group with a control
condition that received no treatment. Subjects were moderately and mildly
retarded adults (8 in the treatment group, 8 in the control group). Since
treatment was carried out sequentially and cumulatively across four behav-
iors (introductions and small talk, asking for help, differing with others,
handling criticism) following initial assessment, a multiple baseline analysis
was possible in addition to a controlled group evaluation.
A 16-item role-play testwas the dependent measure, with subjects receiving
interpersonal skills training for eight of these scenarios. The remaining eight,
for which subjects received no training, served as a measure of transfer of
training. (But this was only accomplished on a pre-post basis.) Skills training
was conducted thrice weekly and consisted of modeling, behavior rehearsal,
coaching, feedback, incentives, and homework assignments. After each set of
three training sessions an assessment was performed.
Results of this analysis appear in Figure 7-6. As the reader will note,
improvements in each of the four targeted behaviors occurred in time-lagged
fashion only when treatment was specifically applied to each. Thus there was
no evidence of correlated baselines. Data indicate that interpersonal skills
training was effective in bringing about behavioral change. Further, results of
the group comparison indicated that there were statistically significant dif-
ferences in favor of the experimental condition.
Although these data are impressive, we would like to identify a few
problems. First, baseline assessment for introductions and small talk should
have been extended to three points, despite the apparent stability. Second a y
three-point assessment in the treatment phase for handling criticism is war-

ranted considering that there is the beginning of a downward trend in the
data. If this trend were to continue, unequivocal statements about the treat-
ment's controlling effects over that behavior could not be made. Thirds
presentation of data for individual subjects in a table would have been useful
from the single-subject perspective.
This can be a very useful design, but in co-opting behavior analytic

INTRODUCTIONS
AND SMALL TALK GROUP INSTRUCTION (B)
10-r BASELINE (A)
8--
uj Z 6--
O J^
< ^ 4--
>
<
u;<
00
CO "=
I
I
ASKING FOR HELP

: i _ I I
o<
(/) Q.
CO LU
Po I
1
DIFFERING WITH OTHERS

Hrit.
u. 3 6--
UJ
_ CO
^
4--
Z u.
SO 2--
Z
o$
U
UJ
00
UJ 10
0-1-
I-
HANDLING
H
CRITICISM
h - J- -
>Q
I- UJ
8
6-|-
< Z 4--
2--
U 0--
H 1 1 h-i-f
PREl PRE 2 WK 1 WK 2 WK 3 WK 4 POST-
TEST
SITUATION ROLE PLAY ASSESSMENTS
FIGURE 7-6. A multiple baseline analysis of the influence of interpersonal skills training on Exp.
I's cumulative content effectiveness score average across four social skill areas. (Figure 1, p. 244,
from: Bates, P. The effectiveness of interpersonal skills training on the social skill
[1980].
acquisition of moderately and mildly retarded adults. Journal of Applied Behavior Analysis, 13,
237-248. Copyright 1980 by Society for Experimental Analysis of Behavior. Reproduced by
permission.)
procedures, one must be careful to present as much individual data as

possible. For example, all of the problems of averaging apply to these data.
That is, some subjects could show the very steady changes apparent in the
group data across measurement sessions, whereas others might demonstrate

very cyclic types of patterns. Presenting data in this way does not allow one
the option of examining sources of variability where it might be important.
Finally, since it is not clear how many individuals changed in clinically
significant ways, estimates of the replicability of these procedures across
individuals and identification of individual predictors of success and failure

are not possible (see chapter 10). Thus, when proceeding in this manner,
presentation of as much individual data as possible is strongly recommended.
In an interesting solution to the problem of averaging when a number of
subjects are treated simultaneously, Kelly (1980) argued for application of a
design referred to as the Simultaneous Replication Design. This design is used
within a multiple baseline format. The specific example cited involves applica-
tion of social skills training in group format to 6 subjects for three compo-
nents of social skill on a time-lagged basis. However, although applied on a
group basis, behavioral assessment of each subject follows each group ses-
sion. Thus individual data for each treated subject are available and can be
plotted individually (see Fig. 10-6). As noted by Kelly (1980):
The use of this group multiple baseline-simultaneous replication design is parti-

cularly useful in applied clinical settings for several reasons. First, it eliminates
the need for elaborate and/or untreated control groups to establish group
treatment effects and rule out many alternative hypotheses which cannot be
adequately controlled by other one group designs. Second, by analyzing the
social skills behavior change effects of a group treatment procedure, it is possible
to demonstrate more compellingly cost- or time-effectiveness than if each subject
had been laboriously handled as an individually treated case study using single
subject procedures. Because subjects all received the same group training but are
individually evaluated after each group, it is possible to examine "within subject"
response to group treatment with greater specificity than in "between groups"
designs. Since data for each subject in the training group is individually measured
and graphed, each subject also serves as a simultaneous replication for the
training procedure and provides important information on the generality (or
specificity) of the treatment, (pp. 206-207)
(See also section 10.2 for a discussion of issues arising from this strategy
relevant to replication.)
Although the multiple baseline design is frequently used in clinical research
when withdrawal of treatment is considered to be detrimental to the patient,
on occasion withdrawal procedures have been instituted following the se-
quential administration of treatment to target behaviors, particularly when
reinforcement techniques are being evaluated (e.g., Russo & Koegel, 1977). If
treatment is reintroduced after a withdrawal, a powerful demonstration of its
controlling effects can be documented. This type of multiple baseline strategy

was used by Russo and Koegel (1977) in their evaluation of behavioral
techniques to integrate an autistic child into a normal pubhc school class-
room. The subject was a 5 -year-old girl who previously had been diagnosed as
autistic. She evinced limited verbal behavior, failed to respond to the initia-
tives of others, and, when she did verbalize, her comments reflected pronoun
INTEGRATING AN AUTISTIC CHILD
FIGURE 7-7. Social behavior, self-stimulation, and verbal response to command in the normal
kindergarten classroom during baseline, treatment by the therapist, and treatment by the trained
kindergarten teacher. All three behaviors were measured simultaneously. (Figure 1, p. 585, from:
Russo, D. C, & Koegel, R. L. [1977]. A method for integrating an autistic child into a normal
public school classroom. Journal of Applied Behavior Analysis, 10, 579-590. Copyright 1977 by
Society for Experimental Analysis of Behavior. Reproduced by permission.)
reversal. Classroom behavior was characterized by inappropriate actions,

tantrums, bizarre mannerisms, and general aloofness.
Three behaviors were targeted for modification by Russo and Koegel (1977)
in one of the multiple baseline analyses performed: social behavior, self-
stimulation, and verbal response to command. They were all assessed and
treated within the context of the child's kindergarten classroom. Examination
of Figure 7-7 indicates that rate of social behavior was uniformly low, self-
stimulation was quite high, and appropriate responses were low but increas-
ing. Treatment consisted of token reinforcement paired with verbal praise,
feedback, and response cost (removal of tokens) for self-stimulation. Tokens
were earned contingently upon occurrence of each instance of social behavior
and appropriate responses, and they were systematically removed for each
occurrence of self-stimulatory behavior. At the end of each training session
the child had the opportunity to trade remaining tokens for a menu of backup
reinforcers. Three pretraining sessions were carried out to estabhsh the rein-
forcing value of tokens.
Initial treatment by the therapist for social behaviors resulted in a marked
increase in responsivity for that 3 -week period. There were no substantial

changes in self-stimulatory behavior. However, there was some concurrent
increase in rate of appropriate responses, which then decreased somewhat. In
Weeks 7-9 the reinforcement contingency for social behaviors was with-
drawn, resulting in a marked decrease. However, when reinstated in Weeks
10-15, there once again was a substantial improvement in social responding,
thus confirming the controlHng effects of reinforcement in A-B-A-B fashion.
Concurrent with retreatment of social behavior Weeks 10-15 was applica-
in
tion of the contingency for self-stimulation. This led to marked diminution in
such behaviors, with no concurrent changes in the third baseline (appropriate
responses). In Weeks 13-16, when treatment was directed specifically to
appropriate responses, a marked improvement was observed.
In Weeks 14 and 15 the therapist began training the teacher to apply
treatment. From Week 16 through Week 25 the teacher carried out treatment
under the supervision of the initial therapist. Over the course of this time
period the reinforcement schedule was gradually thinned. Data for Weeks
16-25 indicate that initial improvement was either maintained or enhanced.
In summary, this study illustrates the use of the multiple baseline design
across behaviors in a single subject, demonstrating general independence of
target behaviors. Sequential application of a reinforcement contingency to
individual behaviors showed the controlling effectsof the contingency. Addi-
tional experimental manipulations (withdrawal and reintroduction of the
contingency) for the first baseline (social behavior) further confirmed the
controlling effects of the treatment. Finally, data indicate that treatment
procedures were effectively taught to the teacher, who was able to maintain
the child's improved performance in the last phase of the study.
In our final example of a multiple baseline design across behaviors, the
effects of booster treatment subsequent to deterioration during follow-up
(after initial successof social skills training) and documented (Van Hasselt,
Hersen, Kazdin, Simon, & Mastantuono, 1983). The subject was a Wind
female child attending a special school for the blind. Baseline assessment of
social skills through role playing revealed deficiencies in posture and gaze, a
hostile tone of voice, inability to make requests for new behavior, and a
general lack of social skills (see Figure 7-8).
The sequential and cumulative application of social skills training resulted
in marked improvements in role-played performance, thus documenting the
controlling effects of the treatment. However, data for the 4- week posttreat-
TRAINING SCENES
Follow-up
Baseline Social Skills Training
9i^
7-
1
I I I I I I I J_J L
I
1.0
.8
V
<p .6
4
.2
J-l
5- %
I I I I I I I I I i I I
V
I I I
w 12
^ S 8
cS 4
z t I ' ' :::^ 1-1

1.0
^8
.0
V
' ' ' lilt L r
'
I
'
I I
1 3 5 7 9 11 13 15 17 4 6 8 10

FIGURE 7-8. Probe sessions during baseline, social skills treatment, follow-up, and booster
assessments for training scenes for SI. A multiple baseline analysis of posture, gaze, hostile tone,
requests for new behavior, and overall social skill. (Figure 1, p. 201, from: Van Hasselt, V. B.,
Hersen, M., Kazdin, A. E., Simon, J., & Mastantuono, A. K. [1983]. Social skills training for
blind adolescents. Journal of Visual Impairment and Blindness, 75, 199-203. Copyright 1983.
ment follow-up revealed a decrement for gaze and requests for new behavior.
Examination of Figure 7-8 shows that retreatment in booster sessions for
those behaviors resulted in a renewed improvement, extending through the 8-
and 10- week follow-up assessments. Thus our multiple baseline analysis
permitted a clear assessment of which behaviors were maintained after treat-
ment in addition to those requiring booster treatment.
Multiple baseline across subjects
Our first example of the multiple baseline strategy across subjects is taken
from the clinical child literature. Barmann, Katz, 0*Brien, and Beauchamp
(1981) examined the sequential application of overcorrection training for
three developmentally disabled children who were diagnosed as irregular
enuretics. These children (4-, 7-, and 8-years-old, respectively) had IQs that
ranged from 23-41. The first 2 subjects lived at home and the third resided in
a home care facility for the developmentally disabled. Subjects 1 and 3 were
20 BASL TRT FOLLOW UP home

16 I ^c hool
12
8
4
:-t I- 1^ 4 4^ r:f-|-4
20
16
12
%^^ 4^^^ tiix

20
16
12
4 8 12 16 20 24 28 S2 36 40 44 48 52 56 60 64 68 72 76 80 84 88
4 CAY BLOCKS
FIGURE 7-9. Total number of accidents at home and school during baseline, treatment, and
follow-up conditions. NOTE: Data are collapsed over 4-day periods. (Figure 1, p. 344, from:
Barmann, B. C, Katz, R. C, O'Brien, E, & Beauchamp, K. L. 11981]. TVeating irregular
enuresis in developmentally disabled persons: A study in the use of overcorrection. Behavior
Modification, 5, 336-346. Copyright 1981 by Sage Publications. Reproduced by permission.)
,
CHILD 1
NoDalar Delay Oalay Oalay
/
/^
\
CNIL0 2
Oalay
100
eo
eo
40
\
y\
J
20
CNILO 3
AA
u*iay >to 0lir
4<y
20-
O-
BLOCKS OF TEN TRIALS
FIGURE 7-10. Results of the multiple baseline analysis with subsequent repeated reversals of the
influence of a response-delay requirement of the correct responding of autistic children. (Figure 1
p. 235, from: Dyer, K., Christian, W. P., & Luce, S. C. [1982]. The role of response delay in
improving the discrimination performance of autistic children. Journal of Applied Behavior
Analysis, 15, 231-240. Copyright 1982 by Society for Experimental Analysis of Behavior.
enuretic at night at encopretic during the day, in addition to evincing diurnal

enuresis. Subject 2 only evidenced diurnal enuresis.
During baseline, hourly pants checks were performed by parents and the
teacher, at home and at school respectively. Instances of dry pants were
praised at home and at school. Inspection of Figure 7-9 indicates that baseline
levels of accidents ranged from 10-15 per child over a 4-day period.
After stable baselines were observed, overcorrection treatment was applied
sequentially and cumulatively to the three children. Treatment involved resti-

tution overcorrection when the pants were found to be wet at home. (No
treatment was administered at school as this served as a measure of general-
ization.) Restitutional overcorrection "... required the child to (a) obtain a
towel, (b) clean up all traces of the accident, (c) go to the bedroom and put on
clean pants, and (d) dispose of the wet pants in the diaper pail" (Barmann et
al., 1981, p. 341). This was followed by 10 repetitions of positive practice
overcorrection in which the child practiced the correct sequence of toileting

behavior.
Results of this multiple basehne analysis clearly documented the control-
ling effects of the treatment, but only when it was directly applied to each
child. Indeed, treatment reduced enuretic accidents to near zero levels for
each subject and was maintained in a lengthy follow-up evaluation period.
Moreover, the effects of treatment generalized from the home to the school
setting.
As in the multiple baseline across behaviors, baseline and treatment phases
for each subject in this study can be conceptualized as separate A-B designs,
with the length of baselines increased for each succeeding subject used in the
The controlling effects of the contingency are
multiple baseline analysis.
inferredfrom the rate changes in the treated subject, while rates remain
unchanged in untreated subjects. When rate changes are sequentially ob-
served in at least 3 subjects, but only after the treatment variable has been
directly applied to each, the experimenter gains confidence in the efficacy of
the procedure (i.e., overcorrection). Thus we have a direct replication of the
basic A-B design in 3 matched subjects exposed to the same environment
under "time-lagged" contingency conditions.
Dyer, Christian, and Luce (1982) used an interesting variation of a multiple
baseline strategy across subjects in their assessment of response delay to
improve the discrimination performance of three autistic children (two 13-
year-old girls and one 14-year-old boy). Discrimination tasks for the three
children were as follows: Child 1
pointing to a male or female figure; Child

2 describing function of two objects (e.g., a towel and a fork); Child 3
discriminating between right and left. Responses to these tasks were obtained
during no-delay and delay conditions, with all experimental sessions con-
ducted in each child's classroom. Treatment (delay) was introduced, with-
drawn, and reintroduced, following an initial no-delay condition for each
child. This, of course, was conducted sequentially under time-lagged condi-
tions for the three children. Delay consisted of having one child withhold his
or her response for 3 to 5 seconds.
Inspection of Figure 7-10 shows that improved performance only occurred
when the contingency (i.e., delay) was directly applied to each child, thus
documenting the controlling effects of treatment. Data clearly indicate that
the three baselines were independent of one another. Moreover, additional
confirmation of the controlling effects of delay were noted when introduction
TRAIN POST FU RETRAINING
100
^
100
50

z 100
u
DON
so
100
so
JOMN
V
1 2 3 20-22 2
SESSIONS
FIGURE 7-11. Percentage of correct emergency escape responses. Baseline first 3 days of
performance from original baseline phase. Training last 3 days of training from original
intervention phase. Post postcheck assessment 2 weeks after training was terminated. Follow-
up 1-5 month follow-up (FU) reassessment when no intervention Retraining- in effect.
reinstatement of original training program. Follow-up 2-9 month follow-up (FU) reassessment
after original training and 4-month follow-up after retraining. (Figure 1, p. 718, from: Jones, R.
T, Kazdin, A. E., & Haney, J. L. [1981]. A follow-up to training emergency skills. Behavior
Therapy, 12, 716-722. Copyright 1981 by Association for Advancement of Behavior Therapy.
of the delay contingency resulted in improved performance, followed by

deterioration when withdrawn and renewed improvement when reinstated.
Thus, for each child we have an A-B-A-B demonstration, but carried out
sequentially and cumulatively across the three. In short, the study by Dyer et
al. (1982) is an excellent example of the combined use of the A-B-A-B design
in multiple baseline fashion across subjects.
R. T. Jones, Kazdin, and Haney (1981b) used a multiple baseline design

across subjects (5 third-grade children) to assess the effects of training (in-
structions, shaping, modeling, feedback, external, and self-reinforcement) in
emergency fire escape skills. The training package in that study proved to be
quite effective, as indicated by the increased percentage of correct emergency
escape responses accrued by subjects in time-lagged fashion. A portion of
these data (first 3 days of performance from original baseline, last 3 days of
training from original treatment, and a 2-week follow-up) is presented in the
left-hand side of Figure 7-11 for four of these five children. However, a 5-
month follow-up (Sessions 1 for Dana, Lisa, Don, and John on the right-
hand side of Figure 7-11) indicates some decrement in responding. Therefore,
the 5-month reassessment was extended (3 sessions for Dana, 6 for Lisa, 8 for
Don, and 10 for John) under time-lagged conditions, in order to evaluate the
effects of retraining (R. T. Jones et al., 1981a).
As can be seen in Figure 7-11, such retraining did result in improved

performance, but only when treatment was directly applied to each child,
thus reconfirming its controlling effects. However, an additional follow-up 4
months after retraining again indicated decrements in performance, particu-
larly for Don and John. R. T. Jones et al. (1981a), on the basis of these
results, argue that:
The present follow-up study has several implications for future research. First,
conclusions about the effectiveness of particular procedures need to be tempered
unless accompanied by evidence showing maintenance of behavior. The implica-
tion of many demonstrations is that an important applied problem has been
solved by application of behavioral (or other) procedures. However, durability of
behavior change is not an ancillary measure of treatment effects, (p. 721)
Our illustration shows how the muhiple basehne strategy allows for (1) an
initial demonstration of the controlling effects of a treatment, (2) an assess-
ment at follow-up, (3) a second demonstration of the controlling effects of
the treatment, and (4) a second follow-up assessment showing differential
responding among subjects.
A three-group application of the multiple baseline strategy across subjects
(groups of children with insulin dependent diabetes) was provided by Epstein
et al. (1981). The effects of a behavioral treatment program to increase the
percentage of negative urine tests were examined in 19 families of such
diabetic children. Treatment was directed to decrease intake of simple sugars
and saturated fats, decrease stress, increase exercise, and adjust insulin
intake. Parents were taught to use praise and token economic techniques to
reinforce improvements in the child *s self-regulating behavior. When treat-
ment began, 10 of the children (ages 8 to 12) were self-administering their
insulin; the remaining 9 were receiving shots from their parents.
The major dependent measure involved a biochemical determination of

any glucose in the urine. As noted by Epstein et al. (1981), this "... suggests
that greater than normal glucose concentrations are present in the blood, and
the renal threshold has been exceeded" (p. 367). Such testing was carried out
on a daily basis during baseline, treatment, and follow-up.
The 19 families were assigned on a random basis to three groups, with
treatment begun under time-lagged conditions 2, 4, or 6 weeks after initiation
BASELINE TREATMENT FOLLOW-UP
50 -

A
40 ^
_ -L V_ ^/\-/ - - - - - - - -- -
30 -
/v GROUP 1
;:::;..
:.vy^
,,,y...
20 - .
50
% NEGATIVE 40 -
URINES
30 "0^ GROUP 2
20 -i i__i I 1 1 I I I I I i__j I '
50 -1
40 -
30 -,
20 -I
WEEKS
FIGURE 7-12. Percentage of 0% urine concentration tests weekly for children in each group. The
mean and standard error of the mean for all the observations in each phase by group are
represented by a solid and dotted line, respectively. (Figure 1, p. 371, from: Epstein, L. H., Beck,
S., Figueroa, J., Farkas, G., Kazdin, A. E., Daneman, D., & Becker, D. [1981]. The effects of
targeting improvements in urine glucose on metabolic control in children with insulin dependent
diabetes. Journal of Applied Behavior Analysis, 14, 365-375. Copyright 1981 by Society for
Experimental Analysis of Behavior. Reproduced by permission.)
of the 12- week program. Examination of Figure 7-12 indicates that percent-
age of negative urines was relatively low for each of the three groups during
baseline. Institution of treatment resulted in marked improvements in per-
centage of negative urines, indicating the controlling effects of the strategy.
Moreover, it appears that these gains were maintained posttreatment, as
indicated by the follow-up assessment at 22 weeks.
In summary, Epstein et al. (1981) presented a powerful demonstration of
the effects of a behavioral treatment over a biochemical dependent measure
(that has serious health implications). From a design standpoint, this study is
an excellent illustration of the multiple baseline strategy across small groups

of subjects, suggesting how the particular experimental strategy can be used
to evaluate treatments in the area of behavioral medicine. However, from the
design standpoint, the cautionary note articulated with respect to averaging
of data in Bates (1980) certainly applies here.
Sulzer-Azaroff and deSantamaria (1980) also used a multiple baseline
strategy across subjects (groups) in their assessment of feedback procedures
to prevent and decrease occupational accidents in a small industrial organiza-
tion. Six departments were evaluated during baseline for frequency of haz-
ards: (1) screen printing, (2) heat sealing, (3) cutting and assembly, (4) credit
and ID card manufacturing, (5) packing, and (6) receiving and distributing.
mean frequency of hazards
Inspection of Figure 7-13 reveals that, in baseline,
in Departments 1 and 2 was 30.1 and 28.8, respectively; 13.2 and 14.8 for
Departments 4 and 5; and 38.6 and 14.0 for Departments 3 and 6.
The experimental intervention consisted of providing twice-weekly feed-
back, specific suggestions for improvement, and positive comments for ac-
complishments in the area of safety to supervisors for each of the six
departments. This, of course, was carried out in time-lagged fashion 3 weeks
after baseline for Departments 1 and 2, 6 weeks after baseline for Depart-
ments 4 and 5, and 9 weeks after baseline for Departments 3 and 6.
The effects of the intervention were considerable, resulting in a 60% drop
in accidents averaged across departments. The specific controlling effects of
the feedback strategy were documented, in that decreased rates occurred in
those departments only when the intervention was directly applied. For
Department 1, feedback appeared to yield continued improvement, which
originally seemed to be occurring during baseline (i.e., downward trend in the
data). However, data are more convincing for application of the intervention
for Department 2, where such a downward trend was not observed in baseline
data.
Data also indicate that the effects of this intervention were maintained
during the follow-up phase (2 and 6 weeks and 4 months).
An important feature of the Sulzer-Azaroff and deSantamaria (1980)
presentation is that data for each supervisor's department are presented
rather than being collapsed across groups. Such data are important, as it is
^Hkick/Si|ftttii Mm !
^fv/VV//>,Vv;^
^^^'^^'^..Jy^ -
Dtp! 4 !
To JO ^0
JO 'sr*i
sessions 'S.Si
FIGURE 7-13. Frequency of hazards across department as a function of the introduction of the
"feedback package." Data for days following unplanned safety meetings are indicated by an open
circle. At point "a" there was a change in supervisors. (Figure 1, p. 293, from: Sulzer-Azaroff,
B., &
deSantamaria, M. C. [1980]. Industrial safety hazard reduction through performance
feedback. Journal of Applied Behavior Analysis, 13, 287-295. Copyright 1980 by Society for
Experimental Analysis of Behavior. Reproduced by permission.)
conceivable (as frequently occurs when a group comparison design is used)

that some subjects may be unaffected by the contingency in force. Therefore,
once again, we recommend that investigators employing group variations of
multiple baseline strategies provide data showing the efficacy of their proce-
dures in a majority of individual subjects in each respective group.
Multiple baseline across settings
Our first example of a multiple baseline strategy across settings involves

treatment of eye twitching in an 11 -year-old white male (David) whose
disorder had been ongoing since age 5 (Ollendick, 1981). Eye twitching began
when David entered kindergarten, which was concurrent with his mother's
being admitted to a hospital for glaucoma treatments. The child was
described as "mommy's boy" and apparently was very dependent on her.
During baseline, David's tics were surreptitiously observed in school by the
teacher and at home by his mother. This was accomplished in 20-minute
sampling periods. Following a 5 -day observation period at school, David was
Self- Self-Moniforing
Baseline Moniforing Self-Overcorrecrion Follow-up
- Tochr
Oovid
- o
8
lu p
13 15 17 19 21 23 25 27 29 31 33 35 37 39 3-6-l^
Days A^nfhs
FIGURE 7-14. Effects of self-monitoring and self-administered overcorrection in the school and
home: David. (Figure I, p. 81, from: Ollendick, T. H. [1981]. Self-monitoring and self-adminis-
tered overcorrection: The modification of nervous tics in children. Behavior Modification, 5,

75-84. Copyright 1981 by Sage Publications. Reproduced by permission.)
,
taught to self-monitor and record rate of tics. On Day 1 1 self-overcorrection

procedures were added to self-observation. This involved practicing the tens-
ing of muscles that were antagonistic to the tic. Throughout the entire study
period, the teacher continued to monitor tic behavior, thus providing a
reliability check for David's self-observations.
As can be seen in Figure 7-14, similar self-monitoring and self-overcorrec-
tion procedures were carried out by David in the home following 15 days of
initial observation by the mother. Here too, mother continued to monitor tic
behavior when David began to self-monitor (Day 16) and self-overcorrect

(Day 21).
The results of this multiple baseline analysis indicate that self-monitoring
resulted in modest improvements followed by marked improvements when
overcorrection was added (school). However, there appeared to be no change
in tic frequency at home until self-monitoring was specifically applied there
(i.e., baselines are independent from one another). Also, application of
overcorrection in the home led to a continuation of the downward trend to a

zero level. Three-, 6- and 12-month follow-ups indicated a complete main-
tenance of gains.
This study is from a design standpoint for two reasons. First
interesting
the successive controlling effects of two strategies are nicely documented.
Second, excellent reliability (teacher and David; mother and David) for the
self-monitoring of tics appears for both the school (r=.88) and the home
(r=.89) settings.
Singh, Dawson, and Gregory (1980) employed the withdrawal strategy (A-
B-A-B) an application of the multiple baseline design across settings in a
in
17'/2-year-old profoundly retarded female. She suffered from epilepsy (con-
trolled pharmacologically) and had a 6-year history of hyperventilation.
Apparently, prior attempts to deal with her symptoms (defined as a single
instance of deep, heavy breathing, accompanied by a grunting noise and up-
and-down head movements) had failed. Such symptoms were observed in
four separate settings (classroom, dining room, bathroom, dayroom) in the
residential unit of the state facility in which she lived. Data were recorded in
10-second intervals throughout 30-minute sessions.
Baseline data were obtained for 5 sessions in the classroom, 10 in the dining
room, 15 in the bathroom, and 20 in the dayroom. Then, under time-lagged
conditions, treatment (B) was introduced. Subsequently it was removed and
reintroduced in each setting. (This constitutes the A-B-A-B part of the de-
sign). Treatment consisted of the application of response-contingent aromatic
ammonia whenever an instance of hyperventilation was observed: ". .a vial .
of aromatic ammonia was crushed and held under her nose for more
. . .
than 3 sec" (Singh et al., 1980, p. 563). Finally, during the 8 weeks of the
genralization phase, ward nurses were requested to carry out the punishment
procedure on an 8-hour-per-day basis. This is in contrast to original treatment
that was carried out for only four 30-minute sessions per day.
Results of this single-case analysis appear in Figure 7-15. Data clearly
indicate the controlling effects of the treatment, both in terms of its initial
on a time-lagged basis (baselines were independent) and when it

application
was removed and reintroduced simultaneously in all four settings. Rate of
hyperventilation episodes increased dramatically when the punishment con-
tingency was removed in the second baseline and decreased to near zero levels
B LINE 1
PUNISHMENT 1
8LINE M PUNISHMCNT II
GfNfRAHZATlON
X = 10 82
14 X-0 20 X = 30l X.0 34 X.014
Cl ASS ROOM WARD -WIDE
/
14
X = 9 95 A X =
0I8J X--3 73 X = 023 X=0 08
12 CXNING BOOM
I"
oc

r
a:
9
<
X=6 75 X = 26 XO 97 X=013 x.QIS
BATH ROOM
I
4
5 = 748 i.on x=ê,6| x = OtS i=01S
-vwA^ DAY ROOM
2 4 10 12 14 1* U 20 22 24 2 20 JO J2 34 30 30 40 42 44 2 4
^""5
.fSSlONS
FIGURE 7-15. Number of hyperventilation responses per minute and condition means across
experimental phases and settings. (Figure 1, p. 565, from: Singh, N. N., Dawson, J. H., &
Gregory, P. R. [1980]. Suppression of chronic hyperventilation using response-contingent dra-
matic ammonia. Behavior Therapy, 11, 561-566. Copyright 1980 by Association for Advance-
ment of Behavior Therapy. Reproduced by permission.)
when it was reintroduced. Moreover, the positive effects of treatment were

prolonged and enhanced as a resuh of the more extensive punishment ap-
proach followed in the generalization phase.
Fairbank and Keane (1982) present an interesting application of the multi-
ple baseline design across settings (i.e., imaginal scenes) in a 31 -year-old
divorced male veteran suffering from a posttraumatic stress disorder follow-
ing his serving 20 months of combat duty in Vietnam. This subject com-
plained of chronic anxiety, nightmares, and flashback of traumatic events that
had occurred during the course of combat. Through careful interviewing,
four particularly traumatic scenes were selected as stimulus material for
assessment and treatment. During baseline these scenes were presented ver-
bally (with one considerable detail) to the subject in 5- to 10 minute probe
evaluations. During presentation of each scene the subject was asked to self-
rate the discomfort elicited by the material (0 = lowest, 10 = highest). This is
referred to as a SUDS rating. The highest of four such SUDS ratings per
scene was recorded. Concurrently, heart rate and skin conductance responses
to scenes were obtained.
Treatment (i.e., flooding) was applied sequentially and cumulatively to
each of the four scenes. Flooding consisted of 60- to 120 minute sessions in
which "Stimulus and response cues relevant to the scene were slowly and
gradually presented by the therapist, who regularly elicited feedback regard-
ing the next chronological event in the sequence" (Fairbank & Keane, 1982,
p. 503). During the course of a session the subject's anxiety level first in-
creased considerably and then dissipated toward the end.
Data in Figure 7-16 clearly confirm the controlling effects of flooding
treatment on SUDS ratings. This is indicated by the fact that decreases in
SUDS ratings were noted only when treatment was directly applied to each
traumatic scene. Moreover, these data are confirmed by concurrent diminu-
tion in skin conductance responses during probe sessions following direct
application of treatment. Further confirmation of these results was obtained
by replicating the procedure with 2 additional posttraumatic stress-disordered
patients.
From it would have been preferable if the
a design perspective, however,
experimenters had obtained more probe measures in Scenes 1 and 2 (i.e., a
minimum of three data points for Scene 1) and additional probe measures in
treatment for Scenes 3 and 4. This, of course, is in direct reference to the
point raised in chapter 3 with regard to obtaining three measurements in
order to determine a trend in the data.
A particularly socially relevant example of a multiple baseline design across
settings (two high density residential areas) was provided by R. E. Kirchner et
al. (1980) (see Figure 7-17). This study also contains A-B-A withdrawal
features. In the portion of the study we are to describe, two high-population
density areas in Nashville were targeted for study (9.82 and 14.7 square miles;
ANXIETY AND TRAUMATIC MEMORIES

Baseline Treatment
Scene 1
in
Kf

-
8
6 -
v
4 ^0 Â
-
2
1 1 1 . 1
12 3 4
Probe Assessrr^nt Sessions
5
FIGURE 7-16. Maximum SUDS ratings during probe sessions (Subject 2). (Figure 2, p. 505,
from: Fairbank, J. A., & Keane, M. [1982]. Flooding for combat-related stress disorders:
Assessment of anxiety reduction across traumatic memories. Behavior Therapy, 13, 499-510.
Copyright 1982 by Association for Advancement of Behavior Therapy. Reproduced by permission.)
populations 49,978 and 65,910). During baseline, the mean number of home
burglaries committed per day was computed for each area (Xs = 2.83 and
2.25).
After 17 days of baseline in Area 1 of standard police patrolling, an
.
HIGH DENSITY AREA
BASELINE INTERVENTION
1.22 par Oty
FIGURE 7-17. Number of home burglaries in two high-density areas over baseline and interven-
tion conditions. (Figure 1, p. 145, from: Kirchner, R. E., Schnelle, J. E, Domash, M., Larson,
L., Carr, A., & McNees, M. P. [1980]. The applicability of a helicopter patrol procedure to
diverse areas: A cost-benefit evaluation. Journal of Applied Behavior Analysis, 13, 143-148.
Copyright 1980 by Society for Experimental Analysis of Behavior. Reproduced by permission.)
intervention consisting of close scrutiny with a helicopter patrol was added.

This resulted in a decrease in home burglaries to 1 .22 per day. However, when
the helicopter patrol was discontinued on Day 29, the home burglary rate
increased to 1 .91 per day. Thus, from the A-B-A aspect of this study, it is clear
that the helicopter patrol served to reduce home burglaries in Area 1
Similarly, on Day 33, when the helicopter patrol was introduced in Area 2,
home burglaries dropped from 2.25 to 1.16 per day, but rose to 2.85 per day
when it was discontinued on day 52 (control demonstrated in A-B-A fashion
forArea 2).
The A-B-A confirmation of the controlling power of the intervention adds
substantially to documentation of the time-lagged contingency. That is, for
Area 2, change only occurred when the helicopter intervention was directly
applied. Baselines were completely independent. R. E. Kirchner et al. (1980)
presented yet additional evidence for the efficacy of this intervention. From
the cost effectiveness perspective, in baseline, daily burglary costs were
$1,376 and $1,094 respectively for the two areas. When the helicopter inter-
vention was instituted, daily burglary costs diminished to $823 and $815.
Thus we have a very powerful demonstration of this contingency in a multiple
baseline design across settings that incorporates A-B-A withdrawal features.
7.3 VARIATIONS OF MULTIPLE BASELINE DESIGNS

Nonconcurrent multipjebiaseline design
As noted in section 7.2, in the multiple baseline design across subjects, each
individual targeted for treatment is exposed to the same environment. Treat-
ment is delayed for each successive subject in time-lagged fashion because of
the increased length of baselines required for each. The functional relation-
ship between treatment and behavior change can be determined
selected for
only when such treatment is applied to each subject in succession. Thus, since
subjects (at least two but usually three or more) are simultaneously available
for assessment and treatment, this design is able to control for history (cf.
Campbell & Stanley, 1963), a possible experimental contaminant.
There are times, however, when one is unable to obtain concurrent obser-
vations for several subjects, in that they may be available only in succession
(e.g., less frequently seen diagnostic conditions such as hysterical spasmodic
torticollis). Following strictures of the multiple baseline strategy across sub-
jects, this design ordinarily would not be considered appropriate under these
circumstances. However, more recently Watson and Workman (1981) have
proposed an alternative the nonconcurrent multiple baseline across individ-
uals.
In this . . . design, the researcher initially determines the length of each of several
baseline designs (e.g., 5, 10, 15 days). When a given subject becomes available
(e.g., a client referred who has the target behavior of interest, and is amenable to
the use of a specific treatment of interest), s(he) is randomly assigned to one of

the pre-determined baseline lengths. Baseline observations are then carried out;
and assuming the responding has reached acceptable stability criteria, treatment
is implemented at the pre-determined point in time. Observations are continued
through the treatment phase, as in a simple A-B design. Subjects who fail to
display stable responding would be dropped from the formal investigation;
however, their eventual reaction to treatment might serve as useful replication
data.
The logic of this variation is graphically portrayed in Figure 7-18. Of

course, the major problem with this strategy is that the control for history
(i.e., the ability to assess subjects concurrently) is greatly diminished (see also
Mansell, 1982). Thus we view this approach as less desirable than the stan-
dard multiple baseline design across subjects. It should be employed only
when the standard approach is not feasible. Moreover, under such circum-
stances, an increased number of replications (i.e., number of subjects so

treated) might enhance the confidence one has in the results. But in the case of
rare disorders this may not be possible. In any event, use of this variant is not
defensible when it is possible to run all of the subjects concurrently in time-
lagged fashion.
1
Baseline Treatnnent
Subject 3
10 days
Baseline Tredtment
Subject 2
5 days
Baseline Treatnnent
Subject I
15 days
Days
FIGURE 7-18. Hypothetical data obtained through use of a nonconcurrent multiple baseline
design. (Figure 1, p. 258, from: Watson, P. J., & Workman, E. A, [1981]. The nonconcurrent
multiple baseline across-individuals design: An extension of the traditional multiple baseline
design. Journal of Behavior Therapy and Experimental Psychiatry, 12, 257-259. Copyright 1981
by Pergamon. Reproduced by permission.)
Multiple-probe technique
To this point in our descriptions of multiple baseline strategies, baseline

measurement has been continuous for all designs, including the nonconcur-
rent multiple baseline design. However, as noted by Horner and Baer (1978),
there are situations in which repeated measurements will result in reactivity
(i.e., a change simply as a result of repetition of the assessment). When
treatment is subsequently introduced under these circumstances, changes may

not be detected or may be masked, due to the inflated or deflated baseline as a
function of reactivity. In addition, there are some instances when continuous
measurement is not feasible and when (on the basis of prior experimentation)
an ''a priori assumption of stability can be made" (Homer & Baer, 1978,
p. 193). This being the case, instead of having 6, 9, and 12 assessments in
three successive baselines, these can be more interspersed, resulting in two,
three, and four measurement points. An example of this approach is pre-
sented in Figure 7-19. Probes (hypothetical) in our example are represented
by closed triangles, whereas actual reported data appear as open circles.
In commenting on this graph, Horner and Baer (1978) argued that:
SCED
o
15
]
Tom
10 1 Hypothetical h-obes
1
Reported Data o
,
(Horner &KeilitzJ975)
5
A 1
1 I
Michael
15
10
.r
5ll__ n
^ 15
Larry
CO
A<^ ^
10 I
Russell
A Sm) A jcA. A cAyjO A
5 10 15
BASELINE SESSIONS
FIGURE 7-19. Number of toothbrushing steps conforming to the definition of a correct response
across 4 subjects. (Figure 2, p. 194, from: Horner, R. D., & Baer, D. M. [1978]. Multiple-probe
technique: A variation of the multiple baseline. Journal of Applied Behavior Analysis, 11,
189-196. Copyright 1978 by Society for Experimental Analysis of Behavior. Reproduced by
permission.)
The multiple-probe technique, with probes every five days, would have provided
one, two, three, and five probe sessions to establish baselines across the four
subjects. The multiple-probe technique probably could have provided a stable
baseline with five or fewer probe sessions for the subject who had 15 days of
continuous baseline in the original study. The use of the multiple-probe proce-
dure might have precluded the increase in irrelevant and competing behaviors by
this subject because such behavior began to increase after the tenth baseline
session, (p. 195)
It should be noted that, over the years, a variety of researchers have applied
this variant of baseline assessment in the multiple baseline design (Baer &
Guess, 1971; Schumaker & Sherman, 1970; Striefel, Bryan, & Aikins, 1974;
Striefel & Wetherby, 1973). In each of these studies the design used was the
multiple baseline design across behaviors. But, as in Figure 7-19, it could be
across subjects, and it certainly might also be across settings.
If reactivity is the primary reason for using probe tech-
this variant, the
nique should be continued when treatment is However, if feasibil-

instituted.
ity is questionable in baseline or if an a priori assumption of baseline stability
can be made, more frequent measurements during treatment may be desir-

able.
Kazdin (1982b) recommended use of the probe technique for assessment of
(i.e., evaluation of generaliza-
behaviors that were not targeted for treatment
tion or transfer of treatment effects, say, in the naturalistic environment). Use
of probes here is particularly valuable if reactivity is to be avoided. This was
specifically carried out in a multiple baseline design across behaviors evaluat-
ing generalization effects of social skill training in three chronic schizo-
phrenics (Bellack, Hersen, & Turner, 1976). In each case, baseline assessment
involved evaluation of verbal and nonverbal behaviors from video taped role-
play scenarios requiring assertive responding. One set of eight scenarios
(Training Scenes) was repeatedly used for assessment during baseline, treat-
ment, and follow-up phases. This also served as the training vehicle (see left
side of Figure 7-20). A second set of eight scenarios (Generalization Scenes)

also was repeatedly used for assessment during baseline, treatment, and
follow-up phases, but the patient did not receive training here (see right side
of Figure 7-20). However, since the patient was repeatedly exposed to Gener-
alization Scenes, reactivitywas considered a good possibility. Therefore, a
third set of eight scenarios (Novel Scenes)was used for an additional general-
ization assessment during baseline, treatment, and follow-up phases on a
probe basis (see open circles on the right side of Figure 7-20).
Examination of Figure 7-20 confirms the controlling effects of treatment
on individual behaviors in Training Scenes, with the exception of "ratio of
words spoken to speech duration." Data also confirm transfer of training
from Training to Generalization Scenes, but again with the exception of
!
TRAINING SCENES GENERALIZATION SCENES
5 100
8 5 80
v^ S
60
5^ > 40
: o 20 /v:.
2 "
z 36
'''''*'
;&.1
8 8 2< -v^^'
?^ '2 **
S
4 I-
'':'''''
2 ':
*
il y-i * ^ *
i 1 i 1 ill I >
1 > 1 1 1 1 1 1 1 iji 1 1 1 1 I 1 t ''''.'''' I
o ?
2 ./>
v.:
o^
-"< /J
I > I I 1 111 1 i 1 > I I i I I 1 > i|i 1 1 III 1 t
-:
lit--
I III 1 > ill i I
.:t
II
?s iLLlI
I 3
''''''''
5 7 9 II 13 15 17 19 >4-IO
1
1
> I
3
III
5
I I
7
1 t
9 II 13
I
15
I I
17
I III I >
19 2-4-10
Preb* SMieni Wkt Preb* Si*ior>i Wkt
FIGURE 7-20. Probe sessions during baseline, treatment, and follow-up for Subject 3. (Figure 3,
p. 396, from: Bellack, A. S., Hersen, M., & Tlirner, S. M. [1976]. Generalization effects of social
skills training in chronic schizophrenics: An experimental analysis. Behaviour Research and
Therapy, 14, 391-398. Copyright 1976 by Pergamon. Reproduced by permission.)
"ratio of words spoken to speech duration." Probe data (open circles) suggest
that there was further evidence of transfer of training to the Novel Scenes,
with the exception of "ratio of words spoken to speech duration." Finally, for
the three sets of scenes, data indicate that gradual improvements in overall
assertiveness were noted throughout treatment, which appeared to be main-
tained in follow-up.
As we have seen, the probe technique can be most useful in a number of
instances. However, as in the case of the nonconcurrent multiple baseline
design, it should not be employed as a substitute for continuous measurement
when that is feasible. That is, data accrued from use of probe measures are
suggestive rather than confirmatory of the controlling effects of a given
treatment.
7.4 ISSUES IN DRUG EVALUATIONS

With the exception of the multiple baseline across subjects, the multiple
baseline strategies are generally unsuitable for the evaluation of pharmacolo-
gical agents on behavior. For example, it will be recalled that, in the multiple
baseline design across behaviors, the same treatment is applied to indepen-
dent behaviors within the same individual under time-lagged conditions.
Clearly, in the case of drug evaluations this is an impossibility, as no drug is so
specific in its action that it can be expected to effect changes in this manner.
However, it would be possible to apply different drugs under time-lagged
conditions to separate behaviors following baseline placebo administrations
for each. But this kind of design would involve a radical departure from the
basic assumptions underlying the multiple baseline strategy across behaviors
and would only permit very tentative conclusions based on separate A,-B
designs for each targeted behavior. In addition, the possible interactive effects
of drugs might obfuscate specific results. Indeed, the interaction design (see
chapter 6) is better suited for evaluation of combined effects of therapeutic
strategies.
Similarly, the use of the multiple baseline across different settings in drug
evaluations would prove difficult unless the particular drug being applied
worked immediately, had extremely short-term effects, and could be rapidly
eliminated from body tissues. However, as most drugs used in controlling
behavior disorders do not meet these three requirements, this kind of design
strategy is not useful in drug research.
Of the three types of multiple baseline strategies currently in use, the
multiple baseline across subjects is most readily adaptable to drug evalua-
tions. The appHcation of the multiple baseline design across subjects in drug
evaluations could be most useful when withdrawal procedures (return to
A, basehne placebo) are unwarranted for either ethical or clinical consider-
ations. Using this type of strategy across matched subjects, baseline adminis-
tration of a placebo (A,) could be followed by the sequential administration
(under time-lagged conditions) of an active drug (B). Thus a series of A,-B
(quasi-experimental) designs would result, with inferences made in accord-
ance with changes observed when the B (drug) condition was applied. Al-
though an approximation of a double-blind procedure is feasible (observer
and patient blind to conditions in force), it is more likely that single-blind
(patient only) conditions would prevail.
Many other design options are possible in the application of the multiple
baseline design across subjects when evaluating pharmacological effects. For
example, V. J. Davis, Poling, Wysocki, and Breuning (1981) looked at the
effects of decreasing phenytoin drug dosage on the workshop performance of
three mentally retarded individuals. Thus one can use the multiple baseline
O S-12
S-15
D S-16
^ 70
LU60
O 50 J5CX
u 40- ^j>V^ o S-14
^ 30
S-17
20-
10
15
I 1 I I I I 1 I I' I
10
I I I I I
15
WEEKS
I' I I I I
20
I I 1 I I
25
I I I I I
30
FIGURE 7-21, Frequencies of inappropriate behaviors for Subjects 12-18 plotted as total
occurrences per week (summed daily interval totals). During the D condition, the subjects
received their drug; during the P no longer
condition, the subjects received a placebo, were
receiving their drug, and the response cost procedure was not in effect. Drugs were discontinued
during the first 3 weeks of the P condition. During the RC condition, the response cost procedure
was in effect, and the subjects were not receiving their drug. The dotted vertical lines separate the
conditions. (Figure 2, p. 261, from: Breuning, S. E., O'Neill, M. J., & Ferguson, D. G. [1980].
Comparison of psychotropic drug, response cost, and psychotropic drug plus response cost
procedures for controlling institutionalized mentally retarded persons. Applied Research in
Mental Retardation, 1, 253-268. Copyright 1980. Reproduced by permission.)
design across subjects to examine the effects of drug withdrawal in discrete

steps. Another possibility is to evaluate the addition of a behavioral regime to
pharmacological maintenance followed by withdrawal of the drug. This
results in a B-BC-C design, with drug as B, drug plus behavioral intervention

as BC, and the behavioral intervention alone as C (of. Breuning, O'Neill, &
Ferguson, 1980).
Breuning et al. (1980) followed yet a different option of the multiple
baseline design across subjects (small groups) in their successive evaluation of
drug, placebo, and response cost conditions. This yields a B (drug), A'
(placebo), C (response cost) design. Let us consider this study in some detail
(see Figure 7-21). Subjects were institutionalized mentally retarded individu-
als evincing inappropriate behavior. After 3 weeks on active neuroleptic
drugs. Subjects 12, 15, and 16 were switched to placebo for 10 weeks. After 6
weeks on active neuroleptic drugs. Subjects 13 and 19 were switched to
placebo for 7 weeks. Finally, after 9 weeks on active neuroleptic drugs.
Subjects 14 and 17 were switched to placebo for 7 weeks. Examination of
drug and placebo data reveals no apparent improvements in inappropriate
behavior. However, as might be expected, the switch to placebo for Subject 18
led toan increase in inappropriate behavior, suggesting at least some control-
ling effectsof the drug. When response-cost procedures were instituted in
Week 14 for Subjects 12, 13, 15, 16, and 18, and in Week 17 for Subjects 14
and 17, marked improvements in appropriate behavior were observed, begin-
ning almost immediately. Thus this rather complicated experimental analysis
confirmed the efficacy of response cost procedures under time-lagged condi-
tions (baseline 3 versus baselines 1 and 2), but only when the contingency was
directly applied. However, both neuroleptic drugs and placebo generally

seemed to be ineffective.
In this type of drug evaluation it is important to underscore that the
prolonged placebo phases are important in that they provide a needed "wash-
out" period for possible carryover effects of drugs. This, of course, would
have been much more critical had neuroleptic drugs substantially decreased
the behavior targeted for change (i.e., inappropriate behavior).
CHAPTER 8
Alternating Treatments Design
8.1. INTRODUCTION
Few areas of single-case experimental designs have advanced as much as the
design strategies to be discussed in this chapter. The strength and underlying
logic of these strategies, as well as the fact that some specific questions can
only be answered using these approaches, have ensured the rapid develop-
ment and increasing use of this design, particularly during the last 5 years.
The major question addressed by this design is the relative effectiveness of
two (or more) treatments or conditions. The most common experimental
approach employed to address this question until now has been the tradi-
tional between-group comparison. In this strategy, each of two or more
treatments is usually administered to a separate group of subjects, and the
outcome of the treatments is compared between groups. Since considerable
intersubject variability exists in each group (some subjects change and some
do not), inferential statistics are necessary to determine if an effect exists.
This leads to problems in generalizing results from the group average to the
individual subjects, as discussed in chapter 2. To avoid intersubject variabil-
ity, an ideal solution would be to divide the subject in two and apply two
different treatments simultaneously to each identical half of the same individ-

ual. This would eliminate intersubject variability and allow effects, if any, to
be directly observed. In fact, this strategy provides one of the most elegant
controls for most threats to internal validity or the ability of an experimental
design to rule out rival hypotheses in accounting for the difference between
the two treatments (Campbell & Stanley, 1966; Cook & Campbell, 1979).
Statements about external validity or the generalizability of findings observed
in one subject to other similar subjects must be made, of course, through the
252
Alternating Treatments Design 253
more usual process of replication and "logical generalization" (Edgington,

1966; see also chapters 2 and 10).
The name that has come to be employed for the experimental design that
accomplishes this goal is the alternating treatments design (ATD) (Barlow &
Hayes, 1979). As the name implies, the basic strategy involved in this design is
the rapid alternation of two or more treatments or conditions within a single
subject. Rapid does not necessarily mean rapid within a fixed period of time;
as, for example, every hour or every day. In applied research, rapid might
mean is seen he or she would receive an alternative
that each time the client
treatment. For example, an experimenter were comparing treatments A and
if
B in a client seen weekly, he or she might apply Treatment A one week and
IVeatment B the next. If the client were seen monthly, alternations would be
monthly Contrast this with the usual A-B-A withdrawal design where, after a
baseline, an experimenter would need at least three, and usually more,
consecutive data points measuring the effect of Treatment A in order to
examine any trends toward improvement. For a client seen weekly, at least 3
weeks would be needed to establish the trend.
Since one is alternating two or more treatments, an experimenter is not
interested simply in the trend toward improvement over time. Therefore, one
would not plot the data simply by connecting data points for Weeks 1, 2, 3,
and so on. Rather, what one is interested in is comparing treatments A and B.
Therefore, in order to examine visually the experimental effects, one would
connect all the data points measuring the effects of TVeatment A and then
connect all the data points measuring the effects of TVeatment B. If, over
time, these two series of data points separated (i.e., TVeatment B, for exam-
ple, produced greater improvement than TVeatment A), then one could say
with some certainty that TVeatment B was the more effective. Naturally, these
results would then need replication on additional clients with the same
problem. Such hypothetical data are plotted in Figure 8-1 for a client who was
treated and assessed weekly.
Of course, one would not want to proceed in a simple A-B-A-B-A-B-A-B
fashion. Rather, one would want to randomize the order of introduction of
the treatments to control for sequential confounding, or the possibility that
introducing Treatment A first, for example, would bias the results in favor of
Treatment A. Therefore, notice in the hypothetical data that A and B are
introduced in a relatively random fashion. Thus, if one were seeing a client in
an office or a child in a school setting, one might administer the treatments in
an A-B-B-A-B-A-A-B fashion, as in the hypothetical data. For a client in an
office setting, these treatment occasions might be twice a week, with the
experiment taking a total of 4 weeks. For a child in a school setting, one
might alternate treatments 4 times a day, and the experiment would be
completed in a total of 2 days. Randomizing introduction of treatments and
SCED !
100
90
80
70
g
cz>
60
^ 50
Treatment B
30
I Treatment A
20
10
WEEKS
FIGURE 8-1. Hypothetical example of an ATD comparing treatments A and B.
Other procedural considerations will be discussed more fully in section 8-2.

The basic logic of this design, then, requires the comparison of two separate
series of data points. For this reason, this experimental design has also been
described as falling within a general strategy referred to as between-series,
where one is comparing results between two separate series of data points. On
the other hand, A-B-A withdrawal designs, described in chapters 5 and 6,
look at data mthin the same series of data points, and therefore the strategy
has been described as within-series (Barlow et al., 1983).
Tenninology
While this basic research strategy has been used for years within a number
of experimental contexts, a confusing array of terminology has delayed a
widespread understanding of the basic logic of this design. In the first edition
of this book, we termed schedule design. Others have
this strategy a multiple
termed the same design a multi-element baseline design (Sidman, 1960;
Ulman & Sulzer-Azaroff, 1973, 1975), a randomization design (Edgington,
1967), and a simultaneous treatment design (Kazdin & Hartmann, 1978;
McCuUough, Cornell, McDaniel, & Meuller, 1974). These terms were origina-
ted for somewhat different reasons, reflecting the multiple historical origins
of single-case research. For example, several proponents of the term multiple

schedule were associated in Vermont in the late 1960s in an effort to apply
operant procedures and methods to clinical problems (e.g., Agras et al., 1969;
Leitenberg, 1973). These procedures and terminology were derived directly

from operant laboratories.
The term multiple schedule implies not only a distinct reinforcement sched-
ule as one of the treatments, but also a distinct stimulus or signal that will
allow the subjects to discriminate as to when each of the two or more
conditions will be in effect. However, in recent years it has become clear
(particularly in applied research with human subjects) that signs or signals
functioning as discriminative stimuli (SDs) are either an inherent part of the
treatment, and therefore require no further consideration, or are not needed.
For example, alternating a pharmacological agent with a placebo, using at
ATD design, would be perfectly legitimate, but each drug would not require a
discriminative stimulus. In fact, this would be undesirable; hence, the usual
double-blind experimental strategies in drug research (see chapter 6). For this
reason, the more appropriate analogy within the basic operant laboratories
would be a mixed schedule rather than a multiple schedule, since a mixed
schedule does not have discriminative stimuli. But the term schedule itself
implies a distinct reinforcement schedule associated with each treatment, and
there is no reason to think that specific treatments under investigation would
contain schedules of reinforcement. Thus the terms multiple schedule and
mixed schedule are not really appropriate.
Ulman and Sulzer-Azaroff (1975) used one of Sidman's terms, multi-
element baseline design, to describe this strategy. Sidman himself (1960) used
the term multi-element manipulation to describe this particular design. Thus
some researchers have settled on the term multi-element design (Bittle &
Hake, 1977), but these terms also are derived directly out of the basic
research laboratories and in their original usage have little applicability to
applied situations (Barlow & Hayes, 1979).
Edgington (1966, 1972), from a somewhat different perspective, originated
the term randomization design to describe his variation of a time series
approach amenable to statistical analysis. He was most interested in exploring
statistical procedures applicable to randomly alternated treatments. In this
respect he continued a tradition begun by R. A. Fisher (1925), who explored

the abilities of a lady to discriminate tea prepared in two different ways.
Edgington emphasized the randomness of the alternation as well as the
number of alternations in developing his statistical arguments. While these
and other statistical approaches discussed below are useful and valuable, they
are not essential to the logic of the design in our view.
The final alternative mentioned above that is sometimes used to describe
alternating treatments designs is the term simultaneous treatment design. But
this is a bit confusing because there is, in fact, a little-used design in which
two or more treatments are actually available simultaneously. Since the

treatments are presented simultaneously, what happens is that the subject
"chooses" a preferred treatment or condition. Furthermore, this design has
also been called the simultaneous treatment design (Browning, 1967). In fact,
the design has little application in applied research and has not been used
since 1967. Therefore, it will be described only briefly at the end of this
chapter (see section 8-6).*
The basic feature of this design, under its various names, then, is the
"rapid" alternation of two or more different treatments or conditions. For
this reason, we suggested in 1979 the term alternating treatments design
(Barlow & Hayes, 1979), which, most likely because of its descriptive proper-
ties, has been widely adopted (see Table 8-1). Although we use the term
alternating treatments ^ we pointed out in 1979 that treatments refers to the
particular condition in force, not necessarily therapy. Baseline conditions can
be alternated with specific therapies as easily as two or more distinct therapies
can be alternated. Whether or not this is needed, of course, depends on the
specific question one is asking. The use of the term treatment in this way
continues a long tradition in experimental design of referring to various
conditions as treatments.
8.2. PROCEDURAL CONSIDERATIONS

In a single-case design, most procedures utilized in an ATD are similar to
those described earlier for other designs. However, because of the unique
purpose of this design (comparing two treatments or conditions in a single
subject)and because of the strategy of rapid alternation, some distinct
procedural issues arise that the experimenter will want to consider.
Multiple-treatment interference
Multiple-treatment interference (Barlow & Hayes, 1979; Campbell & Stan-

ley, 1963) raises the issue: Will the results of Treatment B, in an ATD where it
is alternated with Treatment A, be the sameas when Treatment B is the only

treatment used? In other words, is Treatment A somehow interfering with
TVeatment B, so that we are not getting a true picture of the effects of
treatment? This notion enjoys much common sense, because at first glance
Kazdin (1982b) has used the term multiple-treatment designs very accurately, in our
view, to subsume both and simultaneous treatment designs. However,
alternating
since simultaneous treatment designs are so rare and would seem to have such little
applicability in applied research, this book will concentrate on the description and
illustration of alternating treatment designs.
there are few strictly "applied" situations where treatments are ever alter-
nated. Thus it is not immediately apparent to practitioners how these results

could generalize to their own situations.
On closer analysis, however, we will suggest that this is a relatively small
problem, and in some cases not a problem at all, for applied researchers
(although it is a major issue in basic research). Also, there are steps applied
researchers can take to minimize multiple-treatment interference. After a
discussion of the nature of multiple-treatment interference, the remainder of
this section will describe procedures for minimizing it.
In a sense, all applied research is fraught with potential multiple-treatment

interference. Unlike with the splendid isolation of the experimental animal
laboratories where rats are returned to their cages for 23 hours to await the
next session, the children and adults who are the subjects of applied research
experience a variety of events before and between treatment sessions. A
college student on the way to an experiment may have just failed an examina-
tion. A subject in a fear-reduction experiment may have been mugged on the
way to the session. Another experimental patient may have lost a family
member in recent weeks or just had sexual intercourse before the session. It is
possible that these subjects respond differently to the treatment than
otherwise would have been the case, and it is these historical factors that
account for some of the enormous intersubject variability in between-group
designs comparing two treatments. ATDs, on the other hand, control for this
kind of confounding experience perfectly by "dividing the subject in two"
and administering two or more treatments (to the same subjects) within the
same time period. Thus, if a family member died during the previous week,
that experience would presumably affect each rapidly alternated treatment
equally. But the one remaining concern is the possibility that one experimen-
tal treatment is interfering with the other within the experiment itself. Essen-
tially, there are three related concerns: sequential confounding, carryover
effects, and alternation effects (Barlow & Hayes, 1979; Ulman & Sulzer-
Azaroff, 1975).
We confounding as referring to the fact that
earlier discussed sequential
Treatment B might be always followed Treatment A. Another
different if it
name for sequential confounding is order effects. That is, much of the benefit
of Treatment B might be due simply to the order in which it is administered
vis-a-vis other treatments. Sequential confounding with A-B-A withdrawal
designs has been discussed in section 5.3. The solution, of course, is to
arrange for a random (or semirandom) sequencing of treatments. One can
view this random order of sequencing treatments in a typical ATD in the
hypothetical data presented in Figure 8-1. Such counterbalancing also allows
for statistical analyses of ATDs for those who so desire (see chapter 9).
Carryover effects, on the other hand, refer to the influence of one treat-
ment on an adjacent treatment, irrespective of overall sequencing. Terms such
as induction and, more G. S. Reynolds,

frequently, contrast (Rachlin, 1973;
1968), are used to describe these phenomena. Several of these terms carry
specific theoretical connotations. For our purposes, it will be enough to speak
of positive carryover effects and negative carryover effects. To return to the
hypothetical data in Figure 8-1 as an example, positive carryover effects
would occur if Treatment B were more effective, because it was alternated
with Treatment A than it would be if it were the only treatment adminis-
tered. Negative carryover effects would occur if Treatment B were less
effective because it was alternated with Treatment A than if it were adminis-
tered alone. In other words. Treatment A is somehow interfering with the
effects one would from Treatment B if it were administered in isolation.
see
Recent basic research has shed more light on the nature and parameters of
carryover effects. In basic research laboratories, where the understanding of
carryover effects is very important to various theories of behavior, investiga-
torshave discovered that such effects are almost always transient and due
mostly to the inability of the subject to discriminate among two treatments
(Blough, 1983; Hinson & Malone, 1980; Malone, 1976; McLean & White,
1981). Fortunately for us, the types of experimental situations where car-
ryover effects are observed in basic research rarely occur in applied research.
In basic research, treatments (schedules of reinforcement in this particular
context) are often alternated by the minute. Furthermore, the treatments
themselves are almost impossible to discriminate as they are occurring. For
this reason, signs or signals (discriminative stimuli), referred to as SDs, are
associated with each treatment. As these signals themselves become harder to
discriminate (for example, increasingly closer wavelengths of light), carryover
effects occur (Blough, 1983). But even with these difficult-to-discriminate
treatments and signals, carryover effects eventually disappear as discrimina-
tions are learned. Recently, Blough (1983) has proposed that in situations
where carryover effects are more permanent within this context, individual
differences in ability to learn discrimination may be the reason. That is, those
subjects (pigeons or rats) that are slower in learning the discriminations are
associated with longer periods of carryover effects, whereas subjects learning
the discriminations quickly evidence very short and transient carryover ef-
fects.
When carryover effects have been noticed in humans (e.g., Waite & Os-
borne, 1972), experimental operations similar to those employed in the
laboratories of basic research were in operation. Presumably the same lack of
discriminability was occurring.
In applied research, this would imply that carryover effects of the type
discussed here are a possibility only when learning is occurring. This would
exclude most biological treatments, such as pharmacotherapy, where no real
learning occurs (although biological multiple-treatment interference will oc-
cur if drugs are alternated too quickly, depending on the half-life of the
particular drug, see chapter 6). On the other hand, almost all psychosocial
interventions do involve some learning. But treatments are usually so distinct

that they are very easily discriminated even without any sign or signal. In fact,
in the examples to be described below, adults are usually told which treatment
is in effect from session to session, and therefore discriminations are perfect.
Similarly, children of all ages are certainly capable of discriminating different
treatments (e.g., time-out versus praise in the classroom) very quickly.
Nevertheless, until we know even more about carryover effects, it would
be prudent to consider the following procedures when implementing an ATD.
First, counterbalancing the order of treatments should minimize carryover
effects and control for order The remaining steps involve ensuring
effects.
that treatments are discriminable. Second for example, separating treatment
y
sessions with a time interval should reduce carryover effects. Powell and
Hake (1971) minimized carryover effects in this way in a study comparing
two reinforcement conditions by presenting only one condition per session.
Fortunately, in applied research it is the usual case that only one treatment per
session is administered even if several sessions are held each day (e.g., Agras
et al., 1969; McCullough et al., 1974). Similar procedures have been sug-
gested to minimize carryover effects in the traditional, within-subjects, group

comparison approaches (Greenwald, 1976). Third the speed of alternations ,
seems to increase carryover effects, at least until discriminations are formed.

This is particularly true in basic research, as noted above, where treatments
may be alternated by the minute. Slower and, once again, more discriminable
alternations should minimize carryover effects (Powell Hake, 1971; Waite &
& Osborne, summary, based on what we now know about carryover
1972). In
effects, counterbalancing and insuring discriminability of treatments will
minimize this problem. In appHed research, where possible, simply telling the
subjects which treatment they are getting should be sufficient.
Finally^ in the event that some carryover effects may be occurring even with
the procedural cautions mentioned above in place, there is no reason to think
that these carryover effects would reverse the relative positions of the two
treatments. Returning to the hypothetical data in Figure 8-1, Treatment B is
seen as better than Treatment A. In this particular ATD, B may not be as

effective as it would be if it were the only treatment administered, and A may
be more effective, but it is extremely unlikely that carryover effects would
make A better than B. Thus, even if carryover effects were observed in the
major comparison of treatments, the experimenter would have clear evidence
concerning the effectiveness of Treatment B, but would have to emphasize
caution in determining exactly how effective Treatment B would be if it were
not alternated with Treatment A.
Assessing multiple-treatment interference. For those investigators who are

interested, it is and sometimes desirable to assess directly the extent
possible
to which carryover effects are present. Sidman (1960) suggested two methods.
One is termed independent verification and essentially entails conducting a
controlled experiment in which one or another of the component treatments

in the ATD is administered independently. For example, returning to Figure
8-1 once again, Treatments A and B would be compared using an ATD in the
manner presented in Figure 8-1, and this experiment would be replicated
across two subjects. The investigator could then recruit 3 more closely
matched subjects to receive a baseline condition, followed by Treatment A in
an A-B fashion. Treatment B could be administered to a third trio of subjects
in the same manner. Any differences that occur between the treatment
administered in an ATD or independently could be due to carryover effects.
Alternatively, these subjects could receive treatment A alone, followed by the
ATD which alternated Treatments A and B, returning to Treatment A alone.
An additional 3 subjects could receive Treatment B same manner.
in the
Trends and levels of behavior during either treatment alone could be com-
pared with the same treatment in the ATD. Obviously, this type of strategy
would also be very valuable for purposes of replication and for estimating the
generalizability or external validity of either treatment.
A more elegant method was termed functional manipulation by Sidman
(1960). In this procedure the strength of one of the components is altered. For
example, comparing imaginal flooding versus reinforced practice in the
if
treatment offear, the amount of time in flooding could be doubled at one
point. Changes in fear behavior occurring during the second unchanged

treatment (reinforced practice) could be attributed to carryover effects.
In an important, more recent example using these types of strategies, E. S.
Shapiro, Kazdin, and McGonigle (1982) examined the possible multiple-
treatment interference in an experiment with five retarded, behaviorally dis-
turbed children. The target behavior in this particular experiment was on-task
behavior in a classroom located in a children's psychiatric unit. With a very
clever and elegant variant of the method of independent verification, the
effectsof two treatments and a baseline condition were examined within the
context of an ATD for increasing on-task behavior. One treatment was token
reinforcement for on-task behavior, the second treatment was response cost
where tokens were removed for off-task behavior. l\vo 25-minute sessions
were held per day: one in the morning and one in the afternoon. On any one
day, two treatments would be administered, and these would be counterbal-
anced over a number of days. After a 4-day phase in which baseline condi-
tions were in effect during both time periods, baseline and token
reinforcement were alternated over a 6-day phase. This was followed by the
alternation of token reinforcement and response cost over a 10-day period.
The investigators then returned to the baseline versus token reinforcement
phase for 6 more days, followed by a return to the token reinforcement versus
response cost phase for yet another 6-day period. Finally, this was followed
by a phase where token reinforcement was administered during both time
periods.
The experimental design and the results are represented in Figure 8-2,
where the average responses of the five subjects are presented. (Individual
data were also presented, but this figure will suffice for purposes of illustra-
tion.) Thus this experiment really consisted of four separate ATDs after the
baseline condition, in which token reinforcement was alternated with either
baseline or response costs. Each of these ATDs was repeated twice. The
elegance of this design for examining multiple-treatment interference is found
in the fact that one can examine the effects of token reinforcement when
alternated with either another treatment or baseline. If multiple-treatment
interference is evident when token reinforcement is alternated with the other
treatment, response cost, then the effects of token reinforcement should be
different during that part of the experiment from when token reinforcement
is alternated with baseline.
First, it is important to note here that both token reinforcement and
response costs produced strong and comparable effects in increasing on-task
behavior, and that token reinforcement was clearly effective when compared
to baseline. The investigators decided, however, that token reinforcement was
the preferable treatment because they noticed that more disruptive behavior
occurred during the response-cost procedure than during the token reinforce-
ment procedure. Thus token procedures were continued during both sessions
in the last phase.
The investigators reported three different sets of findings from their ex-
amination of potential multiple-treatment interference. First, no evidence was
BL BL Tkn/BL Tkn/RC Tkn/BL Thn/RC Tkn/Tk

100
PERCENT
INTERVALS
ON TASK
-
A- - - A BL or MnponM Cmi
FIGURE 8-2. Group mean percentages of on-task behavior. Paired interventions in each phase
consisted of Baseline/Baseline; Token Reinforcement/Baseline; Token Reinforcement/Response
Cost; Token Reinforcement/Baseline; Token Reinforcement/Response Cost; Token Reinforce-
ment/Token Reinforcement. (Figure 1, p. 110. from: Shapiro, E. S., Kazdin, A. E., &
McGonigle, J. J. (1982). Multiple-treatment interference in the simultaneous- or alternating-
treatments design. Behavioral Assessment, 4, 105-115. Copyright 1982 by Association for
Advancement of Behavior Therapy. Reproduced by permission.)
found that the overall level of on-task behavior was different when it was
alternated with either baseline or response cost. This, of course, is an ex-

tremely important finding, particularly in terms of estimating what the effects
of token reinforcement in this context would be when applied in isolation;
that is, without the potentially interfering effects of another treatment. In
other words, the investigator or clinician can feel somewhat safe in determin-
ing that the effects of token reinforcement, when alternated with response
costs, are about what they would be if response cost were not present. Of
course, this still is not a "pure" test because it is possible that alternating
token reinforcement with baseline in an ATD yields a somewhat different

effect from token reinforcement administered in isolation. Strict adherence to
Sidman*s method of independent verification would be necessary to estimate
if any carryover effects were present when a treatment was alternated with a
baseline condition.
Nevertheless, the investigators do point out that on-task behavior was
more variable during token reinforcement when alternated with response cost
than when alternated with baseline. Visual inspection of the data indicates
that this was particularly true in 3 out of 5 subjects. While this finding in no
way effects the interpretation of the results, it is an interesting observation in
itself that could be followed up in a number of ways. It is possible, for
example, that "disruptiveness" noted during response cost temporarily car-

ried over into the next token phase, thereby causing some of the variability. A
greater spacing of sessions and subsequent sharpening of stimulus control
might have decreased this variability.
Also, the investigators observed a sequence effect, in that token reinforce-
ment was more effective when applied in the morning session than in the
afternoon session. Once again, this demonstrates the importance of counter-
balancing. Finally, the investigators observed another possible example of
multiple-treatment interference not directly connected with the comparison
of the two treatments. In the first phase, where token reinforcement and
baseline were alternated, on-task behavior averaged 14 percent during the
baseline condition. In the second phase, where this same alternation oc-
curred, however, on-task behavior averaged approximately 30 percent during
the baseline session. Inspection of individual data revealed that this trend
occurred in four out of five children. This may represent a positive carryover
or a generalization of treatment effects to the baseline condition; thus, the
first phase probably presents a truer picture of baseline responding. Studies of
this type will be very critical in the future in mapping out the exact nature of
multiple-treatment interference and improving our ability to draw causal
inferences from ATDs.
The study of carryover effects, or treatment interactions, when they occur,
can be interesting in its own right (Barlow & Hayes, 1979; Sidman, 1960). For
example, it is possible that carryover effects might increase the efficacy of
some treatments. In an early study of fantasy alteration in a sadistic rapist,

Abel, Blanchard, Barlow, and Flanagan (1975) alternated orgasmic recondi-
tioning daily, first using a sadistic fantasy and then a desired heterosexual
fantasy. It is important to note that treatments were not counterbalanced and
alternations were rather rapid. Sexual arousal to the heterosexual fantasy
increased more quickly during the fast alternation than during orgasmic
reconditioning to the appropriate fantasy alone. More recently, Leonard and
Hayes (in press) have also demonstrated that fantasy alternation produces
stronger changes in sexual arousal patterns when alternations are fast rather
than when alternations are slow. This may represent a carryover effect or
simply a sharpening of stimulus control.
Counterbalancing relevant experimental factors
If certain factors extraneous to the treatments themselves might influence

treatment, then these factors should be counterbalanced. Actually, this
should be quite obvious to any investigator designing an experiment. For
example, if Treatments A and B in Figure 8-1 referred to two distinct
manipulations within a classroom, and two classrooms were involved, then it
would be important that one treatment did not always occur in the same
classroom. For example, in McCuUough et al (1974) ATD examining the
effects of two treatments on disruptive behavior in a 6-year-old boy, two
factors were counterbalanced (see Table 8-1). In this particular experiment the
first treatment was social reinforcement for cooperative behavior and ig-
noring of uncooperative behavior. The second treatment was social reinforce-

ment for cooperative behavior plus time-out for uncooperative behavior, in
this case removal from the classroom for 2 minutes. A teacher and a teacher's
aide administered the treatments, with the teacher administering TVeatment A
the first two days and Treatment B the last two days. Thus the two people
Table 8-1
TREATMENT
TIME
DAY 1 DAY 2 DAY 3 DAY 4
AM AT-1 BT-2 AT-2 BT-1

PM BT-2 AT-1 BT-1 AT-2
NOTE: T-1 = teacher, T-2 = teacher's aide
Redrawn Table 1, p. 260 from McCullough, J. P., Cornell, J. E., McDaniel, M. H., & Mueller, R.
K. (1974). Utilizational of the simultaneous treatment design to improve student behavior in a
first-grade classroom. Journal of Consulting and Clinical Psychology, 42, 288-292. Copyright
1974 by the American Psychological Association. Reproduced by permission.
administering treatments were counterbalanced because, of course, differen-

tial effectiveness might have something to do with the person administering
the treatments. In addition, treatments were administered during both a
morning session and an afternoon session. Once again, rather than the
experimenters offering Treatment A only in the morning and IVeatment B
only in the afternoon, treatments were alternated such that administration of
them was counterbalanced across morning and afternoon. In the example
described above (E. S. Shapiro et al., 1982), the investigators observed
greater effectiveness of token reinforcement sessions in the morning than with
afternoon sessions, underscoring once again the need for counterbalancing.
Of course, what should and should not be counterbalanced will be up to
the investigator. Naturally, if different therapists, teachers, or other practi-
tioners are involved in administering the treatments, then they must be
counterbalanced. Some investigatorsmay also want to counterbalance times
of day if these differ, whereas others may not consider this important,
depending on the question asked. Most investigators will have a good feel for
this.
Number and sequencing of alternations
The major question one must consider in determining the number of

alternations is the potential for determining differences among two or more
treatments. In determining behavior trends within a baseline phase or one of
the phases of an A-B-A withdrawal design, we suggested that three data
points were the minimum necessary to determine a trend. In the ATD,
however, when one comparing two treatments, a minimum number of two
is
data points for each treatment would be necessary, although a higher number
would, of course, be much more desirable. TWo data points per treatment
would allow an examination of the relative position of each treatment and
some tentative conclusions on treatment efficacy. However, returning to
Figure 8-1 once again, few investigators would be convinced of the superior-
ity of Treatment B if the experiment were stopped after Week 4. Nevertheless,
if other practical considerations prevented continuation, the findings might

be potentially important, pending replication.
Naturally, frequency of alternations will be limited and other
by practical
considerations. It is and meaningful
possible, for example, that treatment
measurement opportunities would occur only once a month. Once again, one
could conceive of this situation occurring in the alternation of two drugs with
long half-lives, where a meaningful measurement of behavioral or mood
changes could occur only after one month; this might consist of two weeks of
treatment with the drug and two weeks of consolidation of drug effects.
Similar situations might obtain for two different physical interventions in a
rehabilitation setting.
Finally, in arranging for random alternation of treatments to avoid order

effects,one must be careful not to bunch too many administrations of the
same treatment together in a row. For example, in determining the random
order of two treatments by coin toss or a random-numbers table, it is
conceivable that one might arrive by chance at an order that dictates four
administrations of Treatment A in a row. If only one has time for only eight
alternations altogether, then this would not be desirable. Thus the investiga-
tor must move to a "semirandom" order with an upper limit on the number
of times a treatment could be administered consecutively. The investigator
will make this determination based on the total number of alternations
available. For example, if eight alternations were available, as in the hy-
pothetical data in Figure 8-1, then the investigator might want to set an upper
limit of three consecutive administrations of one treatment.
8.3. EXAMPLES OF ALTERNATING TREATMENTS

DESIGNS
ATDs have been used in at least two ways: to compare the effect of
treatment and no treatment (baseline) and to compare two distinct treat-
ments. Some examples of ATDs with specification of the experimental com-
parison are presented in Table 8-2.
Comparing treatment and no-treatment conditions
Several investigators have compared treatment and no treatment in an

ATD. Among and Henson (1969) compared
early examples, O'Brien, Azrin,
the effect of following and not following suggestions made by chronic mental
patients in a group setting on the number of suggestions made by these
patients. Doke and Risley (1972) alternated daily the presence of three
teachers versus the usual one teacher and noted the effect on planned activi-
ties in the classroom (contingencies on individual versus groups were also
compared in an ATD later in the experiment). Redd and Birnbrauer (1969), J.

Zimmerman, Overpeck, Eisenberg, and Garlick (1969), and Ulman and
Sulzer-Azaroff (1975) also reported early examples comparing treatment and
no treatment in an ATD.
A particularly good example of this strategy was reported by Ollendick,
Shapiro, and Barrett (1981). In this experiment the effects of two treatments
(physical restraint and positive-practice overcorrection) were compared to no
treatment in the reduction of stereotypic behavior in three mentally retarded
emotionally disturbed children. The investigators targeted stereotypic behav-
iors for reduction involving bizarre hand movements, such as repetitive hair
twirling and repetitive hand posturing. In a very important consideration
4> 4>
c -^ IS
B B
Is o o^ I Is
2> fi-O c O C
4> C 4> O 1>
^
o
o o
0*0 O o 2 o 2
60 J w tS 'a i> 13
-o '53
=1
tJOe S C C
.5 g ^ -s 'a
SO-53ooo-cooc_gotio-Ca
^!r^#-^7^r!/-^/vS?50n
1111
i6>sâo =
cs ^ cj
>
g o
1
^ 1
-
I
>^
"
^
S ^ 2 -
^3 T.-- o
2i .2 -c
6 -s > 3
a '^ 0.5 5'
T3
I I
a
3
Tt -H -,
1^
U
i3
Tj-
c^
^.K,
<N 2;
S <5 s 5 N

CL, w ^^ S t ^.
^^ Q
.
iJ I
t/3
CQ 2 <
oa
^ =
NO 4> O
06 ^ 0Q
266
8 -
c o
a- o- cu
Q -5 z Pu cu z f2^
ucO^ctfX) ucq(jct]coX)u
I
1
o 00
= 1 > .S V.
1
-a
c >O
I 2 60 U
J j3
2 let
O
CO i/J
CJ
'g
C3
>.
4>
-^
^'^^ o o 3
J^ Si 14> s
(t:
C
(35 ^ 55 (5 o
60 iS
g "2 c TJ*' g
5 -S
I .a >>
60 4) t
i2
-a
c
4> 1
S aj a ^1
^ .2
2-2 :5 o <i>2
.S
60 O "S
c3
-1
II ^'
3 : <N H : o2
C S2 |g S3 g SJ)
rj ^' ON =^
o a ^ 5>
t
c U C
OO
On
1 to
?-
ô ^
a
CO
ON l->
9 J O
O 1
o 1
1o 1 1 1
1
H ^^
u o.
^ c
o u i to
^ t ^ rj"
w s~ .^ O -^ 1
g o
21o o'
tz
CO
x: 1
J
d"
u-
1 <u
5
. 5
^'2 'S.
CO
s_
C/3
'5.
CO
c4 jiC
u
i
1 t^ oo ^B COCS
J=00 -^s
ON .2 On ^'^ -*
S
(73
1" C/D
1 o ui o 3: pg
267
<S
el ill? I
d c8 e oj
C/3 esbesb ^
fill 3'^3-?!:ic
r\ M rs
1 1
*
"i
12
tary tary
id
s ^ s
Son 6o
<^ rt**^ ^
^g.*^^^
U U
I
fS
^^ ^
^ 1
O oc
C; g
4>
3> ^
03
a &
<^
b ^

2 c
2 ? o 1
-"
1 ?
X I jD ^
268
before beginning the experiment, the investigators ruled out the use of an A-
B-A withdrawal design because even temporary increases in stereotypic be-
havior during withdrawal phases were unacceptable in this setting.
Furthermore, previous experience of these investigators suggested that there
was a chance the two treatments might be equally effective. Thus a no-
treatment condition might be necessary to determine if these treatments were
effective at all.Of course, this problem also arises in between-group research
because, iftwo treatments were equally effective (on the average) in two
groups, a control group would be necessary to determine if any clinical
effects occurred over and above no treatment.
In this procedure, three 15-minute sessions were administered by the same
experimenter each day. Individual sessions were separated by at least one
hour. Following baseline conditions for all three time periods, the two treat-
ments and the no-treatment conditions were administered in a counterbal-
anced order across sessions. When one of the treatments produced a zero or
near-zero rate of stereotypic behavior, that treatment was then selected and
implemented across all three time periods during the remainder of the study.
During sessions, each child was escorted to a small table in a classroom and
instructed to work on one of several visual motor tasks. One treatment was
physical restraint, consisting of a verbal warning and manual restraint of the
child's hand on the tabletop for 30 seconds contingent on each occurrence of
stereotypic behavior. The second treatment, positive-practice overcorrection,

involved the same verbal warning but was followed by manual guidance in
appropriate manipulation of the task materials for 30 seconds. Measures
taken included number of stereotypic behaviors during each session and
performance on the task.
The results for two of the three subjects are presented in Figures 8-3 and 8-
4. In Figure 8-3 it is apparent during the ATD phase of this experiment that
physical retraint was the superior treatment for John. Therefore, this treat-
ment was chosen for the remainder of the experiment. Task performance
increased rather steadily throughout the experiment, but was greatest during
physical restraint. On the other hand. Figure 8-4 shows that positive practice
intervention was the superior treatment for Tim.
Several features of this noteworthy experiment are worth mentioning.
First, the ATD part of this experiment was concluded in 3 or 4 days (three
sessions per day), and proper determinations of the effective treatment in
each case were made. This is a relatively brief amount of time for an
experiment in applied research, and yet it is typical of ATDs, particularly in
this context (e.g., McCullough et al., 1974). Second the addition of a,
baseline phase prior to introduction of the ATD allowed further identification

of the naturally occurring frequencies of the target problem and the absolute
amount of reduction in the target problem when treatments were instigated.
Of course, this is not necessary in order to determine which of three condi-
ALTERNATING ^^,^,^ A^-ri/-c

BASELINE TREATMENTS POSITIVE PRACTICE
20
18
16
14 NO INTERVENTION
POSITIVE PRACTICE
12 ^ ^PHYSICAL RESTRAINT
10
8
6 -
UJ
O
li
OC (/)
itifl:
SESSIONS
ÎGURE 8^ Stereotypic hair twirling and accurate task performance for John across experi-
rnenfaTconHitions. The data are plotted across the three alternating time periods according to the
schedule that the treatments were in effect. The three treatments were presented only during the
alternating-treatments phase. During the last phase, physical restraint was used during all three
time periods. (Figure 1, p. 573, from Ollendick, T. H., Shapiro, E. S., & Barrett, R. P. (1981).
Reducing stereotypic behaviors: An analysis of treatment procedures utilizing an alternating
treatments design. Behavior Therapy, 12, 570-577. Copyright 1981 by Association for Advance-
ment of Behavior Therapy. Reproduced by permission.)
tions was more effective, but it provides important additional information to

the investigator. Third, The ATD in this case also served as a clinical assess-
ment procedure for each client, since the most was imme-
effective treatment
diately applied to eliminate the problem behavior. The rapidity with which the
ATD can be implemented makes this design very useful as a clinical assess-
ALTERNATING
BASELINE TREATMENT PHYSICAL RESTRAINT
John
NO INTERVENTION
"POSITIVE PRACTICE
^ - *^ PHYSICAL RESTRAINT
'IGURE 8-4 Stereotypic hand posturing and accurate task performance for Tim across experi-
menlatconditions. The data are plotted across the three alternating time periods according to the
schedule that the treatments were in effect. The three treatments were presented only during the
alternating-treatments phase. During the last phase, positive practice overcorrection was used
during all three time periods. (Figure 2, p. 574, from OUendick, T. H., Shapiro, E. S., & Barrett,
R. P. (1981). Reducing stereotypic behaviors: An analysis of treatment procedures utilizing an
alternating treatments design. Behavior Therapy, 12, 570-577. Copyright 1981 by Association for
Advancement of Behavior therapy. Reproduced by permission.)
ment tool as well as an experimental strategy (see Barlow et al., 1983).

Fourth, John did better with physical restraint, whereas Tim did better with
positive practice intervention. The third subject also did better with positive
practice intervention. This is a good example of the handling of intersubject
variability in an ATD design. As discussed in chapter 2, a between-group
strategy would average out, rather than highlight, these individual differences
in response to treatment. By demonstrating this intersubject variability, how-
ever, the investigators were in a position to speculate on the reasons for these
differences, which in fact they did. Because of this, they were in a position to
examine more carefully client-treatment interactions that would predict
which treatment would be successful in an individual case. Once again,
highlighting intersubject variability in this way can only increase the precision
with which one can generalize the effects of these specific treatments to other
individual clients (see chapter 2).
Finally, the discerning reader will notice that posturing during the no-
treatment condition of the ATD is
somewhat higher with John and Tim than
during baseline, where the same condition was in effect across all three time
periods (but this increased response during no treatment was not true for the
third subject). It is possible that this is an example of negative carryover
effects, because responding during no treatment was worse when it was
alternated with treatment than it was alone; that is, in baseline. In this
experiment the authors purposefully blurred the discriminability of the three
conditions as part of their experimental strategy, which may account, in part,
for the carryover effects. This finding, once again, occurred in baseline and
did not affect the ability of the investigators to determine the most effective
treatment and then to apply it successfully during the last phase.
Of course, determination of the effectiveness of a single treatment com-
pared to no treatment can also be examined via the most common A-B-A-B
withdrawal design (see chapter 6, section 6-3). In this particular experiment,
however, the authors were interested in comparing the effects of two treat-
ments with each other as well as the effects of each compared to no treat-
ment, and thus the ATD was the only choice. Furthermore, they had
determined clinically that it was not possible to allow an increase in stereotyp-
ic responding in the absence of treatment, a condition that would obtain
during the withdrawal phase of any A-B-A design. Nevertheless, when one
wishes to compare treatment with no treatment, one has a choice between a
more standard withdrawal design and an ATD. The advantages of the ATD
have already been mentioned. In addition to not requiring a withdrawal of
treatment for a period of time, the comparison within the ATD can usually be
made more quickly, and it can proceed without a formal baseline if this is
necessary. On the other hand, there no single phase in the ATD where
is
treatment is applied in isolation as it would be in a clinical situation. There-

fore, estimating the generalizability of any given treatment is less certain if
one has any reason to worry about multiple-treatment interference effects.
Investigators will have to weigh these advantages and disadvantages in choos-
ing a particular design to compare treatment and no treatment.
Ollendick and his colleagues have also produced two other excellent exam-
ples of ATDs comparing three conditions. In each case two treatments were
compared to no treatment (Barrett, Matson, Shapiro, & Ollendick, 1981;
Ollendick, Matson, Esveldt-Dawson, & Shapiro, 1980). In the Barrett et al.
study, punishment and DRO procedures were compared to no treatment in
dealing with stereotypic behavior of mentally retarded children. In the OUen-

dick et al. (1980) study, two spelling remediation procedures were compared to
no treatment. Unlike the Ollendick et al. (1981) study reported earlier, the
investigators chose to make each condition clearly discriminable through
either instructions at the beginning of each session or other clear signs and
signals. There is little or no evidence of multiple-treatment interference in
either of these experiments. Once again, if one wants to eliminate the possibil-
ity of multiple-treatment interference, it would seem advisable to make
conditions as discriminable as possible. The easiest method is to use simple
instructions announcing what condition the subject is in.
Comparing multiple treatments
The majority of ATDs compare the effects of two treatments rather than
the effects of treatment with no treatment. An early example in an adult
clinical situation examined the effects of two fear-reduction procedures
(Agras et al., 1969, see Figure 8-5). This study examined the effects of two
forms of exposure-based therapy. The subject was a 50-year-old female with
severe claustrophobia. Her fears had intensified following the death of her
husband some 7 years before admission to the treatment program. When
admitted, the patient was unable to remain in a closed room for longer than
one minute without experiencing considerable anxiety. As a consequence of
this phobia, her activities were seriously restricted. During the study she was
asked four times daily to remain inside a small room until she felt she had to
come out. Time in the room was the dependent measure. During the first four
data points, representing treatment, she kept her hand on the doorknob.
Before the fifth treatment data point (sixth block of session), she took her
hand off the doorknob, resulting in a considerable drop in times. During one
treatment she was simply exposed to the closet, with the therapist nearby
(outside the door). In the second treatment the therapist administered social
praise contingent on her remaining in the room foran increasing period of
time. The two therapists alternated sessions with one another. In the original
experimental phase the therapists switched roles, but they returned to their
original reinforcing or nonreinforcing roles in the third phase. The data
indicate that reinforced sessions were consistently superior to nonreinforced
sessions.
Several procedural considerations deserve comment. First, the counterbal-
ancing was rather weak because the therapists switched roles only twice
during the whole experiment. Ideally, a more systematic counterbalancing
strategy would have been planned. Second, the treatments were not adminis-
tered randomly. Sessions involving exposure without contingent praise always
preceded exposure with contingent praise. Despite this fact, a clear superior-
ityof one treatment over the other emerged. Nevertheless, the experiment
600 -I
Experimental phases 1
550
5 500
O
450
^
O ^^
-
4- NRT
? 350
/)
300
o
< 250
2 200
Z 150
lU
a.
to
RT = Reinforcing therapist
100 Nonreinforcing therapist
NRT
n Therapist 1
- 50
Baseline o -o Therapist 2
4 5 6 7 8 9 10 11 12 14
BLOCKS OF FOUR SESSIONS
FIGURE 8-5. Comparison of effects of reinforcing and nonreinforcing therapists on the modi-
fication of claustrophobic behavior. (Figure 3, p. 1438, from: Agras, W. S., Leitenberg, H.,
Barlow, D. H., & Thomson, L. E. (1969). Instructions and reinforcement in the modification of
neurotic behavior.American Journal of Psychiatry, 125, 1435-1439. Copyright 1969 by the
American Psychiatric Association. Reproduced by permission.)
would be stronger with counterbalancing. Finally, one data point representing

a block of four sessions served as a baseline comparison. While formal
baseline phases are not necessary for ATD comparisons, and one baseline
point is perhaps better than none, the examination of trends is always more
informative than having simply a one-point pretest (or posttest).
The one indication of how far we have come in using the ATD to its fullest
potential can be found comparing the effectiveness of

in the next illustration,
two treatments for depression in an adult clinical population (McKnight,
Nelson, Hayes, & Jarrett, in press). Nine women diagnosed as depressed,
based on a Schedule for Affective Disorders and Schizophrenia (SADS)
interview, were included in this project. Subjects with strong suicidal tenden-
cies or on medication at the time of the initial interview were excluded from
the project, but all who eventually participated were severely depressed.
While depression is a problem with multiple components, two components

that play a prominent role in many depressed cases are irrational cognitions
and deficient social skills. In fact, treatment modalities with proven effective-
ness have concentrated on one or another of these problem areas. For
example. Beck's approach (A. T. Beck, Rush, Shaw, & Emery, 1979) concen-
trated on cognitive aspects of depression, and Lewinsohn, Mischel, Chaplin,
and Barton's (1960) concentrated on deficient social skills.
Careful assessment revealed that 3 depressive subjects were primarily de-
ficient in social skills, with few if any problems with irrational cognitions.
Another 3 subjects presented with clear difficulties with irrational cognitions
but few, if problems with social skills, while yet a third trio of subjects
any,
had difficulties inboth areas.
An ATD was used to compare social skills training and cognitive therapy in
each of the three sets of 3 subjects. The two therapies were randomly assigned
to 8 weeks of therapy such that each subject received four sessions of
cognitive therapy and four sessions of social skills therapy. Appropriate
counterbalancing was employed. The results for the first 2 trios of subjects
displaying either difficulties with irrational cognitions or difficulties with
social skills are presented in Figures 8-6 and 8-7.
One will notice, upon examining these figures, another experimental design
feature that adds to the elegance of this experiment.Not only were treatments
compared an ATD, but in each trio of three
in individual subjects with
subjects a multiple baseline across subjects design was implemented in order
to observe the effects of treatment, compared to the initial baseline, and to
insure that the effects of any treatment occurred only when that treatment
was introduced. This strategy, of course, controls for potential confounds
that are a function of multiple meaures and other conditions present during
baseline (see chapter 7). Thus this experimental design allows a determination
of the effects of treatment over baseline by means of a multiple baseline
across subjects design as well as a comparison of two treatments within the
ATD portion of the experiment.
Examining Figure 8-6, one can see that social skills training was the more
effective treatment for depression in each of the 3 subjects presenting with
social skills deficits, as indictedby scores on the Lubin Depression Adjective
Checklist. Social skills training was also significantly better on a measure of
social skills, the Interpersonal Events Schedule, than was cognitive therapy, as
would be expected. These findings were statistically significant. No significant
differences emerged on measures of irrational cognitions as assessed by the
Personal Beliefs Inventory.
In Figure 8-7, on the other hand, which presents data for the 3 subjects
experiencing primarily cognitive deficits, cognitive therapy was clearly supe-
rior to social skills training, on both measures of depression and measures of
irrational cognitions. These findings were also statistically significant. No
SOCIAL SKILL CROUP

Mifim fiiiisiii .umiii IlillSllt
ll
I

12341234S67I
>
WMkS
t I I > I t I I I < I
IMillli flllfSllf
mjicii
FIGURE 8-6. The effects of each treatment (COG = cognitive treatment; SS = social skill
treatment) in a multiple baseline design across the 3 subjects experiencing difficulties in social
skills on the weekly dependent measures administered. (Total score on the Lubin Depression
Adjective Checklist; Average score on the Personal Beliefs Inventory; Mean cross-product score
on the Interpersonal Events Schedule.) (Figure 2 from: McNight, D. L., Nelson, R. O., Hayes, S.
C, & Jarrett, R. B. (in press). Importance of treating individually assessed response classes in the
amelioration of depression. Behavior Therapy. Copyright 1984 by Association for Advancement
of Behavioral Therapy. Reproduced by permission.)
^
COGNITIVE GROUP
llSfllll llfllMflf llfllll TIEI1MIIT
2t mufCT 1 muiai
M
^T^:;
O
<2I
-M- - I - -I I 14
, IniJEci ,1 I
-I I I I t <
5 4
UIJICI

I
1 1 1 -I f^H <
yi
-I
KBIJECT

I
I
t > I r
I I I
>4
3
2 -
* J 'I
l^4'l234S(7l
12S41234S678
I
WMkS
I } 1 I I 1 t I > * I t
limill TIEITMIIT
1
(UlJECf
FIGURE 8-7. The effects of each treatment (COG = cognitive treatment; SS = social skill
treatment) in a multiple baseline design across the 3 subjects experiencing difficulties in irrational
cognitions on the weekly dependent measures administered. (Total score on the Lubin Depression
Adjective Checklist; Average score on the Personal Beliefs Inventory; Mean cross-product score
on the Interpersonal Events Schedule.) (Figure 4, from: McKnight, D. L., Nelson, R. O., Hayes,
S. C, & Jarrett, R. B. (in press). Importance of treating individually assessed response classes in
the amelioration of depression. Behavior Therapy.

statistically significant differences emerged on the measure of social skills,

however, for people with primarily cognitive deficits.
This very elegant experiment a model in many ways for the use of the
is
ATD The major conclusions derived from these

in adult clinical situations.
data concern the importance of carefully and specifically assessing depression
and all of its multiple components in order to tailor appropriate treatments to
the individual. While these data were not necessary for this presentation, the
third trio of subjects, displaying both irrational cognitions and social skill
deficits, benefited from both treatments. Furthermore, consistent with the
advantages of ATDs in investigating other problems, the results were apparent
rather quickly after a total of eight treatment sessions. Also, the two treat-
ments require the presentation of somewhat different therapeutic rationales
to the patients, but this does not present a problem in our experience, and it
did not in this experiment. Usually clients are simply told, correctly, that each
treatment is directed at a somewhat problem and/or
different aspect of their
that the experimenters are trying to determine which of two treatments
might be best for them. Contrast this experiment with the early example of an
ATD with adult clinical problems described earlier (Agras et al., 1969), and
one can see how far we have advanced our methodology. The elegant experi-
mental manipulations and the wealth of information available due to com-
bining the ATD with a multiple baseline across subjects make these data very
useful indeed.
In one final, good example of an alternating treatment design comparing
two treatments, Kazdin and Geesey (1977) investigated two different forms of
token reinforcement in a special education classroom. Two mentally retarded
children could earn tokens exchangeable for backup events for themselves or
for the entire class. Tokens were contingent on attentive behavior in the
classroom. Data from one of the children are presented in Figure 8-8. Data
on attentive behavior were collected in the classroom during two different
time periods each day. The two different conditions, earning tokens for
oneself or for the entire class, were counterbalanced across these time peri-
ods. Data from the lower panel illustrate the ATD. During baseline, rates of
attending behavior were essentially equal across time periods. During the
ATD, attentive behavior was higher when the subject could earn backup
reinforcers for the whole class. This condition was then implemented in the
final phase across both time periods. As indicated in the figure caption, data
were averaged in the upper panel to convey an overall level of attending

behavior during these phases. As in the Ollendick et al. (1981) experiment
described above, the baseline phase of this experiment provides the investiga-
tor with information on the naturally occurring frequency of the behavior
and therefore allows an estimate of the absolute extent of improvement, as
well as the relative effectiveness of the two conditions. In this experiment, the
ô
TOKEN RFT TOKEN RFTj

BASE (SELF 8 CLASS) (CLASS)
100
(C 80
o
> 60
<
X 40
UJ
OD
UJ
20
>
z
UJ
h- 100
H ^^P^
< 80
H
Z 60
UJ
o
C
UJ
40
Q.
20 SELF
CLASSo
15 10
DAYS
15 20
FIGURE 8-8. Attentive behavior of Max across experimental conditions. Baseline (base) no
experimental intervention. Token reinforcement (token rft) implementation of the token pro-
gram where tokens earned could purchase events for himself (selO or the entire class (class).
Second phase of token reinforcement (token rft 2) implementation of the class-exchange inter-
vention across both time periods. The upper panel presents the overall data collapsed across time
periods and interventions. The lower panel presents the data according to the time periods across
which the interventions were balanced, although the interventions were presented only in the last
two phases. (Figure 2, p. 690, from: Kazdin, A. E., & Geesey, S. (1977). Simultaneous-treatment
design comparisons of the effects of earning reinforcers for one's peers versus for oneself.
Behavior Therapy, 8, 682-693. Copyright 1977 by Association for Advancement of Behavior
ATD also served as a clinical assessment procedure, in that the investigators

were then able to implement the most successful treatment during the last
phase. Finally, the ATD phase of this experiment took only 8 days, demon-
strating once again the relative rapidity with which conclusions can be drawn
using this design. Naturally, this feature depends on the frequency of poten-
tial measurement occasions. With institutionalized patients or subjects in a
classroom, several experimental periods per day are possible. In outpatient

settings, however, measurement occasions might be limited to once a week, or
perhaps even once a month. Of course, the frequency of measurement
occasions is also the function of the particular behavior under study.
In the examples provided thus far, times of treatment administration and,
in some cases, therapists, have been counterbalanced so that the effects of the
treatments themselves become clear. Naturally, the ATD also makes it very
easy to examine directly the effects of different therapists, times of treatment
administration, or settings on a particular intervention. For example, two
therapists could alternately (and randomly) administer a treatment for gener-
alized anxiety disorder from a relatively fixed treatment protocol. Weinrott,
Garrett, and Todd (1978) examined the effects of the presence or absence of
an observer on social aggression in six elementary schoolchildren. The results
of the ATD demonstrated minimal observer reactivity in the situation.
Finally, as mentioned above, E. S. Shapiro et al. (1982) discovered that token
reinforcement was more effective in the morning than in the afternoon.
In some cases the setting in which treatment is administered becomes an
important question. Bittle and Hake (1977) discovered comparable rates of
reduction of self-stimulatory behavior in both experimental and natural
settings during the administration of a given treatment. In other contexts, the
implication of this work is that treatment can then be administered in the
natural setting, where less experimental or therapeutic control exists.
8.4. ADVANTAGES OF THE ALTERNATING

TREATMENTS DESIGN
The various and weaknesses of the ATD have been reviewed
strengths
before (Barlow & Hayes, 1979; Barlow et al., 1983; Ulman & Sulzer-Azaroff,
1975) and mentioned throughout this chapter. The major advantages and
disadvantages will be listed briefly once again. First, the ATD does not require
withdrawal of treatment. If two or more therapies are being compared,
questions on relative effectiveness can be answered without a withdrawal
phase at all. Ifcomparing treatment with no treatment, then one still
one is
would not require a lengthy phase where no treatment was administered.

Rather, no-treatment sessions are alternated with treatment sessions, usually
within a relatively brief period of time.
Second, an ATD will produce usful data more quickly than a withdrawal
design, all things being equal. This is because the relatively lengthy baseline,
treatment, and withdrawal phases necessary to establish trends in A-B-A
withdrawal designs are not important in an ATD design. The examples
provided in this chapter illustrate this point. In fact, the relative rapidity of an
ATD will often make it more suitable in situations where measures can be
taken only infrequently. For example, if it is only practical to take measures
infrequently, such as monthly, then an ATD will also result in a considerable

saving of time. In an example provided in Barlow et al., (1983), it was noted
that it often requires several hours and careful testing by two professional
staff in a physical rehabilitation center to work up a stroke patient's muscular
functioning. Obviously these measures cannot be taken frequently. If one were
testing a rehabilitation treatment program using an A-B-A-B design, with at
least three data points in each phase, then 12 months would be required to
evaluate the treatment, assuming that measures could be taken no more
frequently than monthly. On the other hand, if one month of treatment were
ahernated with one month of maintenance, then useful data within the ATD
format would begin to emerge after four months.
Third, trends that are extremely variable or rapidly rising or falling present
some problems for other single-case designs where interpretation of results is
based on levels and trends in behavior. But the ATD design is relatively
insensitive to background trends in behavior because one is comparing the
results of two treatments or conditions in the context of whatever background
trend is occurring. For example, if a specific behavioral problem is rapidly
improving during baseline, it would be problematic to introduce a treatment.
But in an ATD, two treatments could be alternated in the context of this
improving behavior, with the potential for useful differences emerging.
Finally, no formal baseline phase is required.
Naturally, these advantages vis-a-vis other design choices, apply only to
situations where other design choices are indeed possible. There are many
situations where other experimental designs are more appropriate for ad-
dressing the question at hand. Furthermore, the ATD suffers from the, as
yet, unknown effects of multiple-treatment interference, and although recent
research indicates that this problem may not be a great as once feared, we
must still await systematic investigation of this issue to proceed with certainty.
In any case, when it comes to generalizing the results of single-case experi-
mental investigations to applied situations, there seems little question that the
first treatment phase of an A-B-A-B design (or a multiple baseline design) is
closer to the applied situation than is a treatment that is rapidly alternated

with another treatment or with no treatment. Thesf* are only a few of the
many factors the investigator must consider when choosing an appropriate
experimental design.
8.5. VISUAL ANALYSIS OF THE ALTERNATING

TREATMENTS DESIGNS
If enough data points have been collected for each treatment, and if one is
so inclined, a variety of statistical procedures are appropriate for analyzing

alternating treatment designs (see chapter 9). However, visual analysis should
suffice for most ATDs. Throughout this book, the visual analysis of single-
case designs is discussed in terms of observation of both levels of behavior

and trends in behavior across a phase. Within at ATD, as noted above, levels
and trends in behavior are not necessarily relevant because the major com-
parison is between two or more series of data points representing two or more
treatments or conditions. To date, most investigators have been relatively
conservative, in that very clear divergence among the treatments has been
required. In most cases the series have been nonoverlapping. For example,
with the exceptions of Points 1 and Points 1 1 which represented data points
,
immediately following the switch in therapists, the Agras et al. (1969) ATD
presented nonoverlapping series (see Figure 8-5).
Kazdin and Geesey (1977) also presented two series of data from the two
treatments tested in their experiment which do not overlap, with the exception
of one point very early in the ATD experiment (see Figure 8-8). Also, these
data diverge increasingly as the ATD proceeds. Finally, Ollendick, Shapiro,
and Barrett (1981) demonstrated a clear divergence between treatment and no
treatment (see Figures 8-3 and 8-4). When one examines the effects of the two
treatments, several data points overlap initially, but the two series increasingly
diverge as the ATD proceeds. One must also remember that in this particular
experiment (Ollendick et al., 1981) there were no clear signs or signals
discriminating the treatments, and therefore this overlap may reflect some
confusion about which treatment was in effect early in the experiment.
If overlap among the series occurs, then there is little to choose among the
treatments or conditions, and most investigators say so. For example,
Weinrott et al. (1978) observed considerable overlap between observer-present
and observer-absent conditions in their experiment and concluded that obser-
ver reactivity was not a factor. Last, Barlow and O'Brien (1983) also ob-
served overlap between two cognitive therapies and concluded that each was
effective. Of course, when some overlap does exist, it is possible to utilize
statistical procedures to estimate if any differences that do exist are due to
chance or not (e.g., McKnight et al., 1983, Figure 8-7; E. S. Shapiro et al.,
1982, Figure 8-2). However, as discussed in chapter 9, one must then decide if
these rather small effects, even if statistically significant, are clinically useful.
Our recommendation for these designs, and throughout this book, is to be
conservative and to look for large visually clear, clinically significant effects.
On the other hand, the ATD lends itself to a wide number of statistical tests,
as outlined by Edgington (1984) and reviewed in chapter 9. Many of these
tests require relatively few data points in each series. For example, using some
of the examples presented in this chapter, Edgington (1984) has demonstrated
how a variety of tests would be applicable to these data sets.
8.6. SIMULTANEOUS TREATMENT DESIGN

In the beginning of the chapter we noted the existance of a little-used design
that actually presents two or more treatments simultaneously to an individual
subject. In the first edition of this book, this design was referred to as a
10 total frequency
o
B (B) positive attention
(C) verbal admonishn>ent
9 purposely ignore
U^
hZ 9
(D)
-> 8
S
"S ^* /
hs
^ s 6
tX
5
s.^
H
^
B
s
S.
4
3 3
rS" 2
8 9 10 11
uncontrolled Icontrolledl B.CD I

D treatment
baseline | baseline jtroitmontij
WEEKS
FIGURE 8-9. Total mean frequency of grandiose bragging responses throughout study and for
each reinforcement contingency during experimental period. (Figure 3, p. 241, from: Browning,
R. M. (1967). A same-subject design for simultaneous comparison of three reinforcement
contingencies. Behaviour Research and Therapy, 5, 237-243. Copyright 1967 by Pergamon Press.
concurrent schedule design. But the implication that a distinct schedule of

reinforcement is attached to each treatment produces the same unnecessary
narrowness as calling an alternating treatments design a multiple schedule
design. Browning's (1967) term, simultaneous treatment design, seems both
more descriptive and more suitable. Nevertheless, both terms adequately
describe the fundamental characteristic of this design the concurrent or
simultaneous application of two or more treatments in a single-case. This
contrasts with the fast alternation of two or more treatments in the AID. The
only example of the use of this design in applied research of which we are
aware is the original Browning (1967) experiment, also described in Browning
and Stover (1971). In this experiment. Browning (1967) obtained a baseline
on incidences of grandiose bragging in a 9-year-old child. After 4 weeks, three
treatments were used simultaneously: (1) positive interest and praise con-
on bragging, (2) verbal admonishment, and (3) ignoring. Each treat-
tingent
ment was administered by a team of two therapists who were staff in a
residential college for emotionally disturbed children. To control for possible

differential effects with individual staff, each team administered each treat-
ment for one week in a counterbalanced order. For example, the second group
of two therapists admonished the first week, ignored the second week, and
praised the third week. All six of the staff involved in the study were present
simultaneously to administer the treatment. Browning hypothesized that the
boy "... would seek out and brag to the most reinforcing staff, and shift to
different staff on successive weeks as they switched to S's preferred reinforce-
ment contingency" (p. 241). The data from Browning's subject (see Figure 8-
9) indicate a preference for verbal admonishment, as indicated by frequency
and duration of bragging, and a lack of preference for ignoring. Thus
ignoring became the treatment of choice and was continued by all staff.
In this experiment the effects of three treatments were observed, but it is
unlikely that a subject would be equally exposed to each treatment. In fact,

the very structure of the design ensures that the subject won't be equally
exposed to all treatments because a choice is forced (except in the unlikely
event that all treatments are equally preferred). Thus this design is unsuitable
for studying differential effects of treatments or conditions.
The STD might be useful anytime a question of individual preferences is
important. Of course, in some cases preferences for a treatment may be an

important component of its For example, if one is
overall effectiveness.
treating a phobia, one of two cognitive procedures combined with
and either
exposure-based therapy is equally effective, the client's preference becomes
very important. Presumably a client would be less likely to continue using,
after treatment is terminated, a fear-reduction strategy that is less preferred
or even mildly aversive. But the more preferred or least aversive treatment
procedure would be likely to be used, resulting most likely in a more favor-
able response during follow-up. Similarly, one could use an STD to determine
the reinforcing value of a variety of potential consequences before introduc-
ing a program based on selective positive reinforcement. But it is also possible
that a particular subject might prefer reinforcing consequences or treatments
that are less effective in the long run. The investigator must remember that
preference does not always equal effectiveness. The STD, then, awaits imple-
mentation by creative investigators studying areas of behavior change or
psychopathology where strong experimental determinations of behavioral
preference are desired. Presumably, these situations will be such that the self-
report resulting from asking a subject about his or her preference will not be
sufficient, for a variety of reasons. When these questions arise, the STD can
be a very powerful tool for studying preference in the individual subject. But
the STD is not well suited to an evaluation of the effectiveness of behavior
change procedures.
CHAPTER 9
Analyses for
Statistical
Single-case Experimental Designs
by Alan E. Kazdin*
9.1. INTRODUCTION
Data evaluation consists of methods that are used to draw conclusions about
behavior change. In applied research where single-case designs are used,
experimental and therapeutic criteria are invoked to evaluate data (Risley,
1970). The experimental criterion refers to the way in which data are evaluated
to determine if an intervention has had a reliable or veridical effect on behav-
ior. The experimental criterion is based on a comparison of behavior under
different conditions, usually during intervention and nonintervention (base-

line) phases. To the extent that performance reliably varies under these separate
conditions, the experimental criterion has been met.

The therapeutic criterion refers to whether the effects of the intervention are
important. This criterion entails a comparison between behavior change that
has been accomplished and the level of change required for the client's ade-
quate functioning in society. Even if behavior change is reliable and clearly
related to the experimental intervention, the change may not be of clinical or
applied significance. To achieve the therapeutic criterion, the intervention
needs to make an important change in the client's everyday functioning.
Completion of this chapter was facilitated by a Research Scientist Development

Award (MH00353) from the National Institute of Mental Health.
Please address all correspondence to: Alan E. Kazdin, Department of Psychiatry,
University of Pittsburgh School of Medicine, Western Psychiatric Institute and Clinic,
3811 O'Hara Street, Pittsburgh, PA 15213.
SCED J*
285
Within single-case research, data can be evaluated in different ways to

address the experimental and therapeutic criteria. Visual inspection is the most
commonly used method of evaluating the experimental criterion and consists
of examining a graphic display of the data (see Baer, 1977a; Michael, 1974).
The data are plotted across separate phases of the single-case design. A
judgment is made about whether the requirements of the design have been met,
to draw a causal relationship between the intervention and behavior change. To
those unfamiliar with the method, visual inspection seems to be completely
subjective and free from specifiable criteria that guide decision making. Yet for
visual inspection to be applied, special data requirements need to be met. Also,
the data are visually inspected according to specific criteria (e.g., changes in
trend, latency of the change at the point of intervention) to indicate whether the
changes are reliable (see Kazdin, 1982b; Parsonson & Baer, 1978).
Statistical analysis represents another method of data evaluation in single-
case research. Statistical tests provide a quantitative method and a set of rules
to determine if a particular experimental effect is reliable. Statistical tests do
not eliminate judgment from data evaluation. Rather, they provide replicable
methods of evaluating information and reaching a conclusion about the ex-
perimental criterion. For statistical evaluation, a level of confidence (signifi-
cance), decided by consensus, is used as a criterion to define whether a change
in behavior is reliable (i.e., meets the experimental criterion). Judgment still
enters into data analysis in terms of defining the datum, selecting the unit of
analysis, identifying the statistical test, and so on. But the analyses themselves
consist of replicable computational methods and rules for making decisions
about the data.
Visual inspection and statistical data evaluation address the experimental
criterion for single-case research. The applied, or clinical, significance of the
change also is important. The therapeutic criterion has been addressed in
different ways (Kazdin, 1977; Wolf, 1978). One method is to evaluate if the
changes in the client's behaviors bring him or her within the level of his or her
peers who are functioning adequately in society. For example, in the case of
treatment for deviant behavior, a clinically significant change is achieved if the
client's behavior after treatment falls within the range of persons who have not
been identified as having problems. Another method is to have various persons
and other people in everyday life) evaluate the
(the client, relatives, experts,
magnitude of change achieved by the client. If such persons perceive a distinct
improvement in behavior or qualitative differences before and after treatment,
the results suggest that the change is of applied significance.
The purpose of the present chapter is to detail statistical analyses for single-
case experimental designs. The statistical analyses need to be viewed in the
context of other methods of data evaluation to which they are compared. In
between-group research, statistical analysis obviously has been widely adopted
and accepted as the method of data evaluation. Even though questions are
Statistical Analyses for Single-case Experimental Designs 287
occasionally raised about whether statistical significance is an appropriate
criterion,whether certain types of tests should be used, and so on, they remain
in the background in terms of the actual conduct of research. Within single-
case research, application of statistical tests is far less well developed or
established. The types of statistical tests available are not widely familiar, and
their appropriate application has relatively few exemplars (Kratochwill, 1978b;
Kratochwill & Brody, 1978). More basic than the application of the tests is the
question of whether such tests should be used at all in single-case research. The
present chapter discusses issues regarding the use of statistical analyses in
single-case research. However, major emphasis will be given to various tests
themselves and how they are applied. Advantages and limitations in applying
particular tests will be presented as well.
9.2. SPECIAL DATA CHARACTERISTICS

Most research between-group designs,
in the behavioral sciences utilizes
where multiple subjects are observed one or a few points in time. Parametric
at
statistical analyses are applied that invoke several assumptions about the
nature of the data and the population from which subjects are drawn. In single-
case research, one or a few individuals are observed at several different points
in time. Statistical tests applicable to group studies may not be appropriate for
single cases where data are collected over time.
Serial dependency
In applications of analyses of variance in group research, researchers are

familiar with the fact that the tests are "robust" and can handle the violation of
various assumptions (e.g., Atiqullah, 1967; G. V. Glass, Peckham, & Sanders,
1972; Scheffe, 1959). There is one assumption which, if violated, seriously
affects analysis of variance and makes t or F tests inappropriate. The assump-
tion is the independence-of-error components. The assumption refers to the
correlation between the error {e) components of pairs of observations (within
and across conditions) for andy subjects. The expected value of the correla-
/
tion for pairs of observations is assumed to be zero (i.e. r^^ = 0). Typically, in
, .
between-group designs, independence-of-error components are assured by

randomly assigning subjects to conditions. In the case of continuous or re-
peated measures over time, the assumption of independence-of-observations
often is not met. Successive observations in a time series tend to be correlated,
inwhich case the data are said to be serially dependent. The correlation among
successive data points means that knowing the level of performance of a
subject at a given time allows one to predict subsequent points in the series.
The extent to which there is dependency among successive observations can
be assessed by examining autocorrelation in the data. Autocorrelation refers to

a correlation(r) between data points separated by different time intervals (lags)
in the series. An autocorrelation of lag 1 (or r,) is computed by pairing the

initial observation with the second observation, the second with the third, the
third with the fourth, and so on throughout the time series. Autocorrelation of
lag 1 yields the correlation coefficient that reflects serial dependency. If the
correlation is significantly different from zero, this indicates that performance
at a given point in time can be predicted from performance on the previous
occasion (the direction of the prediction determined by the sign of the autocor-
relation).
Generally, autocorrelation of lag 1 is sufficient to reveal serial dependency in
the data. However, a finer analysis of dependency may be obtained by comput-
ing several autocorrelations with different time lags (e.g., autocorrelations of
and so on). For the general case, an autocorrelation of the lag t is
lags of 2, 3, 4,
computed by pairing observations / data points apart. For example, autocorre-
lation of lag 2 is computed by pairing the initial observation in the series with
the third, the second with the fourth, the third with the fifth, and so on.
Serial dependency throughout the time series is clarified by computing and
plotting correlations of different lags.^ The plot of the autocorrelations is
referred to as a correlogram. Figure 9-1 provides correlograms (i.e., autocorre-
lations plotted as a function of different lags) for two hypothetical sets of data.
In each correlogram, the point that is plotted reflects the correlation coefficient
for observations of a given lag. As can be seen for the data in the upper portion
of the figure, the correlations with short lags are positive and relatively high. As
the lag (i.e., the distance between the data points) increases, the autocorrela-
tion approaches zero and eventually becomes negative. The hypothetical data
in the upper portion of Figure 9-1 reflect serial dependency because the
autocorrelation of lag 1 is likely to be significantly different from 0.^ Moreover,
the correlogram reveals that the dependency continues beyond lag 1 until the
autocorrelation approaches 0. In contrast, the lower portion of Figure 9-1
reveals a hypothetical correlogram where the observations in the time series are
not dependent. The autocorrelations do not significantly deviate from 0. The
lack of dependence signifies that the errors of successive observations are
"random," that is, a data point below the "average" value is just as Hkely to be
followed by a high value as by another low value. Time series data that reveal
this latter pattern can be treated as independent observations and can be
subjected to conventional statistical analyses.
When autocorrelation is significant, serious problems occur if conventional
analyses are used (Scheff^, 1959). Initially, serial dependency reduces the
number of independent sources of information in the data. The degrees of
freedom based upon the actual number of observations is inappropriate be-
cause it assumes that the observations are independent. Any F test is likely to
overestimate the true F value because of an inappropriate estimate of the

13 15
+I.Or
.8-
.6-
.4-
'
0-
-.2-
-.4-
-.6-
LAG
-.8-
-I.OL
FIGURE 9-1. Correlograms for data with (upper portion)
and without serial dependency (lower portion).
degrees of freedom. For the appropriate application of / and F tests, the

degrees of freedom must be independent (uncorrelated) sources of informa-
tion. A second and related problem associated with dependency is that the
autocorrelation spuriously reduces the variability of the time series data. Thus,
error terms derived from the data underestimate the variability that would
result from independent observations. The smaller error term inflates or

positively biases E In general, significant autocorrelation can greatly bias
and t
Ftests. Use of these tests when the data are serially dependent can lead to Type
I and Type II errors, and simple corrections to avoid these biases (e.g.,
adjustment of probability level) do not address the problem. (In passing, it may
be important to note as well that serial dependency in the data can also bias the
conclusions reached through visual inspection as well as statistical analyses [see
R. R. Jones, Weinrott, & Vaught, 1978].)
General comments
Serial dependency is not a necessary characteristic of single-case data or

observations over time. However, significant autocorrelation is a likely charac-
teristic of continuous data and is a central consideration in deciding if particu-
lar statistical tests should be applied to single-case data. Several statistical tests
for single-case data, including variations of t and F, are presented below. The
tests vary as to whether they acknowledge, take into account, or are influenced
by serial dependency in the data.
9.3. THE ROLE OF STATISTICAL

EVALUATION IN SINGLE-CASE RESEARCH
Sources of controversy
The use of statistical analyses has been a major source of controversy

because the approach embraced by such analyses appears to conflict with the
purposes of single-case research and the criteria for identifying effective inter-
ventions. To begin with, identifying reliable intervention effects does not
necessarily require statistical evaluation, as implicitly assumed in between-
group research. In single-case research, demonstration of a reliable effect (i.e.,
meeting the experimental criterion) is determined by replication of intervention
and baseline levels of performance over the course of an experiment, as is
commonly illustrated in A-B-A-B designs. Other single-case experimental de-
signs replicate intervention effects in different ways and permit comparisons to
be made between what performance would be with and without treatment. In
practice, whether the results clearly meet the experimental criterion depends
upon the pattern of the data in light of the requirements of the specific design.
Several characteristics such as changes in means or slope across phases, abrupt
shifts or repeated changes in performance as an intervention is presented and
withdrawn, and similar characteristics can be used to evaluate intervention
effects without inferential statistics (Kazdin, 1982b).
Statistical criteria are objected to in part because of the goal of applied
single-case research. The goal is to identify and evaluate potent interventions
(Baer, 1977a; Michael, 1974). Visual inspection, the method commonly used to
evaluate single-case data, is viewed as a relatively /^sensitive method for

determining an intervention has been effective. Only marked effects are
if
likely to be regarded as reliable through visual inspection. In contrast, statisti-
cal analyses may identify as significant subtle changes in performance. The

tests may detect changes in performance that are not replicable. Indeed, within
statistical evaluation, the possibility exists that the findings were obtained by
"chance."
Single-case research designs do not necessarily require visual inspection or
statistical analysis as a method of data evaluation. However, applied research
where single-case designs are used (applied behavior analysis) has emphasized
the importance of searching for potent intervention effects and subjecting the
data to visual inspection rather thanstatistical evaluation. The two different
methods are not fundamentally different, but they do vary in the sorts of
effects that are sought and the manner in which decisions are reached about
intervention effects.'
Some of the objections to statistics in single-case research have stemmed
from the focus on groups of subjects in between-group research. Within-group
variability is often a basis for evaluating the effect of interventions in group
research. Yet, within-group variability is not part of the behavioral processes of
individual subjects and perhaps should not be included
in the evaluation of
performance (Sidman, 1960; also see chapter 2). Related group research often
obscures the performance of the individual subject. Statistical analyses usually
reflect the performance of the group as a whole with data characteristics
(means, variances) that do not bear on the performance of any single subject. It
remains unclear how the intervention affects individuals and the extent to
which group performance represents individual subjects. As these objections
illustrate, concerns over statistical analyses extend beyond the manner in which
data are evaluated. The objections pertain to fundamental issues about experi-
mental design and the approach toward research more generally (J. M. John-
ston & Pennypacker, 1981;Kazdin, 1978).
Potential contributions
Statistical analyses in single-case research may provide a valuable supple-

ment rather than an alternative to visual inspection. In many applications,
inferences about the effects of the intervention can be readily drawn through
visual inspection. Statistical analyses in such situations may not add an incre-
ment of useful information unless a specific question arises about a particular
facet of the data at a given point in time. In many situations, the pattern of data
required for visual inspection may not be met, and statistical tests may provide
important advantages.
Evaluation of intervention effects can be difficult when performance during
baseline is systematically improving. An intervention may still be required to
accelerate the rate of change. For example, self-destructive behavior of an
autistic child might be decreasing gradually during baseline but an intervention

may be required to achieve more rapid progress. Visual inspection is often
difficult to invoke with a baseline trend reflecting improvement. Selected
statistical analyses (discussed later in the chapter) can readily examine whether
a reliable intervention effect has been achieved over and above what would be
expected by continuation of the initial trend. Thus statistical analyses provide
an evaluative tool in cases where visual inspection may be difficult to invoke.
Apart from trend in baseline, visual inspection is also difficult to invoke if
data show relatively high variability within and across phases. Single-case
research designs have been applied in a variety of settings such as psychiatric
hospitals, institutions, classrooms, and others. In such settings, investigators
have frequently been able to control several features of the environment such
as staff behavior and activities of the clients, in addition to the intervention.
Because extraneous factors are held relatively constant for purposes of experi-
mental control, variability in subject performance can be held to a minimum.
Visual inspection is more readily applied to single-case data when variability is
small.
Over the years, single-case research has been extended to several community
or open-field settings (Geller, Winett, & Everett, 1982; Kazdin, in press). In
such extensions, control over extraneous factors in the situation may be
minimal. Moreover, the persons who serve as subjects may change over the
course of the project, so that the effect of the intervention is evaluated against
the backdrop of intrasubject and intersubject variability. Increased variability
inperformance decreases the likelihood of demonstrating marked effects in
performance and the ability of visual inspection to detect reliable changes.
Statistical evaluation may provide a useful aid in detecting if the intervention
has produced a reliable effect.
Proponents of applied single-case research have stressed the need to investi-
gate interventions that produce potent effects. Yet there may be different
situations where it is important to detect reliable intervention effects, even if
relatively small. To begin with, investigators may embark on new lines of
research where the interventions are not well developed. The interventions may
not be potent at this stage because of lack of information about the interven-
tion or the conditions that maximize its efficacy. Statistical analyses at this
initial stage of research may help identify interventions and variables that
produce reliable effects. More stringent criteria of visual inspection might lead
to abandonment of interventions that do not produce marked effects at the
outset. Yet identification of procedures through statistical analyses may help
screen among variables that warrant further pursuit. Interventions identified in
this fashion might be developed further through subsequent research and
perhaps eventually produce large effects that meet the criteria of visual inspec-
tion. But, at the initial stage of research, statistical analyses may serve a useful
purpose in identifying variables that warrant further scrutiny and develop-
ment.
Statistical Analyses for Single<ase Experimental Designs 293
It may be important As applied

to detect small effects in other situations.
research has been extended to community settings, small changes in the behav-
iors of individual subjects have become increasingly important. These changes,
when accrued across many persons, become highly significant. For example,
small changes in energy consumption within individuals are important because
such effects become socially significant when extended on a larger scale. Also,
in community applications, small changes in performance may be important to
detect because of the significance of the behaviors. For example, interventions
designed to reduce violent crimes in the community may produce minute
effects that do not pass the test of visual inspection. Yet small but reliable
changes are important to detect because of the significance of any change in
such behaviors.
General comments
The controversy over statistical analyses is not whether all data in single-case
research should be evaluated statistically. Single-case research designs, the
tradition from which they derive, and the dual concerns in applied work for
experimental and therapeutic criteria for evaluating change all place limits on
the role of statistical analysis. Within the approach of single-case research, the
question is whether statistical tests can be of use in situations where visual
inspection might be difficult to apply. There are different reasons for posing an
affirmative answer. Although visual inspection can be readily applied to many
investigations, the method has its own weaknesses. In a variety of circum-
stances, researchers often have difficulty in judging (via visual inspection)

whether reliable effects have been produced and disagree in their interpreta-
tions of the data (DeProspero & Cohen, 1979; Gottman & Glass, 1978; R. R.
Jones et al., 1978). Also, systematic biases may operate when invoking visual
inspection criteria, such as ignoring the impact of autocorrelation and being
influenced by the metric by which data are graphed (R. R. Jones et al., 1978;
Knapp, 1983; Wampold & Furlong, 1981a). An attractive feature of statistical
analyses is that once the statistic is decided, the results are (or should be)
consistent among different investigators. Judgment plays less of a role in
applying a statistical analysis to the data. Thus statistical analyses can be a
useful tool in cases where the idealized data patterns required for visual
inspection are not obtained.
9.4. SPECIFIC STATISTICAL TESTS
There are a large number of statistical tests that can be applied to data
obtained from a single subject over time. The range of available tests has not
been conveniently codified or illustrated. Indeed, the task is rather large
because a given test might be applied in a variety of different ways depending
on the specific variant of single-subject designs and the statement the investiga-
tor wishes to make about the intervention. Several tests discussed below
illustrate major variants currently available but do not exhaust the range of
appropriate tests.
Conventional t and F tests

Although many different statistical tests are available for single-case de-
signs, certainly themost familiar are / and F tests. Each single-case design
includes two or more phases that can be compared with a / or Ftest depending,
of course, on the number of different conditions or phases. For example, in an
A-B-A-B design, comparisons can be made over baseline (A) and intervention
(B) phases. An obvious test would be to compare A and B phases (/ test) or to
compare the four A-B-A-B phases (analysis of variance). The test would
evaluate whether the difference(s) between (or among) means is statistically
significant.
If the single-case design is applied to a group of subjects, correlated /-test or
repeated-measures analyses of variance can be performed. For data from an
individual subject, t and F tests may not be appropriate if the data are serially
dependent. A test is appropriate if autocorrelation is computed and shown to
be nonsignificant.
Consider, as an example, hypothetical data for a socially withdrawn child
who received reinforcing consequences at school for interacting with peers.
Consider data from the first two (AB) phases of an A-B-A-B design. The
change from baseline to intervention phases can be evaluated with a t test.

Table 9-1 presents the data for each day, where the numbers reflect the percent-
age of intervals of appropriate social interaction. The baseline phase tends to
show lower rates of performance than the intervention phase, but are the
differences statistically significant?
To first assess if the data are serially dependent, autocorrelations are com-
puted for the separate phases. The autocorrelations are computed within each
phase rather than for the data across both phases, because the intervention may
influence the relation of data points to each other (i.e., their dependency). As
shown in the table, neither autocorrelation is statistically significant. The data
appear to meet the independence-of-error assumption and can be subjected to
conventional / testing. The results of a / test for independent observations (or
groups) and for unequal sample sizes indicate that A and B phases were
significantly different (/(25) = 6.86, /?<.01). Thus the differences in social
behavior between the two phases are reliable.
Variations of t and F tests

Variations of t and F have been suggested for situations where autocorrela-
tion is significant and the data are dependent. Prominent among the sugges-
TABLE 9-1. r test Comparing Hypothetical Data

for A and B Phases for One Subject
BASELINE (A) INTERVENTION (B)
DAYS DATA DAYS DATA
1 12 13 88
2 10 14 28
3 12 15 40
4 22 16 63
5 19 17 86
6 10 18 90
7 14 19 82
8 29 20 95
9 26 21 39
10 5 22 51
11 11 23 56
12 34 24 86
25 31
26 77
27 76
Mean (A) = 17.00 Mean (B) = 65.87
Autocorrelation r = .005 --
Autocorrelation r = .010
(lagl) (lag 1)
tions is the analysis proposed by Gentile, Roden, and Klein (1972). When
autocorrelation exists, these investigators suggested that nonadjacent phases
that employed the same treatment can be combined and will reduce the effect
of serial dependency. For example, in an A-B-A-B design, the two A phases are
not adjacent and could be combined and compared with the two B phases. The
rationale for combining phases is based on the fact that autocorrelations tend
to decrease as the lag between observations increases. Assuming serial depen-
dency in the data. Observation 1 in phase A, would be more highly correlated
with Observation 1 in Phase B, (i.e. the immediately adjacent phase) than with
,
Observation 1 in phase A2 (i.e., a nonadjacent phase). Since the error compo-

nents of all more like the components for the observa-
observations in A, are
tions in B, than in A2, assumed that combining treatments separated in time
it is
will reduce the dependency. Combining phases that are not adjacent should
make A and B treatments more dissimilar, due to dependency in the data. The
resulting t (or F) should be reduced because the dependency of adjacent
observations will minimize treatment differences. Additional variations of /
and Fhave been proposed, some of which attempt to address the issue of serial
dependency by developing special error terms to make statistical comparisons
of treatment effects (see Gentile et al., 1972; Shine & Bower, 1971).
Considerations and limitations of t and F tests
Appropriateness of the There is considerable agreement that t and F tests

Tests.
are not appropriate if from a single subject are serially dependent

the data
(Hartmann, 1974; Kratochwill et al., 1974; Thoresen & Elashoff, 1974). The
variations alluded to above do not clearly resolve the issues. The effects of
trying to compensate for serial dependency (e.g., by combining phases) are not
easily estimated and no doubt vary with different patterns of autocorrelation.
The safest approach is to precede / and F tests with an analysis of serial
dependency. If significant autocorrelation exists, alternative statistical tests
should be considered.
Evaluation of Means. Another issue may influence selection of / or F tests.

Typically, these analyses, when appropriate, are applied to test whether or not
there are significant changes in means between or among phases. Trends in the
data are ignored. It is an accelerated slope in
possible, for example, that
baseline and intervention phases is apparent, in which case each data point may
exceed the value of the preceding point. A simple test of means across A and B
phases could reflect a statistically significant effect, but the effect might be
accounted for by the trend. Alternatively, the data might show an increasing
slope in baseline and a decreasing slope in treatment, with no overall mean
differences. A test of means in both the above instances would lead to interpre-
tive problems if the trends were ignored. The need to consider trend and mean
changes as well as other data parameters is clarified in the discussion of time
series analysis.
9.5. TIME SERIES ANALYSIS

Time series analysis compares data over time for separate phases for an
individual subject or group of subjects (see G. V. Glass et al., 1974; Gottman,
1981; Hartmann et al., 1980; R. R. Jones, Vaught, & Weinrott, 1977). The
which alternative phases (e.g.,
analysis can be used in single-case designs in
baseline and intervention) are compared. There are two important features of
time series analysis for single-case research. First, the analysis provides a t test
that is appropriate when dependency in the data. Second, the

there is serial
analysis provides important information about different characteristics of

behavior change across phases. The notion of serial dependency has been
addressed already. The different features of the data that time series analysis
reveals require a brief digression.
Patterns of change in time-series data
Continuous observations across separate phases may indicate change along

several dimensions. Three dimensions that are especially relevant in under-
standing time series analysis include change in level, change in slope, and
presence or absence of slope in a given phase (R. R. Jones et al., 1977). A

change in change at the point in which the intervention is made.
level refers to a
If data at the end of baseline and the beginning of intervention phases show an
abrupt departure or discontinuity, this would reflect a change in level. A change

in slope refers to a change in trend between or among phases.
The notion of a change in level warrants further mention because it differs
from the more familiar concern of a change in mean across phases. A change in
mean across phases refers to differences in the average performance. A change
in level does not necessarily entail a change in mean, and vice versa. However, a
change in one does entail a change in the other when there is no slope in the data
in either baseline or intervention phases. Applied researchers are concerned
primarily with a change in means. Whether or not there is a change in the
precise point of intervention (i.e., beginning of the B phase) is not necessarily
crucial as long as behavior shows a marked overall increase or decrease.
Time series analysis provides separate tests of a change in level and a change
in slope. A change in mean can be inferred from these other parameters. For
example, a very gradual change in behavior after the intervention is applied
might be detected as a significant change in slope but no change in level. The
absence of change in level indicates that behavior did not change abruptly at
the point of intervention. The significant change in the slope would imply a
change in the means across phases. An advantage of time series analysis is that
the nature of the change across phases is examined in a more analytic fashion
than by merely evaluating overall means. Because separate tests are provided
for changes in slope and level, there is no requirement that baseline phases
show little or no trend in the data. The test allows one to evaluate whether any
trend in an intervention phase departs from the slope in baseline, if one exists.
To convey how changes in level and slope can appear in single-case data,
several different data patterns are illustrated in Figure 9-2. The figure provides
hypothetical data over two phases (AB) of a larger design. The data patterns
some of the relationships among changes in
illustrate level and slope and in
means across phases. Also, some of the data patterns (e.g.. Figures 9-2a, 9-2b,
and 9-2c) represent instances where visual inspection presents problems be-
cause of the presence of an overall trend across baseline and intervention
phases. Conventional / and F tests that examine changes in means might
overlook important changes when means do not change (as in Figure 9-2d), or
they may indicate a significant change when in fact level or slope have not
changed (e.g., as in Figure 9-2b).
Data analysis
The actual analysis itself cannot be outlined in a fashion that permits simple
computation. Time depends upon more than entering raw data
series analysis
models of time series analysis exist that make

into a single formula. Several
different assumptions about the data and require different equations to
B B
Q 1
>
<
X
1
' ^^--^^^
^^ i
1
^
^^^
UJ ^ ^^^0*^^"^ ^^^
<D x^
j
U.
O
\
^^
^y'^x
UJ ^^.^""^j
^^^^^'"''^ y^
^0^ \
l- 1 1
< 1
(T
a. Change in level; no b. No change in level

change in slope or slope
A , B A , B
tr.
o /
>
[
/
/
< 1
/
X 1
/
UJ '
(D \
/ L
u. \/ >^ 1 >v
o
^^X
1 ^N^
UJ 1 >>^^
H ^^^,,,*
-^^ 1
/^ 1 N,^
< .^^"^'''''^ 1
q:
c. Change in level and d. No change in level;
slope change in slope
A ^ B A B
q:
o [ y
j/^
> 1
< 1 >^
X 1
>^
>^
>^
y
UJ
X
1
CD 1
y/^
j/^
U.
O
1
y^
Ul
1-
<
or
e. No change in level, f Change in level an

change in slope slope
FIGURE 9-2. Examples of selected patterns of data

over two phases (AB), illustrating changes in level and/or trend.
achieve the final The analysis begins by evaluating serial dependency

statistics.
in the data. Different patterns of dependency may emerge that depend upon
the pattern of autocorrelations, which are computed with different lags or
intervals, as noted earlier. Once the pattern of serial dependency is identified, a
model is applied to the data. The analysis consists of several steps, including
adoption of a model that best fits the data, evaluation of the model, estimation
of parameters for the and generation of t for level and slope changes
statistic,
(G. V. Glass et al., 1974; Gorsuch, 1983; Gottman, 1981; Home, Yang, &
Ware, 1982; Stoline, Huitema, & Mitchell, 1980). Computer programs are
available to handle these steps (see Gottman, 1981; Hartmann et al., 1980).
It is useful to examine the results of a time series analysis for illustrative
purposes and to evaluate the results in light of the characteristics of the data
that might be inferred from visual inspection. As an illustration, one program
focused on the frequency of inappropriate talking in a second-grade classroom
(C. Hall et al., 1971, Exp. 6). Although there were many children in class, the
class as a whole was treated as a single subject. The intervention consisted of
praise and other reinforcers provided to children for their appropriate class-
room behavior. The effects of the intervention, evaluated in an A-B-A-B
design, are plotted in Figure 9-3. The results suggest that inappropriate talking
out was generally high during the two different baseline phases and was much
lower during the different reinforcement phases (praise, tokens plus a sur-
prise). The first two phases (AB) have been analyzed using time series analysis
(R. R. Jones, Vaught, & Reid, 1975). Through a computer program, the
analyses revealed that the data were serially dependent, that is, the adjacent
points were significantly correlated. Indeed, autocorrelation for lag 1 was .96
(p<.01). Thus conventional t and F test analyses would be inappropriate.

Time series analyses revealed a significant change in level across the first two
phases (AB) (/(39) = 3.90, p < .01) but no significant change in slope. change A
in level with no change in slope suggests also a change in mean performance,
obvious from visual inspection of the graphical display of the data. The data
analysis only addresses the changes in the first two phases of the design. In
principle, comparisons could be made across the other phases as well, although
restrictions on the number of data points in this particular study present a
limiting condition, discussed later.
The analysis is not restricted to variations of an A-B-A-B design. In any
design where there is a change across phases, time series analysis provides a
potentially useful tool. For example, in multiple baseline designs, time series
analysis can evaluate change from baseline to intervention phases for each of
the responses, persons, or situations, depending upon the precise design.
Considerations and limitations
Among the available statistical analyses, time series analysis is recom-

mended because of manner in which serial dependency is handled. With
the
conventional / and Ftests and many variations, dependency in the data is either
Straws
plus
(Grade 2) Baseline Praise plus a favorite activity
I
surprise Bi Praise
25
vv^'V/u
V
J L
20
_l I
30
I
40
L \AI 60
10 15 25 35 45 50 55
Days
FIGURE 9-3. Daily number of talk-outs in a second-grade classroom. Baseline before experi-
mental conditions. Praise plus a favorite activity systematic praise and permission to engage in
a favorite classroom activity contingent on not talking out. Straws plus surprise systematic
praise plus token reinforcement (straws) backed by the promise of a surprise at the end of the
week. Bi
withdrawal of reinforcement. Praise systematic praise and attention for handraising
and ignoring of talking out. (From: Hall, R. V, Fox, R., Willard, D., Goldsmith, L., Emerson,
M., Owen, M., Davis, F, & Porcia, E. [1971]. The teacher as observer and experimenter in the
modification of disputing and talking-out behaviors. Journal of Applied Behavior Analysis, 4,
141-149. Copyright 1971 The Society for the Experimental Analysis of Behavior, Inc. Repro-
duced by permission.)
ignored, assumed to be present but disregarded, or recognized and handled in a

relativelycumbersome (and controversial) fashion. In contrast, time series
analysis dependsupon the serial dependency in the data, adjusts to the specific
dependency relationships among data points, and provides separate analyses
for level and slope changes in light of special characteristics of the data.
Another important feature of the analysis is that it does not depend upon stable
baselines. Evaluation of single-case designs through visual inspection is facili-
tated when there is no slope in baseline or even a slope in the direction opposite
to that predicted by the intervention effects. In contrast, time series analysis
can be readily applied even when there is a trend toward improved perfor-
mance in baseline, as illustrated earlier. The separate analyses of the changes in
level and slope provide a reliable criterion where visual inspection may
in cases
be particularly difficult to invoke. Notwithstanding the desirable features of
time series analysis, several issues need to be considered before using the
analysis in applied research.
Number of Data Points. Time series analysis depends on a relatively large

number of data points to identify the model that best describes the data (Box &
Jenkins, 1970). The nature of the underlying data is revealed through autocor-
relations of different lags. In conventional analyses, large sample sizes are
important to achieve statistical power. In time series analysis, the large sample
(of data points) is necessary to identify the processes within the series itself and
to select amodel that fits the data.

Precisely what constitutes a large or sufficient number of observations
depends on several factors such as the nature of the data, the types of changes
across phases, variability within a phase, and other parameters that character-
ize a given series. However, the number of data points usually advocated is
much greater than the number typically available in applied or clinical investi-
gations. For example, various authors have suggested that at least 50 (G. V.
Glass et al., 1974), and preferably 100 (Box & Jenkins, 1970), observations are
required for estimating autocorrelations. Fewer observations have been used
(e.g., data with 10 to 20 observations) in applied research and have detected
statistically significant changes (R. R. Jones et al., 1977). Yet applied investiga-
tions often employ relatively short phases lasting only a few days to demon-
strate intervention effects. In such cases, time series analyses will not be
applicable.
Prevalence of Serial Dependency in Single-Case Data. Time series analysis in

behavioral research has been advocated because of the concern over serial
dependency in the data for a single subject. Intuitively one might expect serial
dependency because multiple data points are generated by the same subject
over time and because any influence on a particular occasion may spread (i.e.,
continue) to other occasions as well. Thus data from one occasion to the next
are likely to be correlated, and the correlation is likely to attenuate over time as
new factors impinge on the subject. In the middle and late 1970s, when time
series analyses began to receive attention in single-case research, it seemed as if
serial dependency were likely to be the rule rather than the exception (e.g.,
Hartmann, 1974; Kratochwill et al., 1974; Thoresen & Elashoff, 1974; R. R.

Jones et al., 1977). Moreover, empirical evaluation of published single-case
data indicated that the prevalence of serial dependency was quite high (e.g.,
83^0 of nonrandomly selected instances) (R. R. Jones et al., 1977). However, in
recent years questions have been raised about the prevalence of significant
autocorrelation and hence the need for time series, as opposed to conventional,
analyses. For example, one evaluation of applied research has suggested that
only a minority of studies (less than 30%) shows serial dependency (Kennedy,
1976). The basis for the discrepancy in the prevalence of serial dependency is
not readily clear, particularly since R. R. Jones et al. (1977) and Kennedy (1976)
selected published investigations from the same journal. In general, whether
data from a particular subject are serially dependent should not be assumed
but should be tested directly. The difficulty is that computing autocorrelation
itself requires multiple data points to detect a statistically significant effect, and
a small number of data points may not permit precise evaluation of the
processes involved in the data.
General Comments. Time series analysis has been used increasingly within the
last several years. The increased availability of publications on the topic (e.g.,
Gottman, 1981; McCleary & Hay, 1980) and several computer programs
(Hartmann et al., 1980; Home et al., 1982) may be fostering increased use of
time series analyses. Nevertheless, use of the analysis has been relatively limited
for several reasons. The tests are complex and involve multiple steps that are
not easily described in terms familiar to most researchers. For example, serial
dependency and autocorrelation, two of the less esoteric notions underlying
time series analysis, are not part of the usual training of researchers who
conduct group studies in the social sciences. More in-depth examination of
time series analysis and its underlying rationale introduces many concepts that
depart from conventional statistical techniques and training (see Gottman,
1981). In addition, requirements for conducting time series analysis may not
foster widespread adoption within applied behavioral research. The relatively
brief phases typically used in single-case experimental designs make the test
difficult to apply and perhaps, simply, inappropriate. Recent controversy over
whether single-case data as a rule are serially dependent raises questions for
some about the need for time series analysis. Nevertheless, time series analyses
have been appropriately applied in several demonstrations and provide a
valuable addition to statistical analyses of single-case data.
9.6. RANDOMIZATION TESTS

Several different tests useful for single-case experiments are based on the
notion of assigning treatments randomly to different occasions (e.g., days or
sessions) (Edgington, 1980b, 1984; Levin, Marascuilo, & Hubert, 1978; Wam-
pold & Furlong, 1981b). At
two treatments, or conditions, are required;
least
one of which may be baseline (A) and the other an intervention (B), and
therefore these tests are useful for evaluating ATDs (see chapter 8). Prior to the
experiment, the total number of occasions that the treatments will be imple-
mented must be specified, along with the number of occasions on which each
specific condition will be applied. Once these decisions are made, A and B (or
A, B, C . . .n) conditions are assigned randomly to each session or day of the
experiment, with the restriction that the number of occasions for each meets
the prespecified totals. Each
one of the conditions is administered accord-
day,
ing to the randomized schedule planned in advance.
The null hypothesis of the randomization test is that the client's response on
the dependent measure(s) is not influenced by the condition in effect on that
occasion (e.g., baseline or intervention). If the condition makes no difference.
performance on any particular day will be a function of factors unrelated to the

condition in effect.The random assignment of treatments to occasions in
effect randomly assigns responses of the subject to the treatments. The ob-
tained data are assumed to be the same as those that would have been obtained
under any other random ordering of the treatments to occasions. Thus the null
hypothesis attributes differences between conditions to the chance assignment
of one condition rather than the other to particular occasions. To test the null
hypothesis, a sampling distribution of the differences between the conditions
under every equally likely assignment of the same response measures to occa-
sions of A and B is computed. From this distribution, one can determine the
probability of obtaining a difference between treatments as large as the one
that was actually obtained.'*
Data analysis
Consider as an illustration an investigation designed to evaluate the effect of

teacher praise on the attentive behavior of a disruptive student. To use the
randomization test, the investigator must decide in advance the number of days
of the study and the number of days that each of two (or more) conditions will
be administered. Assume for present purposes that the investigator wishes to
compare the effects of ordinary classroom practices (baseline or A Condition)
with a reinforcement program based on praise (intervention or B Condition).
To facilitate computations, suppose that the duration of the study is decided in
advance to be 8 days and that each condition be in effect for 4 days. (The
will
statistical test does not require an equal number of days for each condition.)
On each of the 8 days, either condition A or condition B is in effect, until each is

administered for 4 different days. Each day, observations of teacher and child
performance are made, and they provide the data to evaluate the effects of the
different conditions.
The prediction is that praise (Condition B) will lead to higher levels of
attentive behavior than ordinary classroom practices (Condition A). Stated as
a one-tailed (directional) hypothesis, Condition B is expected to lead to higher
scores than Condition A. Under the null hypothesis, any difference between
means for the two conditions is due solely to chance differences in performance
on the occasions to which A and B conditions were randomly assigned. To
determine whether the differences are sufficient to reject the null hypothesis,
the mean level of performance is computed separately for each condition, and
the difference between these means is derived.
Hypothetical data for the example appear in Table 9-2 (upper portion). The
mean difference between A and B Conditions is 43.75, also shown in the table
(lower portion). Whether this difference is statistically significant is determined

by estimating the probability of obtaining scores this discrepant in the pre-
dicted directionwhen conditions have been assigned randomly to occasions.
TABLE 9-2. Percentage of Intervals of Attentive Behavior

Across Days and Treatments (Hypothetical Data)
ABAABABB
20 50 15 10
DAYS
60 25 65 70
COMPARING TREATMENT MEANS
A B
20 50
15 60
10 65
25 70
EA = 70 EB = 245
Xa = 17.50 Xb = 61.25
Xb >Xa == 43.75
The random assignment of conditions to occasions makes several combina-

tions of the obtained data equally probable. Actually, 70 different combina-
tions (8!/4!4!) are possible. The question for computing statistical significance
is: What proportion of the different combinations (of assigning conditions to

occasions) would provide as large a difference between means as 43.75?'
A critical region of the sampling distribution is identified to evaluate the
statistical significance of the obtained difference. The critical region is based on
a =
the level of confidence the investigator selects for the statistical test (e.g.,
.05) and the number of combinations of data At the .05 level of
possible.
confidence for the present example, the critical level would be .05 x 70 (or the
level of confidence times the number of possible combinations). The result
would be 3.5. When a critical region is not an integer, selection of the larger
whole number is recommended (Conover, 1971). In the present example, the
larger whole number would be 4. With this critical region, the four combina-
tions of the obtained data that are the least likely under the null hypothesis
must be found. The least likely combination of data of course is one in which
the A and B mean difference in the predicted direction is the greatest possible
given the obtained scores. For the present example, the critical region consists
of the four combinations of the obtained data allocated to A and B conditions
that maximize the difference between the two means. The four data permuta-
tions that constitute the critical region are obtained by reallocating the ob-
tained data to A and B conditions in such a way that the differences between
conditions are the greatest in the predicted direction.
Table 9-3 presents permutations of the obtained data that reflect the four
least likely combinations. The table was derived by first reallocating data points
TABLE 9-3. Critical Region for the Obtained Data from the Hypothetical Example
TOTAL FOR TOTAL FOR

A B
A OCCASIONS Xa B OCCASIONS Xb Xa>Xb
20 10 15 25 (70) 17.50 50 60 65 70 (245) 61.25 43.75
20 10 15 50 (95) 23.75 25 60 65 70 (220) 55.00 31.25
50 10 15 25 (100) 25.00 20 60 65 70 (215) 53.75 28.75
60 10 15 20 (105) 26.25 25 50 65 70 (210) 52.50 26.25
Note. All other combinations of the obtained data (allocated to A

and B treatments) are not in the
critical region using .05 as a level of significance for a one-tailed test.
to conditions that yielded the greatest difference between A and B, then the
combination of data points that could show the next greatest difference, and so
on. A total of four combinations was selected because this is the number of
combinations that reflects the critical region for the .05 level of confidence.
Thus the critical region consists of the n set of data combinations in the
predicted direction that are the least likely to have occurred by chance (where n
= the number of combinations that constitutes the critical region). The
question for the randomization test is whether the difference between means
obtained in the original data is equal to or greater than one of the mean
differences included in the critical region. The obtained mean difference
(43.75) equals the most extreme value in the critical region and hence is a
statistically significant effect. The actual probability of the difference being
this large, given random assignment of conditions to occasions, is 1/70 or p =
.014. When the data represent the least probable combination of data (given a
one-tailed null hypothesis), the probability equals 1 divided by the total num-
ber of possible data combinations.
In the above example, a one-tailed test was performed. For a two-tailed test,
the critical region is at both ends (tails) of the distribution. The number of data
combinations that constitute the critical region unchanged for a given level of
is
confidence. However, the is divided among the two

number of combinations
tails. Because of the division of the critical region into two tails, the probability
level of an obtained mean difference is doubled. Thus, if the above example

utilized a two-tailed test, the probability level of the obtained difference would
be 2/70 or/? = .028.
Special Features.An advantage of randomization tests is that they do not rely

on some of the assumptions of conventional tests such as random sampling of
subjects from a population or normality of the population distribution. Also,
serial dependency is not a problem that affects application of the tests. Depen-
dency may exist in the data. Yet the test is based on the null hypothesis that
there would be identical responses across occasions if the conditions were
presented in a different order. Every order of presenting treatments should lead
to an identical pattern of data (assuming the null hypothesis). Serial depen-
dency does not affect the estimation of the sampling distribution of the statistic
from which the inference of significance is drawn.
Computational Difficulties. An important issue regarding the use of randomi-

zation tests is the computation of the critical region. For a given confidence
level, the investigator must compute the number of different ways in which the
obtained scores could result from random assignment of conditions to occa-

sions. When the number of occasions for assigning treatments exceeds 10 or 15,
even obtaining the possible arrangements of the data by computer becomes
monumental (Conover, 1971; Edgington, 1969). Thus, for most appUcations
of randomization tests in single-case research, computation of the statistic in
the manner described above may be prohibitive.

Fortunately convenient approximations of the randomization test are avail-
able that permit use of the test without the cumbersome computation of the
critical region. The approximations depend on the same conditions as the
randomization test does, namely, the random assignment of treatments to

occasions. The approximations include the familiar / and F tests for two or
more conditions, respectively. The t and F tests are identical in computation to
conventional t anddiscussed earlier. Yet there is one important difference in
F,
the test dependency makes conventional / and F tests inappro-

itself. Serial
priate. The use of / or Fas an approximation of randomization tests avoids the

problem of serial dependency. Because the treatments are assigned to occasions
in a random order across all occasions, t and F provide a close approximation
to the randomization distribution (Box & Tiao, 1965; Moses, 1952). Serial
dependency does not interfere with this approximation.
For example, in the earlier example (Table 9-2), a / test for independent
groups could be applied to approximate the randomization distribution where
degrees of freedom is based on the number of A and B occasions {df = n^ + 2
- 2). The data yield a /(6) = 8. 1 7, /?<. 001), which is less than the probability
obtained with the exact analysis from the randomization test (p = .014). In
cases in which the critical region is not easily computed, / and F can provide
useful approximations if the conditions are randomly assigned to occasions in
the design.
An alternative to the use of the / test is to approximate the randomization
distribution with theMann-Whitney t/Test. To employ this test, the A and B
data points are ranked from 1 to n (the number of treatment occasions) without
reference to the treatment conditions from which each value is derived. The
null hypothesis of no difference between treatments may be rejected if the
ranks associated with one treatment tend to be larger than the values of the
Other treatment. The distribution from which this determination is made is

available in published tables (Conover, 1971) and need not be computed for
each set of data unless A plus B occasions are relatively large (e.g., over 20).
The Mann- Whitney Cîs a convenient test that may be used in place of t and has
been described in other sources (see Conover, 1971; Kirk, 1968).
Practical Restrictions. A few practical considerations influence the utility of

randomization tests (Kazdin, 1980a; see also chapter 8). First, the use of the
tests as described here requires that the subject's performance change rapidly
(or reverse) across conditions. Thus, when conditions are changed from one
day to the next (from A to B or B to A), performance must respond quickly to
reflect treatment effects. Although rapid shifts in performance are often found
when conditions are withdrawn or altered in applied research, this is not always
the case. Without consistently rapid reversals in performance, differences
between A and B conditions may not be detected. In situations where perfor-
mance does not reverse, where there is a carryover effect from one condition to
the next, or where attempting to reverse behavior is undesirable for clinical or
ethical reasons, use of the randomization test may be limited.
A second and related issue involves the fact that it may not be feasible to
allow different conditions such as baseline (A) and treatment (B) or multiple
treatments (C, D, etc.) to vary on a daily basis. Such conditions cannot be
implemented and shifted rapidly in applied settings to meet the requirements of
the statistic. For example, a randomization test might be used to compare
baseline (A) and token economy (B) conditions among patients on a psychiatric
ward. Because of random assignment of conditions to days, the AB conditions
will be alternated frequently to meet the requirements of the design. Yet to
alternate conditions on a daily basis would be extremely difficult in most
settings. One cannot easily implement an intervention such as a token economy
for 1 or 2 days, remove it on the next, implement it again for 1 or 2 days, and so
on, as dictated by the design.
There is a solution that overcomes this practical obstacle. Rather than
alternation ofconditions on a daily basis, a fixed block of time (e.g., 3 days or 1
week) could serve as the unit for alternating treatment. Whenever A is imple-
mented, it would be in place for 3 consecutive days or a week; when B is
assigned, the time period would be the same. The mean (or total) score for each
period (rather than for each day) serves as the unit for computing the randomi-
zation test. The AB conditions are still assigned in a random order, but a given
condition stays in effect whenever it is assigned for a period longer than one
day. Thus the different conditions need not be shifted daily. Moreover, because
of random assignment, a given condition is likely to be assigned for two or
more consecutive occasions (periods). This would increase the length of the
period in which a particular condition is in effect (e.g., 6 days if two consecu-
tive 3-day periods of a particular condition are assigned). Thus the problem of
rapidly shifting treatments would be partially ameliorated. If fixed blocks of

several days rather than single days constitute the occasions, the mean score for
a block as a whole is the datum used to compute the test. Because a block of
days of a condition counts as only one occasion, several blocks will be required
to achieve a relatively large number of occasions. A small number of occasions
may restrict the possibility statistically significant effects when
of obtaining
treatments differ in their effects. Thus, when fixed blocks of several days are
used to define the occasion, the number of days of the investigation will be
longer than if individual days are used as the occasion. The practicality of
extending the duration of time that defines an occasion needs to be weighed
against the feasibility of extending the overall duration of the project.
In general, randomization tests provide a useful set of statistical techniques
for single-case research. The availability of convenient (and familiar) approx-
imations to the randomization distribution makes the tests more readily acces-
sible to most users than such tests as time series analysis. The major problems
delimiting use of the tests pertain to the need to assign conditions to occasions
on a random basis and to show that treatment effects can be reversed rapidly as
the conditions are changed.
9.7. THE R TEST OF RANKS

A test of ranks, referred to as R, has been proposed for evaluating data
obtained in multiple baseline designs (Revusky, 1976; Wolery & Billingsley,
1982). The test requires that data be collected across several different base-
lines (e.g., different individuals, behaviors, or situations). Whether the inter-
vention produces a statistically reliable effect is determined by evaluating the

performance of each of the baselines at the point when the intervention is
introduced. For example, in a multiple baseline design across individuals, the

statistical comparison completed by ranking scores of each subject at the
is
point when the interventionis introduced for any one of the subjects. Each
individual is considered a subexperiment. When Condition B is introduced

for a subject, the performance of all subjects (including those for whom
treatment is withheld) is ranked. The sum of the ranks across all subexperi-
ments each time the treatment is introduced constitutes the statistic R;,-
An essential feature of the test is that the intervention is applied to different
baselines in a random order. Thus the rationale underlying R^^ follows that of
randomization tests as outlined earlier. Because the basehne (e.g., person or
behavior) that receives the intervention is determined randomly, the combina-
tion of ranks at the point of intervention for all subjects will be randomly
distributed if the intervention has no effect. On the other hand, if the
behavior of the client who receives the intervention changes at the point of
intervention, compared with persons who have yet to receive the intervention.
this should be reflected in the ranks. If each subject in turn shows a change
when the intervention is introduced, this would be reflected in the sum of the
ranks (or R) across all subjects, and it suggests that the ranks are not the
of random factors. R requires several different baselines or
likely result
subexperiments to evaluate whether change at the point of treatment is
reliable. At the .05 level of confidence the minimum requirement for detecting
a statistically significant effect is four baselines (i.e., persons, behaviors, or
situations).
Data analysis
Application of the R can be illustrated in a hypothetical example in which

an intervention is applied to increase the amount of time that five aggressive
children engage in appropriate and cooperative play during recess at school.
To fulfill the requirements of the multiple baseline design, data are gathered
for the target behaviors. For present purposes, assume that the data consist of
the percentage of intervals (e.g., 30 sec) observed during recess in which the
child engages in appropriate play. Treatment is introduced to different
children at different points in time. The child who receives treatment first,
second, and so on is always determined randomly.

Table 9-4 provides hypothetical data on the percentage of intervals of
appropriate play across 10 days. As is evident in the table, baseline is in effect
for everyone for 5 days. On the sixth day, one child is randomly selected to
receive the intervention (B), whereas all other children continue under base-
line (A) conditions. On successive days, a different child is exposed to the
intervention. The ranking procedure is applied to each subexperiment at the
point when the intervention is introduced. On each occasion that the interven-
tion is introduced (which includes Days 6-10 in the example), the children are
ranked. The lowest rank is given to the child who has the highest score (if a
high score is in the desired direction).^ In the example, on Days 6-10, the
child with the highest amount of appropriate play at each point of interven-
tion receives the rank of 1, the next highest the rank of 2, and so on. When
the intervention is introduced to the first child, all children are ranked. When
the intervention is introduced on subsequent occasions, all children except
those who previously received the intervention are ranked. Even though all
subjects are ranked when the intervention is introduced, not all ranks are
used. R consists of the sum of the ranks for those subjects who receive the
intervention at the point that the intervention is introduced. If treatment is
ineffective, the ranks of these persons should be randomly distributed, i.e.,
include numbers ranging from 1 to the n number of baselines. If treatment is
effective, the point of intervention should result in low ranks for each subject
at that point (if low numbers are assigned to the most extreme score in the
predicted direction of change).
TABLE 9-4. Percentage of Intervals of Appropriate Play

for Five Children Studied in a Multiple Baseline Design (Hypothetical Data)
DAYS
1 2 3 4 5 6 7 8 9 10
1 45 30 35 50 40 30a 70b
g 2 60 75 80 60 50 70a 50a 65a 80b
2 3 20 20 25 10 30 80b
8^
^ 5
55 60 40 45 50 40a 75a 90b
30 25 20 30 20 30a 30a 40a 35a 50b
Ranks = 1 2 1 1 1 ER == 6
Note. Days 1 through 5 served as baseline (a) days for all subjects and are unmarked,
a = control or baseline, b = experimental or intervention point for a child.
As is evident in Table 9-4, hypothetical data show that the child who
receives the intervention at a given point in time, with the exception of
Subject 1, receives the lowest rank (i.e., 1 or 1st place) for performance on
that occasion. Summing the ranks for all children exposed to the intervention
yields R = 6. The significance of the ranks for designs employing different
numbers of subjects (or baselines) can be determined by examining Table 9-5.
The table provides a one-tailed test for R. (A two-tailed test, of course, can
be computed by doubling the probabiHty level for the tabled columns.) To
return to the above example, R,2 = 6 for 5 subjects (one-tailed test) is equal to
the tabled value required for the .05 level (see arrow). Thus the data in the
hypothetical example permit rejection of the null hypothesis of no treatment
effect.
Rapidity of Behavior Change. In the above example, the rankings were

assigned to the different baselines (children) at the point when the interven-
tion was introduced (i.e., on the first day). However, it is quite possible, and
indeed likely, would not be evident on the
that intervention effects first day
that the intervention was applied. With some interventions, slow and gradual
improvements may be expected, or performance may even become slightly
worse before becoming better. The statistic can still be used without necessar-
ily applying the ranks on the first day of the intervention for each baseline.
The intervention can be evaluated on the basis of mean performance for a

given person (behavior or situation) across several days rather than on the
basis of a change in level (at the point of intervention) on the first day that the
intervention is introduced. For example, the intervention could be introduced
for one person and withheld from others for several days or a week. The
1
TABLE 9-5. Maximum values of R significant
at the indicated one-tailed probability levels when the

experimental scores tend to be smaller than the control scores.
NO. OF SIGNIFICANCE LEVEL

SUBJECTS 0.05 0.025 0.02 0.01 0.005
4 4
5 6 5 5 5
6 8 7 7 7 6
7 11 10 10 9 8
8 14 13 13 12 11
9 18 17 16 15 14
10 22 21 20 19 18
11 27 25 24 23 22
12 32 30 29 27 26
Note. Table provides significance for a one-tailed test. The num-

ber of subjects in the table also can be used to denote the number
of responses or situations across which baseline data are
gathered, depending on the variation of the multiple baseline
design. (From Revusky, S. H. 11967]. Some statistical treaments
compatible with individual organism methodology Journal of
Experimental Analysis of Behavior, 10, 319-330. Copyright 1976
Society for the Experimental Analysis of Behvior, Inc. Repro-
duced by permission.)
rankings could be made on the basis of the mean performance across the
entire week while the intervention was in effect. Mean performance of the
target child would be compared with the mean of the other persons, and
ranks would be assigned on the basis of each person's mean for that time
period. Using means across days is likely to provide a more stable estimate of
actual performance, to allow the intervention to operate on behavior, and
consequently to reflect intervention effects more readily than evaluation
based on the first day that the intervention is applied. Also, by using averages,
the statistic takes into account the usual manner in which multiple basehne
designs are conducted where the intervention is continued for several days for
one person (baseline) before being introduced to the next person.'
If ranks are to be based on several days rather than a single day, additional
considerations become important. First, the duration employed to evaluate
treatment changes within subjects should be specified in advance. If interven-
tion effects are expected to take a certain period of time, the precise number
of days (or a conservative estimate) should be specified. The mean for that
period is then used when the ranks are assigned. Second the duration for ,
introducing the treatment and for computing mean performance should be

constant across all subjects. These two features ensure that randomness will
not be influenced by post hoc treatment of the data and capitalization on
chance fluctuations in performance.
Differences in Responses Across Baselines. If the scores across the different

baseHnes vary markedly from each other in absolute magnitude, it may be
change using R. The scores may vary so much that when

difficult to reflect
the intervention is introduced to one subject, and change occurs, the amount
of change does not bring the person's score higher (or lower) than the level of
another person who has continued in baseline conditions. The intervention
may have led to change, but this is not reflected in the rankings because of
discrepancies in the magnitude of scores across subjects.
For example, in Table 9-4, compare the hypothetical performance of Child
2 and Child 5. The performance of Child 2 was higher during baseline than
was the performance of Child 5 when treatment was introduced. Had treat-
ment been introduced to Child 5 before Child 2, the rank assigned to Child 5
would not have been as low as it was in the example. This would have been an
artifact of the differences in absolute levels of performance of the subjects
rather than of the ineffectiveness of the intervention. In general, the ranking
procedure, as described thus far, does not take into account the differences in
basehne magnitudes.
A simple data transformation can be used to ameliorate the problem of
different response magnitudes. The transformation corrects for the different
initial levels of baseline responding (Revusky, 1967). The formula for the
transformation is
B/ - A/
A/
Where B/ = performance level for Subject / when the experimental inter-

vention is introduced, and A/ = mean performance across all
baseline days for the same subject.
Use of the transformation is the same as examination of the change in

The raw scores for each
percentage of responding from baseline to treatment.
subject (i.e., for each baseline) are transformed when the intervention is
introduced to any one subject. The ranks are computed on the basis of the
transformed scores. In general, the transformation might be used routinely
because of its simplicity and the likelihood that responses would have dif-
ferent magnitudes that could obscure the effects of treatment. Where re-
sponse levels are widely discrepant during basehne, the transformation will be
especially useful.
9.8. THE SPLIT-MIDDLE TECHNIQUE

The split-middle technique provides a method of describing the rate of
behavior change over time for a single individual or group (White, 1971,
1972, 1974). The technique is designed to reveal a linear trend in the data, to
characterize present performance, and to predict future performance. By

describing the rate of behavior change, one can estimate the likelihood that
the client's behavior will attain a particular goal. The technique permits
examination of the trend or slope within phases and comparison of slopes
across phases. Rate of behavior (frequency/time) has been advocated as the
most useful measure for this method. The advantage of rate for purposes of
plotting trends is that no upper limit exists. Theoretically at least there is no
ceiling effect that can limit the slope of the trend. Yet the method can be
applied to other performance measures than rate that are often used in
applied research such as intervals, discrete categorization, and duration.
Special charting paper has been advocated for the use of the split-middle
techniques that allows graphing of performance in semilog units.* The special
charting paper increases the linearity of the data, may enhance predictive
validity, and is easily employed by practitioners (White, 1972, 1974). How-
ever, the split-middle technique can be used with ordinary graph paper with
arithmetic (equal interval) units rather than log units on the ordinate.
The split-middle technique has been proposed primarily to describe the
process of change within and across phases rather than to be used as an
inferential statistical technique. The descriptive purposes are achieved by
plotting trends within baseline and intervention phases to characterize client
progress. Statistical significance can be examined once the trend lines have
been determined.
Data description
The split-middle technique involves multiple steps. The technique begins

with graphically plotting the data. From the data within a given phase, a
trend, or celeration line, is constructed to characterize the rate of perfor-
mance over time. (The term celeration derives from the notions of accelera-
tion and deceleration if the trend is ascending or descending, respectively.)
The celeration line predicts the direction and the rate of change.
To illustrate computation of the celeration line, consider hypothetical data
plotted in Figure 9-4. (The example will utilize rate of performance and
semilog units to illustrate recommended use of the method.) The data in the
upper panel are from one phase of an A-B-A-B (or other) design plotted on a
semilog chart. The manner in which the celeration line is computed will be
conveyed with data from only one phase, although in practice celeration lines
would be computed and plotted separately for each phase.
The first step for computing a celeration line in a phase is to divide the
phase in half by drawing a vertical line at the median number of sessions (or
days). The second step is to divide each of these halves in half again. (When
there is an uneven number of days, the vertical line is drawn through the data
point that is the median day rather than between two data points.) The
dividing lines should always result in an equal number of points on each side
* I
314 Single<ase Experimental Designs
50
40
30
20 h
10 I 'III! 1 1 L
XT
g 50
40 t..
X
UJ
GO
30 !

\-
O 20
LU
b
10 J I I L J L_l I L
50 "
40
30
-
>*
^>*^
slope=l.65
level =39
20 h

1 1 1 1 1 1 1 1 1 1
DAYS 10
FIGURE 9-4. Hypothetical data during one phase of an A-B-A-B design {top panel a), with
steps to determine the median data points in each half of the phase {middle panel b), and with
the original data (dashed) and adjusted (solid) celeration line {bottom panel c).
of the division. The next step

is to determine the median rate of performance
for the and second halves of the phase. This median refers to the data
first
points that form the dependent measure rather than to the number of
sessions.
T\vo potentially confusing points should be resolved. First, although the
sessions are divided into quarters, only the first division (halves) is employed
at this stage. Second, the median data value within each half of the sessions is
selected.These medians are based on the ordinate (dependent variable values)

rather than the abscissa (number of days). To obtain the data point that is the
median within each half, one merely counts from the bottom (ordinate) up
toward the top data point for each half. The data point that constitutes the
median value within each half is selected. A horizontal fine is drawn through
the median at each half of the phase until the line intersects the vertical line
dividing each half.
Figure 9-4b shows the above three steps, namely, a division of the data into
quarters and the selection of median values within each half. Within each half
of the data, a vertical and horizontal line intersect. The next step is finding the
slope, which entails drawing a line connecting the points of intersection
between the two halves.
The final step is to determine whether the line that results "splits" all of the
data, in other words, is the split-middle line or slope. The split-middle slope is
that line that is situated so that 50% of the data fall on or above the line and
50% fall on or below the line. The line is adjusted to divide the data in this
fashion. In practice the line is moved up or down to the point at which all of
the data are divided. The adjusted line remains parallel to the original line.
Figure 9-4c shows the original line (dotted) and the line (solid) after it has
been adjusted to achieve the split-middle slope. Note that the original line did
not divide the data so that an equal number of points fell above and below the
line. The adjustment achieves this "middle" slope by altering the level of the
line (and not the slope). (In some cases, the original line may not have to be
adjusted.)
The celeration line reflects the rate of behavior change, which can also be
expressed numerically. White (1974) has used the weekly rate of change as the
basis of calculating rate, although any time period that might be more
meaningful for a given situation can be employed. To calculate the rate of
change, a point of the celeration line (Day;^) that passes through a given value
on the ordinate is determined. The data value on the ordinate for the
celeration line 7 days later (i.e., Day;^-+7) is obtained. To compute the rate of
change, the numerically larger value (either Day;^: or Dayj^ +7) is divided by the
smaller value.
The procedure can be applied to the data in Figure 9-4c. At Day 1, the
celeration line is at 20. Seven days later, the line is at approximately 33.
Applying the above computations, the ratio for the rate of change is 1.65.
Because the celeration line is accelerating, this indicates that the average rate
of responding for a given week is 1.65 times greater than it was for the prior
week. The ratio merely expresses the slope of the line.
The level of the slope can be expressed by noting the level of the celeration
line on the last day of the phase. In the above example, the level is approxi-
mately 39. When separate phases are evaluated (e.g., baseline and interven-
tion), the levels of the celeration
lines refer to the last day of the first phase
and the first day of the second phase, as will be discussed below.
For each phase in the experimental design, separate celeration lines are
drawn. The slope of each line is expressed numerically. The change across
phases is evaluated by comparing the levels and slopes. Consider hypothetical
data for A and B phases, each with its separate celeration line, in Figure 9-5.
To estimate the change in level, a comparison is made between the last data
point in baseline (approximately 22) and the first data point during the
intervention (approximately 28). The larger value is divided by the smaller
value, yielding a ratio of 1.27. The ratio merely expresses how much higher
(or lower) the intersection of the different celeration lines is. Similarly, for a
change in slope, the larger slope is divided by the smaller slope, yielding a
value in the example of 1.52. The change in level and slope summarizes the
differences in performance across phases.
Statistical analysis
It should be reiterated that the split-middle procedure has been advocated

as a technique to describe the process of change in an individual's behavior
rather than as a tool to assess statistical significance. However, statistical
significance of change across phases can be evaluated once the celeration lines
have been calculated.
To determine whether there is a statistically significant change in behavior
across phases, a simple statistical test has been proposed (White, 1972).
Again, consider change across A and B phases in an A-B-A-B design. The null
hypothesis upon which the test is based is that there is no change in perfor-
mance across A and B phases. If this hypothesis is true, then the celeration
line of the baseline phase should be a valid estimate of the celeration line of
the intervention phase. Assuming the intervention had no effect, the split-
middle slope of baseline should be the spHt-middle slope of the intervention
phase, as well. Thus 50<yo of the data in the intervention or B phase should
fallon or above and 50ô of the data should fall on or below the slope of
baselinewhen that slope is projected into the intervention phase.
To complete the statistical test, the slope of the baseline phase is extended
or projected through the intervention phase. Consider the example of hy-
pothetical data in Figure 9-5, which shows the celeration line computed and
BASELINE INTERVENTION
100
Slope = X 1.05 Slope = x 1.60
Level = 22 Level = 28
g (line at (line at
last day) first day)
50
I 40
UJ
CD
-
30
Ll
O
20 -
UJ
I-
<
or
10
Change in level = x 1.27

Change inslope=x 1.52
FIGURE 9-5. Hypothetical data across baseline (A) and intervention (B) phases, with separate
celeration lines for each phase (solid lines). The dashed line represents an extension of the
celeration line for the baseline phase.
extended from baseline into the intervention phase. For purposes of the
statistical test, it is assumed that the probability of a data point during the
intervention phase falling above the projected celeration line of baseline is
50% (i.e., p = .5), given the null hypothesis of no change across phases. A
binomial test can be used to determine if the number of data points that are
above the projected slope in the intervention phase is of a sufficiently low

probability to reject the null hypothesis.'
Using this procedure for the data in Figure 9-5, 10 of 10 data points during
the intervention phase fall above the projected slope of baseline. Applying the
binomial test to determine the probability of obtaining all 10 data points
above the slope, p = (io)'/2' yields ap< .001. Thus the null hypothesis can
be rejected; the data in the intervention phase are significantly different from
the data of the baseline phase. The results do not convey whether the level
and/or slope account for the differences but only that the data overall depart
from one phase to another.
Utility of the The primary purpose of the split-middle technique is to

Test.
describe the data in a summary fashion and to predict the outcome given the
rate of change. The utility of the test is that it provides a computationally
simple technique for characterizing data and for examining if trends change
across phases. In the usual case of data presentation in single-case research,
summary statistics are often restricted to describing mean changes across
phases (see Kazdin, 1982b). The split-middle technique can provide addi-
tional descriptive information on the level, slope, and changes in these
characteristics over time (see Wolery & Billingsley, 1982).
Since a major purpose of the technique is to predict behavior rather than to
determine statistical significance of change, it is appropriate to examine the
extent to which this purpose is adequately achieved. White (1974) presented
data based upon "several thousand" analyses of classroom performance. The
analyses determined the accuracy of predicting behavior using the split-
middle procedure at different points in the future. As might be expected, the
extent to which the predictions approximated the actual data depended upon
the number of data points upon which the prediction was based and upon the
amount of time into the future that was predicted. For example, on the basis
of 7 days of data, performance one week into the future would be success-
fully predicted (with a narrow margin of error) 64% of the time; for perfor-
mance 3 weeks into the future, predictions were successful SO^Vo of the time.
With 1 1 days of data, predictions one week into the future were successful
89<7o of the time; for performance 3 weeks into the future, predictions were
successful 81% of the time.
The predictive uses of the split-middle technique have been accorded
important applied significance. If the data suggest that behavior is not
changing at a sufficient rate to obtain a particular goal, the intervention can
be altered. Thus the technique may provide useful information that leads the
investigator to change the intervention as needed.
Statistical Inferences. Several different tests have been proposed to assess

change based on information obtained from plotting slope and level (see
White, 1972; Wolery & Billingsley, 1982). Most of these tests also rely on the
binomial as illustrated above. As E. S. Edgington (personal communication,
August, 1974) has noted, the binomial may not be valid when apphed to data
that show a trend during baseline. Consider the following circumstances in
which the binomial might lead to misinterpretation. A random set of num-
bers could be assigned randomly as data points to baseline and intervention
phases. On the basis of chance alone, baseline occasionally would show an
accelerating or decelerating slope. If the data points in the A phase show a
slope, it is unlikely that the data points in the B phase will show the same
slope.The randomness of the process of assigning data points to phases

would make identical trends possible but very unlikely. Hence if there is an
initial trend in baseline, it is quite possible that data in the intervention phase
on the basis of chance alone would
above or below the projected slope of
fall
baseline. The binomial test might show a statistically significant effect even
though the numbers were assigned randomly and no intervention was imple-
mented. Thus problems may exist in drawing inferences using the binomial
test when trend is evident in baseline (or the condition from which a projected
celeration line is made).

The split-middle technique has been infrequently reported in published
investigations as either a descriptive or an Thus impor-
inferential procedure.
tant questions about the statistical techniquesand the problems they may
introduce remain to be elaborated. The conditions in which the binomial test
represents the probability of the distribution of data points across phases,
given the null hypothesis, are not well explored. Nevertheless, as a descriptive
tool, the split-middle technique provides important information about level
and slope changes that is usually not reported.
9.9. EVALUATION OF STATISTICAL TESTS:

GENERAL ISSUES
Single-case designs provide a wide array of options for the applied re-
searcher. Statistical techniques available for such designs are numerous.
Selected tests were reviewed to convey the breadth of options available.
Additional variations of these analyses, as well as different tests, have also
been described (e.g., Edgington, 1982; Tryon, 1982).
Some of the analyses discussed have wider applicability than others. Single-
case designs generally involve a comparison of two or more phases. This one
characteristic raises the possibility of time series, split-middle,randomiza-
tion,and t tests. The options were illustrated and discussed in the context of
A-B-A-B and multiple baseline designs, but they can also be applied to other
designs such as the changing-criterion designs, and alternating or simulta-
neous treatment designs.'" Despite the flexibility of various tests, several
considerations and sources of caution warrant mention.
First, statistical evaluation of single-case (or any other) data only addresses
the issue of whether the change is statistically significant over the course of
separate conditions. When statistical significance is obtained, this does not of
course provide any necessary clues about the basis for a change in behavior.
Conclusions about the basis for the change derive from the experimental
design rather than from the mere demonstration of statistical significance.
Thus statistical evaluation of an A-B design does not elevate the sophistica-
tion of the comparison. Drawing conclusions between the effect of an inter-
vention and behavior assumes an adequate design independent of the

techniques to evaluate the data.
Second, the analyses outlined above only addresses the statistical significance
and not the clinical significance of the changes. Although rules of science
have depended upon levels of confidence as a criterion to decide veridical
effects, no leap is warranted from levels of confidence to the applied value of
the finding. Clinical significance, as noted earlier, refers to the importance of
the change and entails different criteria from those invoked for statistical
analyses.
Clinical significance is usually viewed as a more stringent criterion than
statistical significance because many statistically reliable effects can be ob-
tained without clear or detectable impact on everyday client functioning. It is
generally true that, with clinically significant effects, behavior change is
especially marked and hence typically statistically significant. There are also
cases, however, where clinically significant effects might be evident where
statistical tests might not be applicable and or where statistical significance is
not For example, for clinical cases where complete amelioration of the
clear.
problem is achieved in one trial (e.g., Creer, Chai, & Hoffman, 1977),
statistical significance would be difficult if not impossible to demonstrate with
conventional techniques. The main point is that statistical and clinical

significance need to be kept distinct in applied research. A statistically signifi-
cant difference obtained in applied single-case research may lead the investi-
gator to conclude that the intervention was effective. In this context, effective
refers to effective in producing a statistically reliable change and not necessar-
ily effective in ameliorating the clinical problem to which the intervention was
applied.
Finally, the statistical techniques mentioned above invoke special condi-
tions that may limit their use in many
applied investigations. For example, a
randomization test of means and R require assigning conditions randomly
(to occasions or baselines). Yet it is easy to consider many situations in
hospitals, classrooms, or institutional settings where this requirement could
not be invoked. Different sorts of problems are raised with other statistical
tests. For example, protracted baseline phases are difficult to justify but could
be essential in order to apply such tests of time series analyses.
An important characteristic of single-case designs is that they are quite
flexible. Design changes are made in part as a function of the client*s re-
sponses to alternative interventions. This is unlike between-group studies,

where designs are usually worked out well in advance and subjects are run in
a predetermined fashion. There are important implications for the applicabil-
ity of statistical tests to these different design practices. The statistical analy-
ses reviewed earlier often entail conditions that must be planned in advance of
the study. Insofar as these conditions restrict the flexibility of the investigator,
their application in any given case may present problems. Experimental
design considerations already constrain clinical applications in some instances

because of temporary suspensions of treatment (reversal phases) or delays in
introducing treatment (multiple baseline designs). Statistical analyses need to
be considered carefully in advance because they may place additional restric-
tions on the manner in which treatment is implemented.

Statistical analyses should not be viewed as practical obstacles for the
The tests can assist and overcome many problems of evaluation.
investigator.
For example, when ideal conditions for data evaluation through visual in-
spection are not obtained, descriptive and inferential statistics may greatly
facilitate interpretation of outcome. A prime example would be where there is
initial trend in baseline. An investigator ordinarily might hope and wait for
an asymptote to be reached to facilitate subsequent evaluation of intervention
effects. Yet alternative statistical analyses such as time series analyses and
split-middle techniques can be quite helpful because they examine interven-
tion effects in light of prior trends in the data. Thus statisticaltechniques can
also make important practical contributions to applied research.
9.10. CONCLUSIONS
The present chapter has discussed specific statistical tests for single-case
experimental designs and considerations dictated by their use. The availability
of multiple statistics provides the investigator with diverse options for the
single-case. A few salient considerations underlying all of the tests warrant
reiteration. To begin with, the appropriateness of utilizing statistical criteria
for the evaluation of applied behavioral interventions remains a major source
of controversy. Statistical analysis is seen by many proponents of single-case
research as a violation of the rationale for conducting research with the
individual subject. Thus whether statistical tests should be used to draw
inferences from single-case research remains an issue.
On this issue, it is important to distinguish experimental designs (e.g.,
single-case and between-group designs), methods of data evaluation (e.g.,
visual inspection and statistical analyses), and types of research (e.g., basic or
applied). There are no necessary connections between particular types of
research, designs, and analyses. Thus use of statistical analyses does not
necessarily conflict with single-case designs or their purposes. When research
attempts to develop a technology of behavior change and to achieve clinically
important effects, statistical analyses will definitely be of limited value. Small
effects that pass beyond a threshold of traditional levels of confidence may
not address the priorities of applied research. Yet there are several uses of
statistics, detailed earlier, that may contribute to the goals of applied re-
search.
Another issue important to mention is that the use of statistical tests may
have implications for the manner in which a particular intervention needs to

be implemented. For example, the random assignment of treatment to occa-
sions or subjects may compete with clinical priorities. Exigencies of clinical
settings may delimit the applicability of diverse procedures upon which
various statistical tests depend. Yet in many situations, there is flexibility in
on the part of the
deciding the research design. Awareness of statistical tests
investigator may lead to different arrangements of the intervention that do
not impact on clinical care. In some cases, the investigator may have other
options for data evaluation in addition to visual inspection.
Statistical analyses for single-case research have been used relatively infre-
quently. Their use is likely to increase, albeit slowly, for different reasons.
Concerns over the interjudge reliability of visual inspection and increased
dissemination of statistical analyses for single-case designs and the computer
programs for their execution are two influences pointing in the direction of
increased utilization. Interventions are applied in increasingly diverse set-
tings, and experimental control over factors that minimize variability is more
difficult to obtain. Statistical analyses may be helpful in evaluating interven-
tionswhere data requirements for visual inspection are not readily obtained.
The present chapterillustrated several options for statistical analyses and the
problems attendant upon their use.
NOTES
1. As the lag increases, the correlation becomes somewhat less stable, in part,
because of the decrease in the number of pairs of observations upon which the
coefficient can be based (Holtzman, 1963).
2. Although the statistical significance of autocorrelations can be approximated by

testing them as correlations in the usual manner, Anderson (1942) has provided
tables for the exact test. (See also Anderson, 1971, and Ezekiel & Fox, 1959.)
3. Baer (1977a) has articulately stated the similarities and differences in the ra-
tionales underlying statistical analysis and visual inspection. Both methods of data
evaluation attempt to avoid Type I and Type II error. Type I error refers to
concluding that the intervention produced a veridical effect when in fact the
results are attributed to chance. Type II errors refers to concluding that the
intervention did not produce a veridical effect when in fact it did. Typically,
researchers give a higher priority to avoiding a Type I error. In statistical analyses,
the probability of committing a Type I error is specified (by the level of confidence
of the statistical test or a). With visual inspection, the probability of a Type I error
is not known. Hence, to avoid chance effects, the investigator searches for highly
consistent effects that can be readily seen. By minimizing the probability of a Type
I error, researchers increase the probability of making a Type II error. Investiga-
tors who rely on visual inspection are more likely to commit Type II errors than
investigators who rely on statistical analyses. Thus reliance on visual inspection
will tend to overlook and discount many reliable but weak effects. From the
standpoint of developing an effective applied technology of behavior change,
Baer (1977a) has argued persuasively that minimizing Type I errors leads to
few variables whose effects are consistent and potent across a
identification of a
wide range of conditions. Thus visual inspection may be suited for the special
goals of applied research. For other research purposes (e.g., testing of alternative
theories), weak but reliable effects may be important to detect, and the priorities
of erring in one direction rather than another might change.
4. The randomization test discussed and illustrated here is one of many available
tests (see Edgington, 1969, 1984). The specific one selected, which compares
means from different conditions, is likely to be of special interest in single-case
experiments where performance is compared across phases.
5. The example selected here is devised for computational simplicity. It is unlikely

that an investigator would be interested in only eight occasions for evaluating two
different phases (baseline and intervention). In addition, it is also unlikely that the
nonoverlapping distributions of the magnitude included in the example would be
subjected to a statistical test.
6. As a general guideline, ranks are assigned so that the lowest number is given to the
baseline that shows the highest level of performance in the desired direction. An
easy rule of thumb is to assign "first place" (a rank of 1) to the highest or lowest
score that represents the "best" performance in terms of the dependent measure.
Thus 1 might be assigned to the highest performance of social skills or the lowest
performance of self-abusive behavior. Second, third, and subsequent ranks are

assigned accordingly for lower scores in the therapeutic direction.
7. In addition to the use of R to evaluate changes in means, a recent extension has

illustrated evaluation of changes in trends combining R and split-middle tech-
niques (see Wolery & Billingsley, 1982).
8. The semilog units refer to the fact that the scale on the ordinate is logarithmic but
the scale on the abscissa is not. The effect of this arrangement is to ensure that
there is no zero origin on the graph and that low and high rates of performance
can be readily represented. The chart can be used for behaviors with extremely
high or low rates. Rates of behavior can vary from .0006944 per minute (i.e., one
every 24 hours) to 1000 per minute. (The semilog chart paper has been developed
by Behavior Research Company, Kansas City, KS.) Adoption of the charting
procedure has not been widespread in applied research. Hence it is useful to note
that the split-middle technique can be used with ordinary graph paper.
9. The binomial applied to the split-middle slope test would be the probability of
attaining x data points above the projected slope:
Ax) = " p^g"- Hot simply " p").
Where n = the number of total data points in Phase B

X = the number of data points above (or below) the projected slope
p = ^ = .5 by definition of the split-middle slope
p and q = the probability of data points appearing above or below the slope given
the null hypothesis
10. Other design options may raise special issues for statistical tests. For example, in a
changing criterion design, the intervention may be introduced in such a way that
only gradual and small changes in behavior are sought. Obviously, one might not
wish to test for changes in level in such instances, because abrupt changes at the
point of introducing the intervention might not be expected. In an alternating- or
simultaneous-treatment design of special interest, it is not the change from one
phase to another but rather whether separate interventions implemented in the
same phase differ significantly. Analyses discussed previously can be adopted to
these circumstances (e.g., see Edgington, 1982; Kratochwill & Levin, 1980).
CHAPTER 10
Beyond the Individual:

Replication Procedures
10.1 INTRODUCTION
Replication is at the heart of any science. In all sciences, replication serves at
least two purposes: first, to establish the reliability of previous findings; and,
second, to determine the generality of these findings under differing condi-
tions. These goals, of course, are intrinsically interrelated. Each time that
certain results are replicated under different conditions, this not only es-
tablishes generality of findings, but also increases confidence in the reliability
of these findings. The emphasis of this chapter, however, is on replication
procedures for establishing generality of findings.
In chapter 2 the difficulties of establishing generality of findings in applied
research were reviewed and discussed. The problem in generalizing from a
heterogeneous group to an individual limits generality of findings from this
approach. The problem in generalizing from one individual to other individu-
als who may differ in many ways limits generality of findings from a single-
case. One answer to this problem is the replication of single-case experiments.
Through this procedure, the applied researcher can maintain his or her focus
on the individual, but establish generality of findings for those who differ
from the individual in the original experiment. Sidman (1960) has outlined
two procedures for replicating single case experiments in basic research: direct
replication and systematic replication. In applied research a third type of
replication, which we term clinical replication, is assuming increasing impor-
tance.
The purpose of this chapter is to outline the procedures and goals of
replication strategies in applied research. Examples of each type of replication
series will be presented and criticized. Guidelines for the proper use of these
325
procedures in future series will be suggested from current examples judged to

be successful in establishing generality of findings. Finally, the feasibility of
large-scale replication series will be discussed in light of the practical Hmita-
tions inherent in applied research.
10.2 DIRECT REPLICATION

Direct replication of single-case experiments have often appeared in profes-
sional journals. As noted above, these series are capable of determining both
reliability of findings and generality of findings across clients. In most cases,
however, the very important issue of generality of findings has not been
discussed. Indeed, it seems that most investigators employing single-case
methodology, as well as editors of journals who judge the adequacy of such
endeavors, have been concerned primarily with reliability of findings as a goal
than generality of findings. That is, most investiga-
in replication series rather
tors have been concerned with demonstrating that certain results can or
cannot be replicated in subsequent experiments rather than with systemati-
cally observing the replications themselves to determine generality of find-
ings. However, since any attempt to establish reliability of a finding by
replicating the experiment on additional cases also provides information on
generality, many applied researchers have conducted direct replication series
yielding valuable information on client generality. Examples of several of
these series will be presented below.
Definition of direct replication
For our purposes, we agree basically with Sidman's (1960) definition of

direct replication as ". replication of a given experiment by the same
. .
investigator" (p. 73). Sidman divided direct replication into two different
procedures: repetition of the experiment on the same subject and repetition
on different subjects. While repetition on the same subject increases con-
fidence in the reliability of findings and is used occasionally in applied
research (see chapter 5), generality of findings across cHents can be ascer-
tained only by replication on different subjects. More specifically, direct
replication in applied research refers to administration of a given procedure
by the same investigator or group of investigators in a specific setting (e.g.,
hospital, clinic, or classroom) on a series of clients homogeneous for a
particular behavior disorder (e.g., agoraphobia, compulsive hand washing).
While it is recognized that, in applied research, clients will always be more
heterogeneous on background variables such as age, sex, or presence of
additional maladaptive behaviors than in basic research, the conservative
approach is to match clients in a replication series as closely as possible on
Beyond the Individual: Replication Procedures 327
these additional variables. Interpretation of mixed results, where some clients
benefit from the procedure and some do not, can then be attributed to as few
differences as possible, thereby providing a clearer direction for further
experimentation. This point will be discussed more fully below.
Direct replication as we define can begin to answer questions about
it
generality of findings across clients but cannot address questions concerning

generality of findings across therapists or settings. Furthermore, to the extent
homogeneous on a given behavior disorder (such as agorapho-
that clients are
cannot answer questions on the results of a
bia), a direct replication series
given procedure on related behavior disorders such as claustrophobia, al-
though successful results should certainly lead to further replication on
related behavior disorders. A close examination of several direct replication
series will serve to illustrate the information available concerning generality
of findings across clients.
Example one: Two successful replications
The example concerns one successful experiment and two successful

first
examined the
replications of a therapeutic procedure. This early clinic2il series
effects of social reinforcement (praise) on severe agoraphobic behavior in
three patients (Agras et al., 1968). This series was also one of the first
evaluations of direct-exposure-based treatments for phobia that have become
the treatment of choice today (Mavissakalian & Barlow, 1981b). This proce-
dure has also come to be known as reinforced practice (Leitenberg, 1976) and
self-observation therapy (Emmelkamp, 1982). The procedure was straight-
forward.
All patients were hospitalized. Severity of agoraphobic behavior was
measured by observing the distance the patients were able to walk on a course
from the hospital to a downtown area. Landmarks were identified at 25 -yard
intervals for over one mile. The patients were asked two or more times a day
to walk as far as they could on the course without feeling "undue tension."
Their report of distance walked was surreptitiously checked from time to time
by an observer to determine reliability, precise feedback of progress in terms
of increases in distance was provided, and this progress was socially rein-
forced with praise and approval during treatment phases and ignored during
withdrawal phases. In the first patient, increases in time spent away from the
center were praised first, but as this resulted in the patient simply standing
outside the front door of the hospital for longer periods, the target behavior
was changed to distance. Because baseline procedures were abbreviated, this
design is best characterized as a B-A-B design (see chapter 5). The compari-
son, then, is between treatment (praise) and no treatment (no praise).
For purposes of generality across clients, it is important to note that the
patients in this experiment were rather heterogeneous, as is typically the case
in applied research. Although each patient was severely agoraphobic, all had
numerous associated fears and obsessions. The extent and severity of
agoraphobic fears differed. One subject was a 36-year-old male with a 15-year
agoraphobic history. He was incapacitated to the extent that he could manage
a 5-minute drive to work in a rural area only with great difficulty. A second
subject was a 23 -year-old female with only a one-year agoraphobic history.
This patient, however, could not leave her home unaccompanied. The third
subject, a 36-year-old female, also could not leave her home unaccompanied,
but had a 16-year agoraphobic history. In fact, this patient had to be sedated
and brought to the hospital in an ambulance. In addition, these 3 patients
presented different background variables such as personality characteristics
and cultural variations (one patient was European).
The results from one of the cases (the male) are presented in Figure 10-1.
Reinforcement produced a marked increase in distance walked, and with-
drawal of reinforcement resulted in a deterioration in performance. Reintro-
duction of reinforcement in the final phase produced a further increase in
distance walked. These results were replicated on the remaining 2 patients.
At least three conclusions can be drawn from these data. The first conclu-
sion is that the treatment was effective in modifying agoraphobic behavior.
The second conclusion is that within the limits of these data, the results are
reliable and not due to idiosyncracies present in the first experiment, since two
replications of the first experiment were successful. The third conclusion,
however, is of most interest here. The procedure was clearly effective with 3
patients of different ages, sex, duration of agoraphobic behavior, and cultural
backgrounds. For purposes of generality of findings, this series of experi-
ments would be strengthened by a third rephcation (a total of 4 subjects). But
the consistency of the results across 3 quite different patients enables one to
draw initially favorable conclusions on the general effectiveness of this proce-
dure across the population of agoraphobic clients through the process of
logical generalization (Edgington, 1967).
On the other hand, if one client had failed to improve or improved only
slightly such that the result was clinically unimportant, an immediate search
would have had to be made for procedural or other variables responsible for
the lack of generality across clients. Given the flexibility of this experimental
design, alterations in procedure (e.g., adding additional reinforcers, changing
the criterion for reinforcement) could be made
an attempt to achieve
in
clinically important results. If mixed results such as these were observed,
further replication would be necessary to determine which procedures were
most efficacious for given clients (see section 2.2, chapter 2).
In this series, however, these steps were not necessary due to the uniformly
successful outcomes, and some preliminary statements about client generality
were made. The next step in this series, then, would be an attempt to replicate
the results systematically, that is, across different situations and therapists. It
1200
1000
'^ 800 -
600 -
400
200 -
10 12 16
BLOCKS OF 5 TRIALS
FIGURE 10-1. The effects of reinforcement and nonreinforcement upon the performance of an
agoraphobic patient (Subject 2). (Figure 2, p. 425, from: Agras, W. S., Leitenberg, H.,and
Barlow, D. H. [1968]. Social reinforcement in the modification of agoraphobia. Archives of
General Psychiatry, 19, 423-427. Copyright 1968 by American Medical Association. Reproduced
by permission.)
is evident that the preliminary series, which was carried out in Burlington,
Vermont, does not address questions on effectiveness of techniques in dif-
ferent settings or with different therapists. It is entirely possible that charac-
teristics of the therapist or the particular structure of the course that the
agoraphobic walked facilitated the favorable results. Thus these variables
must be systematically varied to determine generality of findings across all
important clinical domains. In fact, this step was taken many times. Using
procedures that were operationally quite similar to those described above, but
carrying different labels, Marks (1972) successfully treated a variety of severe
agoraphobics in an urban European setting (London) using, of course,
different therapists, and Emmelkamp (1974, 1982) treated a long series of
Dutch agoraphobics.
In fact, further experimentation over a period of 10 years revealed that

while this intervention was repeatedly successful with thousands of cases,
reinforcement, feedback, and other techniques served primarily to motivate
practice with or exposure to feared objects or situations and that this was the
primary therapeutic ingredient (see Mavissakalian and Barlow, 1981b, for a
review). One strong cue was the rising baseline in Figure 10-1 where
agoraphobics' behavior was improving with practice or exposure alone.
Ideally, of course, reinforcement should not have been introduced until the
baseline stabilized (see section 3, chapter 3). When this was tested properly in
subsequent single-case experimentation, the power of pure exposure, even in
the absence of external motivating variables such as praise, was demonstrated
(Leitenberg, Agras, Edwards, Thomson, &
Wincze, 1970). But the purpose
of these illustrations is to examine the process of establishing generality of
findings through replication and it is to this topic that we now return.
Example two: Four successful replications

with design alterations during replications
A second rather early example of a direct replication series will be pre-

sented because the behavior is clinically important (compulsive rituals), and
the issue of client generality within a direct replication series is highlighted
because 5 patients participated in the study (Mills, Agras, Barlow, & Mills,
1973). In this experiment, what was a new treatment at the time response
prevention
was tested. The basic strategy in this experiment and its replica-
tions was an A-B-A design: baseline, response prevention, baseHne. During
replications, however, the design was expanded somewhat to include controls
for instructional and placebo effects. For example, two of the replications
were carried out in an A-B-BC-B-A design, where A was baseline, B was a
placebo treatment, and C was response prevention.
The addition of new control phases during subsequent replication is not an
uncommon strategy in single-case design research because each replication is
actually a separate experiment that stands alone. When testing a given

treatment, however, new variables interacting within the treatment complex
that might be responsible for improvement may be identified and "teased
out" in later replications. It was noted in chapter 2 that such flexibility of
single-case designs allows one to alter experimental procedures within a case.
Within the context of replication, if a procedure is effective in the first
experiment, one has the flexibiUty to add further, more stringent controls
during repHcation to ascertain more specifically the mechanism of action of a
successful treatment. But, to remain a direct replication series within our
definition, the major purpose of the series should be to test the effectiveness
of a given treatment on a well-defined problem in this case compulsive

rituals administered by the same therapeutic team in the same setting. Thus
the treatment, if successful, must remain the same, and the comparison is
between treatment and no treatment or treatment and placebo control.

The first 4 subjects in this experiment were severe compulsive hand
washers. The fifth subject presented with a different ritual. All patients were
hospitalized on a hand washers encountered articles or
research unit. All
situations throughout the experiment that produced hand washing. Response
prevention consisted of removing the handles from the wash basin wherein all
hand washing occurred. The placebo phase consisted of saline injections and
oral placebo medication with instructions suggesting improvement in the
rituals, but no response prevention. Once again, the design was either A-B-A,
with A representing baseline and B representing response prevention, or A-B-
BC-B-A, where A was baseline, B was placebo, and C was response preven-
tion. Both self-report measures (number of urges to wash hands) and an
objective measure (occasions when the patient approached the sink, recorded

by a washing pen see chapter 4) were administered.
As in the previous series, the patients were relatively heterogeneous. The
first subject was a 31 -year-old woman with a 2-year history of compulsive
hand washing. Previous to the experiment, she had received over one year of
both inpatient and outpatient treatment including chemotherapy, individual
psychotherapy, and desensitization. She performed her ritual 10 to 20 times a
and rinsings with
day, each ritual consisting of eight individual washings
alternating hot and cold water. The
was contamination of
associated fear
herself and others through contact with chemicals and dirt. These rituals
prevented her from carrying out simple household duties or caring for her
child.
The second subject was a 32-year-old woman with a 5-year history of hand
washing. Frequency of hand washing ranged from 30 to 60 times per day,
with an average of 39 during baseline. Unlike with the previous subject, these
rituals had strong religious overtones concerning salvation, although fear of
contamination from dirt was also present. Prior treatments included two
series of electric shock treatment, which proved ineffective.
A third was a 25-year-old woman who had a 3-year history of the
subject
hand-washing compulsion. Situations that produced the hand washing in this
case were associated with illness and death. If an ambulance passed near her
home, she engaged in cleansing rituals. Hand washings averaged 30 per day,
and the subject was essentially isolated in her home before treatment.
The fourth was a 20-year-old male with a history of hand washing
subject
for 1 Vi years. He had
been hospitalized for the previous year and was hand
washing at the rate of 20 to 30 times per day. The fifth subject, whose rituals
differed considerably from the first 4 subjects, will be described below.
Representative results from one case are presented below. Hand washing
remained high during baseline and placebo phases and dropped markedly
after response prevention. Subjective reports of urges to wash declined
slightly during response prevention and continued into follow-up. This de-
cline continued beyond the data presented in Figure 10-2 until urges were
minimal. These results were essentially replicated in the remaining three hand
washers.
Before discussion of issues relative to replication, experimental design
considerations in this series deserve comment. The dramatic success of re-
sponse prevention in this series is obvious, but the continued reduction of

hand washing after response prevention was removed presents some prob-
lems in interpretation. Since hand washing did not recover, it is difficult to
attribute its reduction to response prevention using the basic A-B-A with-
c
Baseline Placebo Response Placebo Baseline
80 1 Prevention
I 4 5 ,
Begin + Placebo
C Exposure 3
xeoA
CO
i^
O
B
c
s
40
V
20-
f
0-
45.
I
A
1.35
'5
O
15
25
/-
KV
I I I ^T 1 1 r-
1 9 11 12 18 19 21 22 28
Two-Day Blocks
FIGURE upper half of the graph, the frequency of hand washing across treatment
10-2. In the
phases is Each point represents the average of 2 days. In the lower portion of the
represented.
graph, total urges reported by the patient are represented. (Figure 3, p. 527, from: Mills, H. L.,
Agras, W. S., Barlow, D. H., and Mills, J. R. [1973], Compulsive rituals treated by response
prevention: An experimental analysis. Archives of General Psychiatry, 28, 524-529. Copyright
1973 by American Medical Association. Reproduced by permission.)
drawal design. From the perspective of this design, it is possible that some
correlated event occurred concurrent with response prevention that was ac-
tually responsible for the gains. Fortunately, the aforementioned flexibility in
adding new control phases to replication experiments afforded an experimen-

tal from a different perspective. In all patients, hand washing was
analysis
reasonably stable by history and through both baseline and placebo phases.
Hand washing showed a marked reduction only when response prevention
was introduced. In these cases, baseline and placebo phases were adminis-
tered for differing amounts of time. In fact, then, this becomes a multiple
baseline across subjects (see chapter 7), allowing isolation of response preven-
tion as the active treatment.
Again, this series demonstrates that response prevention works, and repli-
cations ensure that this finding is reliable. In addition, the clinical significance
of the result is easily observable by inspection, since rituals were entirely
elminated in all 4 patients. More importantly, however, the fact that this
was consistently present across 4 patients lends considerable
clinical result
confidence to the notion that this procedure would be effective with other
patients, again through the process of logical generalization. It is common
sense that confidence in generality of findings across clients increases with
each replication, but it is our rule of thumb that a point of diminishing
returns is reached after one successful experiment and three successful repli-
cations for a total of 4 subjects. At this point, it seems efficient to publish the
results so that systematic replication may begin in other settings.
An alternative strategy would be to administer the procedure in the same
setting to clients with behavior disorders demonstrating marked differences
from those of the first series. Some behavior disorders such as simple phobias
lend themselves to this method of replication since a given treatment (e.g., in
vitro exposure) should theoretically work on many different varieties of
simple phobia. Within a disorder such as compulsive rituals, this is also
feasible because several different types of rituals are encountered in the clinic
(Mavissakalian & Barlow, 1981a; Rachman & Hodgson, 1980). The question
that can be answered in the original setting then is: Will the procedure
work on other behavior disorders that are topographically different but
presumably maintained by similar psychological processes? In other words,
would rituals quite different from hand washing respond to the same proce-
dure? The fifth case in this series was the beginning of a replication along
these lines.
The fifth was a 15-year-old boy who performed a complex set of
subject
when retiring
rituals at night and another set of rituals when arising in the
morning. The night rituals included checking and rechecking the pillow
placement and folding and refolding pajamas. The morning rituals were
concerned mostly with dressing. This type of ritual has come to be known as
checking as opposed to previous washing rituals. The rituals were extremely
time consuming and disruptive to the family's routine. After a baseline phase
in which rituals remained relatively stable, the night rituals were prevented,
but the morning rituals were allowed to continue. Here again, response
prevention dramatically eliminated nighttime rituals. Morning rituals gradu-
ally decreased to zero during prevention of night rituals.
The experiment further suggests that response prevention can be effective

in the treatment of ritualistic behavior. The implications of this replication,
however, are somewhat different from the previous three replications, where
the behavior in question was topographically similar. Although the treatment
was administered by the same therapists in the same setting, this case does not
represent a direct replication because the behavior was topographically dif-
ferent. To consider this case as part of a direct replication series, one would
have to accept, on an a priori basis, the theoretical notion that all compulsive
rituals are maintained by similar psychological processes and therefore will
respond to the same treatment. Although classification of these under one
name (compulsive rituals) implies this, in fact there is some evidence that
these rituals aresomewhat different and may react differently to response

prevention treatments (Rachman & Hodgson, 1980). As such, it was probably
inappropriate to include the fifth case in the present series because the clear
implication is that response prevention is applicable to all rituals, but only
one case was presented where rituals differed.
From the perspective of sound replication procedures, the proper tactic
would be to include this case in a second series containing different rituals.
This second series would then be the first step in a systematic repHcation
series, in that generality of findings across different behaviors would be
established in addition to generality of findings across clients. In fact, re-

sponse prevention and exposure, combined occasionally with medication, has
become the treatment of choice for obsessive-compulsive disorders, based on
an extended systematic and clinical replication series that began in the early
1970s (Rachman & Hodgson, 1980; Steketee & Foa, in press; Steketee, Foa, &
Grayson, 1982). This series, relying on individual experimental analyses and
close examination of individual data from group studies, has also begun to
identify patient characteristics that predict failure (e.g., Foa, 1979; Foa et al.,
1983), a critical function of any replication series (see section 10.4).
Example three: Mixed results in three replications
The goal of this experiment was an experimental analysis of a new proce-

dure for increasing heterosexual arousal in homosexuals desiring this goal
(Herman et al., 1974b). A chance finding in our laboratories suggested that

exposure to an explicitly heterosexual film increased heterosexual arousal in
separate measurement sessions (see section 2.3, chapter 2). Subsequently, this
was tested in an A-B-C-B design, where A was baseline, B was exposure to
heterosexual films (the treatment), and C

was a control procedure in which
the subject was was homosexual.
also exposed to erotic films, but the content
The measures included changes in penile circumference to homosexual and
heterosexual slides (recorded in sessions separate from the treatment sessions)
as well as reports of behavior outside the laboratory setting. The purpose of
the experiment was on heterosexual arousal of exposure
to analyze the effect
to films with heterosexual content over and above the effects of simply
viewing erotic films, a condition obtaining in the control procedure. Thus the
comparison was between treatment and placebo control.
Again, the patients were relatively heterogeneous. The first patient was a
24-year-old male with an 11 -year history of homosexuality. During the year
preceding treatment, homosexual encounters averaged one to three per day,
usually in public restrooms. Also, during this period, the patient had been
mugged once, had been arrested twice, and had attempted suicide. The
second patient was a 27-year-old homosexual pedophile with a 10-year history
of sexual behavior with young boys. The third patient was an 18-year-old
male who had not had homosexual relations for several years but complained
of a high frequency of homosexual urges and fantasies. The fourth patient, a
38-year-old male, reported a 26-year history of homosexual contacts. Homo-
sexual behavior had increased during the previous 4 years, despite the fact
that he had recently married. None of the patients reported previous hetero-
sexual experience with the exception of the fourth subject, who had sexual
intercourse with his wife approximately twice a week. Intercourse was
successful if he employed homosexual fantasies to produce arousal, but he
was unable to ejaculate during intercourse. All patients were seen daily, with
the exception of the fourth patient, who was seen approximately three times
per week.
Representative results from one case, the first patient, are presented in
Figure 10-3. Heterosexual arousal, as measured in separate measurement
sessions, increased during exposure to the female (heterosexual) film,
dropped considerably when the homosexual film was shown, and rose once
again when the female film was reintroduced. The results in this case repre-
sent clear and clinically important changes in heterosexual arousal, and the
experimental analysis isolated the viewing of the heterosexual film as the
procedure responsible for increases. Changes in arousal in the laboratory
were accompanied by report of increased heterosexual fantasies and behavior.
These results were replicated on Subjects 2 and 3, where similar increases in
heterosexual arousal and reports of heterosexual behavior were noted. But
the results from the fourth case differed somewhat, thereby posing difficulties
in interpretation in this direct replication series (Figure 10-4).
In this case, heterosexual arousal increased somewhat during the first
treatment phase, but the increase was quite modest. Withdrawing treatment
resulted in a slight drop in heterosexual arousal, which increased once again
MALE
BASELINE . FEMALE EXPOSURE EXPOSURE FEMALE EXPOSURE
62.5-
Circumference change tO:
Females
Males
50-
37.5-
25
123 456789 10 11 12 13 14 15
BLOCKS OF THREE SESSIONS
{ Circumference Change to Males Averaged Over Each Phase )
FIGURE 10-3. Mean penile circumference change, expressed as a percentage of full erection, to
nude female (averaged over blocks of three sessions) and nude male (averaged over each phase)
slides. (Figure 1, p. 338, from: Herman, S. H., Barlow, D. H., and Agras, W. S. [1974]. An
when the heterosexual film was reinstated. This last increase, however, does
not become clear until the last point in the phase, which represents only one
session. Subsequently, the patient was unable due to to continue treatment
prior commitments precluding an extension of which would have this phase,
confirmed (or disconfirmed) the increase represented by that one point.
Reports of sexual fantasies and behavior were consistent with the modest
increases in heterosexual arousal. While some increase in heterosexual fanta-
sies was noted, the patient continued to employ homosexual fantasies occa-
FEMALE EXPOSURE
MALE FEMALE
EXPOSURE EXPOSURE
75
Circumference change to:

Females
Males
Z LL
oO 50
I.I LU
1 point
iZ O
S^
^^
LU <
= q25-
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
BLOCKS OF TWO SESSIONS
( Circumference Change to Males Averaged Over Each Phase )
FIGURE 10-4. Mean penile circumference change, expressed as a percentage of full erection, to
nude female (averaged over blocks of two sessions) and nude male (averaged over each phase)
slides. (Figure 4, p. 342 from: Herman, S. H., Barlow, D. H., and Agras, W. S. [1974]. An
sionally during sexual intercourse with his wife and was still unable to
ejaculate.
Again, conclusions in three general areas can be drawn from these data.
First, exposure to explicit heterosexual films can be an effective variable for
increasing heterosexual arousal, as demonstrated by the experimental analysis
of the first patient. Second to the extent
, that the results were replicated
directly on three patients, the data are reliable and are not due to idiosyncra-
cies in the first case. It does not follow, however, that generality of findings
across patients- has been firmly established. Although the results were clear
and clinically significant for the first 3 patients, results from the fourth patient
cannot be considered due to the weakness of the effect. In

clinically useful
this case, a clear distinction arises between the establishment of functional
relationships and the establishment of cHnically important generality of find-
ings across clients. As in the first 3 patients, a functional relationship between
treatment and heterosexual arousal was demonstrated in the fourth patient.
This finding increases our confidence in the reliabihty of the result. Unlike the
first 3 patients, however, the finding was not clinically useful. The conclusion,
then, is that this procedure has only limited generality across clients, and the
task remains to pinpoint differences between this patient and the remaining
patients to ascertain possible causes for the limitations on client generality.
The authors (Herman et al., 1974b) noted that the fourth patient differed
two ways from the remaining three. One difference falls under the
in at least
heading of background variables and the other is procedural. First, the
patient was married and therefore was required to engage in heterosexual
intercourse before heterosexual arousal or interest was generated. In fact, he
reported this to be quite aversive, which may have hampered the development
of heterosexual interest during treatment. The remaining patients had expe-
rienced no significant heterosexual behavior prior to treatment. Second, this
patient was seen less frequently than other patients. At most he was seen three
times a week, rather than daily. At times, this dropped to once a week and
even once every 3 weeks during periods when other commitments interfered
with treatment. It is possible that this factor retarded development of hetero-
sexual interest. To the extent that this was a procedural problem, rather than
a variable that the patient brought with him to the experiment, it would have
been possible to alter the procedure prior to the beginning of the experiment
or even during the experiment (i.e., require daily attendance). If this altera-
tion had been undertaken and similar results (the weak effect) had ensued, it
might have limited the search for causes of the weak effect to just the
background variables, such as the ongoing aversive heterosexual behavior. Of
course, this procedural variable was not thought to be important when the
experiment was designed. In fact, failures to replicate are always occurring in
direct replication series. Another good example was presented in the study by
Ollendick et al. (1981) in chapter 8 (Figures 8-3 and 8-4). In this comparison
of two treatments in an ATD, one treatment was more effective than another
for the first subject, but just the opposite was true for the second subject.
Because the investigators were close to the data, they speculated on one
seemingly obvious reason for this discrepancy. Thus, pending a subsequent
test of their hypothesis, they have already taken the first step on the road to
tracking down intersubject variability and establishing guidelines for general-
ity of findings. The investigators themselves are always in the best position to
identify, and subsequently test, putative sources of lack of generality of
findings.
The issue of interpreting mixed results and looking for causes of failure
illustrates an important principle in replication series. We noted above that

subjects in a direct replication series should be as homogeneous as possible. If
subjects in a series are not homogeneous, the investigator is gambling (Sid-
man, 1960). If the procedure is effective across heterogeneous subjects, he or
she has won the gamble. If the results are mixed, he or she has lost. More
specifically, if one subject differs in three or four definable ways from
previous subjects, but the data are similar to previous subjects, then the
experimenter has won the gamble by demonstrating that a procedure has
client generality despite these differences. If the results differ in any signifi-
cant manner, however, as in the example above, the experimenter cannot

know which of the three, four, or more variables was responsible for the
differences. The task remains, then, to explore systematically the effects of
these variables and track down causes of intersubject variability.
In basic research with animals, one seldom sees this type of gamble in a
direct replication series, because most variables are controlled and subjects
are highly homogeneous. In applied research, however, clients always bring to
treatment a variety of historical experiences, personality variables, and other
background variables such as age and sex. To the extent that a given treat-
ment works on 3, 4, or 5 clients, the applied researcher has already won a
gamble even in a direct replication series, because a failure could be attributed
to any one of the variables that differentiate one subject from another. In any
event, we recommend the conservative approach whenever possible, in that
subjects in a direct replication series should be homogeneous for aspects of
the target behavior as well as background variables. The issue of gambling
arises again when one starts a systematic replication series because the re-
searcher must decide on the number of ways he or she wishes the systematic
replication series to differ from the original direct series.
Example four: Mixed results in nine replications
Although all subjects demonstrated some improvement in the study

described above, the data are more variable in a direct replication series. Such
is the case in the following study, where attempts to modify delusional speech
in 10 paranoid schizophrenics produced mixed results (Wincze et al., 1972).
In this procedure the effects of feedback and token reinforcement on delu-
sional speech were evaluated. Feedback consisted of reading sentences with a
high probability of eliciting a particular patient's delusional behavior. If the
patient responded delusionally, he or shewould be informed that the response
was incorrect and given the correct response. For instance, one patient
thought he was Jesus Christ. If he answered affirmatively when asked this
question, he would be told that he was not Jesus Christ, who lived 2,000 years
ago, but rather Mr. M., who was 40 years old. If he answered correctly, he
would be so informed. During token reinforcement phases, the patient re-
ceived tokens redeemable for food and recreational activities, contingent on

nondelusional speech in the sessions. Sessions consisted of 15 questions each
day. Tokens were also administered to some patients for nondelusional talk
on the ward in addition to the contingencies within sessions; but, for our
purposes, we will discuss only the effects of feedback and token reinforce-
ment on delusional talk within sessions.
All patients were chronic paranoid schizophrenics who had been hospital-
ized at least 2 years (the range covered from 2 to 35 years). Six males and four
females participated, with an age range from 25 to 67. Level of education
ranged from eighth grade through college. Thus these patients were, again,
heterogeneous on many background variables.
The experimental design for the first 5 patients consisted of baseline
procedures followed by feedback and then token reinforcement. In some
cases, token reinforcement on the ward, in addition to tokens within sessions,
was introduced toward the end of the experiment. Additional baseline phases
were introduced whenever feedback or reinforcement produced marked de-
creases in delusional talk. For Subjects 6 through 10, the first feedback and
token reinforcement in-session phases were withdrawn, to examine the effects
of token reinforcement when it was presented first in the treatment sequence.
All data were presented individually in the experiment so that any func-
tional relations between treatments and delusional speech were apparent.
Individual data from the first patient are presented in Figure 10-5 to illustrate
the manner of presentation. In this particular case, the baseline phase follow-
ing the first feedback phase was omitted because no improvement was noted
during feedback. Results from all patients are summarized in Table 10-1.
In 5 out of 10 cases, feedback alone produced at least a 20% decrease in
delusional speech within sessions. In two cases, this decrease in delusional
speech was clinically impressive both in magnitude and in the consistent trend
in behavior throughout the phase (Subjects 2 and 8). In the remaining 3
patients, the magnitude of the decrease and/or the behavior trend across the
feedback phase was relatively weak. For instance. Table 10-1 indicates that
the last two data points in the feedback phase for Subject 9 were considerably
lower than the last two data points in the preceding baseline phase (a drop of
49.8%). But the extreme variability in data across the feedback phase indi-
cates that this was a weak effect. A withdrawal of feedback and return to
baseline procedures was not associated with a clear reversal in delusional
speech (at least a 20% increase) in any of the 5 patients who improved,
although the finding is particularly important for those 2 patients who
demonstrated improvement of Thus it was not demon-
clinical proportions.
strated that feedback was the variable responsible for improvement within
treatment sessions.
If the marked improvement of Subjects 2 and 8 had been replicated on
additional patients, onewould be tempted to undertake a further experimen-
1
I 1 1 1 1 1 1 1 1 1 I 1 1 1
14 15 25 26 32 33 43 44 50
DAYS
FIGURE 10-5. Percentage delusional talk of Subject 1 during therapist sessions and on ward for
each experimental day. (Figure 1, p. 254, from: Wincze, J. P., Leitenberg, H., and Agras, W. S.
[1972]. The effects of token reinforcement and feedback on the delusional verbal behavior of
chronic paranoid schizophrenics. Journal of Applied Behavior Analysis 5, 247-262. Copyright y
1972 by Society for Experimental Analysis of Behavior. Reproduced by permission.)
tal analysis to determine which variables were responsible for the improve-
ment. The lack of replication, however, suggests that this would not be a
fruitful line of inquiry.
The results from token reinforcement were quite different. This procedure
was administered to 9 patients. Six (Subjects 1, 2, 4, 5, 8 and 9) improved
an improvement that was confirmed by a return of delusional speech when
token reinforcement was removed. Subject 7 also improved, but delusional
speech did not reappear when token reinforcement was removed. In all of
these patients, the decrease was substantial both in percentage of delusional
speech and in trends across the token phase.
Several conclusions can be drawn from these data. In terms of reduction of
delusional speech within sessions, the experimental analysis demonstrated
that token reinforcement was effective, and replication indicated that the
finding had some reliability. Generality of findings across clients, however, is
limited. Two improve during administration of token rein-

patients did not
forcement. As Sidman (1960) noted, the failure to replicate on all subjects
does not detract from the successes in the remaining subjects. Token rein-
forcement is clearly responsible for improvement in those subjects to the
P r^ -^ On n o
fn />
00 <N t^ <N
O I I I I
O vO ON o^^ O NO w-> -^
? S5 iS
t^ NO f^ ni:
ro
I I I I I I
>V Z v;
r- Tf Tj- <s
SQc^
O o< CO
-^ oo O O O 00
VO "O 0\ vo -H m <r>
I I
f<|ob
NO '^
O 2;
d d <N m I
_ 55 N6obo\odd/^No^-^
-^ <N r- NO OO O 00 ON o <N
I I
ob TJ- ij
NO <N NO r~
o
C/5
ooTtNoc^pa^TtNon<s r-r>oow-^rofnfnONTt-^
cKo-^w^rnCrvNOfS-ÔN
C/5
U
u
z
D
a
u '^<NpNOf^P'<tpr<^fn r~<SNOOfnONt^Ti-ONO
CO ooNOf^No-^r^vor^-NOoo
< vOr>J00-Ô\<Nr~-<S00rf
Qu
.ii
C/)
73 .S T3
l_i (/I U;
.ii
(/)
T3
b;
b.
2
CA
T3
kZ
*t^
(^
jo
4>>
ij^ ^ j^4)>.S w>r
a>> <j^
Djs:
IS
^ ^ v-i r> vONor-r-ooooONON

(/) O) C/3 C/5 (/5C/5C/3C/5C/5(/3a)C/3C/3C/5
342
extent that the experimental design was sound (internally valid). However,
applied researchers cannot stop here, satisfied that the procedure seems to
work well enough on most cases, since the practicing clinician would be at a
loss to predict which cases would improve with this procedure. In fact,
because the authors (Wincze et al., 1972) noted that these two cases actually
deteriorated on the ward during this treatment, the search for accurate
predictions of success becomes all the more important to the clinician. Thus a
careful search for differences that might be important in these cases should
ensue, leading to a more intensive functional investigation and experimental
manipulation of those factors that contribute to success or failure.
In view of the additional fact that all subjects in this series demonstrated
little generalization of improvement from session to ward behavior, analysis
of this treatment is in a very preliminary state and, as Wincze et al. (1972)
pointed out, "... much work needs to be done in order to predict when a
given type of behavioral intervention is likely to succeed in a given case"
(p. 262).
Finally,seems important to make a methodological point on the size of
it
While the nine replications in this series yielded a wealth of data, a

this series.
more efficient approach might have been to stop after four or five replications
and conduct a functional analysis of failures encountered. In the unlikely
event that failures did not occur in the initial replication series, the results
would be strong enough to generate systematic replication in other research
settings, where failures would almost certainly appear, leading to a search for
critical differences at this point. If failures did appear in this shorter series,
the investigators could immediately begin to determine factors responsible for
variant data rather than continue direct replications that would only have a
decreasing yield of information as subjects accumulated. Perhaps for this
reason, one encounters few direct replication series with an N of seven or
more. One notable exception is a multiple-baseline-across-subjects experi-
ment on seven anorexics, where, unfortunately for both experimental and
cHnical reasons, all patients improved substantially (Pertschuk, Edwards, &
Pomerleau, 1978).
Example five: Simultaneous replication
Finally, a method of conducting simultaneous replications has been sug-

gested by J. A. Kelly (Kelly, 1980; Kelly, Laughlin, Clairborne, & Patterson,
1979). This procedure is very useful when one is intervening with a coexisting
group. Examples would be group therapy for any of a number of problems
such as phobia and assertiveness, or interventions in a classroom or on a
hospital ward. In this procedure, any number of subjects in the group can be
treated simultaneously in a particular experimental design, but individual
data would be plotted separately. Figure 10-6 illustrates this strategy with
hypothetical data originally presented by J. A. Kelly (1980). In this hypotheti-
cal strategy, the experimental design was a multiple baseline across behaviors
for six subjects. Three different aspects of social skills were repeatedly
assessed by role playing. Intervention then proceeded for all six subjects on
the first social skill, followed by the second social skill, and so on. In this
hypothetical example, of course, all subjects did very well, with particular
aspects of social skills improving only when treated. Naturally, this strategy
need not be limited to a multiple-baseline-across-behaviors design. Almost
any single-subject design, such as an alternating treatments design or a
standard withdrawal design, could be simultaneously replicated.
From the point of view of replication, this is a very economical and
conservative way to proceed. It is economical because it is less time consum-
ing to treat six clients in a group than it is to treat six clients individually. But
one still has the advantage of observing individual data repeatedly measured
from six different subjects. Naturally, this is only possible where opportuni-
ties for group therapy exist. Furthermore, the procedure is conservative
because fewer variables are different from client to client. The gamble taken
by the investigator in a replication series with increasing heterogeneity or
diversity of subjects or settings was mentioned above. To repeat, if a replica-
tion fails, the more differences there are in subjects, settings, timing of the
intervention, and so forth, the harder it is to track down the cause of the
failure for replication during subsequent experimentation. If all subjects are
treated simultaneously in the same group, at the same time, then one can be
relatively sure that the intervention procedures, as well as setting and tem-
poral factors, are identical. If there is a failure to replicate, then the investiga-
tor should look elsewhere for possible causes, most likely in background
variables or personality differences in the subjects themselves.
Of course, treating clients in group therapy has its own special kind of
setting. If one were interested in the generality of these findings to individual
treatment settings, the first step in a systematic replication series would be to
test the procedure in subjects treated individually. Also, when groups of
individuals are treated simultaneously, one cannot stop the series at just any
time to begin examining for causes of failures if they occur. However, this is
not really a problem as long as the groups remain reasonably small (e.g.,
three to six), such that the investigator would be unlikely to accumulate a

large number of failures before having an opportunity to begin the search for
causes. Other examples of simultaneous replication can be found in an
experiment by E. B. Fisher (1979) mentioned in chapter 8.
Guidelines for direct replication
Based on prevailing practice and accumulated knowledge on direct replica-

tion,we would suggest the following guidelines in conducting a direct replica-
tion series in applied research:
(RATINGS OF EACH SUBJECT'S INDIVIDUAL SOCIAL SKILLS ROLE-PLAVS)

BASE LINE ;GROUP TRAININGIGROUP TRAINING GROUP TRAINING '
ON 1st SKILL I ON 2nd SKILL ' ON 3rd SKILL
FREOUFNCY OF
FIRST
COMPONENT SKILL
IN ROLE PLAY
FREQUENCY OF
SECOND
COMPONENT SKILL
IN ROLE PLAY
FREQUENCY OF
THIRD
COMPONENT SKILL
IN ROLE PLAY
DAYS
FIGURE 10-6. Graphed hypothetical data of simultaneous replications design. (Figure 2, p. 306
from: Kelly, J. A., Laughlin, C, Claiborne, M., & Patterson, J. [1979]. A group procedure for
teaching job interviewing skills to formerly hospitalized psychiatric patients. Behavior Therapy,
10, 299-310. Copyright 1979 by Association for Advancement of Behavior Therapy, Reproduced
by permission.)
1. Therapists and settings should remain constant across replications.

2. The behavior disorder in question should be topographically similar across
clients, such as a specific phobia.
3. Client background variables should be as closely matched as possible,
although the ideal goal of identical clients can never be attained in applied
research.
4. The procedure employed (treatment) should be uniform across clients,
until failures ensue. If failures are encountered during replication, at-
tempts should be made to determine the cause of this intersubject variabil-
ity through improvised and fast-changing experimental designs (see
section 2.3, chapter 2). If the search is successful, the necessary alteration
in treatment should be tested on additional clients who share the charac-
teristics or behavior of the first client who required the alteration. If the
search for sources of variability is not successful, differences in that
particular client from other successful clients should be noted for future
research.
5. One successful experiment and three successful replications are usually
sufficient to generate systematic replication of topographically different
behaviors in the same setting or of the same behavior in different settings.
This guideline is not as firm as those preceding, because results from a
study containing one unusual or significant case may be worth publishing,
or an investigator may wish to continue direct repHcation if experimentally
successful but clinically "weak" results are obtained. Generally, though,
after one experiment and three successful replications, it is time to go on to
systematic replication.
On the other hand, if direct replication produces mixed success and
failure, then investigators must decide when to stop the series and begin to
analyze reasons for failure in what is essentially a new series, because the
procedure or treatment presumably will change. If one success is followed
by two or three failures, then neither the reliability of the procedure nor
the generality of the finding across clients has been established, and it is
probably time to find out why. If two or three successes are mixed in with
one or two failures, then the reliability of the procedure would be es-
tablished to some extent, but the investigator must decide when to begin
investigating reasons for lack of client generality. In any case, it does not
appear to be sound experimental strategy to continue a direct replication
series indefinitely, when both successes and failures are occurring.
6. Broad client generality cannot be established from one experiment and
three replications. Although a practitioner can observe the extent to which
an individual client who responded to treatment in a direct replication
series is similar to his or her client and can proceed accordingly with the
treatment, chances are the practitioner may have a client with a topo-
graphically similar behavior disorder who is different in some clinically
important way from those in the series. Fortunately, as clinical and

systematic replication ensues with other therapists in other settings, many
more clients with different background variables are treated, and con-
fidence in generality of findings across clients, which was established in a
preliminary manner in the first series, is increased with each new replica-
tion.
10.3 SYSTEMATIC REPLICATION

Sidman (1960) noted that where direct replication helps to establish
generality of findings among members of a species, "... systematic replica-
tion can accomplish this and at the same time extend its generality over a wide
range of situations" (p. 111). In applied research, we have noted that direct
replication can begin to establish generality of findings across clients but
cannot answer questions concerning applicability of a given procedure or
functional relationship in different therapeutic settings or by different thera-
pists. Another limitation of the initial direct replication series is an inability to
determine the effectiveness of a procedure proven effective with one type of
behavior disorder on a related but topographically different behavior disor-
der.
Definition of systematic replication
We can define systematic replication in applied research as any attempt to

replicate findings from a direct replication series, varying settings, behavior
change agents, behavior disorders, or any combination thereof. It would
appear that any successful systematic replication series in which one or more
of the above-mentioned factors is varied also provides further information on
generality of findings across clients because new clients are usually included in
these efforts.
Example: Differential attention series
There are now many examples of mature, important, systematic replication

series in appHed research. Extant series on time-out procedures (see J. M.
Johnston &
Pennypacker, 1980), exposure-based treatments for phobia (see
Mavissakalian & Barlow, 1981) and social skills training with a variety of
populations (e.g., Bornstein, Bellack & Hersen, 1980; Hersen & Bellack,
1976; Turner, Hersen, & Bellack, Him-
1978; Wells, Hersen, Bellack, &
melhoch, 1979), among others, have established broad generality for what are
now common therapeutic interventions. But one of the most extensive and
advanced systematic replication series has been in progress since the early
1960s. The purpose of this series has been to determine the generality of the
effectiveness of a single intervention technique, often termed differential

attention. Differential attention consists of attending to a client contingent on
the emission of a well-defined desired behavior. Usually such attention takes
theform of positive interaction with the client consisting of praise, smiling,
and so on. Absence of the desired behavior results in withdrawal of attention,
hence "differential" attention. This series, consisting of over 100 articles, has
provided practitioners with a great deal of specific information on the effec-
tiveness of this procedure in various settings with different behavior disorders
and behavior change agents. Preliminary success in this area has generated a
host of books advocating use of the technique in various settings, particularly
with children in the home or classroom, most often in combination with other
procedures such as other types of reinforcing or mildly punishing conse-
quences including time-out (e.g., Forehand & McMahon, 1981; Patterson,
1982; Ross, 1981; Sulzer-Azaroff & Mayer, 1977). What is perhaps more
important is that articles in this series have noted certain occasions when the
procedure fails, leading to a clearer delineation of the generality of this
technique in all relevant domains in the applied area. A brief review of
findings from this series in the various important domains of applied research
will illustrate the process of systematic repHcation.
Differential attention: Adult psychotic behaviors
One of the first reports on differential attention appeared in 1959 (Ayllon &
Michael). This report contained several examples of the application of dif-
ferential attention to institutionalized patients in a state hospital. The thera-
pists in all cases were psychiatric nurses or aides. The purpose of this early
demonstration was to illustrate to personnel in the hospital the possible
clinical benefits of differential attention. Thus differential attention was
applied to most cases in an A-B design, with no attempt to demonstrate
experimentally its controUing effects. In several cases, however, an experi-
mental analysis was performed. One patient was extremely aggressive and
required a great deal of restraint. One behavior incompatible with aggression
was or lying on the floor. Four-day baseline procedures revealed a
sitting
relatively low rate of being on the floor. Social reinforcement by nurses

increased the behavior, resulting in decreased aggression. Subsequent with-
drawal of social reinforcement produced decreases in the behavior and in-
creases in aggression. Unfortunately, ward personnel could not tolerate this,
and the patient was restrained once again, aborting a return to social rein-
forcement. The resultant A-B-A design was sufficient, however, to demon-
strate the effects of social reinforcement in this setting for this class of
behavior.
This early experiment suggested that differential attention could be effec-
tive when applied by nurses or aides as therapists. These successes sparked
replication by these investigators in additional cases. Other psychotic behav-

ior in adult psychiatricwards modified by differential attention or a combina-
tion of differential attention and other procedures included faulty eating
behavior (Ayllon & Haughton, 1964) and towel hoarding (Ayllon, 1963).
These early studies were the beginning of the systematic replication series, in
that topographically different behavior responded to differential attention.
Another problem behavior in adult psychiatric wards considered more
central to psychiatric psychopathology is psychotic verbal behavior such as
delusions or hallucinations. An early example of the application of differen-
tial attention to delusions was reported by Rickard, Dignam, and Horner
(1960), who attended (smiled, nodded, etc.) to a 60-year-old male during

periods of nondelusional speech and withdrew attention (minimal attention)
during delusional speech. Therapists were psychologists. Initially, nondelu-
sional speech increased to almost maximal levels (9 minutes out of a 10-
minute session) during periods of attention and decreased during the minimal
attention condition. Later, even minimal attention was sufficient to maintain
nondelusional speech. A 2-year follow-up (Rickard & Dinoff, 1962) revealed
maintenance of these gains and reports of generalization to hospital settings.
Unfortunately, only one patient was included in this experiment, precluding
any preliminary conclusion on generality of findings across other patients.
Ayllon and Haughton (1964) followed this up with a series of 3 adult
patients in a psychiatric ward who demonstrated bothersome delusional or
psychosomatic verbal behavior. In all three cases, differential attention was
effective in controlling the behavior, as demonstrated by an A-B-C-B design,
where A was baseline, B was social attention, and C was withdrawal of
attention. Here, as in other reports by Ayllon and his associates, therapists
were nurses or aides. This early experiment was a good direct replication
own right but, more importantly, served to systematically replicate
series in its
findings from the single-case reported by Rickard, Dignam, and Horner
(1960). In Ayllon and Haughton 's experiment, therapists were nurses or aides,
rather than psychologists, and the setting was, of course, a different psychia-
tric ward. Despite these factors, differential attention again produced control
over deviant behavior in adults on a psychiatric ward. This independent,

systematic replication provides a further degree of confidence in the effective-
ness of the technique with psychotic behavior and in its generality across
therapists and settings.
After these early attempts to control psychotic behavior of adults on
psychiatric wards through differential attention, Ayllon and his associates
moved on and developed the token economy (Ayllon
to stronger reinforcers
& Azrin, 1968), abandoning for the most part their work on the exclusive use
of differential attention. The impact of this early work was not lost on clinical
investigators, however, and the importance of differential attention on adult
wards of hospitals was once again demonstrated in a very clever experiment
SCED L
by Gelfand, Gelfand, and Dobson (1967). These investigators observed six

psychotic patients on an inpatient psychiatric ward, to determine sources of
social attention contingent on disruptive or psychotic behavior. At the same
time, they noted who was most successful in ignoring behaviors among the
groups on the ward (i.e., other patients, nurses' aides, or nurses). Results
indicated that other patients reinforced these behaviors least and ignored
them the most effectively, followed by nurses' aides and nurses. Thus the
personnel most responsible for implementing therapeutic programs, the
nurses, were providing the greatest amount of social reinforcement con-
tingent on undesirable behavior. This study does not, of course, demonstrate
the controlling effects of differential attention. But, growing out of earlier
experimental demonstrations of the effectiveness of this procedure, this study
highlighted the potential importance of this factor in maintaining undesirable
behavior on inpatient psychiatric units and led to further replication efforts
on other wards.
After the appearance of these early studies analyzing the effects of dif-
most investigators working in these settings moved on to
ferential attention,
more comprehensive, multifaceted treatment programs incorporating a va-
riety of treatment components in addition to differential attention (e.g.,
Liberman, Neuchterlein, & Wallace, 1982; Monti, Corriveau, & Curran,

1982; Paul & Lentz, 1977). For example, the well-known and very successful
program devised and described by Paul and Lentz (1977) included a compre-
hensive point system, or token economy, as well as other structured training
procedures.
The exciting therapeutic program devised by Liberman, Wallace, and their
colleagues (Wallace et al., in press) emphasized a very detailed and meticulous
approach to training in social and life skills necessary for functioning outside
of the institutional setting. Some of these skills include recreational planning,
food preparation, locating and moving into an apartment, money manage-
ment, job interviews, anger and stress control, long-term planning, and
dealing with friendship or dating situations. While a token economy or point
system is not part of this program, differential attention in terms of praise for
completion of assignments and so forth is woven throughout the various
modules or treatment components. Largely as a result of this integration,
few, if any, studies analyzing the effects of differential attention in isolation
with this population have appeared recently.
Comment on replication procedures
It is safe to say that the impact of this work on adult wards has been
substantial, and differential attention to psychotic behavior is now a common
therapeutic procedure on many wards. More has been thor-
importantly, it
oughly integrated into comprehensive psychosocial treatment programs for

these populations (e.g., Paul & Lentz, 1977; Wallace et al., in press). In
retrospect, however, there are many methodological faults with this series,
leading to large gaps in our knowledge, which could have been avoided had
replication been more systematic.
While differential attention was successfully administered on psychiatric
wards in several different parts of the country across the range of therapists
or ward personnel typically employed in these settings and across a variety of
psychotic behaviors, from motor behavior through inappropriate speech,
only a few studies contained experimental analyses. On the other hand, many
of the reports would come under the category of case studies (A-B designs
with measurement). Certainly, this preliminary series on institutionalized
patients would be much improved had each class of behavior (e.g., verbal
behavior, withdrawn behavior, inappropriate behavior, aggressive or other
motor behaviors) been subjected to a direct replication series with three or
four patients and then systematically replicated in other settings with other
therapists.
This procedure most likely would have produced some failures. Reasons
for these failures could then have been explored, providing considerably more
information to clinicians and ward personnel on the limitations of differential
attention. As it stands, Ayllon and Michael (1959) reported a failure but did
not describe the patient in any detail or the circumstances surrounding the
failure. This type of reporting leads to undue confidence in a procedure
among when failures do occur, disappointment is followed
naive clinicians;
by a tendency to eliminate the procedure entirely from therapeutic programs.
In this specific case, however, what has happened is that differential attention
has been incorporated into more comprehensive programs without adequate
analysis of its contribution. With some cases or in some settings it may be
either important or superfluous. In other cases it may even be detrimental (see
Herbert et al., 1973).
This early series also illustrated a second use of the single-case study (A-B).
In chapter 1 we noted that case studies can suggest initially that a new
technique is clinically effective, which can lead to more rigorous experimental
demonstration and direct replication. In a systematic replication series the
single-case study makes another appearance. Many reports are published that
include only one case, but replicate an earlier direct replication series in either
an experimental or an A-B form. Usually the reports are from different
settings and contain a slight twist, such as a new form of the behavior
disorder or a slight modification of the procedure. While these reports are less
desirable from the larger viewpoint of a systematic replication series, the fact
is that they are published. When a sufficient number accumulate, these
reports can provide considerable information on generality of findings. We
will return to this point later.
Differential attention: Other adult behaviors
The early success of differential attention and positive reinforcement pro-

cedures in general with institutionalized patients led to application of this
procedure to other adult behavior disorders in other settings.
Most of these examples were published as single-case reports. Some of
these single-cases contain a functional analysis of differential attention;
others are A-B designs wth measurements. For instance, Brookshire (1970)
eliminated crying in a 47-year-old male suffering from multiple sclerosis by
attending to incompatible verbal behavior. Other single-case examples include
Brady and Lind's (1961) modification of hysterical blindness through dif-
ferential attention to a visual task in a hospital setting. A hospital setting was
also utilized to test the effectiveness of differential attention on a conversion
reaction, specifically astasia-abasia, or stumbling and falling while walking
(Agras, Leitenberg, Barlow, & Thomson, 1969). Praise combined with ig-
noring stumbling resulted in improvement in this case. In another setting,

these procedures also proved effective on a similar case (Hersen, Gullick,
Matherne, & Harbert, was treated in a hospital
1972). Psychogenic vomiting
settingby Alford, Blanchard, and Buckley (1972) who ignored vomiting and
withdrew social contact immediately after vomiting. Therapists in this case
were nurses. The authors cite success of this procedure on vomiting in a child
(Wolf, Birnbrauer, Williams, & Lawler, 1965) as a rationale for attempting it
with an adult. More recently. Redd has extended this work by demonstrating
the usefulness of differential attention in controlling retching and vomiting in
cancer patients undergoing chemotherapy Redd, 1980). Specifically,
(e.g.,
nurses seem able to manage the well-known conditioned nausea response

using differential attention.
Various other case studies along these lines were published. Many of the
studies describe slight modification of the procedure or some variation in the
behavior disorder. As in the treatment of psychotic patients, differential
attention also was combined with other treatment variables such as other
forms of positive reinforcement or punishment in many research reports,
making it difficult to specify the exclusive effects of differential attention.
From an historical viewpoint, one of the more interesting studies on
differential attention was reported by Truax (1966), who reanalyzed tape
recordings of Carl Rogers' therapy sessions. He discovered that Rogers
responded differently (i.e., positively) to five classes of verbal behavior over a
number of therapy sessions, and four of these classes increased in frequency.
This is reminiscent of the verbal conditioning studies (e.g., Greenspoon,
1955) and suggests, in a non-experimental A-B fashion, that differential
attention is operative in a variety of different psychotherapeutic approaches.
But, once again, few if any studies examining the effects of differential
attention in isolation with non-psychotic adult populations have occurred in

recent years.
The reasons for this seem to be very similar to those described above in
series on institutionalized psychotic patients. That is, differential attention
has been "co-opted" into larger treatment packages without further analysis
of its effects. One good example is marital therapy. In a large, early series
Goldstein (1971) used differential attention procedures with 10 women who
were experiencing marital difficulties. Specifically, these women were in-
structed on attending to desired behaviors emitted by their husbands and
ignoring undesirable behaviors. Using a time series analysis, statistically
significant changes occurred in eight out of ten cases. To the extent that these
changes were clinically as well as statistically significant, these uncontrolled
case studies suggested that differential attention was effective in this context.
Since that time, marital therapies based broadly on social learning principles
have become well developed and are widely used for the treatment of marital
distress & MargoHn, 1979; Liberman et al., 1980; 0*Leary &
(Jacobson
Tbrkewitz, 1981).Most of these programs contain a variety of interventions,
including comunications training, problem solving, and instructions on al-
tering various dyadic patterns of behavior. Embedded within these ap-
proaches, however, is a strong differential attention component. For
example, when leading marital therapists describe their actual approaches in
great detail (e.g., L. F. Wood & Jacobson, 1984), these treatments include
training in expressions of appreciationand praise contingent on desirable
partner behavior. Often this most prominent in the early stages of therapy.
is
For example, during "caring days" husbands and wives are taught to express
appreciation for positive qualities or behaviors of their spouses. Ways in
which spouses would like their partners to express appreciation are carefully
explored in the therapy session. These types of expressions, most often
including positive verbal feedback of some sort or another, are then inte-
grated into the couples' daily lives. Unfortunately, this treatment component
has never been evaluated systematically, and thus, once again, we are not sure
of the specific conditions in which it succeeds or fails.
Comment on replication procedures
Thus the deficits and faults in this area are similar to those encountered in
the series with psychotic adults described above. Evidence exists that differen-
tial number of settings (e.g., inpatient, outpa-
attention can be effective in a
tient, when applied by different therapists (e.g., doctors, nurses, or
or home)
wives) on a number of different behavioral problems. The difficulty here is
with the dearth of experimental analyses and direct replication in each new
setting or with each new problem. Nevertheless, clinical investigators have for
the most part not followed the type of detailed technique-building approach
described in chapter 2 that would ensure that treatment programs, such as
marital therapy, be as powerful as they might be.
Differential attention: Children's behavior disorders
In fact, differential attention procedures applied to adults, whether psy-

chotic or nonpsychotic, comprise only a small part of the work reported in
this area. The greatest number of experimental on the effectiveness
inquiries
of differential attention have been conducted with children, and this series
represents what is probably the most comprehensive systematic replication
series to data. One of the earliest studies on the application of differential
attention to behavior problems of a child was reported by C. D. Williams
(1959), who instructed parents to withdraw attention from nighly temper
tantrums. When an aunt unwittingly attended to tantrum behavior, tantrums
increased and were extinquished once again by withdrawal of attention.
Table 10-2 presents summaries of replication efforts in this series since that
time. Studies reported in this table used differential attention as the sole or, at
least, a very major treatment component. Studies where differential attention
was a minor part of a treatment package, such as parent training, were for the
most part omitted. It is certainly possible that a few additional studies were
inadvertently excluded. In the table, it is important to note the variety of
clients, problem behaviors, therapists, and settings described in the studies,
because generality of findings in all domains is entirely dependent on
relevant
the diversity of settings, clients, and the rest employed in such studies. One
should also note that the bulk of this work occurred in the late 1960s and
early 1970s, with a decrease in published research since that time. Unlike the
examples above, this is due to the fact that many of the goals of this
systematic replication series were completed. We will discuss this issue further.
Most replication efforts through 1965 presented an experimental analysis
of results from a single-case (see Table 10-2). A good example of the early
studies was presented by Allen et al. (1964), who reported that differential
attention was responsible for increased social interaction with peers in a
socially isolated preschool girl. The setting for the demonstration was a
classroom, and the behavior change agent, of course, was the teacher. While
most of the early studies contained only one case, the experimental demon-
stration of the effectiveness of differential attention in different settings with
different therapists began to provide information on generality of findings
across all-important domains. These replications increased confidence in this
procedure as a generally effective clinical tool. In addition to isolate behavior,
the successful treatment of such problems as regressed crawling (Harris,
Johnston, Kelley, & Wolf, 1964), crying (Hart, Allen, Buell, Harris, & Wolf,
1964), and various behavior problems associated with the autistic syndrome
(e.g., Davison, 1965) also suggested that this procedure was applicable to a
wide variety of behavior problems in children while at the same time provid-
ing additional information on generality of findings across therapists and
settings.
Although studies of successful application of differential attention to a
single-case demonstrated that this procedure is applicable in a wide range of
situations, a more important development in the series was the appearance of
direct replication efforts containing three or more cases within the systematic
Although reports of single-cases are uniformly successful,
replication series.
or they would not have been published, exceptions to these reports of success
can and do appear in series of cases, and these exceptions or failures begin to
define the limits of the applicability of differential attention.
For this reason, it is particularly impressive that many series of three or
more cases reported consistent success across many different clients, with
such behavior disorders as inappropriate social behavior in disturbed hospi-
talized children (e.g., Laws, Brown, Epstein, & Hocking, 1971), disruptive
behavior in the elementary classroom (e.g., Cormier, 1969; R. V. Hall et al.,
1971; R. V. Hall, Lund, & Jackson, 1968) or high school classroom (e.g.,
Schutte & Hopkins, 1970), chronic thumb-sucking (Skiba, Pettigrew, &
Alden), disruptive behavior in the home (Veenstra, 1971; Wahler, Winkel,
Peterson, & Morrison, and disruptive behavior in brain-injured
1965),
children (R. V. Hall & Broden, 1967). These improvements occurred in many
different settings such as elementary and high school classrooms, hospitals,
homes, kindergartens, and various preschools. Therapists included profes-
sionals, teachers, aides, parents, and nurses (see Table 10-2).
The consistency of their success was impressive, but as these series of cases
accumulated, the inevitable but extremely valuable reports of failures began
to appear. Almost from the beginning, investigators noted that differential
attention was not effective with self-injurious behavior in children. For
instance, Tate and Baroff (1966) noted that in the length of time necessary for
differential attention to work, severe injury would result. In place of differen-
tial attention, a strong aversive stimulus
electric shock
proved effective in
suppressing this behavior. Later, Corte, Wolf, and Locke (1971) found that
differential attention was totally ineffective on mild self-injurious behavior in
retarded children but, again, electric shock proved effective. Because there
are no reports of success in the literature using differential attention for self-
injurious behavior, it is unlikely that these cases would have been published at
all if differential attention had not proven effective on other behavior disor-
ders. Thus this is an example of a systematic replication series setting the

stage for reports of limitations of a procedure.
More subtle limitations of the procedure are reported in series of cases
wherein the technique worked in some cases, but not in others. In an early
series, Wahler et al. (1965) trained mothers of young, oppositional children in
is o o s s s
ZZ :5i ;5J :5i
(4-1
{3 2 2 o o
0U 0U
9i
g
1 I 8
8
ill
C 4> t_
CO
I Si 6 1 1 6.9
4> w a o. o I
X 0^ u
C 3 C
:3 c PU 11
x> CO
2 P
^ i.
16
5 Êô. .S.I-5
(u O -r "2 "o > >
a>
g
x> wj "3 x; Si
'^
- o 2 .2 '? I ^
|Ie| a llllil S to C3 4> dO
(S^ ci "O a,
4)
^_^
w
13 4>
6l
g u 13
g i2i
1 S E 13 6
u T3 I B E T3
n 9 2 a o 9 S vi ^Z t5
u ? O "o T3 1*U
E
I
7* *= cl<=^
G
^
,cO CO "O "is "K
00
4 i i:J 4 rA $ vi.E
<^
^ <^ c
=
On ( X "S 4> Co
On c
w ^
2 Si t
CO
S
E b ffl
g^ 3:
f^ N
.2E
t;5
c
= ^^ a5
OU On
S 1 11
I if x:
o
o
J c Jo o=a 4J
O
11
</5
. t/T
DC 2
OX g ^ -C ^4- W' Jj-J
.2 -it c K Harri
Allen
11
ui 115 M.
356
2
u
I JS I I
ft-
<
X
Si wh
o 21 o
Pu -S
O
!3
o Se
I I 8 -1
cj O ^ J -I
o <^ .9 <*H o .9
*>
? (u a t: jc
- ^3 > g ^
6 c 3 4>
-?>-*' t_ "O o, .^ .> 60 u, o .9 2^ -B IJ C > >
^ a > o c
^1 ?^y 2- fc
I 1 ^ ^' o Q -o X> C/5
J i I iI X g 8-2
o ^ a> (U 2
? 73 |
t-i ,4>
^1 G 6
- E 2o T3
^ "o
r^ 9 4>
J2
^ i
1
^ ?
^
1
4 m vo o S a4 vi 00
c
<^
S"
CQ
(/T
c .-s
si o 2i o<
c
^ CQ

c
X PQ </5 '=y
^^ -^
oo
o
T3 CO 0\ CO 4> O
0- ?i
J
-^ 2O
T3 en < CO pQ Ql^
o < oo
(55 2i c c 2i
en
H 2H
lT
c P ^^
^ ^"^
^ K- OO w ^ ^ oo
gC^ &
S^ lu r" CO
e3 5 2 ^ Os
CO O oa 05 USw Q^Q i=4 ^
357
V3 l/i
<
z
ft, <
X
*- JT! fl>
O CO CO
SfSi
â 8
le^6-2 6 2
c o
a> 2 "C 2 11 1
S
s P 4>
s > t 'o
t 5 G > </5
o o H iS o o (/5
j2 o ^ a JS
o-a t''
DC X SU u X PC "o PU "o (/) "o
o
s ^ <=>
W5 ,
CO 3t.S
us
U
1> CO
^^
c -. > o 5 ^ o g <u 2 .52 u
OO
1
rr Jrt
'K 'S ^ rs cj -rc S CO -->

.J-. '5b
00
S
H 5 =
T3 o g 60O.S2 go.-ga.fe '^
2 g 8 2 5 O 6 '%. =1
E CO Q (5 e w s <2 fS II
4>
"cO
8 e a
1 M a
b^ O ^ *?"
s (u -75
<ii ^
o 2o "S
"o a -s A a
2
9 o o 00 ^ a >.
2
T3
io Ô ^ OTN
VO (U
I
CO a M ^ a^ CO
.
vi
4> A s
-o I
T3
5 CO
T3 CL) J
SO "o PC 73 <s f^ a On A
60
ca
^S
|i
y a
-2 ir^y
a
o
E C.
-^ - oa
a
=2 s| a E
*i 5 j^ . =3 o S
53 as
I
2 > o a
b c5? ^ ^ OQ U OQ UÔ Oc)5 C/5
358
la I I
I
I
^ 4>
Q CO
73
I
2 o
w
s^
o
CO to iS
oo.su U
r- 4> 3 ^ g
I -B 9 S bo 2p 2 fe
.S .S H .2
.2 ^ T .2 2 .S ^
S"^
43 CO ffl <iJ W
S H U fed- ^
O r-
Tt n vd
^ 5 1 ^
rS s,
CO c3
6 3 C
5
9
T. -O
1 s
b-o 2 S 2 ^ o 1 1lîa9l?l
o p
o E
II
<^ t ^_^
S
B r~
g ON
O < 4>
^ ii .2?
CO
i 1
m 00
c 3 1 2
. & CP *;< (U
0 W3
ii Q
Jl c/o 1 :5 ;2
359
s
Ou <
m
(t^ C
(U
C4
*-
Mi
"O
3
4-<
13 C/9
(/3
-
4)
c3
a> a>
a>
Xo J= J= o 3
o u DC (2 "^
3 3 B3
CO o I-.
0^ C &H H
C? H
L H
L ^'
<N o
OS 4D
"o O
73 o 13 o S
CO 4> i-i a> SI
c ^ 6 ^ E .2 ^ fe
D C O a> o -3 4> !
S 5 ii -fi
o
:s 8DC UK QC
2i
aps5
iJ
aw S^
-
O t ,^ iS
-H <s
.
Dli J3
T3
"2 " ^
I
1.2
g ^ -^ > I boS
-a (4-1 M i> ** S -3 t- t,
> o - 5
4> -C .> .^ S3 6 I I -S
S k w CO 1/3 > 00 ^ ^ 8 -s 8 -^
11 .S.ti g.S 2 ^ S ^.S ^ ?
^ s
^ .2> ' c > -a P a>
o </3 fto a>
^ "".SO
<N -H
1)
13
^
13
"O
"o -a
c
i c 2 13
O O 2 13 13 2 13 5 2i
E E Â E 2
Tt g 00 c
2 S-^-g ^ (/TS ? 2o
i^ o o ^w TH 9 9 -a c3 5
C CO >^
^ ^
I
c '. '. '.
"'g^'OE^Ê^ ^13 ^2-0 13 ^ E ^^

m cO-^ cO4-iOOr(<--^ <r^ t4_ 00 m
0^
^-s
oo
o
r-
o\ 2
o ^-^ o (L>
2
13 ^
g >%
s ^3
cia
o 13 3
3
- ^
I (S
2 > y 1
'^ ^ 3 <
5;:
o o^ 1^ o
CO ^ .
a sa^ Oh 03 S >O
360
differential attention procedures. The setting was an experimental preschool.

In two out of three cases the mothers were quite successful in modifying
oppositional behavior in their children, and an experimental analysis isolated
differential attention as the important ingredient. In a third child, however,
this procedure was not effective, and an additional punishment (time-out)
procedure was necessary. The authors did not offer any explanation for this
discrepancy, and there were no obvious differences in the cases that could
account for the failure based on descriptions in the article. The authors did
not seem concerned with the discrepancy, probably because it was an early
effort on the replication series, and the goal was to control the oppositional
behavior, which was accomplished when time-out was added. This study was
important, however, for it contained the first hint that differential attention
might not be effective with some cases of oppositional behavior.
In a later series, after differential attention was well established as an
effective procedure, further failures to replicate did elicit concern from the
investigator (Wahler, 1968, 1969a). Wahler trained parents of children with
severe oppositional behavior in differential attention procedures. Results
indicated that differential attention was ineffective across five children, but
the addition of time-out again produced the desired changes. Replication in
two more cases of oppositional behavior confirmed that differential attention
was only effective when combined with a time-out procedure.
In the best tradition of science, Wahler (1969a) did not gloss over the
failure of differential attention, although his treatment "package" was ulti-
mately successful. Contemplating reasons for the failure, Wahler hypothe-

sized that in cases of severe oppositional behavior, parental reinforcement
value may be extremely low; that is, attention from parents is not as reinforc-
ing. After treatment using the combination of time-out and differential
attention, oppositional behavior was under control, even though time-out
was no longer used. Employing a test of parental reinforcement values,
Wahler demonstrated that the treatment package increased the reinforcing
value of parental attention, allowing the gain to be maintained. This was the
first clear suggestion that therapist variables are important in the application
of differential attention, and that with oppositional children particularly,
differential attention alone may be ineffective due to the low reinforcing
value of parental attention.
Although differential attention occasionally has been found ineffective in
other settings, such as the classroom (O'Leary et al., 1969), other investiga-
tors actually observed deleterious effects under certain conditions (e.g., Her-
bert et al., 1973; Sajwaj &
Hedges, 1971). For example, Herbert et al. (1973)
trained mothers in the use of differential attention in two separate geographi-
cal locations (Kansas and Mississippi). Although preschools were the settings
in both locations, the design and function of the preschools were quite
dissimilar. Clients were children with a variety of disruptive and deviant

behaviors, including hyperactivity, oppositional behavior, and other inappro-
These young children presented different background
priate social behaviors.
variables, from familial retardation through childhood autism and Down's
syndrome, and they came from differing socioeconomic backgrounds. The
one similarity among the six cases (two from Mississippi, four from Kansas)
was that differential attention from parents was not only ineffective but
detrimental in many cases, in that deviant behavior increased, and dangerous
and surprising side effects appeared. Deleterious effects of this procedure
were confirmed in extensions of A-B-A designs, where behavior worsened
under differential attention and improved when the procedure was with-
drawn.
These results were, of course, surprising to the authors, and discovery of
similar results in two through personal communication prompted the
settings
combining of the data into a single publication. In this particular report the
investigators were unable to pinpoint reasons for these failures. As the
authors note, "... the results were not peculiar to a particular setting,
certain parent-child activities, observation code or recording system, experi-
menter or parent training procedure. Subject characteristics also were not
predictive of the results obtained" (Herbert et al., 1973, p. 26). But in one
case where time-out was added, disruptive behavior declined. In fact, Sajwaj
and Dillon (1977) analyzed a large portion of their systematic replication
series and found a ratio of 87 individual successes to only 27 individual
failures. In many of the cases that failed, the addition of another procedure,
such as time-out, quickly converted the failure to a success. More recent
studies have continued to find that adding time-out corrects differential
attention failures (Roberts, Hatzenbuehler, & Bean, 1981).
As noted above, the number of articles analyzing the effects of differential
attention with children has dropped off markedly in recent years, as is evident
in Table 10-2. Most likely this is due to widespread confidence in its general
applicability. But another reason is that the field has moved on. As was the
case with various adult behaviors, differential attention has been fully incor-
porated into a package treatment, usually referred to sls parent training (e.g.,
Forehand & McMahon, 1981). This package consists of additional compo-
nents to differential attention, such as time-out and training in the discrimina-
tion of certain instructions or commands. Since this package has been well
worked out, the field is now more concerned with results from a clinical
replication analysis of the treatment package than with continued systematic
replications of the differential attention procedure attempting to determine
what conditions predict failure. Yet, in 1979 Wahler, Berland, and Coe
referred to these occasional failures of differential attention as one of the
anomalies of operant interventions.
Comment on replication
In our view, data on failures are a sign of the maturity of a systematic

replication series.Only when a procedure is proven successful through many
replications, do negative results assume this importance. But these failures do
not detract from the successful replications. The effectiveness of differential
attention has been established repeatedly. These data do, however, indicate
that there are conditions that even today are not fully understood that limit
generality of effectiveness and that practitioners must proceed with caution
(Wahler et al., 1979).
In conclusion, this advanced systematic replication series on differential

attention has generated a great deal of confidence among practitioners. The
evidence indicates that it can be effective with adults and children with a
most any
variety of behavioral problems in setting. The clinically oriented
books and monographs widely advocating its most often in combination
use,
with other procedures as part of a treatment package (Forehand & McMa-
hon, 1981; Jacobson & Margolin, 1979; Patterson, 1982; Paul & Lentz,
1977), have made this procedure available to numerous professionals con-
cerned with behavior change, as well as to the consuming public. In fact,
most editors of appropriate journals probably would not consider accepting
another article on differential attention unless it illustrated a clear exception
to the effectiveness of this procedure, as did the Herbert et al. (1973) report.
However, the process of establishing generality of findings across all rele-
vant domains is a slow one indeed, and it will probably be years before we
know all we should about this treatment or other treatments currently under-
going systematic replication. As we pointed out in the context of adult
psychotic behavior, investigators probably proceeded too quickly to incor-
porating differential attention into various package treatments without fully
understanding the limits of its effects. Even with the very informative and
complete systematic replication series on childhood problems, we do not yet
know what predicts failure from differential attention. In fact, there are
many promising hypotheses to account for these failures (Paris & Cairns,
1972; Sajwaj & Dillon, 1977; Wahler, 1969a; Warren & Cairns, 1972). But
these have not yet been explored in the applied setting. Until the time that the
process of systematic replication reveals the precise limitations of a proce-
dure, clinicians and other behavior change agents should proceed with cau-
tion, but also with hope and confidence that this powerful process will
ultimately establish the conditions under which a given treatment is effective
or ineffective.
Guidelines for systematic replication
The formulation of guidelines for conducting systematic replication is
more difficult than for direct replication due to the variety of experimental
efforts that comprise a systematic replication series. However, in the interest

of providing some structure to future systematic replication, we will attempt
to provide an outline of the general procedures necessary for sound system-
atic replication in applied research. These procedures or guidelines fall into
four categories.
1. Earlier we defined systematic replication in applied research as any at-

tempt to replicate findings from a direct replication series, varying set-
tings, behavior change agents, behavior disorders, or some combination
thereof. Ideally, then, the systematic replication should begin with sound
direct replication where the rehability of a procedure is established and the
beginnings of client generality are ascertained. If results in the initial
experiment and three or more replications are uniformly successful, then

the important work of testing the effectiveness of the procedure in other
settings with other therapists and so on can begin. If a series begins with a
report of a single case (as it often does), then the first order of business is
to initiate a direct replication series on this procedure, so that the search

for exceptions can begin.
2. Investigators evaluating systematic replication should clearly note the
differences among their clients, therapists, or settings from those in the
original experiment. In a conservative systematic replication, one, or
possibly two, variables differ from the original direct replication. If more
than one or two variables differ, this indicates that the investigator is
"gambhng" somewhat (Sidman, 1960). That is, if the experiment suc-

ceeds, the series will take a large step forward in establishing generality of
findings. If the experiment fails, the investigator know which of the
cannot
differing variables or combination of variables was responsible for the
change and must go back and retrace his or her steps. Whether scientists
take the gamble depends on the setting and their own inclinations; there is
no guideline one could suggest here without also limiting the creativity of
the scientific process. But it is important to be fully aware of previous
efforts in the series and to list the number of ways in which the current
experiment differs from past efforts, so that other investigators and
clinicians can hypothesize along with the experimenter on which dif-
ferences were important in the event of failure. In fact, most good
scientists do this (e.g., Herbert et al., 1973).
3. Systematic replication is essentially a search for exceptions. If no excep-
tions are found as replications proceed, then wide generality of findings is
established. However, the purpose of systematic replication is to define the
conditions under which a technique will succeed or fail, and this means a
search for exceptions or failures. Thus any experimental tactics that hinder
the finding and reporting of exceptions are of less value than an experi-
mental design that highlights failure. Of those experimental procedures
typically found in a systematic replication series (e.g., see Table 10-2), two
fall into this category: the experimental analysis containing only one case
and the group study.
As noted above, the report of a single-case, particularly when accompa-
nied by an experimental analysis, can be a valuable addition to a series in
that it describes another setting, behavior disorder, or other item where the
procedure was successful. Reports of single-cases also may lead to direct
and systematic repHcation, as in the differential attention series. Unfortu-
nately, however, failures in a single-case are seldom published in journals.
Among the numerous successful reports of single-case studies contained in
the differential attention series, very few reported a failure, although it is
our guess that differential attention has failed on many occasions, and
these failures simply have not been reported.
The group study suffers from the same limitation because failures are
lost in thegroup average. Again, group studies can play an important role
in systematic replication in that demonstration that a technique is success-
ful with a given group, as opposed to individuals in the group, may serve
an important function (see section 2.9). In the differential attention series,
several investigators thought it important to demonstrate that the proce-
dure could be effective in a classroom as a whole (e.g., Ward & Baker,
1968). These data contributed to generality of findings across several
domains. The fact remains, however, that failures will not be detected
(unless the whole experiment fails, in which case it would not be
published), thus leading us no closer to the goal of defining the conditions
in which a successful technique fails. In clinical replication, ox field testing,
described below, one has more flexibility in examining results from large
groups of treated clients as long as it is possible to pinpoint individuals
who succeed or fail.
Finally, the question arises: When is a systematic replication series over?

For direct replication series, it was possible to make some tentative recom-
mendations on a number of subjects, given experimental findings. With
systematic replication, no such recommendations are possible. In applied
research, we would have to agree with Sidman's (1960) conclusion con-
cerning basic research that a series is never over, because scientists will
always attempt to find exceptions to a given principle, as well they should.
It may be safe to say that a series is over when no exception to a proven
therapeutic principle can be found, but, as Sidman pointed out, this is
entirely dependent on the complexity of the problem and the inductive
reasoning of clinical researchers who will have to judge in the light of new
and emerging knowledge which conditions could provide exceptions to old
principles. Of course, series will eventually begin to "fade away," as with
the differential attention series, when wide generality of applicability has
been established.
Fortunately, practitioners do not have to wait end of a series to

for the
apply interim findings to their knowledge is cumula-
clients. In these series,
tive. A clinician may apply a procedure from an advanced series, such as
differential attention, with more confidence than procedures from less ad-
vanced series (Barlow, 1974). However, it is still possible through inspection
of these data to utilize those new procedures with a degree of confidence
dependent on the degree to which the experimental clients, therapists, and
At the very least, this is a
settings are similar to those facing the clinician.
good beginning to the often discouraging and sometimes painful process of
clinical trial and error.
10.4 CLINICAL REPLICATION

A somewhat different type of replication process occurs only in applied
research. We have termed this process clinical replication (Hersen & Barlow,
1976). Clinical replication an advanced replication procedure in which a
is
treatment package containing two or more distinct procedures is applied to a

succession of clients with multiple behaviors or emotional problems that
cluster together; in other words, the usual and customary types of multiface-
ted problems that present to practitioners such as conduct problems in
children, depression, schizophrenia, or autism.
Direct replication was defined as the administration of a given procedure
by the same investigator or group of investigators in a specific setting (e.g.,
hospital, clinic, classroom) on a series of clients homogeneous for a particular
behavior disorder such as agoraphobia or compulsive hand washing. As this
definition implies, one component of a treatment procedure is applied to one
well-defined problem in succeeding clients. Similarly, systematic replication
examines the effectiveness of this functional relationship across multiple
settings, therapists, and (related) behaviors. Most often, direct and systematic
replications are testing only one component of what eventually becomes a
treatment package, as in the examples above.
In constructing an effective treatment package, however, it is very impor-
tant that one develop and test treatments for one problem at a time, with the
eventual goal of combining successful treatments for all coexisting problems.
This is the technique-building strategy suggested by Bergin and Strupp (1972).
For example, one of the direct replication series described above tested the
effects of a specified treatment on delusional speech, which, of course, is
often one component of schizophrenia (Wincze et al., 1972). If this series
were consistently successful, the applied researcher might begin to test treat-
ments for coexisting problems in these patients, such as social isolation or
thought disorders, if these were present. When successful procedures had
been developed for all coexisting problems, the next step would be to estab-
lish generality of findings by replicating this treatment package on additional
patients who present a similar combination of problems. This would be

Wallace, 1982). The insertion of differential atten-
clinical replication (e.g.,
and other well-tested procedures into a "parenting" package is
tion, time-out,
a good example of technique building resulting in a treatment ready for
clinical replication.
Another name for clinical replication, then, could be field testing, because
this iswhere clinicians and practitioners take newly developed treatments or
newly modified treatments and apply them to the common, everyday prob-
lems encountered in their practice. While this process can be carried out by
either full-time clinical investigators or scientist-practitioners (Barlow et al.,
1983), establishing the widest possible client and setting generality would
require substantial participation by full-time practitioners. The job of these
practitioners, then, would be to apply these treatments to large numbers of
their clients while observing and recording successes and failures and analyz-
ing through experimental strategies, where possible, the reasons for this
individual variation. But even if practitioners are not inclined to analyze
causes for failures in the application of a particular treatment package, full
descriptions of these failures will be extremely important for those investiga-
tors who are in a position to carry on this search (Barlow et al., 1983).
Thus, while all facets of single-case experimental research are much closer
to the procedures in clinical or applied practice than to other types of research
methodology (see below), chnical replication in its most elementary form
becomes almost identical with the activities of practitioners.
Definition of clinical replication
We would define chnical replication as the administration by the same

investigator or practitioner of a treatment package containing two or more
distinct treatment procedures. These procedures would be administered in a
specific setting to a series of clients presenting similar combinations of multi-
ple behavioral and emotional problems. Obviously, this type of replication
process advanced in that it should be the end result of a systematic,
is
technique-building applied research effort, which should take years.

Of course, there are many clinical replication series in the literature describ-
ing the apphcation of comprehensive treatments that did not benefit from
careful technique-building strategies. One good example is the Masters and
Johnson series describing the treatment of sexual dysfunction. Because of this
weakness, this treatment approach, which enjoys wide application, is now
coming under increasing attack one that does not have wide generality of
as
effectiveness (Zilbergeld & Evans, 1980). And, since no technique-building
strategy preceded the introduction of this treatment, we have no idea why.
Example: Clinical replication with autistic children
One of the best examples of a clinical replication series is the work of

Lovaas and his colleagues with autistic children (e.g., Lovaas, Berberich,
Perloff, & & Simmons, 1965; Lovaas &
Schaeffer, 1966; Lovaas, Schaeffer,
Simmons, The diagnosis of autism fulfills the requirements of clinical
1969).
replication in that it subsumes a number of behavioral or emotional problems
and is a major clinical entity. Lovaas, Koegel, Simmons, and Long (1973)
listed eight distinct problems that may contribute to the autistic syndrome: (1)
apparent sensory deficit, (2) severe affect isolation, (3) self-stimulating behav-
ior, (4) mutism, (5) echolalic speech, (6) deficits in receptive speech, (7) deficits
in social and self-help behaviors, and (8) self-injurious behavior. Step-by-
step, they developed and tested treatments for each of these behaviors, such
as self-destructive behavior (Lovaas & Simmons, 1969), language acquisition
(e.g.,Lovaas et al., 1966), and social and self-help skills (Lovaas, Freitas,
Nelson, & Whalen, 1967). These procedures were tested in separate direct
replication series on the initial group of children. The treatment package
constructed from these direct replication series was administered to subse-
quent children presenting a sufficient number of these behaviors to be labeled
autistic.
Lovaas et al. (1973) presented the results and follow-up data from the
initial clinical replication series for 13 children. Results were presented in
terms of response of the group as a whole, as well as of individual improve-
ment across the variety of behavioral and emotional problems. While these
data are complex, they can be summarized as follows. All children demon-
strated increases in appropriate behaviors and decreases in inappropriate
behaviors. There were marked differences in the amount of improvement. At
leastone child was returned to a normal school setting, while several children
improved very little and required continued institutionalization. In other
words, each child improved, but the change was not clinically dramatic for
several children.
Because clinical replication is similar to direct replication, it can be ana-
lyzed in a similar fashion, and conclusions can be made in two general areas.
First, the treatment package can be effective for behaviors subsumed under
the autistic syndrome. This conclusion is based on (1) the initial experimental
analysis of each component of the treatment package in the original direct
replication series (e.g., Lovaas & Simmons, 1969) and (2) the withdrawal and
reintroduction of this whole package in A-B-A-B fashion in several children
(Lovaas et al., 1973). Second, replication of this finding across all subjects
indicates that the data are reliable and not due to idiosyncracies in one child.
It does not follow, however, that generality across children was established.
As in example 3 in the section on direct replication (10.2), the results were
clear and clinically significant for several children, but the results were also
weak and clinically unimportant for several children. Thus the package has
only limited generality across clients, and the task remains to pinpoint dif-
ferences between children who improved and those who did not improve.
From these differences, possible causes for limitations on client generality
should emerge.
In fact, children in this series were quite heterogeneous. In many respects,
this was due to an inherent difficulty in clinical replication the vagueness
and unreliability of many diagnostic categories. As Lovaas et al. (1973)
pointed out, "... the delineation of 'autism' is one area that will demand
considerably more work. It has not been a particularly useful diagnosis. Few
people agree on when to apply it" (p. 156). It follows that heterogeneity of
clients will most likely be greater than in a direct replication series, where the
target behavior is well defined and clients can be matched more closely.
Thus the causes of failure in a series with mixed results are more difficult to
ascertain, due to the greater number of differences among individuals. Never-
theless, it is necessary to pinpoint these differences and begin the search for
intersubject variability. As Lovaas et al (1973) concluded:
Finally a major focus of future research should attempt more functional descrip-
tions of autistic children. As we have shown, the children responded in vastly
different ways to the treatment we gave them. We paid scant attention to
individual differences when we treated the first twenty children. In the future, we
will assess such individual differences, (p. 163)
In the meantime, child clinicians would do well to examine closely the

exemplary series by Lovaas and his associates to determine logical generaliza-
tion to children under their care.
Taking cues from this initial clinical replication series, the investigators in
this research group have since improved their treatment package, based on a
long-term analysis of individual differences, and hypothesized reasons for
failure or minimal success. Subsequent experimental analyses have isolated
procedures and strategies that seem to improve the training program as a
whole (e.g., Koegel & Schreibman, 1982; Schreibman, Koegel, Mills, &
Burke, in press). These innovations, with particular emphasis on parent
training, combined with new and more valid measures of overall change,
have made possible another more advanced clinical replication series cur-
rently under way.
Guidelines for clinical replication are similar to those for direct replication
when series are relatively small and contain four to six cUents. A detailed
discussion of series containing 20, 50, or even 100 clients was presented in
Barlow et al. (1983).
10.5. ADVANTAGES OF REPLICATION

OF SINGLE-CASE EXPERIMENTS
In view of the reluctance of clinical researchers to carry out the large-scale
replication studies required in traditional experimental design (Bergin &
Strupp, 1972), one might be puzzled by the seeming enthusiasm with which
investigators undertake replication efforts using single-case designs, as evi-
denced by the differential attention series and other less advanced series. A
quick examination of Table 10-2 demonstrates that there is probably little or
no savings in time or money when compared to the large-scale collaborative
factorial designs initially proposed by Bergin and Strupp (1972). No fewer
clients are involved and, in all likeHhood, more applied researchers and
settings are involved. Why, then, does this replication tactic succeed when
Bergin and Strupp concluded that the alternative could not be implemented?
In our view, there are four very important but rather simple reasons.
First y the effort is decentralized. Rather than in the type of large collabora-
tive factorial study necessary to determine generality of findings at a cost of
millions of dollars, the replication efforts are carried out in many settings
such that funding, when available, is dispersed. This, of course, is more
government or other funding sources, who are not reluctant to
practical for
award $10,000 to each of 100 investigators but would be quite reluctant to
award $100,000 to one group of investigators. Often, of course, these small
studies involving three or four subjects are unfunded. Also, rather than
administering a large collaborative study from a central location where all
scientists or therapists are to carry out prescribed duties, each scientist

administers his or her own replication effort based on his or her ideas and
views of previous findings (see Barlow et al., 1983). What is lost here is some
efficiency, since there is no guarantee that the next obvious step in the
replication series will be carried out at the logical time. What is gained is the
freedom and creativity of individual scientists to attack the problem in their
own ways.
Second, systematic replication will continue because the professional con-
tingencies are favorable to its The professional contingencies
success. in this
case are publications and the accompanying professional recognition. Initial
efforts in a series experimentally demonstrating success of a technique on a

single case are publishable. Direct replications are pubHshable. Systematic
replications are publishable each time the procedure is successful in a dif-
ferent setting or with a different behavior disorder or whatever. Finally, after
a procedure has been proven effective, failures or exceptions to the success
are publishable. It is a well-established principle in psychology that intermit-
tent reinforcement, preferably on a short-variable interval schedule, is more
effective in maintaining behavior (in this case the replication series) than the
schedule arrangement for a large group study, where years may pass before
publishable data are available.
Third, the experimental analysis of the single-case is close to the clinic. As
noted in chapter 1, thisapproach tends to merge the role of scientist and
practitioner. Many an important series has started only after the clinician
confronted an interesting case. Subsequently, measures were developed, and
an experimental analysis of the treatment was performed (Mills et al., 1973).
As a result, the data increase one's understanding of the problem, but the
client also receives and benefits from treatment. If one plans to treat the
patient, it is an easy enough matter to develop measures and perform the
necesssary experimental analyses. The recent book mentioned above (Barlow
et al., 1983) was designed to explore this potential in our full-time practi-
tioners by demonstrating how they can incorporate these principles into their
practices and thereby participate in the research process. This ability to work
with ease within the clinical setting, more than any other fact, may ensure the
future of meaningful replication efforts.
Finally, as noted above, the results of the series are cumulative, and each
new replicative effort has some immediate payoff for the practicing clinician.
As this is the ultimate goal of the applied researcher, it is far more satisfactory
than participating in a multiyear collaborative study where knowledge or
benefit to the clinician is a distant goal.
Nevertheless, the advancement of a systematic replication series is a long
and arduous road full of pitfalls and dead ends. In the face of the immediate
demands on clinicians and behavior change agents to provide services to
society, it is tempting to "grab the glimmer of hope" provided by treatments
that prove successful in preliminary reports or case studies. That these hopes
have been repeatedly dashed as therapeutic techniques and schools of therapy
have come and gone supplies the most convincing evidence that the slow but
inexorable process of the scientific method is the only way to meaningful
advancement in our knowledge. Although we are a long way from the
sophistication of the physical sciences, the single case experimental design
with adequate replication may provide us with the methodology necessary to
overcome the complex problems of human behavior disorders.
Hiawatha Designs an Experiment
Maurice G. Kendall
(Originally published in The American Statistician, Dec. 1959, Vol. 13,
No. 5. Reprinted by Permission).
Hiawatha, mighty hunter. This, they said, was rather

He could shoot ten arrows upwards doubtful.
Shoot them with such strength and Anyway, it didn't matter
swiftness What resulted in the long run;
That the last had left the bowstring Either he must hit the target
Ere the first to earth descended. Much more often than at present
This was commonly regarded Or himself would have to pay for
As a feat of skill and cunning. All the arrows that he wasted.
One or two sarcastic spirits Hiawatha, in a temper.

Pointed out to him, however. Quoted parts of R. A. Fisher
That might be much more useful
it Quoted Yates and quoted Finney
If he sometimes hit the target. Quoted yards of Oscar Kempthorne
Why not shoot a little straighter Quoted reams of Cox and Cochran
And employ a smaller sample? Quoted Anderson and Bancroft
Practically in extenso
Hiawatha, who at college Trying to impress upon them
Majored in applied statistics. That what actually mattered
Consequently felt entitled Was to estimate the error.
To instruct his fellow men on
Any subject whatsoever. One or two of them admitted
Waxed exceedingly indignant Such a thing might have its uses.
Talked about the law of error. Still, they said, he might do better
Talked about truncated normals, If he shot a little straighter.

Talked of loss of information,
Talked about his lack of bias, Hiawatha, to convince them.
Pointed out that in the long run Organized a shooting contest
Independent observations Laid out in the proper manner
Even though they missed the target By experimental methods
Had an average point of impact Recommended in the textbooks
Very near the spot he aimed at (Mainly used for tasting tea, but
(With the possible exception Sometimes used in other cases)
Of a set of measure zero). Randomized his shooting order
372
Hiawatha Designs an Experiment 373
In factorial arrangements Or from Hiawatha's

Used the theory of Galois (This last point, one should
Fields of ideal polynomials, acknowledge
Got a nicely balanced layout
Might have been much more
And successfully confounded convincing
Second-order interactions. If he hadn't been compelled to
Estimate his own component
All the other tribal marksmen From experimental plots in
Ignorant, benighted creatures, Which the values all were missing.
Of experimental set-ups Still, they didn't understand it
Spent their time of preparation So they couldn't raise objections.

Putting in a lot of practice This is what so often happens
Merely shooting at a target. With analyses of variance.)
Thus it happened in the contest All the same, his fellow tribesmen
That their scores were most Ignorant, benighted heathens.
impressive Took away his bow and arrows.
With one notable exception Said that though my Hiawatha
This (I hate to have to say it) Was a brilliant statistician
Was the score of Hiawatha, He was useless as a bowman.
Who, as usual, shot his arrows As for variance components,
Shot them with great strength and Several of the more outspoken
swiftness Made primeval observations
Managing to be unbiased Hurtful to the finer feelings
Not, however, with his salvo Even of a statistician.
Managing to hit the target.
There, they said to Hiawatha In a corner of the forest
That is what we all expected. Dwells alone my Hiawatha
Permanently cogitating
Hiawatha, nothing daunted. On the normal law of error.
Called for pen and called for paper Wondering in idle moments
Did analyses of variance Whether an increased precision
Finally produced the figures Might perhaps be rather better.
Showing, beyond peradventure. Even at the risk of bias.
Everybody else was biased If thereby one, now and then,
And the variance components could
Did not differ from each other Register upon the target.
References
Abel, G. G., Blanchard, E. B., Barlow, D. H., & Flanagan, B. (1975, December). A controlled
behavioral treatment of a sadistic rapist. Paper presented at the meeting of the Association for
Advancement of Behavior Therapy, San Francisco.
Agras, W. S. (1975). Behavior modification in the general hospital psychiatric unit. In H.
Leitenberg (Ed.), Handbook of behavior modification (pp. 547-565). Englewood Cliffs, NJ:
Prentice-Hall.
Agras, W. S., Barlow, D. H., Chapin, H. N., Abel, G. G., & Leitenberg, H. (1974). Behavior
modification of anorexia nervosa. Archives of General Psychiatry, 30, 279-286.
Agras, W. S., Kazdin, A. E., & Wilson, G. T. (1979). Behavior Thearpy: Toward an applied
clinical science. San Francisco: W. H. Freeman.
Agras, W. S., Leitenberg, H., & Barlow, D. W. (1968). Social reinforcement in the modification
of agoraphobia. Archives of General Psychiatry, 19, Ali-All.
Agras, W. S., Leitenberg, H., Barlow, D. H., Curtis, N. A., Edwards, J. A., & Wright, D. E.
(1971). Relaxation in systematic desensitization. Archivesof General Psychiatry, 25, 511-514.
Agras, W. S., Leitenberg, H., Barlow, D. H., & Thomson, L. E. (1969). Instructions and
reinforcement in the modification of neurotic behavior. American Journal of Psychiatry, 125,
1435-1439.
Alford, G. S., Blanchard, E. B., & Buckley, M. (1972). Treatment of hysterical vomiting by
modification of social contingencies: A case study. Journal of Behavior Therapy and Experi-
mental Psychiatry, 3, 209-212.
Alford, G. S., Webster, J. S., & Sanders, S. H. (1980). Covert aversion of two interrelated deviant
sexual practices: Obscene phone calling and exhibitionism. A single case analysis. Behavior
Therapy. 11. 15-25.
Allen, K. E., & Harris, F. R. (1966). Elimination of a child's excessive scratching by training the
mother in reinforcement procedures. Behaviour Research and Therapy, 4, 79-84.
Allen, K. E., Hart, B. M., Buell, J. S., Harris, F R., & Wolf, M. M. (1964). Effects of social
reinforcement on isolate behavior of a nursery school child. Child Development, 35, 511-518.
Allen, K. E., Henke, L. B., Harris, F R., Baer, D. M., & Reynolds, N. J. (1%7). Control of
hyperactivity by social reinforcement of attending behavior. Journal of Educational Psychol-
ogy, 58, 231-237.
Allison, M. G., & Ayllon, T. (1980). Behavioral coaching in the development of skills in football,
gymnastics, and tennis. Journal of Applied Behavior Analysis, 13, 297-314.
Allport, G. D. (1961). Pattern and growth in personality. New York: Holt, Rinehart and
Winston.
Allport, G. D. (1962). The general and the unique in psychological science. Journal of Personal-
ity, 30, 405-422.
Altman, J. (1974). Observational study of behavior: Sampling methods. Behaviour. 49, 227-267.
American Psychological Association. (1973). Ethical principles in the conduct of research with
human participants. Washington, DC: Author.
Anderson, R. L. (1942). Distribution of the serial correlation coefficient. Annab of Mathematical
Statistics, 13, 1-13.
374
References 375
Anderson, R. L. (1971). The statistical analysis of time series. New York: Wiley.
Arrington, R. E. (1939). Time-sampling studies of child behavior. Psychological Monography, 51
( ).
Arrington, R. E. (1943). Time sampling in studies of social behavior: A critical review of

techniques and results with research suggestions. Psychological Bulletin, 40, 81-124.
Ashem, R. (1963). The treatment of a disaster phobia by systematic desensitization. Behaviour
Research and Therapy 1, 81-84.
AtiquUah, M. (1967). On the robustness of analysis of variance. Bulletin of the Institute of
Statistical Research and Training, 7, 77-81.
Ault, M. E., Peterson, R. E, & Bijou, S. W. (1968). The management of contingencies of
reinforcement to enhance study behavior in a small group of young children. Unpublished
manuscript.
Ayllon, T. (1961). Intensive treatment of psychotic behavior by stimulus satiation and food
reinforcement. Behaviour Research and Therapy, 1, 53-61.
Ayllon, T, & Azrin, N. H. (1965). The measurement and reinforcement of behavior of psy-
chotics. Journal of the Experimental Analysis of Behavior, 8, 357-383.
Ayllon, T, & Azrin, N. H. (1968). The token economy: A motivational system for therapy and
rehabilitation. New York: Appleton-Century-Crofts.
Ayllon, T, & Haughton, E. (1964). Modification of symptomatic verbal behavior of mental
patients. Behaviour Research and Therapy, 2, 87-91.
Ayllon, T, & Michael, J. (1959). The psychiatrist nurse as a behavioral engineer. Journal of the
Experimental Analysis of Behavior, 2, 323-334.
Azrin, N. H., Holz, W., Ulrich, R., & Goldiamond, I. (1961). The control of the content of
conversation through reinforcement. Journal of the Experimental Analysis of Behavior, 4,
25-30.
Baer, D. M. (1971). Behavior modification: You shouldn't. In E. Ramp & B. L. Hopkins (Eds.),
A new direction for education: Behavior analysis. Lawrence, KS: Lawrence University Press.
Baer, D. M. (1977a). "Perhaps it would be better not to know everything." Journal of Applied
Behavior Analysis, 10, 167-172.
Baer, D. M. (1977b). Reviewer's comment: Just because it's reliable doesn't mean that you can use
it. Journal of Applied Behavior Analysis, 10, 117-119.
Baer, D. M., & Guess, D. (1971). Receptive training of adjectival inflections in mental retardates.
Journal of Applied Behavior Analysis, 4, 129-139.
Baer, D.M., Wolf, M. M., & Risley, T R. (1968). Some current dimensions of applied behavior
analysis. Journal of Applied Behavior Analysis, 1, 91-97.
Bailey, J. S., Wolf, M. M., & PhilHps, E. L. (1970). Home-based reinforcement and the
modification of pre-delinquents' classroom behavior. Journal of Applied Behavior Analysis, 3,
223-233.
Bakeman, R. (19*78). Untangling streams of behavior: Sequential analysis of observational data.
In G. P. Sackett (Ed.), Observing behavior: Vol. 2. Data collection and analysis methods (pp.
63-78). Baltimore: University Park Press.
Ban, T. (1969). Psychopharmacology. Baltimore: Williams & Wilkins.
Bandura, A. (1969). Principles of behavior modification. New York: Holt, Rinehart & Winston.
Barker, R. G., & Wright, H. E
Midwest and its children: The psychological ecology of an
(1955).
American town. New York: Harper & Row.
Barlow, D. H. (1974). The treatment of sexual deviation: Towards a comprehensive behavioral
approach. In K. S. Calhoun, H. E. Adams, & K. M. Mitchell (Eds.), Innovative treatment
methods in psychopathology. New York: John Wiley & Sons, Inc., 1974.
Barlow, D. H. (1980). Behavior therapy: The next decade. Behavior Therapy, 11, 315-328.
Barlow, D. H. (Ed.). (1981). Behavioral assessment of adult disorders. New York: Guilford
Press.
Barlow, D. H., Agras, W. S., Leitenberg, H., Callahan, E. J., & Moore, R. C. (1972). The
contributions of therapeutic instructions to covert sensitization. Behaviour Research and
Therapy, 70,411-415.
Barlow, D. H., Becker, R., Leitenberg, H., & Agras, W. S. (1970). A mechanical strain gauge for
recording penile circumference change. Journal of Applied Behavior Analysis, 3, 73-76.
Barlow, D. H., Blanchard, E. B., Hayes, S. C, & Epstein, L. H. (1977). Single case designs and
biofeedback experimentation. Biofeedback and Self-Regulation, 2, 211-236.
Barlow, D. H., & Hayes, S. C. (1979). Alternating treatments design: One strategy for comparing
the effects of two treatments in a single subject. Journal of Applied Behavior Analysis, 12,
199-210.
Barlow, D. H., Hayes, S. C, & Nelson, R. O. (1983). The scientist-practitioner: Research and
accountability in clinical and educational settings. Elmsford, New York: Pergamon Press.
Barlow, D. H., & Hersen, M. (1973). Single case experimental designs: Uses in applied clinical
research. Archives of General Psychiatry, 29, 319-325.
Barlow, D. H., Leitenberg, H., & Agras, W. S. (1969). Experimental control of sexual deviation
through manipulation of the noxious scene in covert sensitization. Journal of Abnormal
Psychology, 74, 596-601.
Barlow, D. H., Leitenberg, H., Agras, W. S., & Wincze, J. R (1969). The transfer gap in
systematic desensitization: An analogue study. Behaviour Research and Therapy, 7, 191-197.
Barlow, D. H., Mavissakalian, M., & Schofield, L. (1980). Patterns of desynchrony in agorapho-
bia: A preliminary report. Behaviour Research and Therapy, 18, 441-448.
Barmann, B. C, Katz, R. C, O'Brien, E, & Beauchamp, K. L. (1981). Treating irregular
enuresis in developmentally disabled persons: A study in the use of overcorrection. Behavior
Modification, 5, 336-346.
Barnes, K. E., Wooton, M., & Wood, S. (1972). The public health nurse as an effective therapist-
behavior modifier of preschool play behavior. Community Mental Health Journal, 8, 3-7.
Barrera, R. D., & Sulzer-Azaroff, B. (1983). An alternating treatment comparison or oral and
total communication training program with echolalic autistic children. Journal of Applied
Barrett, R. R, Matson, J. L., Shapiro, E. S., & Ollendick, T. H. (1981). A comparison of
punishment and DRO procedures for treating stereotypic behavior of mentally retarded
children. Applied Research in Mental Retardation, 2, 247-256.
Barrios, B. A., & Hartmann, D. P. (in press). Traditional assessment's contributions to behavioral
assessment: Concepts, issues, and methodologies. In. R. O. Nelson & S. C. Hayes (Eds.),
Conceptual foundations of behavioral assessment. New York: Guilford Press.
Barrios, B. A., Hartmann, D. P., & Shigetomi, C. (1981). Fears and anxieties in children. In E. J.
Mash & L. G. Terdal (Eds.), Behavioral assessment of childhood disorders (pp. 259-304). New
York: Guilford Press.
Barron, ., & Leary, T. (1955). Changes in psychoneurotic patients with and without psy-
chotherapy. Journal of Consulting Psychology, 19, 239-245.
Barton, E. S., Guess, D., Garcia, E., & Baer, D. M. (1970). Improvement of retardates' mealtime
behaviors by timeout procedures using multiple baseline techniques. Journal of Applied
Bates, P. The effectiveness of interpersonal skills training on the social acquisition of
(1980).
moderately and mildly retarded adults. Journal of Applied Behavior Analysis, 13, 237-248.
Baum, C. G., Forehand, R. L., & Zegiob, L. E. (1979). A review of observer reactivity in adult-
child interactions. Journal of Behavioral Assessment, 1, 167-178.
Beck, A. T, Rush, A. J., Shaw, B. J., & Emery, G. (1979). Cognitive therapy of depression. New
Beck, A. T, Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for
measuring depression. Archives of General Psychiatry, 4, 561-571.
References 377
Beck, S. J. (1953). The science of personality: Nomothetic or idiographic. Psychological Review,

60, 353-359.
Bellack, A. S., & Hersen, M. (1977). The use of self-report inventories in behavior assessment. In
J. D. Cone& R. P. Hawkins (Eds.), Behavior assessment: New direction in clinical psychology
(pp. 52-76). New York: Brunner/Mazel.

Bellack, A. S., Hersen, M., & Himmelhoch, J. M. (1981). Social skills training, pharma-
cotherapy and psychotherapy for unipolar depression. American Journal of Psychiatry, 138,
1562-1567.
Bellack, A. S., Hersen, M., & Turner, S. M. (1976). Generalization effects of social skills training
in chronic schizophrenics: An experimental analysis. Behaviour Research and Therapy, 14,
381-398.
Bellack, L., & Chassan, J. B. (1964). An approach to the evaluation of drug effects during
psychotherapy: A double-blind study of a single case. Journal of Nervous and Mental Disease,
139, 20-30.
Bergin, A. E. (1966). Some implications of psychotherapy research for therapeutic practice.
Journal of Abnormal Psychology, 71, 235-246.
Bergin, A. E., & Lambert, M. J. (1978). The evaluation of therapeutic outcomes. In S. L.
Garfield & A. E. Bergin (Eds.), Handbook of psychotherapy and behavior change: An
empirical analysis (2nd ed.), (pp. 139-191). New York: Wiley.
Bergin, A. E., & Strupp, H. H. (1970). New directions in' psychotherapy research. Journal of
Abnormal Psychology, 76, 13-26.
Bergin, A. E., & Strupp, H. H. (1972). Changing frontiers in the science of psychotherapy. New
York: Aldine.
Berk, R. A. (1979). Generalizabihty of behavioral obvservations: A classification of interobserver
agreement and interobserver reliability.American Journal of Mental Deficiency, 83, 460-472.
Berler, E. S., Gross, A. M., & Drabman, R. S. (1982). Social skills training with children:
Proceed with caution. Journal of Applied Behavior Analysis, 15, 41-53.
Bernard, C. (1957). An introduction to the study of experimental medicine. New York: Dover.
Bernard, M. E., Kratochwill, T R., & Keefauver, L. W. (1983). The effects of rational-emotive
therapy and self-instructional training on chronic hair pulling. Cognitive Therapy and Re-
search, 7, 273-280.
Bickman, L. (1976). Observational methods. In C. Selltiz, L. S. Wrightsman, & S. W. Cook
(Eds.), Research methods in social relations (pp. 251-290). New York: Holt, Rinehart and
Winston.
Bijou, S. W, Peterson, R. E, & AuU, M. E. (1968). A method to integrate descriptive and
experimental field studies at the level of data and empirical concepts. Journal of Applied
Bijou, S. W., Peterson, R. E, Harris, K R., Allen, K. E., & Johnston, M. S. (1969). Methodology
for experimental studies of young children in natural settings. Psychological Record, 19,
177-210.
Birkimer, J. C, & Brown, J. H. (1979). Back to basics: Percentage agreement measures are
adequate, but there are easier ways. Journal of Applied Behavior Analysis, 12, 535-543.
Birnbrauer, J. S., Peterson, C. P., & Solnick, J. V. (1974). The design and interpretation of studies
of single subjects. American Journal of Mental Deficiency, 79, 191-203.
Birney, R. C, & Teevan, R. C. (Eds.). (1965). Reinforcement. Princeton, NJ: Van Nostrand.
Bittle, R., & Hake, D. F. (1977). A multielement design model for component analysis and cross-
setting assessment of a treatment package. Behavior Therapy, 8, 906-914.
Blanchard, E. B. (1981). Behavioral assessment of psychophysiological disorders. In D. H.
Barlow (Ed.), Behavioral assessment of adult disorders (pp. 239-269). New York: Guilford
Press.
Blough, P. M. (1983). Local contrast in multiple schedules: The effect of stimulus discriminability.
Journal of the Experimental Analysis of Behavior, 39, 427-437. ,
Boer, A. P. (1968). Application of a single recording system to the analysis of free-play behavior
in autistic children. Journal of Applied Behavior Analysis, 1, 335-340.

Boice, R. (1983). Observational skills. Psychological Bulletin, 93, 3-29.
Bolger, H. (1965). The case study method. In B B. Wolman (Ed.), Handbook of clinical
psychology (pp. 28-39). New York: McGraw-Hill.
Boring, E. G. (1950). A history of experimental psychology. New York: Appleton-Century-
Crofts.
Bornstein, M. R., Beilack, A. S., & Hersen, M. (1977). Social-skills training for unassertive
children: A multiple-baseUne analysis. Journal of Applied Behavior Analysis, 10, 183-195.
Bornstein, M. R., Beilack, A. S., & Hersen, M. (1980). Social skills training for highly aggressive
children: IVeatment in an inpatient psychiatric setting. Behavior Modification, 4, 173-186.
Bornstein, P. H., Bridgewater, C. A., Hickey, J. S., & Sweeney, T. M. (1980). Characteristics and
trends in behavioral assessment: An archival analysis. Behavioral Assessment, 2, 125-133.
Bornstein, P H., Hamilton, S. B., Carmody, T. B., Rychtarik, R. G., &, Veraldi, D. M. (1977).
Reliability enhancement: Increasing the accuracy of self-report thjough mediation-based pro-
cedures. Cognitive Therapy and Research, 1, 85-98.
Bornstein, P. H., & Rychtarik, R. G. (1983). Consumer satisfaction in aduh behavior therapy:
Procedures, problems, and future perspective. 5e/iav/or Therapy, 14, 191-208.
Bowdlear, C. M. (1955). Dynamics of idiopathic epilepsy as studied in one case. Unpublished
doctoral dissertation. Case Western Reserve University, Cleveland, Ohio.
Box, G. E. P., & Jenkins, G. M. (1970). Time series analysis: Forecasting and control. San
Francisco: Holden-Day.
Box, G. E. P., & Tiao, G. C. (1965). A change in level of non-stationary time series. Biometrika,
52, 181-192.
Boykin, R. A., & Nelson, R. O. (1981). The effects of instruction and calculation procedures on
observers' accuracy, agreement, and calculation correctness. Journal of Applied Behavior
Analysis, 14, 479-489.
Bradley, L. A., & Prokop, C. K. (1982). Research methods in contemporary medical psychology.
In P. C. Kendall & J. N. Butcher (Eds.), Handbook of research methods in clinical psychology
(pp. 591-649). New York: Wiley
Brady, J. P., & Lind, D. L. (1961). Experimental analysis of hysterical blindness. Archives of
General Psychiatry, 4, 331-339.
Brawley, E. R., Harris, F. R., Allen, K. E., Fleming, R. S., & Peterson, R. E (1969). Behavior
modification of an autistic child. Behavioral Science, 14, 87-97.
Breuer, J., & Freud, S. (1957). Studies on hysteria. New York: Basic Books.
Breuning, S. E., O'Neill, M. J., & Ferguson, D. G. (1980). Comparison of psychotropic drugs,
response cost, and psychotropic drug plus response cost procedures for controlling institution-
alized mentally retarded persons. Applied Research in Mental Retardation, 1, 253-268.
Brill, A. A. (1909). Selected papers on hysteria and other psychoneuroses: Sigmund Freud.
Nervous and Mental Disease Monograph Series, 4.
Broden, M., Bruce, C, Mitchell, M. A., Carter, V, & Hall, R. V. (1970). Effects of teacher
attention on attending behavior of two boys at adjacent desks. Journal of Applied Behavior
Analysis, 3, 205-211.
Broden, M., Hall, R. V, Dunlap, A., & Clark, R. (1970). Effects of teacher attention and a token
reinforcement system in a junior high school special education class. Exceptional Children, 36,
341-349.
Brookshire, R. H. (1970). Control of "involuntary" crying behavior emitted by a multiple
sclerosis patient. Journal of Community Disorders, 1, 386-390.
Browning, R. M. (1967). A same-subject design for simultaneous comparison of three reinforce-
ment contingencies. Behaviour Research and Therapy, 5, 237-243.
.
References 379
Browning, R. M., & Stover, D. O. (1971). Behavior modification in child treatment: An

experimental and clinical approach. Chicago: Aldine.
Brunswick, E. (1956). Perception and the representative design of psychological experiments.
Berkeley: University of California Press.
Bryant, L. E., & Budd, K. S. (1982). Self-instructional training to increase independent work
performance in preschoolers. Journal of Applied Behavior Analysis, 15, 259-271.
Budd, K. S., Green, D. R., & Baer, D. M. (1976). An analysis of multiple misplaced parental
social contingencies. Journal of Applied Behavior Analysis, 9, 459-470.
Buell, J. S., Stoddard, P., Harris, E R., & Baer, D. M. (1968). Collateral social development
accompanying reinforcement of outdoor play in a preschool child. Journal of Applied Behav-
ior Analysis, 1, 167-173.
Burgio, L. D., Whitman, T. L., & Johnson, M. R. (1980). A self-instructional package for
increasing attending behavior in educable mentally retarded children. Journal of Applied
Buys, C. J. (1971). Effects of teacher reinforcement on classroom behaviors and attitudes.
Dissertation Abstracts International, 31, 4884A 1-4885 A.
Campbell, D. T. (1969). Reforms as experiments. American Psychologist, 24, 409-429.
Campbell, D. T, & Fiske, D. W. (1959). Convergent and discriminant validation by the multi-
trait-multimethod matrix. Psychological Bulletin, 56, 81-105.
Campbell, D. T., & Stanley, J. C. (1963). Experimental and quasi-experimental designs for
research. In D. T. Campbell and J. C. Stanley, Handbook of Research on Teaching. Chicago:
Rand McNally.
Campbell, D. T., & Stanley, J. C. (1966). Experimental and quasi-experimental designs for
research. Chicago: Rand McNally.
Carey, R. C, & Bucher, B. (1983). Positive practice overcorrection: The effects of duration of
positive practice on acquisition and response relation. Journal of Applied Behavior Analysis,
16, 101-111.
Carlson, C. S., Arnold, C. R., Becker, W. C, & Madsen, C. H. (1968). The elimination of
tantrum behavior of a child in an elementary classroom. Behaviour Research and Therapy, 5,
117-119.
Carver, R. P. (1974). Two dimensions of tests: Psychometric and edumetric. American Psycholo-
gist, 29, 512-518.
Catania, A. C. (Ed.), (1968). Contemporary research in operant behavior. Glenview, IL: Scott,
Foresman.
Chaplin, J. P. (1975). Dictionary of psychology (Rev. Ed.). New York: Dell Publishing.
Chaplin, J. P., & Kraweic, T. S. (1960). Systems and theories of psychology. New York: Holt,
Rinehart and Winston.
Chassan, J. B. (1960). Statistical inference and the single case in clinical design. Psychiatry, 23,
173-184.
Chassan, J. B. (1962). Probability processes in psychoanalytic psychiatry. In J. Scher (Ed.),
Theories of the mind (pp. 598-618). New York: Free Press of Glencoe.
Chassan, J. B. (1967). Research design in clinical psychology and psychiatry. New York: Apple-
ton-Century-Crof ts
Chassan, J. B. (1979). Research design in clinical psychology and psychiatry (2nded.) New York:
Irvington.
Ciminero, A. R., Calhoun, K. S., & Adams, H. E. (Eds.), (1977). Handbook of behavioral
assessment. New York: Wiley.
Coates, T. J., & Thoresen, C. E. (1981). Sleep disturbances in children and adolescents. In E. J.
Mash & L. G. Terdal (Eds.), Behavioral assessment of childhood disorders (pp. 639-678). New
York: Guilford.
Cohen, D. C. (1977). Comparison of self-report and overt-behavioral procedures for assessing
acrophobia. Behavior Therapy, 8, 17-23.
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological
Measurement, 20, 37-46.
Cohen, J. (1968). Weighted kappa: Nominal scale agreement with provisions for scale disagree-
ment or partial credit. Psychological Bulletin, 70, 313-220.
Cohen, L. H. (1976). Clinicians' utilization of research findings. JSAS Catalog of Selected
Documents in Psychology, 6, 116.
Cohen, L. H. (1979). The research readership and information source reliance of clinical
psychologists. Professional Psychology, JO, 780-786.
Coleman, R. A. (1970). Conditioning techniques applicable to elementary school classrooms.
Cone, J. D. (1977). The relevance of reliability and validity for behavior assessment. Behavior
Therapy, 8, 411-426.
Cone, J. D. (1979). Confounded comparisons in triple response mode assessment research.
Behavioral Assessment, I, 85-95.
Cone, J. D. (1982). Validity of direct observation assessment procedures. In D. P. Hartmann
(Ed.), Using observers to study behavior: New directions for methodology of social and
behavioral science (pp. 67-79). San Francisco: Jossey-Bass.
Cone, J. D., & Foster, S. L. (1982). Direct observations in clinical psychology. In P. C. Kendall &
J. N. Butcher (Eds.), Handbook of research methods in clinical psychology (pp. 311-354).
New York: Wiley.
Cone, J. D., & Hawkins, R. P. (Eds.). (1977). Behavior assessment: New directions in clinical
psychology. New York: Brunner/Mazel.
Conger, A. J. (1980). Integration and generalization of kappas for multiple raters. Psychological
Bulletin, 88, 322-328.
Conger, J. C. (1970). The treatment of encopresis by the management of social consequences.
Behavior Therapy, 1, 386-390.
Conover, W. J. (1971). Practical nonparametric statistics. New York: Wiley.
Conrin, J., Pennypacker, H. S., Johnston, J. M., & Rast, J. (1982). Differential reinforcement of
other behaviors to treat chronic rumination of mental retardates. Journal of Behavior Therapy
and Experimental Psychiatry, 13, 325-329.
Cook, T. D., & Campbell, D, T. (Eds.). (1979). Quasi-experimentation: Design and analysis issues
for field settings. Chicago: Rand McNally.
Cormier, W. H, (1969). Effects of teacher random and contingent social reinforcement on the
classroom behavior of adolescents. Dissertation Abstracts International, 31, 1615A-1616A.
Corte, H. E., Wolf, M. M., & Locke, B. J. (1971). A comparison of procedures for eliminating
self-injurious behavior of retarded adolescents. Journal of Applied Behavior Analysis, 4,
201-215.
Cossairt, A., Hall, R. V, & Hopkins, B. L. (1973). The effects of experimenters' instructions,
feedback, and praise on teacher praise and student attending behavior. Journal of Applied
Creer, T. L., Chai, H., & Hoffman, A. (1977). A single application of an aversive stimulus to
eliminate chronic cough. Journal of Behavior Therapy and Experimental Psychiatry, 8,
107-109.
Cronbach, L. J. of psychological testing (3rd ed.). New York: Harper & Row.
(1970). Essentials
Cronbach, L. R. L. Thorndike (Ed.), Educational measurement (pp.
J. (1971). Test validation. In
443-507). Washington: American Council on Education.

Cronbach, L. J., Gleser, G. C, Nanda, H., & Rajaratnam, N. (1972). The dependability of
behavioral measurements: Theory of generalizability for scores and profiles. New York: Wiley.
Cuvo, A. J., Leaf, R. B., & Borakove, L. S. (1978). Teaching janitorial skills to the mentally
retarded: Acquisition, generalization, and maintenance. Journal of Applied Behavior Analy-
sis, U, 345-355.
References 381
Cuvo, A. J., & Riva, M. T. (1980). Generalization and transfer between comprehension and
production: A comparison of retarded and nonretarded persons. Journal of Applied Behavior
Analysis, /5, 215-231.
Dalton, K. (1959). Menstruation and acute psychiatric illness. British Medical Journal, 1,
148-149.
Dalton, K. (1960a). Menstruation and accidents. British Medical Journal, 2, 1425-1426.
Dalton, K. (1960b). School girls' behavior and menstruation. British Medical Journal, 2,
1647-1649.
Dalton, K. (1961). Menstruation and crime. British Medical Journal, 2, 1752-1753.
Davidson, P. O., & Costello, C. G. (1969). N= J: Experimental studies of single cases. New York:
Van Nostrand Reinhold.
Davis, K. v., Sprague, R. L., & Werry, J. S. (1969). Stereotyped behavior and activity level in
severe retardates: The of drugs. American Journal of Mental Deficiency, 73, 721-727.
effect
Davis, V. J., PoHng, A. D., Wysocki, T, & Breuning, S. E. (1981). Effects of Phenytoin
withdrawal on matching to sample and workshop performance of mentally retarded persons.
Journal of Nervous and Mental Disease, 169, 718-725.
Davison, G. C. (1965). The training of undergraduates as social reinforcers for autistic children.
In L. P. UUmann & L. Krasner (Eds.), Case studies in behavior modification (pp. 146-148).
New York: Holt, Rinehart and Winston.
DeProspero, A., & Cohen, S. (1979). Inconsistent visual analysis of intrasubject data. Journal of
Applied Behavior Analysis, 12, 573-579.
Doke, L. A. (1976). Assessment of children's behavioral deficits. In M. Hersen & A. S. Bellack
(Eds.), Behavioral assessment (pp. 493-536). Elmsford, New York: Pergamon Press.
Doke, L. A., fe Risley, T. R. (1972). The organization of day-care environments: Required vs
optional activities. Journal of Applied Behavior Analysis, 5, 405-420.
Dollard, J., Doob, L. W., Miller, N. E., Mowrer, O. H., & Sears, R. R. (1939). Frustration and
aggression. New Haven: Yale University Press.
Domash, M. A., Schnelle, J. E, Stomatt, E. L., Carr, A. E, Larson, L. D., Kirchner, R. E., &
Risley, T. R. (1980). Police and prosecution systems: An evaluation of a police criminal case
preparation program. Journal of Applied Behavior Analysis, 13, 397-406.

Drabman, R. S., Hammer, D., & Rosenbaum, M.S. (1979). Assessing generalization in behavior
modification with children: The generalization map. Behavioral Assessment, 1, 203-219.
Dukes, W. F (1965). N= 1. Psychological Bulletin, 64, 74-79.
duMas, F. M. (1955). Science and the single case. Psychological Reports, 1, 65-75.
Dunlap, G., & Koegel, R. L. (1980). Motivating autistic children through stimulus variation.
Dunlap, K. (1932). Habits: Their making and unmaking. New York: Liverright.
Dyer, K., Christian, W. P., & Luce, S. C. (1982). The role of response delay in improving the
discrimination performance of autistic children. Journal of Applied Behavior Analysis, 15,
231-240.
Edelberg, R. (1972). Electrical activity of the skin. In N. S. Greenfield & R. A. Sternbach (Eds.),
Handbook of psychophysiology (pp. 367-418). New York: Holt, Rinehart and Winston.
Edgington, E. S. (1966). Statistical inference and nonrandom samples. Psychological Bulletin,
66, 485-487.
Edgington, E. S. (1967). Statistical inference from N= 1 experiments. Journal of Psychology, 65,
195-199.
Edgington, E. S. (1969). Statistical inference: The distribution-free approach. New York: Mc-
Graw-Hill.
Edgington, E. S. (1972). N=l experiments: Hypothesis testing. Canadian Psychologist, 13,
121-135.
Edgington, E. S. (1980a). Randomization tests. New York: Marcel Dekker.
Edgington, E. S. (1980b). Validity of randomization tests for one-subject experiments. Journal of
Educational Statistics, 5, 235-251.

Edgington, E. S. (1982). Nonparametric tests for single-subject multiple schedule experiments.
Behavioral Assessment, 4, 83-91.
Edgington, E. S. (1983). Response-guided experimentation. Contemporary Psychology, 28,
64-65.
Edgington, E. S. (1984). Statistics and single case analysis. In M. Hersen, R. M. Eisler, & P. M.
Monti (Eds.). Progress in Behavior Modification (Vol. 16). New York: Academic Press.
Edwards, A. L. (1968). Experimental design in psychological research (3rd ed.). New York: Holt,
Rinehart and Winston.
Egel, A. L., Richman, G. S., & Koegel, R. L. (1981). Normal peer models and autistic children's
learning. Journal of Applied Behavior Analysis, 14, 3-12.
Eisler, R. M., & Hersen, M. (August, 1973). The A-B design: Effects of token economy on
behavioral and subjective measures in neurotic depression. Paper presented at the meeting of
the American Psychological Association, Montreal.
Eisler, R. M., Hersen, M., & Agras, W. S. (1973). Effects of videotape and instructional feedback
on nonverbal marital interaction: An analog study. Behavior Therapy, 4, 551-558.
Eisler, R. M., Miller, P. M,, & Hersen, M. (1973). Components of assertive behavior. Journal of
Clinical Psychology, 29, 295-299.

Elkin, T. E., Hersen, M., Eisler, R. M., & Williams, J. G. (1973), Modification of caloric intake
in anorexia nervosa: An experimental analysis. Psychological Reports, 32, 75-78.
Ellis, D. P. (1968). The design of a social structure to control aggression. Dissertation Abstracts,
29, 672A.
Emmelkamp, P. M. G. (1974). Self-observation versus flooding in the treatment of agoraphobia.
Behaviour Research and Therapy, 12, 229-237.
Emmelkamp, P. M. G. Phobic and obsessive-compulsive disorders: Theory, research and
(1982).
practice. New York: Plenum.
Emmelkamp, P. M. G., & Kwee, K. G. (1977). Obsessional ruminations: A comparison between
thought stopping and prolonged exposure in imagination. Behaviour Research and Therapy,
15, 441-444.
Epstein, L. H., Beck, S. J., Figueroa, J., Farkas, G., Kazdin, A. E., Daneman, D., & Becker, D.
(1981). The effects of targeting improvements in urine glucose on metabolic control in children
with insulin dependent diabetes. Journal of Applied Behavior Analysis, 14, 365-375.
Epstein, L. H., & Hersen, M. (1974). Behavioral control of hysterical gagging. Journal of
Epstein, L. H., Hersen, M., & Hemphill, D. P. (1974). Music feedback in the treatment of tension
headache: An experimental case study. Journal of Behavior Therapy and Experimental Psy-
chiatry, 5, 59-63.
Etzel, B. C, & Gerwitz, J. L. (1967). Experimental modifications of caretaker-maintained
highrate operant crying in a 6- and 20- week-old infant (Infans tyrannotearus): Extinction of
crying with reinforcement of eye contact and smiling. Journal of Experimental Child Psychol-
ogy, 5, 303-317.
Evans, I. M. (1983). Behavioral assessment. In C. E. Wallace (Ed.), Handbook of clinical
psychology: Vol. 1. Theory, research, and practice {pv^. 391-419). Homewood, IL: Dow Jones-
Irwin.
Evans, I. M., & Wilson, F. E. (1983). Behavioral assessment on decision making: A theoretical
analysis. In M. Rosenbaum, C. M. Franks, & Y. Jaffe (Eds.), Perspectives on behavior therapy
in the eighties (Vol. 9, (pp. 35-53). New York: Springer Publishing.
Eyberg, S. M., & Johnson, S. M. (1974). Multiple assessment of behavior modification with
families: Effects of contingency contracting and order of treated problems. Journal of Con-
sulting and Clinical Psychology, 42, 594-606.
Eysenck, H. J. (1952). The effects of psychotherapy: An evaluation. Journal of Consulting
References 383
Eysenck, H. J. (1965). The effects of psychotherapy. International Journal of Psychiatry, 1,
97-178.
Ezekiel, M., & Fox, K. A- (1959). Methods of correlation and regression analysis: Linear and
curvilinear New York: Wiley.
Fairbank, J. A., & Keane, T. M. (1982). Flooding for combat-related stress disorders: Assessment
of anxiety reduction across traumatic memories. Behavior Therapy, 13, 499-510.
Fisher, E. B. (1979). Overjustification effects in token economies. Journal of Applied Behavior
Analysis, 12, 407-415.
Fisher, R. A. (1925). On the mathematical foundations of the theory of statistics. In Cambridge
Phil. Society (Ed.), Theory of statistical estimation (Proceedings of the Cambridge Philosophi-
cal Society) England.
Fjellstedt, N., & Sulzer-Azaroff, B. (1973). Reducing the latency of a child's responding to
instructions by means of a token system. Journal of Applied Behavior Analysis, 6, 125-130.
Fleiss, J. H. (1975). Measuring agreement between two judges on the presence or absence of a
trait. Biometrics, 31, 651-659.
Foa, E. B. (1979). Failure in treating obsessive-compulsives. Behaviour Research and Therapy,
17, 169-175.
Foa, E. B., Grayson, J. B., Steketee, G. S., Doppelt, H. G., Tlirner, R. M., & Latimer, P. R.
(1983). Success and failure in the behavioral treatment of obsessive compulsives. Journal of
Consulting and Clinical Psychology, 51, 287-297.
Forehand, R. L. (Ed.). (1983). Mini-series on consumer satisfaction and behavior therapy.
Behavior Theraoy, 14, 189-246.
Forehand, R. L., & McMahon, R. J. (1981). Helping the noncompliant child: A clinician's guide
to parent training. New York: Guilford Press.
Frank, J. D. (1961). Persuasion and healing. Baltimore: Johns Hopkins University Press.
Freund, K., & Blanchard, R. (1981). Assessment of sexual dysfunction and deviation. In M.
Hersen & A. S. Bellack (Eds.), Behavioral assessment: A practical handbook (2nd ed., pp.
427-455). Elmsford, New York: Pergamon Press.
Frick, X, & Semmel, M.I. (1978). Observer agreement and reliability of classroom observational
measures. Review of Educational Research, 48, 157-184.
Feuerstein, M., & Adams, H. E. (1977). Cephalic vasomotor feedback in the modification of
migraine headache. Biofeedback and Self-Regulation, 3, 241-254.
Garfield, S. L., & Bergin, A. E. (Eds.). (1978). Handbook of psychotherapy and behavior
change: An empirical analysis (2nd ed.). New York: Wiley.
Geer, J. H. (1965). The development of a scale to measure fear. Behaviour Research and Therapy,
13, 45-53.
Gelfand, D. M., Gelfand, S., & Dobson, W R. (1967). Unprogrammed reinforcement of
patients' behavior in a mental hospital. Behaviour Research and Therapy, 5, 201-207.
Gelfand, D. M., & Hartmann, D. P (1975). Child behavior analysis and therapy Elmsford, N.Y.:
Pergamon Press.
Gelfand, D. M., & Hartmann, D. P (1984). Child behavior: Analysis and therapy (2nd ed.).
Elmsford, New York: Pergamon Press.
& Everett, P. B. (1982). Preserving the environment: New strategies
Geller, E. S., Winett, R. A.,
for behavior change. Elmsford, New York: Pergamon Press.

Gentile, J. R., Roden, A. H., & Klein, R. D. (1972). An analysis of variance model for the
intrasubject replication design. Journal of Applied Behavior Analysis, 5, 193-198.
Glass, G. S., Heninger, G. R., Lansky, M., & Talan, K. (1971), Psychiatric emergency related to
American Journal of Psychiatry, 128, 705-711.
the menstrual cycle.
Glass, G. v.,Peckham, P. D., & Sanders, J. R. (1972). Consequences of failure to meet
assumptions underlying the fixed effects analyses of variance and covariance. Review of
Educational Research, 42, 237-288.
Glass, G. v., Willson, V. L., & Gottman, J. M. (1974). Design and analysis of time-series
experiments. Boulder: Colorado Associated University Press.

Goetz, E. M., & Baer, D. M. (1973). Social control of form diversity and the emergence of new
forms in children's blockbuilding. Journal of Applied Behavior Analysis, 6, 209-217.
Goldfried, M. R., & D'Zurilla, T. J. (1969). A behavioral-analytic model for assessing compe-
tence. In C. D. Spielberger (Ed.), Current topics in clinical and community psychology (Vol. 1,
pp. 151-196). New York: Academic Press.

Goldfried, M. R., & Linehan, M. M. (1977), Basic issues in behavioral assessment. In A, R.
Ciminero, K. S. Calhoun, & H. E. Adams (Eds.), Handbook of behavioral assessment (pp.
15-46). New York: Wiley.
Goldstein, M, K. (1971). Behavior rate change in marriages: Training wives to modify husbands'
behavior. Dissertation Abstracts International, 32, 559A.
Goodlet, G. R., Goodlet,M. M., & Dredge, M. (1970). Modification of disruptive behavior of
two young children and follow-up one year later. Journal of School Psychology, 8, 60-63,
Goodman, L. A., & Gilman, A. (1975). The pharmacological basis of therapeutics. New York:
Macmillan.
Gorsuch, R. L. (1983), Three models for analyzing limited time-series (Nof 1) data. Behavioral
Assessment, 5, 141-154.
Gottman, J. M. (1973). N-of-one and N-of-two research in psychotherapy. Psychological Bul-
letin, 80, 93-105.
Gottman, J. M. (1979). Marital interaction: Experimental investigations. New York: Academic
Press.
Gottman, J. M. (1981). Time-series analysis: A comprehensive introduction for social
scientists. Cambridge: Cambridge University Press.
Gottman, J. M., & Glass, G. V. (1978). Analysis of interrupted time-series experiments. In T, R.
Kratochwill (Ed,), Single-subject research: Strategies for evaluating change (pp. 197-237). New
York: Academic Press.
Gottman, J. M., McFall, R. M., & Barnett, J. T (1969). Design and analysis of research using
time series. Psychological Bulletin, 72, 299-306.
Greenfield, N. A., & Sternbach, R. A. (Eds.). (1972). Handbook of psychophysiology. New
York: Holt, Rinehart and Winston.
Greenspoon, J. (1955). The reinforcing effect of two spoken sounds on the frequency of two
responses. American Journal of Psychology, 68, 409-416.
Greenwald, A. G. (1976). Within-subjects designs: To use or not to use? Psychological Bulletin,
1976, 83, 314-320.
Grinspoon, L., Ewalt, J., & Shader, R. (1967). Long term treatment of chronic schizophrenia.
International Journal of Psychiatry, 4, 116-128.
Hall, C, Sheldon- Wildgen, J., & Sherman, J. A. (1980). Teaching job interview skills to retarded
clients. Journal of Applied Behavior Analysis, 13, 433-442.
Hall, R. v., Axelrod, S Tyler, L., Grief, E,, Jones, E C, & Robertson, R, (1972). Modification
of behavior problems in the home with a parent as observer and experimenter. Journal of
Hall, R. v., & Broden, M, (1967). Behavior changes in brain-injured children through social
reinforcement. Journal of Experimental Child Psychology, 5, 463-479.
Hall, R. v., Cristler, C, Cranston, S. S,, & Tlicker, B, (1970). Teachers and parents as researchers
using multiple baseline designs. Journal of Applied Behavior Analysis, 3, 247-255.
Hall, R. v.. Fox, R., Willard, D., Goldsmith, L., Emerson, M., Owen, M., Davis, E, & Porcia,
E, (1971). The teacher as observer and experimenter in the modification of disputing and
talking-out behaviors. Journal of Applied Behavior Analysis, 4, 141-149.
Hall, R. v., Lund, D., & Jackson, D. (1968). Effects of teacher attention on study behavior.
Journal of Applied Behavior Analysis, I, 1-12.
Hall, R. v., Panyan, M., Rabon, D., & Broden, M. (1968). Instructing beginning teachers in
reinforcement procedures which improve classroom control. Journal of Applied Behavior
References 385
Analysis, J, 315-322.
Hallahan, D. P., Lloyd, J. W., Kneedler, R. D., & Marshall, K. J. (1982). A comparison of the
effects of self- versus teacher-assessment of on-task behavior. Behavior Therapy, 13, 715-723.
Halle, J. W., Baer, D. M., & Spradlin, J. E. (1981). Teachers' generahzed use of delay as a
stimulus control procedure to increase language use in handicapped children. Journal of
Harbert, T. L., Barlow, D. H., Hersen, M., & Austin, J. B. (1974). Measurement and modifica-
tion of incestuous behavior: A case study. Psychological Reports, 34, 79-86.
Harris, F. R., Johnston, M. K., Kelley, C. S., & Wolf, M. M. (1964). Effects of positive social
reinforcement on regressed crawling of a nursery school child. Journal of Educational Psychol-
ogy, 55, 35-41.
Hart, B. M., Allen, K. E., Buell, J. S., Harris, F. R., & Wolf, M. M. (1964). Effects of social
reinforcement on operant crying. Journal of Experimental Child Psychology, 1, 145-153.
Hart, B. M., Reynolds, N. J., Baer, D. M., Brawley, E. R., & Harris, F R. (1968). Effect of
contingent social reinforcement on the cooperative play of a preschool child. Journal of
Hartmann, D. P. (1974). Forcing square pegs into round holes: Some comments on "An analysis-
of-variance model for the intrasubject replication design." Journal of Applied Behavior
Analysis, 7, 635-638.
Hartmann, D. P. (1976). Some restrictions in the application of the Spearman-Brown prophecy
formula to observational data. Educational and Psychological Measurement, 36, 843-845.
Hartmann, D. P. (1977). Consideration in the choice of interobserver reliability estimates.
Hartmann, D. P. (1982). Assessing the dependability of observational data. In D. P. Hartmannn
Hartmann, D. P. (1983). Editorial. Behavioral Assessment, 5, 1-3.
Hartmann, D. P., & Gardner, W. (1979). On the not so recent invention of interobserver reliability
statitics: A commentary on two articles by Birkimer and Brown. Journal of Applied Behavior
Analysis, 12, 559-560.
Hartmann, D. P., & Gardner, W (1981). Considerations in assessing the reliability of observa-
tions. In E. E. Filsinger & R. A. Lewis (Eds.), Assessing marriage (pp. 184-196). Beverly Hills:
Sage.
Hartmann, D. P, Gottman, J. M., Jones, R. R., Gardner, W, Kazdin, A. E., & Vaught, R. S.
(1980). Interrupted time-series analysis and its application to behavioral data. Journal of
Hartmann, D. P., & Hall, R. V. (1976). The changing criterion design. Journal of Applied
Hartmann, D. P., Roper, B. L., & Bradford, D. C. (1979). Some relationships between behavioral
and traditional assessment. Journal of Behavioral Assessment, 1, 3-21.
Hartmann, D. R, Roper, B. L., & Gelfand, D. M. (1977). Evaluation of alternative modes of
child psychotherapy. In B. Lahen & A. Kazdin (Eds.), Advances in child clinical psychology
(Vol 1, pp. 1-46). New York: Plenum.
Hartmann, D. R, & Wood, D. D. (1982). Observation methods. In A. S. Bellack, M. Hersen, &
A. E. Kazdin (Eds.), International handbook of behavior modification and therapy (pp.
109-138). New York: Plenum.
Hasazi, J. E., & Hasazi, S. E. (1972). Effects of teacher attention on digit-reversal behavior in an
elementary school child. Journal of Applied Behavior Analysis, 5, 157-162.
Hawkins, R. P. (1975). Who decided that was the problem? Two stages of responsibility for
applied behavior analysis. In W S. Wood (Ed.), Issues in evaluating behavior modification (pp.
95-214). Champaign, IL: Research Press.
Hawkins, R. P. (1979). The functions of assessment: Implications for selection and development
of devices for assessing repertoires in clinical, educational, and other settings. Journal of
Hawkins, R. P. (1982). Developing a behavior code. In D. P. Hartmann (Ed.), Using observers to
study behavior: New directions for methodology of social and behavioral science (pp. 21-35).
San Francisco: Jossey-Bass.
Hawkins, R. P., Axelrod, S., & Hall, R. V. (1976). Teachers as behavior analysts: Precisely
monitoring student performance. In J. A. Brigham, R. P. Hawkins, J. Scott, & J. F.
McLaughlin (Eds.), Behavior analysis in education: Self-control and reading (pp. 274-2%).
Dubuque, lA: Kendall/Hunt.
Hawkins, R. P., & Dobes, R. W. (1977). Behavioral definitions in applied behavior analysis:
Explicit or implicit. In B. C. Etzel, J. M. LeBlanc, & D. M. Baer (Eds.), New directions in
behavioral research: Theory, methods, and applications. In honor of Sidney W. Bijou (pp.
167-188). Hillsdale, NJ: Erlbaum.
Hawkins, R. P., & Dotson, V. A. (1975). Reliability scores that delude: An Alice in Wonderland
trip through the misleading characteristics of interobserver agreement scores in interval record-
ing. In E. Ramp &

G. Semb (Eds.), Behavior analysis: Areas of research and application (pp.
359-376). Englewood CHffs, NJ: Prentice-Hall.
Hawkins, R. P., & Fabry, B. D. (1979). Applied behavior analysis and interobserver reliability: A
commentary on two articles by Birkimer and Brown. Journal of Applied Behavior Analysis,
12, 545-552.
Hawkins, R. P, Peterson, R. F, Schweid, E., & Bijou, S. W. (1966). Behavior therapy in the
home: Amelioration of problem parent-child relations with the parent in a therapeutic role.
Journal of Experimental Child Psychology, 4, 99-107.
Hay, L. R., Nelson, R. O., & Hay, W. M. (1980). Methodological problems in the use of
participation observers. Journal of Applied Behavior Analysis, 13, 501-504.
Hayes, S. C. (1981), Single case experimental design and empirical clinical practice. Journal of
Haynes, S. N. (1978). Principles of behavioral assessment. New York: Gardner Press.
Haynes, S. N., & Wilson, C. C. (1979). Behavioral assessment. San Francisco: Jossey-Bass.
Hendrickson, J. M., Strain, P. S., TVemblay, A., & Shores, R. E. (1982). Interactions of
behaviorally handicapped children: Functional effects of peer social interactions. Behavior
Herbert, E. W, & Baer, D. M. (1972). TVaining parents as behavior modifiers: Self-recording of
contingent attention. Journal of Applied Behavior Analysis, 5, 139-149.
Herbert, E. W, Pinkston, E. M., Hayden, M. L., Sajwaj, T. E., Pinkston, S., Cordua, G., &
Jackson, C. (1973). Adverse effects of differential parental attention. Journal of Applied
Herman, S. H., Barlow, D. H., & Agras, W. S. (1974a). An experimental analysis of classical
conditioning as a method of increasing heterosexual arousal in homosexuals. Behavior
Therapy, 5, 33-47.
Herman, S. H., Barlow, D. H., & Agras, W. S. (1974b). An experimental analysis of exposure to
"explicit" heterosexual stimuli as an effective variable in changing arousal patterns of homo-
sexuals. Behaviour Research and Therapy, 12, 335-345.
Herrnstein, R. J. (1970). On the law of effect. Journal of the Experimental Analysis of Behavior,
13, 243-266.
Hersen, M. (1973), Self-assessment of fear. Behavior Therapy, 4, 241-257.
Hersen, M. (1978). Do behavior therapists use self-report as major criteria? Behavioral Analysis
and Modification, 2, 328-334.
Hersen, M. (1981). Complex problems require complex solutions. Behavior Therapy, 12, 15-29.
Hersen, M, (1982). Single-case experimental designs. In A. S. Bellack, M. Hersen, & A. E.
Kazdin (Eds.), International handbook of behavior modification and therapy (pp. 167-201).
New York: Plenum.
.
References 387
Hersen, M., & Bellack, A. S. (1976). A multiple-baseline analysis of social-skills training in

chronic schizophrenics. Journal of Applied Behavior Analysis, 9, 239-245.
Hersen, M., & Bellack, A. S. (Eds.), (1981). Behavioral assessment: A practical handbook (2nd
ed.). Elmsford, New York: Pergamon Press.
Hersen, M., & Breuning, S. E. (Eds.), (in press). Pharmacological and behavioral treatment: An
integrated approach. New York: Wiley.
Hersen, M., Eisler, R. M., Alford, G. S., & Agras, W. S. (1973). Effects of token economy on
neurotic depression: An experimental analysis. Behavior Therapy, 4, 392-397.
Hersen, M., Eisler, R. M., & Miller, P. M. (1973). Development of assertive responses: Clinical,
measurement, and research considerations. Behaviour Research and Therapy, 11, 505-522.
Hersen, M., Gullick, E. L., Matherne, P M., & Harbert, T. L. (1972). Instructions and
reinforcement in the modification of a conversion reaction. Psychological Reports, 31,
719-722.
Hersen, M., Miller, P. M., & Eisler, R. M. (1973). Interactions between alcoholics and their wives:
A descriptive analysis of verbal and non-verbal behavior. Quarterly Journal of Studies on
Alcohol, 34, 516-520.
Hilgard, J. R. (1933). The effect of early and delayed practice on memory and motor perfor-
mances studies by the method of co-twin control. Genetic Psychology Monographs, 14,
493-567.
Hinson, J. M., & Malone, J. C, Jr. (1980). Local contrast and maintained generalization.
Journal of the Experimental Analysis of Behavior, 34, 263-272.
Hoch, P. H., & Zubin, J. (Eds.). (1964). The evaluation of psychiatric treatment. New York:
Grune & Stratton.
Hollands worth, J. G., Glazeski, R. C, & Dressel, M. E. (1978). Use of social skills training in the
treatment of extreme anxiety and deficit verbal skills in the job interview setting. Journal of
Hollenbeck, A. R. (1978). Problems of reliability in observational research. In G. P. Sackett
(Ed.), Observing behavior: Vol. 1. Data collection and analysis methods (pp. 79-98). Balti-
more: University Park Press.
Hollon, S. D., & Bemis, K. M. (1981). Self-report and the assessment of cognitive funcitons. In
M. Hersen & A. S. Bellack (Eds.), Behavioral assessment: A practical handbook (2nd ed.) (pp.
125-174). Elmsford, New York: Pergamon Press.
Holm, R. A. (1978). Techniques of recording observational data. In G. P. Sackett (Ed.),
Observing behavior: Vol. 2. Data collection and analysis methods (pp. 99-108). Baltimore:
University Park Press.
Holmes, D. S. (1966). The application of learning theory to the treatment of a school behavior
problem: A case study. Psychology in the School, 3, 355-359.
Holtzman, W. H. (1963). Statistical models for the study of change in the single case. In C. W.
Harris (Ed.), Problems in measuring change (pp. 199-211). Madison, WI: University of
Wisconsin Press.
Honig, W. K. (Ed.), (1966). Operant behavior: Areas of research and application. New York:
Appleton-Century-Crofts
Hopkins, B. L., Schutte, R. C, & Garton, K. L. (1971). The effects of access to a playroom on
the rate and quality of printing and writing of first- and second-grade students. Journal of
Home, G. P., Yang, M. C. K., & Ware, W. B. (1982). Time series analysis for single-subject
designs. Psychological Bulletin, 91, 178-189.
Horner, R. D., & Baer, D. M. (1978). Multiple-probe technique: A variation of the multiple
baseline. Journal of Applied Behavior Analysis, 11, 189-1%.
House, A. E., House, B. J., & Campbell, M. B. (1981). Measures of interobserver agreement:
Calculation formulas and distribution effects. Journal of Behavioral Assessment, 3, 31-51.
Hubert, L. J. (1977). Kappa revisited. Psychological Bulletin, 84, 289-297.
Hundert, J. (1982). Training teachers in generalized writing of behavior modification programs

for multihandicapped deaf children. Journal of Applied Behavior Analysis, 15, 111-122.
Hurlbut, B. I., Iwata, B. A., & Green, J. D. (1982). Nonvocal language acquisition in adolescents
with severe physical disabilities: Blissymbol versus iconic stimulus formats. Journal of Applied
Hutt, S. J., & Hutt, C. (1970). Direct observation and measurement of behavior. Springfield, IL:
Charles C Thomas.
Hyman, R., & Berger, L. (1966). Discussion: In H. J. Eysenck (Ed.), The effects of psy-
chotherapy (pp. 81-86). New York: International Science Press.
Inglis, J. (1966). The scientific study of abnormal behavior Chicago: Aldine.
Jact)bson, N. S., & Margolin, G. (1979). Marital therapy: Strategies based on social learning and
behavior exchange principles. New York: Brunner/Mazel.
Jayaratne, S., & Levy, R. L. (1979). Empirical clinical practice. New York: Columbia University
Press.
Johnson, S. M., & Bolstad, O. D. (1973). Methodological issues in naturalistic observation: Some
problems and solutions for field research. In L. A. Homerlynck, L. C. Handy, & E. J. Mash
(Eds.), Behavior change: Methodology, concepts, and practice (pp. 7-67). Champaign, IL:
Research Press.
Johnson, S. M., & Lobitz, G. K. (1974). Parental manipulation of child behavior in home
observations. Journal of Applied Behavior Analysis, 7, 23-31.
Johnston, J. M. (1972). Punishment of human behavior. American Psychologist, 27, 1033-1054.
Johnston, J. M., & Pennypacker, H. S. (1981). Strategies and tactics of human behavioral
research. Hillsdale, NJ: Erlbaum.
Johnston, M. K., Kelley, C. S., Harris, E R., & Wolf, M. M. (1966). An application of
reinforcement principles to development of motor skills of a young child. Child Development,
37, 379-387.
Joint Commission on Mental Illness and Health (1961). Action for mental health. New York:
Science Editions.
Jones, R. R., Reid, J. B., & Patterson, G. R. (1975). Naturalistic observation in clinical
assessment. In P. McReynolds (Ed.), Advances' in psychological assessment (Vo\. 3, pp. 42-95).
San Francisco: Jossey-Bass.
Jones, R. R., Vaught, R. S., & Reid, J. B. (1975). Time-series analysis as a substitute for single-
subject analysis of variance designs. In G. R. Patterson, I. M. Marks, J. D. Matarazzo, R. A.
Myers, G. E. Schwartz, & H. H. Strupp (Eds.), Behavior change, 1974 (pp. 164-169). Chicago:
Aldine.
Jones, R. R., Vaught, R. S., & Weinrott, M, R. (1977). Time-series analysis in operant research.
Jones, R. R., Weinrott, M. R., & Vaught, R. S. (1978). Effects of serial dependency on the
agreement between visual and statistical inference. Journal of Applied Behavior Analysis, 11,
277-283.
Jones, R. T, Kazdin, A. E., & Haney, J. I. (1981a). A follow-up to training emergency skills.

Jones, R. T, Kazdin, A. E., & Haney, J. I. (1981b). Social validation and emergency
training of
fire safety skills for potential injury prevention and life saving. Journal of Applied Behavior
Analysis, 14, 249-250.
Kantorovich, N. V. (1928). An attempt of curing alcoholism by associated reflexes. Novoye v
Refleksologii i Fixiologii Nervnoy Sistemy, 3, 436. Cited by Razran, G. H. S. (1934). Condi-
tional withdrawal responses with shock as the conditioning stimulus in adult human subjects.
Psychological Bulletin, 31, 111.
Kaufman, K. F, & O'Leary, K. D. (1972). Reward cost and self-evaluation procedures for
disrupting adolescents in a psychiatric hospital school. Journal of Applied Behavior Analysis,
4, 77-87.
References 389
Kazdin, A. E. (1973a). The effect of response cost and aversive stimulation in suppressing
punished and non-punished speech dysfluencies. Behavior Therapy, 4, 73-82.
Kazdin, A. E. (1973b). Methodological and assessment considerations in evaluating reinforce-
ment programs in applied settings. Journal of Applied Behavior Analysis, 6, 517-531.
Kazdin, A. E. (1977). Assessing the of behavior change through
clinical or applied significance
social validation. Behavior Modification, 1, 427-453.

Kazdin, A. E. (1978). History of behavior modification: Experimental foundations of contem-
porary research. Baltimore: University Park Press.
Kazdin, A. E. (1979). Unobtrusive measures in behavioral assessment. Journal of Applied
Kazdin, A. E. (1980a). Obstacles in using randomization tests in single-case experimentation.
Journal of Educational Statistics, 5, 253-260.
Kazdin, A. E. (1980b). Research design in clinical psychology. New York: Harper & Row.
Kazdin, A. E. (1981). Drawing valid inferences from case studies. Journal of Consulting and
Kazdin, A. E. (1982a). Observer effects: Reactivity of direct observation. In D. P. Hartmann
Kazdin, A. E. (1982b). Single-case research designs: Methods for clinical and applied settings.
New York: Oxford University Press.
Kazdin, A. E. (1982c). Sympton substitution, generalization, and response covariation: Implica-
tions for psychotherapy outcome. Psychological Bulletin, 91, 349-365.
Kazdin, A. E. (in press). Behavior modification in applied settings, (3rd ed.). Homewood, IL:
Dorsey Press.
Kazdin, A. E., & Bootzin, R. R. (1972). The token economy: An evaluative review. Journal of
Appleid Behavior Analysis, 5, 343-372.
Kazdin, A. E., & Geesey, S. (1977). Simultaneous-treatment design comparisons of the effects of
earning reinforcers for one's peers versus for oneself. Behavior Therapy, 8, 682-693.
Kazdin, A. E., & Hartmann, D. P. (1978). The simultaneous-treatment design. Behavior Therapy,
5, 912-923.
Kazdin, A. E., & Kopel, S. A. (1975). On resolving ambiguities of the multiple-baseline design:
Problems and recommendations. Behavior Therapy, 6, 601-608.
Kelly, D. (1980). Anxiety and emotions: Physiologial basis and treatment. Springfield, IL:
Charles C Thomas.
Kelly, J. A. (1980). The simultaneous replication design: The use of a multiple baseline to
establish experimental control in single group social skills treatment studies. Journal of
Behavior Therapy and Expermental Psychiatry, 11, 203-207.
Kelly, J. A., Laughlin, C, Claiborne, M., & Patterson, J. T. (1979). A group procedure for
teaching job interviewing skills to formerly hospitalized psychiatric patients. Behavior Thearpy,
10, 299-310.
Kelly, J. A., Urey, J. R., & Patterson, J. T. (1980). Improving heterosocial conversational skills of
male psychiatric patients through a small group training procedure. Behavior Therapy, 11,
179-188.
Kelly, M. B. (1977). A review of observational data-collection and reliability procedures reported
in the Journal of Applied Behavior Analysis. Journal of Applied Behavior Analysis, 10,
97-101.
Kendall, P. C, & Butcher, J. N. (1982). Handbook of research methods in clinical psychology.
New York: Wiley.
Kennedy, R. E. (1976). The feasibility of time-series analysis of single-case experiments. Un-
published manuscript.
Kent, R. N., & Foster, S. L. (1977). Direct observational procedures: Methodological issues in
naturalistic settings. In A. R. Ciminero, K. S. Calhoun, & H. E. Adams (Eds.), Handbook of
behavioral assessment (pp. 279-329). New York: Wiley.

Kernberg, O. F. (1973). Summary and conclusions of psychotherapy and psychoanalysis: Final
report of the Menninger Foundation's psychotherapy research project. International Journal
of Psychiatry, 11, 62-77.
Kessel, L., & Hyman, H. T. (1933). The value of psychoanalysis as a therapeutic procedure.
Journal of American Medical Association, 101, 1612-1615.
Kiesler, D. J. (1966). Some myths of psychotherapy research and the search for a paradigm.
Psychological Bulletin, 65, 110-136.

Kiesler, D. J. (1971). Experimental designs in psychotherapy research. In A. E. Bergin & S. L.
Garfield (Eds.), Handbook of psychotherapy and behavior change: An empirical analysis (2nd
ed.) (pp. 36-74). New York: Wiley
Kirby, F. D., & Shields, F. (1972). Modification of arithmetic response rate and attending behavior
in a seventh-grade student. Journal of Applied Behavior Analysis, 5, 79-84.
Kircher, A. S., Pear, J. J., & Martin, G. L. (1971). Shock as punishment in a picture naming task
with retarded children. Journal of Applied Behavior Analysis, 4, 227-233.
Kirchner, R. E., Schnelle, J. F, Domash, M. A., Larson, L. D., Carr, A. F, & McNees, M. P.
(1980). The applicability of a helicopter patrol procedure to diverse areas: A cost-benefit

evaluation. Journal of Applied Behavior Analysis, 13, 143-148.
Kirk, R. E. (1968). Experimental design: Procedures for the behavioral sciences. Glenmont, CA:
Brooks/Cole.
Kistner, J., Hammer, D., Wolfe, D., Rothblum, E., & Drabman, R. S. (1982). Teacher popularity
and contrast effects in a classroom token economy. Journal of Applied Behavior Analysis, 15,
85-96.
Knapp, T. J. (1983). Behavior analysts' visual appraisal of behavior change in graphic display.
Knight, R. P. (1941). Evaluation of the results of psychoanalytic therapy. American Journal of
Psychiatry, 98, 434-466.
Koegel, R. L., & Schreibman, L. (1982). How to teach autistic and other severely handicapped
children. Lawrence, KS: H&H Enterprises.
Kraemer, H. C. (1979). One-zero sampling in the study of primate beahvior. Primates, 20,
237-244.
Kraemer, H. C. (1981). Coping strategies in psychiatric clinical research. Journal of Consulting
and Clinical Psychology, 49, 309-319.
Krasner, L. (1971a). Behavior therapy Annual Review of Psychology, 22, 483-532.
Krasner, L. (1971b). The operant approach in behavior therapy. In A. E. Bergin & S. L. Garfield
(Eds.), Handbook of psychotherapy and behavior change: An empirical analysis (pp.
612-653). New York: Wiley
Kratochwill, T. R. (1978a). Foundations of time-series research. In T. R. Kratochwill (Ed.),
Single-subject research: Strategies for evaluating change (pp. 1-101). New York: Academic
Press.
Kratochwill, T. R. (Ed.) (1978b). Single-subject research: Strategies for evaluating change. New
Kratochwill, T. R., Alden, K., Demuth, D., Dawson, D., Panicucci, C, Arntson, P., McMurray,
N., Hempstead, J., & Levin, J. R. (1974). A further consideration in the application of an
analysis-of-variance model for the intrasubject replication design. Journal of Applied Behavior
Analysis, 7, 629-633.
Kratochwill, T. R., & Brody, G. H. (1978). Single subject designs: A perspective on the con-
troversy over employing statistical inference and implications for research and training in
behavior modification. Behavior Modification, 2, 291-307.
Kratochwill, T. R., & Levin, J. R. (1980). On the applicability of various data analysis procedures
to the simultaneous and alternating treatment designs in behavior therapy research. Behavioral
.
References 391
Lacey, J. I. (1959). Psychophysiological approaches to the evaluation of psychotherapeutic

process and outcome. In E. A. Rubinstein & M. B. Parloff (Eds.), Research in psychotherapy
(pp. 160-208). Washington, DC: National Publishing Co.
Lang, P. J. (1%8). Fear reduction and fear behavior: Problems in treating a construct. In J. M.
Shlien (Ed.), Research in psychotherapy (Vol. 3, pp. 90-102). Washington, DC: American
Psychological Association.
Last, C. G., Barlow, D. H., & O'Brien, G. T. (1983). Comparison of two cognitive strategies in
treatment of a patient with generalized anxiety disorder. Psychological Reports, 53, 19-26.
Laws, D. R., Brown, R. A., Epstein, J., & Hocking, N. (1971). Reduction of inappropriate social
behavior in disturbed children by an untrained paraprofessional therapist. Behavior Therapy,
2, 519-533.
Lawson, D. M. (1983). AlcohoHsm. In M. Hersen (Ed.), Outpatient behavior therapy: A clinical
guide (pp. 143-172). New York: Grune & Stratton.

Lazarus, A. A. (1%3). The results of behavior therapy in 126 cases of severe neurosis. Behaviour
Research and Therapy, 1, 69-80.
Lazarus, A. A. (1973). Multi-modal behavior therapy: Treating the BASIC ID. Journal of
Nervous and Mental Disease, 156, 404-41 1
Lazarus, A. A., & Davison, G. C. (1971). Clinical innovation in research and practice. In A. E.
Bergin & S. L. Garfield (Eds.), Handbook of psychotherapy and behavior change: An
empirical analysis (pp. 196-213). New York: Wiley.
Leitenberg, H. (August, 1973). Interaction designs. Paper read at American Psychological
Association, Montreal.
H. (1973). The use of single-case methodology
Leitenberg, in psychotherapy research. Journal of
Abnormal Psychology, 82, 87-101.
Leitenberg, H. (1976). Behavioral approaches to treatment of neuroses. In H. Leitenberg (Ed.),
Handbook of behavior modification and behavior therapy (pp., 124-167). Englewood CHffs,
NJ: Prentice-Hall.
Leitenberg, H., Agras, W. S., Edwards, J. A., Thomson, L. E., & Wincze, J. R (1970). Practice
as a psychotherapeutic variable: An experimental analysis within single cases. Journal of
Psychiatric Research, 7, 215-225.
Leitenberg, H., Agras, W S., Thomson, L. E., & Wright, D. E. (1968). Feedback in behavior
modification: An experimental analysis of two phobic cases. Journal of Applied Behavior
Analysis, 1, 131-137.
Leonard, S. R., & Hayes, S. C. (1983). Sexual fantasy alternation. Journal of Behavior Therapy
and Experimental Psychiatry, 14, 241-249.
Levin, J. R., Marascuilo, L. A., & Hubert, L. J. (1978). N = Nonparametric randomization
tests. In T. R. Kratochwill (Ed.), Single-subject research: Strategies for evaluating change (pp.
167-197). New York: Academic Press.
Levy, R. L., & Olson, D. G. (1979). The single-subject methodology in clinical practice: An
overview. Journal of Social Service Research, 3, 25-49.
Lewin, K. (1933). Vectors, cognitive processes and Mr. Tolman's criticism. Journal of General
Lewinsohn, P. M., & Libet, J. (1972). Pleasurable events, activity schedules, and depression.
Journal of Abnormal Psychology, 79, 291-295.
Lewinsohn, P. M., Mischel, W, Chaplin, W, & Barton, R. (1980). Social competence and
depression: The roles of illusory self-perceptions. Journal of Abnormal Psychology, 89,
203-212.
Liberman, R. P., Davis, J., Moon, W, & Moore, J. (1973). Research design for analyzing drug-
environment-behavior interactions. Journal of Nervous and Mental Disease, 156, 432-439.
Liberman, R. R, Neuchterlein, K. H., & Wallace, C. J. (1982). Social skills training in the nature
of schizophrenia. In Curran, J. P. & Monti, P. M. (Eds.), Social skills training (pp. 1-56). New
Liberman, R. P., & Smith, V. (1972). A multiple baseline study of systematic desensitization in a
patient with multiple phobias. Behavior Therapy, 3, 597-603.
Liberman, R. P., Wheeler, E. G., DeVisser, L. A., Kuehnel, J., & Kuehnel, T. (1980). Handbook
of marital therapy. New York: Plenum.
Lick, J. R., Sushinsky, L. W., & Malow, R. (1977). Specificity of Fear Survey Schedule items and
the prediction of avoidance behavior. Behavior Modification, /, 195-204.
Light, F. J. (1971). Measures of response agreement for qualitative data: Some generalizations
and alternatives. Psychological Bulletin, 76, 365-377.
Lindsley, O. R. (1%2). Operant conditioning techniques in the measurement of psychopharmaco-
logical response. In J. H. Nodine & J. H. Moyer (Eds.), Psychosomatic medicine: The first
Hahnemann symposium on psychosomatic medicine (pp. 373-383). Philadelphia: Lea &
Febiger.
Linehan, M. M. (1980). Content validity: Its relevance to behavioral assessment. Behavioral
Lovaas, O. I., Berberich, J. P, Perloff, B. E, & Schaeffer, B. (1966). Acquisition of imitiative
speech by schizophrenic children. Science, 161, 705-707.
Lovaas, O. I., Freitas, L., Nelson, K., & Whalen, C. (1967). The establishment of imitation and
its use for the development of complex behavior in schizophrenic children. Behaviour Research
and Therapy, 5, 171-181.
Lovaas, O. I., Koegel, R., Simmons, J. Q., & Long, J. D. (1973). Some generalization and
follow-up measures on autistic children in behavior therapy. Journal of Applied Behavior
Analysis, 5, 131-166.
Lovaas, O. I., Schaeffer, B., & Simons, J. Q. (1965). Experimental studies in childhood
schizophrenia: Building social behaviors using electric shock. Journal of Experimental Re-
search in Personality, 1, 99-109.
Lovaas, O. I., & Simmons, J. Q. (1969). Manipulation of self-destruction in three retarded
children. Journal of Applied Behavior Analysis, 2, 143-157.
P. R. Farnsworth & Q. McNemar (Ed.), Annual review of
Luborsky, L. (1959). Psychotherapy. In
psychology (pp. 317-344). Palo Alto, CA: Annual Review.
Lyman, R. D., Richard, H. C, & Elder, I. R. (1975). Contingency management of self-report
and cleaning behavior. Journal of Abnormal Child Psychology, 3, 155-162.
Madsen, C. H., Becker, W. C, & Thomas, D. R. (1968). Rules, praise, and ignoring: Elements of
elementary classroom control. Journal of Applied Behavior Analysis, 1, 139-150.
Malan, D. H, (1973). Therapeutic factors in analytically oriented brief psychotherapy. In R. H.
Gosling (Ed.), Support, innovation and autonomy (pp. 187-205). London: Tavistock.
Malone, J. C, Jr. (1976). Local contrast and Pavlovian induction. Journal of the Experimental
Analysis of Behavior, 26, 425-440.
Mandell, R. M., & Mandell, M. P. (1967). Suicide and the menstrual cycle. Journal of the
American Medical Association, 200, 792-793.
Mann, R. A. (1972). The behavior-therapeutic use of contingency contracting to control an adult
behavior problem: Weight control. Journal of Applied Behavior Analysis, 5, 99-109.
Mann, R. A., & Baer, D. M. (1971). The effects of receptive language training on articulation.
Mann, R. A., & Moss, G. R. (1973). The therapeutic use of a token economy to manage a young
and assaultive inpatient population. Journal of Nervous and Mental Disease, 157, 1-9.
Mansell, J. (1982). Repeated direct replication of AB designs (Letter to the Editor). Journal of
Behaviour Therapy and Experimental Psychiatry, 13, 261-262.
Marks, I. M. (1972). Flooding (implosion) and allied treatments. In W. S. Agras (Ed.), Behavior
modification: Principles and clinical applications (pp. 151-213). Boston: Little, Brown.
Marks, I. M. (1981). New developments in psychological treatments of phobias. In M. R.
Mavissakalian & D. H. Barlow (Eds.), Phobia: Psychological and pharmacological treatment
(pp. 175-199). New York: Guilford Press.
References 393
Marks, I. M., & Gelder, M. G. (1967). Transvestism and fetishism: Clinical and psychological
changes during faradic aversion. British Journal of Psychiatry, IB, 711-729.
Martin, G., Pallotta-Cornick, A., Johnstone, G., & Celso-Goyos, A, (1980). A supervisory
strategy to improve work performance for lower functioning retarded clients in a sheltered
workshop. Journal of Applied Behavior Analysis, 13, 185-190.
Martin, P. J., & Lindsey, C. J. (1976). Irregular discharge as an unobtrusive measure of . . .
something: Some additional thoughts. Psychological Reports, 38, 627-630.

Mash, E. J., & Makohoniuk, G. (1975). The effects of prior information and behavioral
predictability on observer accuracy. Child Development, 46, 513-519.
Mash, E. J., & Terdal, L. G. (Eds.). (1981). Behavioral assessment of childhood disorders. New
Matson, J. L. (1981). Assessment and treatment of clinical fears in mentally retarded children.
Matson, J. L. (1982). The treatment of behavioral characteristics of depression in the mentally
retarded. Behavior Therapy, 13, 209-218.
Mavissakalian, M. R., & Barlow, D. H. (1981a). Assessment of obsessive-compulsive disorders.
In D. H. Barlow (Ed.), Behavioral assessment of adult disorders (pp. 209-239). New York:
Guilford Press.
Mavissakalian, M. R., & Barlow, D. H. (1981b). Phobia: An overview. In M. R. Mavissakalian
& D. H. Barlow (Eds.), Phobia: Psychological and pharmacological treatment (pp. 1-35). New
Mavissakalian, M. R., & Barlow, D. H. (Eds.). (1981c). Phobia: Psychological and pharmacolo-
gical treatment. New York: Guilford Press.
Max, L. W. (1935). Breaking up a homosexual fixation by the conditioned reaction techique: A
case study Psychological Bulletin, 32, 734.
May, P. R. A. (1973). Research in psychotherapy and psychoanalysis. International Journal of
Psychiatry, 1, 78-86.
McCalHster, L. W., Stachowiak, J. G., Baer, D. M., & Conderman, L. (1969). The application of
operant conditioning techniques in a secondary school classroom. Journal of Applied Behavior
Analysis, 2, 277-285.
McCleary, R., & Hay, R. A., Jr. (1980). Applied time series analysis for the social sciences.
Beverly Hills: Sage.
McCullough, J. P, Cornell, J. E., McDaniel, M. H., & Meuller, R. K. (1974). Utilization of the
simultaneous treatment design to improve student behavior in a first-grade classroom. Journal
of Consulting and Clinical Psychology, 42, 288-292.
McFall, R. M. (1970). Effects of self-monitoring on normal smoking behavior. Journal of
McFall, R.M. (1977). Analogue methods in behavioral assessment: Issues and prospects. In J. D.
Cone & R. P. Hawkins (Eds.), Behavioral assessment: New direction in clinical psychology (pp.
152-177). New York: Brunner/Mazel.
McFall, R. M., & Lillesand, D. B. (1971). Behavior rehearsal with modeling and coaching in
assertion training. Journal of Abnormal Psychology, 77, 313-323.
McFarlain, R. A., & Hersen, M. (1974). Continuous measurement of activity level in psychiatric
Journal of Clinical Psychology, 30, 37-39.
patients.
McKnight, D. L., Nelson, R. O., Hayes, S. C, & Jarrett, R. B. (1983). Importance of treating
individually assessed response classes in the amelioration of depression. Behavior Therapy.
McLaughUn, T. E, & Malaby, J. (1972). Intrinsic reinforcers in a classroom token economy.
McLean, A. P., & White, K. G. (1981). Undermatching and contrast within components of
multiple schedules. Journal of the Experimental Analysis of Behavior, 35, 283-291.
McMahon, R, J., & Forehand, R. L. (1983). Consumer satisfaction in behavioral treatment of
children: Types, issues, and recommendations. Behavior Therapy, 14, 209-225.
McNamara, J. R. (1972). The use of self-monitoring techniques to treat nailbiting. Behaviour

Research and Therapy, W, 193-194.
McNamara, J. R., & MacDonough, T. S. (1972). Some methodological considerations in the
design and implementation of behavior therapy research. Behavior Therapy, 3, 361-378.
Melin, L.,& Gotestam, K. G. (1981), The effects of rearranging ward routines of communication
and eating behaviors of psychogeriatric patients. Journal of Applied Behavior Analysis, 14,
47-51.
Metcalfe, M. (1956). Demonstration of a psychosomatic relationship. British Journal of Medical
Michael, J. (1974). Statistical inference for individual organism research: Mixed blessing or
curse? Journal of Applied Behavior Analysis, 7, 647-653.
Miller, P. M. (1973). An experimental analysis of retention control training in the treatment of
nocturnal enuresis in two institutionalized adolescents. Behavior Therapy, 4, 288-294.
Miller, P. M., Hersen, M., Eisler, R. M., & Watts, J. G. (1974). Contingent reinforcement of
lowered blood/alcohol levels in an outpatient chronic alcoholic. Behaviour Research and
Therapy, 12, 261-263.
Mills, H. L., Agras, W. S., Barlow, D. H., & Mills, J. R. (1973). Compulsive rituals treated by
response prevention: An experimental analysis. Archives of General Psychiatry, 28, 524-529.
Minkin, N., Braukmann, C. J., Minkin, B. L., Timbers, G. D., Timbers, B. J., Fixsen, D. L.,
Phillips, E. L., & Wolf, M. M. (1976). The social validation and training of conversational
skills. Journal of Applied Behavior Analysis, 9, 127-139.
Mischel, W (1968). Personality and assessment. New York: Wiley.
Mitchell, S. K. (1979). Inter observer agreement, reliability, and generalizability of data collected
in observational studies. Psychological Bulletin, 86, 376-390.
Montague, J. D., & Coles, E. M. (1966). Mechanism and measurement of the galvanic skin
response. Psychological Bulletin, 65, 261-279.
Monti, P. M., Corriveau, E. P., & Curran, J. P. (1982). Social skills training for psychiatric
patients: Treatment and outcome. In J. P. Curran & P. M. Monti (Eds.), Social skills training
(pp. 185-223). New York: Guilford Press.

Moses, L. E. (1952). Nonparametric statistics for psychological research. Psychological Bulletin,
49, 122-143.
Munford, P. R., & Liberman, R. P. (1978). Differential attention in the treatment of operant
cough. Journal of Behavioral Medicine, 1, 280-289,
Nathan, R E., Titler, N. A., Lowenstein, L. M., Solomon, R, & Rossi, A. M. (1970). Behavioral
analysis of chronic alcoholism. Archives of General Psychiatry, 22, 419-430.
National Institute of Mental Health. (1980). Behavior therapies in the treatment of anxiety
disorders: Recommendations for strategies in treatment assessment research. (Final report of
NIMH conference ^RFP NIMH ER-79-003). Unpublished manuscript.
Nay, W R. (1977). Analogue measures. In A. R. Ciminero, K. S. Calhoun, & H. E. Adams
(Eds)., Handbook of behavioral assessment (pp. 233-279). New York: Wiley.
Nay, W R. (1979). Mult imethod clinical assessment. New York: Gardner Press.
Neale, J. M., & Oltmanns, T. (1980). Schizophrenia. New York: Wiley.
Neef, N. A., Iwata, B. A., & Page, T J. (1980). The effects of interspersal training versus high-
density reinforcement on spelling acquisition and retention. Journal of Applied Behavior
Analysis, 13, 153-158.
Nelson, R. O. (1977). Methodological issues in assessment via self-monitoring. In J. D. Cone &
R. P. Hawkins (Eds.), Behavioral assessment: New directions in clinical psychology (pp.
217-254). New York: Brunner/Mazel.
Nelson, R. O., & Hayes, S. C. (1979). Some current dimensions of behavioral assessment.
Nelson, R. O., & Hayes, S. C. (1981). Nature of behavioral assessment. In M. Hersen & A. S.
Bellack (Eds.), Behavioral assessment: A practical handbook, (2nd ed.) (pp. 3-37). Elmsford,
References 395
New York: Pergamon Press.

Nietzel, M. T, & Bernstein, D. A. (1981). Assessment of anxiety and fear. In M. Hersen & A. S.
Bellack (Eds.), Behavioral assessment: A practical handbook, (2nd ed.) (pp. 215-245). Elms-
ford, New York: Pergamon Press.
Nordquist, V. M. (1971). The modification of a child's enuresis: Some response-response relation-
ships. Journal of Applied Behavior Analysis, 4, 241-247.
Nunnally, J. (1978). Psychometric theory, (2nd ed.). New York: McGraw-Hill.
O'Brien, E, Azrin, N. H., & Henson, K. (1969). Increased communication of chronic mental
patients by reinforcement and by response priming. Journal of Applied Behavior Analysis, 2,
23-29.
O'Brien, E, Bugle, C, & Azrin, N. H. (1977). Training and maintaining a retarded child's proper
eating.Journal of Applied Behavior Analysis, 10, 465-478.
O'Leary, K. D. (1979). Behavioral assessment. Behavioral Assessment, I, 31-36.
O'Leary, K. D., & Becker, W. C. (1967). Behavior modification of an adjustment class: A token
reinforcement program. Exceptional Children, 9, 637-642.
O'Leary, K. D., Becker, W. C, Evans, M. B., & Saudargas, R. A. (1969). A token reinforcement
program in a public school: A replication and systematic analysis. Journal of Applied Behavior
Analysis, 2, 3-13.
O'Leary, K. D., Kent, R. N., & Kanowitz, J. (1975). Shaping data collection congruent with
experimental hypotheses. Journal of Applied Behavior Analysis, 8, 43-51.
O'Leary, K. D., & Tbrkewitz, H. (1981). A comparative outcome study of behavioral marital
therapy and communication therapy. Journal of Marital and Family Therapy, 7, 159-169.
OUendick, T. H. (1981). Self-monitoring and self-administered overcorrection: Behavior Modi-
fication, 5, 75-84.
OUendick, T. H., Matson, J. L., Esveldt-Dawson, K., & Shapiro, E. S. (1980). Increasing spelling
achievement: An analysis of treatment procedures utilizing an alternating treatments design.
OUendick, T. H., Shapiro, E. S., & Barrett, R. P. (1981). Reducing stereotypic behaviors: An
analysis of treatment procedures utilizing an alternating treatments design. Behavior Therapy,
12, 570-577.
Orne, M. T. (1962). On the social psychology of the psychological experiment: With particular
reference to demand characteristics and their implications. American Psychologist, 17,
776-783.
Paris, S. G., & Cairns, R. B. (1972). An experimental and ethological analysis of social
reinforcement with retarded children. Child Development, 43, 717-719.
Parsonson, B. S., & Baer, D. M. (1978). The analysis and presentation of graphic data. In T. R.
Kratochwill (Ed.), Single-subject research: Strategies for evaluating change (pp. 101-167). New
Patterson, G. R. (1982). Coercive family process. Eugene, OR: Castalia.
Paul, G. L. (1967). Strategy of outcome research in psychotherapy. Journal of Consulting
Paul, G. L. (1969). Behavior modification research: Design and tactics. In C. M. Franks (Ed.),
Behavior therapy: Appraisal and status (pp. 29-62). New York: McGraw-Hill.
Paul, G. L. (1979). New assessment systems for residential treatment, management, research, and
evaluation: A symposium. Journal of Behavioral Assessment, 1, 181-184.
Paul, G. L., & Lentz, R. J. (1977). Psychosocial treatment of chronic mental patients: Milieu
versus social-learning programs. Cambridge: Harvard University Press.
Pavlov, 1. P. (1928). Lectures on conditioned reflexes. (W. H. Gantt, TVans.) New York: Interna-
tional.
Pendergrass, V. E. (1972). Timeout from positive reinforcement following persistent, high-rate
behavior in retardates. Journal of Applied Behavior Analysis, 5, 85-91.
Pertschuk, M. J., Edwards, N., & Pomerleau, O. F. (1978). A
multiple baseline approach to
behavioral intervention in anorexia nervosa. Behavior Therapy, 9, 368-376.

Peterson, L., Homer, A. L., & Wonderlich, S. A. (1982). The integrity of independent variables
in behavior analysis. Journal of Applied Behavior Analysis, 15, 477-492.
Peterson, R. E, & Peterson, L. (1968). The use of positive reinforcement in the self-control of
self-destructive behavior in a retarded boy. Journal of Experimental Child Psychology, 6,
351-360.
Pinkston, E. M., Reese, N. M., LeBlanc, J. M., & Baer, D. M. (1973). Independent control of a
preschool child's aggression and peer interaction by contingent teacher attention. Journal of
Poche, C, Brouwer, R., & Swearingen, M. (1981). Teaching self-protection to young children.
Porterfield, J., Blunden, R., & Blewitt, E. (1980). Improving environments for profoundly
handicapped adults: Using prompts and social attention to maintain high group engagement.
Behavior Modification, 4, 225-241.
Powell, J.,& Hake, D. F. (1971). Positive vs. negative reinforcement: A direct comparison of
effects on a complex human response. Psychological Record, 21, 191-205.
Powell, J., Martindale, A., & Kulp, S. (1975). An evaluation of time-sample measures of
behavior. Journal of Applied Behavior Analysis, 8, 463-469.
Power, C. T. (1979). The Time-Sample Behavioral Checklist: Observational assessment of patient
functioning. Journal of Behavioral Assessment, 1, 199-210.
Powers, E., & Witmer, H. (1951). An prevention of delinquency. New York:
experiment in the
Columbia University Press.

Rachlin, H. (1973). Contrast and matching. Psychological Review, 80, 297-308.
Rachman, S. J., & Hodgson, R. J. (1980). Obsessions and compulsions. Englewood Cliffs, NJ:
Prentice-Hall.
Ramp, E., Ulrich, R., & Dulaney, S. (1971). Delayed timeout as a procedure for reducing
disruptive classroom behavior: A case study. Journal of Applied Behavior Analysis, 4,
235-239.
Rapport, M. D., Sonis, W. A., Fialkov, M. J., Matson, J. L., & Kazdin, A. E. (1983).
Carbamazepine and behavior therapy for aggressive behavior: Treatment of a mentally re-
tarded, postencephalic adolescent with seizure disorder. Behavior Modification, 7, 255-265.
Ray, W. J., & Raczynski, J. M. (1981). Psychophysiological assessment. In M. Hersen & A. S.
Bellack (Eds.), Behavioral assessment: A practical handbook, (2nd ed.) (pp. 175-211). Elms-
ford, New York: Pergamon Press.
Redd, W. H, (1980). Stimulus control and extinction of psychosomatic symptoms in cancer
patients in protective isolation. Journal of Consulting and Clinical Psychology, 48, 448-456.
Redd, W. H., & Birnbrauer, J. S. (1969). Adults as discriminative stimuli for different reinforce-
ment contingencies with retarded children. Journal of Experimental Child Psychology, 7,
440-447.
Redfield, J. P., & Paul, G. L. (1976). Bias in behavioral observation as a function of observer
familiarity with subjects and typicality of behavior. Journal of Consulting and Clinical Psy-
chology, 44, 156.
Rees, L. (1953). Psychosomatic aspects of the prementrual tension system. Journal of Mental
Science, 99, 62-73.
Reid, J. B. (1978). The development of specialized observation systems. In J. B. Reid (Ed.), A
social learningapproach to family intervention: Vol. 2. Observation in home settings (pp.
43-49). Eugene, OR: Castalia.
Reid, J. B. (1982). Observer training in naturalistic research. In D. P. Hartmann (Ed.), Using
observers to study behavior: New directions for methodology of social and behavioral science
(pp. 37-50). San Francisco: Jossey-Bass.
Revusky, S. H. (1976). Some statistical treatments compatible with individual organism method-
ology. Journal of the Experimental Analysis of Behavior, 10, 319-330.
References 397
Reynolds, G. S. (1968). A primer of operant conditioning. Glenview, IL: Scott, Foresman.

Reynolds, N. J., & Risley, T. R. (1968). The role of social and material reinforcers in increasing
talking of a disadvantaged preschool child. Journal of Applied Behavior Analysis, 7, 253-262.
Rickard, H. C, Dignam, P. J., & Horner, R. E (I960). Verbal manipulation in a psychothera-
peutic relationship. Journal of Clinical Psychology, 16, 164-167.
Rickard, H. C, & Dinoff, M. (1962). A follow-up note on "Verbal manipulation in a psy-

chotherapeutic relationship." Psychologicl Reports, 11, 506.
Rickard, H. C, & Saunders, T. R. (1971). Control of "clean-up" behavior in a summer camp.
Risley, T. R. (1968). The effects and side-effects of punishing the autistic behaviors of a deviant
child. Journal of Applied Behavior Analysis, 1, 21-34.
Risley, T. R. (1970). Behavior modification: An experimental-therapeutic endeavor. In L. A.
Hamerlynck, P. O. Davidson, & L. E. Acker (Eds.), Behavior modification and ideal health
services (pp. 103-127). Calgary, Alberta, Canada: University of Calgary Press.
Risley, T. R., & Wolf, M. M. (1972). Strategies for analyzing behavioral change over time. In J.
Nesselroade & H, Reese (Eds.), Life-span developmental psychology: Methodological issues
(pp. 175-183). New York: Academic Press.
Roberts, M. W., Hatzenbuehler, L. C, & Bean, A. W. (1981). The effects of differential attention
and timeout on child noncompliance. Behavior Therapy, 12, 93-99.
Rogers, C. R., Gendlin, E. T, Kiesler, D. J., & Truax, C. B. (1967). The therapeutic relationship
and its impact: A study ofpsychotherapy with schizophrenics. Madison: University of Wiscon-
sin Press.
Rogers- Warren, A., & Warren, S. F. (1977). Ecological perspectives in behavior analysis. Balti-
more: University Park Press.
Rojahn, J., Mulick, J. A., McCoy, D., & Schroeder, S. R. (1978). Setting effects, adaptive
clothing, and the modification of head-banging and self-restraint in two profoundly retarded
adults. Behavioural Analysis and Modification, 2, 185-196.
Rosen, J. C, & Leitenberg, H. (1982). Bulimia Nervosa: Treatment with exposure and response
evaluation. Behavior Therapy, 13, 117-124.
Rosenblum, L. A. (1978). The creation of a behavioral taxonomy. In G. P. Sackett (Ed.),
Observing behavior: Vol. 2. Data collection and analysis methods (pp. 15-24). Baltimore:
University Park Press.
Rosenthal, R. (1976). Experimenter effects in behavioral research (enlarged ed.). New York:
Irvington.
Rosenzweig, S. (1951). Idiodynamics in personality therapy with special reference to projective
methods. Psychological Review, 58, 213-223.
Ross, A. O. (1981). Child behavior therapy: Principles, procedures, and empirical basis. New
York: Wiley
Roxburgh, P. A. (1970). TVeatment of persistent phenothiazine-induced oraldyskinesia. British
Journal of Psychiatry, 116, 277-280.
Rubenstein, E. A., & Parloff, M. B. (1959). Research problems in psychotherapy. In E. A.
Rubenstein & M. B. Parloff (Eds.), Research in psychotherapy, (Vol. 1) (pp. 276-293).
Washington, DC: American Psychological Association.
Rugh, J. E., & Schwitzgebel, R. L. (1977). Instrumentation for behavioral assessment. In A. R.
Ciminero, K. S. Calhoun, & H. E. Adams (Eds.), Handbook of behavioral assessment (pp.
79-113). New York: Wiley
Rusch, F. R., & Kazdin, A. E. (1981). Toward a methodology of withdrawal designs for the
assessment of response maintenance. Journal of Applied Behavior Analysis, 14, 131-140.
Rusch, F. R., Walker, H. M., & Greenwood, C. R. (1975). Experimenter calculation errors: A
potential factor affecting interpretation of results. Journal of Applied Behavior Analysis, 5,
460.
Russell, M. B., & Bernal, M. E. (1977). Temporal and climatic variables in naturalistic observa-
tion. Journal of Applied Behavior Analysis, 10, 399-405.

Russo, D. C, & Koegel, R. L. (1977). A method for integrating an autistic child into a normal
public school classroom. Journal of Applied Behavior Analysis, 10, 579-590.
Sackett, G. P. (1978). Measurement in observational research. In G. P. Sackett (Ed.), Observing

behavior: Vol. 2. Data collection and analysis methods (pp. 25-43). Baltimore: University Park
Press.
St. Lawrence, J. S., Bradlyn, A. S., & Kelly, J. A. (1983). Interpersonal adjustment of a
homosexual adult: Enhancement via social skills training. Behavior Modification, 7, 41-55.
Sajwaj, T. E., & Dillon, A. (1977). Complexities of an "elementary" behavior modification
procedure: Differential adult attention used for children's behavior disorders. In B. C. Etzel, J.
M. LeBlanc, & D. M. Baer (Eds)., New developments in behavioral research: Theory, methods
and application (pp. 303-315). Hillsdale, NJ: Erlbaum.
Sajwaj, T. E., & Hedges, D. (1971). Functions of parental attention in an oppositional retarded
boy In Proceedings of the 79th Annual Convention of the American Psychological Association
(pp. 697-698). Washington, DC: American Psychological Association.
Sajwaj, T. E., TXvardosz, S., & Burke, M. (1972). Side effects of extinction procedures in a
remedial preschool. Journal of Applied Behavior Analysis, 5, 163-175.
Sanson-Fisher, R. W., Poole, A. D., Small, G. A., & Fleming, I. R. (1979). Data acquisition in
real time: An improved system for naturalistic observations. Behavior Therapy, 10, 543-554.
Scheffe, H. (1959). The analysis of variance. New York: Wiley.
Schindele, R. (1981). Methodological problems in rehabilitation research. International Journal
of Rehabilitation Research, 4, 233-248.
Schleien, S. J., Weyman, P., 8c Kiernan, J. (1981). Teaching leisure skills to severely handicapped
adults: An age appropriate darts game. Journal of Applied Behavior Analysis, 14, 513-519.
Schreibman, L., Koegel, R. L., Mills, D. L., & Burke, J. C. (in press). Training parent child
interactions. In E. Scholper & G. Mesibov (Eds.), Issues in autism: Vol. III. The effects of
autism on the family. New York: Plenum.

Schumaker, J., & Sherman, J. A. (1970). Training generative verb usage by imitation and
reinforcement procedure. Journal of Applied Behavior Analysis, 3, 273-287.
Schutte, R. C, & Hopkins, B. L. (1970). The effects of teacher attention following instructions in
a kindergarten Journal of Applied Behavior Analysis, 3, 117-122.
class.
Sechrest, L. (Ed.). (1979). Unobtrusive measurement today: New directions for methodology of
behavioral science. San Francisco: Jossey-Bass.
Shapiro, D. A., & Shapiro, D. (1983). Comparative therapy outcome research: Methodological
implications of meta-analysis. Journal of Consulting and Clinical Psychology, 51, 42-53.
Shapiro, E. S., Barrett, R. P., & Ollendick, T H. (1980). A comparison of physical restraint and
positive practice overcorrection in treating stereotypic behavior. Behavior Therapy, 11,
227-233.
Shapiro, E. S., Kazdin, A. E., & McGonigle, J.J. (1982). Multiple-treatment interference in the
simultaneous- or alternating-treatments design. Behavioral Assessment, 4, 105-115.
Shapiro, M. B. (1961). The single case in fundamental clincial psychological research. British
Journal of Medical Psychology, 34, 255-263.

Shapiro, M. B. (1966). The single case in clinical-psychological research. Journal of General
Shapiro, M. B. (1970). Intensive assessment of the single case: An inductive-deductive approach.
In P. Mittler (Ed.), Psychological assessment of mental and physical handicaps. London:

Methuen.
Shapiro, M. B., & Ravenette, A. T. (1959). A preliminary experiment of paranoid delusions.
Journal of Mental Science, 105, 295-312.
Shine, L. C, & Bower, S. M. (1971). A one-way analysis of variance for single-subject designs.
Educational and Psychological Measurement, 31, 105-1 13.
References 399
Shontz, E C. (1965). Research methods in personality. New York: Appleton-Century-Crofts.

Shrout, P. E., & Eleiss, J. H. (1979). Intraclass correlations: Uses in assessing rater reliability.
Psychological Bulletin, 86, 420-428.
Shuller, D. Y, & McNamara, J. R. (1976). Expectancy factors in behavioral observation.
Sidman, M. (1960. Tactics of scientific research: Evaluating experimental data in psychology. New
York: Basic Books.
Simon, A., & Boyer, E. G. (1974). Mirrors for behavior: Vol. 3. An anthology of observation
instruments. Eyncote, PA: Communication Materials Center.
Simpson, M. J. A. (1979). Problems of recording behavioral data by keyboard. In M. E. Lamb,
S. J. Suomi, & G. R. Stephenson (Eds.), Social interaction analysis: Methodological issues (pp.
137-156). Madison: University of Wisconsin Press.
Singh, N. N., Dawson, J. H., & Gergory, P. R. (1980). Suppression of chronic hyperventilation
using response contingent aromatic ammonia. Behavior Therapy, 11, 561-566.
Singh, N. N., Manning, P. J., & Angell, M. J. (1982). Effects of an oral hygiene punishment
procedure on chronic schizophrenic rumination and collateral behaviors in monozygous twins.
Singh, N. N., Winton, A. S., & Dawson, M. H. (1982). Suppression of antisocial behavior by
facial screening using multiple baseline and alternating treatments designs. Behavior Therapy,
75,511-520.
Skiba, E. A., Pettigrew, E., & Alden, S. E. (1971). A behavioral approach to the control of
thumbsucking in the classroom. Journal of Applied Behavior Analysis, 4, 121-125.
Skinner, B. E (1938). The behavior of organisms. New York: Appleton-Century-Crofts.
Skinner, B. F. (1953). Science and human behavior New York: Macmillan.
Skinner, B. F. (1966a). Invited address to the Pavlovian Society of America, Boston.
Skinner, B. F. (1966b). Operant behavior. In W. K. Honig (Ed.), Operant behavior: Areas of
research and application (pp. 12-32). New York: Appleton-Century-Crofts.
Slavon, R. E., Wodarski, J. S., & Blackburn, B. L. (1981). A group contingency for electricity
conservation in master-metered apartments. Journal of Applied Behavior Analysis, 14,

357-363.
Sloane, H. N., Johnston, M. K., & Bijou, S. W. (1967). Successive modification of aggressive
behavior and aggressive fantasy play by management of contingencies. Journal of Child
Psychology and Psychiatry, 8, 217-226.
Smeets, P. M. (1970). Withdrawal of social reinforcers as a means of controlling rumination and
regurgitation in a profoundly retarded person. Training School Bulletin, 67, 158-163.
Smith, C. M. (1963). Controlled observations on the single case. Canadian Medical Association
Journal, 88, 410-412.
Smith, M. L., & Glass, G. V. (1977). Meta-analysis of psychotherapy outcome studies. American
psychologist, 32, 752-760.
Smith, P. C, & Kendall, L. M. (1963). Retranslation of expectations: An approach to the
construction of unambiguous anchors for rating scales. Journal of Applied Psychology, 47,
149-155.
Sowers, J., Rusch, F. R., Connis, R. T, & Cummings, L. T. (1980). Teaching mentally retarded
adults to time-manage in a vocational setting. Journal of Applied Behavior Analysis, 13,
119-128.
Spitzer, J. B. W, & Nee, J. (1979). DSM-III field trials: I. Initial interrater
R. L., Forman,
American Journal of Psychiatry, 136, 815-817.
diagnostic reliability.
Steinman, W. M. (1970). The social control of generalized imitation. Journal of Applied Behavior
Analysis, 3, 159-167.
Steketee, G., & Foa, E. B. (in press). Obsessive-compulsive disorders. In D. H. Barlow (Ed.),
Behavioral treatment of adult disorders. New York: Guilford Press.
Steketee, G., Foa, E. B., & Grayson, J. B. (1982). Recent advances in the behavioral treatment of
obsessive compulsives. Archives of General Psychiatry, 39, 1365-1371.

Stern, R. M., Ray, W. J., & Davis, C. M. (1980). Psychophysiological recording. New York:
Oxford University Press.
Stilson, D. W. (1966). Probability and statistics in psychological research and theory. San
Francisco: Holden-Day.
Stokes, T. E, & Baer, D. M. (1977). An implicit technology of generalization. Journal of Applied
Stokes, T. E, & Kennedy, S. H. (1980). Reducing child uncooperative behavior during dental
treatment through modeling and reinforcement. Journal of Applied Behavior Analysis, 13,
41-49.
Stoline, M. R., Huitema, B. E., & Mitchell, B. T. (1980). Intervention time-series model with
different pre- and postintervention first-order autoregressive parameters. Psychological Bul-
letin, 88, 46-53.
Stravynski, A., Marks, I. M., & Yule, W. (1982). The sleep of patients with obsessive-compulsive
disorder. Archives of General Psychiatry, 39, 1378-1385.
Striefel, S., Bryan, K. S., & Aikens, D. A. (1974). Transfer of stimulus control from motor to
verbal stimuli. Journal of Applied Behavior Analysis, 7, 123-135.
Striefel, S., & Wetherby, B. (1973). Instruction following behavior of a retarded child and its
controlling stimuli. Journal of Applied Behavior Analysis, 6, 663-670.

Strupp, H. H., & Hadley, S. W. (1979). Specific vs. nonspecific factors in psychotherapy. Archives
of General Psychiatry, 36, 1 125-1 137.
Strupp, H. H., & Luborsky, L. (Eds.) (1962). Research in psychotherapy (Vol. 2). Washington,
DC: American Psychological Association.
Stuart, R. B. (1971). A three-dimensional program for the treatment of obesity. Behaviour
Research and Therapy, 9, 177-186.
Sulzer-Azaroff, B., & deSantamaria, M. C. (1980). Industrial safety hazard reduction through
performance feedback. Journal of Applied Behavior Analysis, 13, 287-295.
Sulzer-Azaroff, B., & Mayer, R. G. (1977). Applying behavior-analysis procedures with children
and youth. New York: Holt, Rinehart and Winston.
Swan, G. E., & MacDonald, M. L. (1978). Behavior therapy in practice: A national survey of
behavior therapists. Behavior Therapy, 9, 799-807.
Taplin, P. S., & Reid, J. B. (1973). Effects of instructional set and experimental influences on
observer reliability. Child Development, 44, 547-554.
Tate, B. B., & Baroff, G. S. (1966). Aversive control of self-injurious behavior in a psychotic boy.
Behaviour Research and Therapy, 4, 281-287.
Taylor, C. B., & Agras, W. S. (1981). Assessment of phobia. In D. H. Barlow (Ed.), Behavioral
assessment of adult disorders (pp. 181-209). New York: Guilford Press.
Thomas, D. R., Becker, W. C., & Armstrong, M. (1968). Production and elimination of
disruptive classroom behavior by systematically varying teachers' behavior. Journal of Applied
Thomas, D. R., Nielsen, T. J., Kuypers, D. S., & Becker, W C. (1968). Social reinforcement and
remedial instruction in the elimination of a classroom behavior problem. Journal of Special
Education, 2, 291-305.
Thomas, J. D., & Adams, M. A. (1971). Problems in teacher use of selected behaviour
modification techniques in the classroom. New Zealand Journal of Educational Studies, 6,
151-165.
Thomson, C., Holmberg, M., & Baer, D. M. (1974). A brief report on a comparison of time-
sampling procedures. Journal of Applied Behavior Analysis, 7, 623-626.
Thoreson, C. E., & Elashoff, J. D. (1974). Some comments on "An analysis-of-variance model
of Applied Behavior Analysis, 7, 639-641.
for the instrasubject replication design." Journal
Thome, F. C. (1947). The clinical method in science. American Psychologist, 2, 161-166.
Tinsley, H. E. A., & Weiss, D. J. (1975). Interrater reliability and agreement of subjective
References 401
judgments. Journal of Counseling Psychology, 22, 358-376.

TVuax, C. B. (1966). Reinforcement and non-reinforcement in Rogerian psychotherapy. Journal
of Abnormal Psychology, 71, 1-9.
Tlruax, C. B., & Carkhuff, R. R. (1965). Experimental manipulation of therapeutic conditions.

Journal of Consulting Psychology, 29, 119-124.
Tyron, W. W. (1982). A simplified time-series analysis for evaluating treatment interventions.
Journal of Applied Behavior Analysis, IS, Ali-Al^.
Tbrkat, I. D., & Maisto, S. (in press). Personality disorders. In D. H. Barlow (Ed.), Behavioral
treatment of adult disorders. New York: Guilford Press.
Tlirner, S. M., Hersen, M., & Alford, H. (1974). Effects of massed practice and meprobamate on
spasmodic torticollis: An experimental analysis. Behaviour Research and Therapy, 12^
259-260.
1\irner, S. M., Hersen, M., & Bellack, A. S. (1978). Social skills training to teach prosocial
behaviors in an organically impaired and retarded patient. Journal of Behavior Therapy and
Experimental Psychiatry, 9, 253-258.
& Capparell, H. V. (1980). Behavioral
Tbrner, S. M., Hersen, M., Bellack, A. S., Andrasik, E,
and pharmacological treatment of obsessive-compulsive disorders. Journal of Nervous <fc
Mental Disease, 168, 651-657.
Twardosz, S., & Sajwaj, T. E. (1972). Multiple effects of a procedure to increase sitting in a
hyperactive, retarded boy. Journal of Applied Behavior Analysis, 5, 73-78.
UUmann, L. P., & Krasner, L. (Eds.) (1965). Case studies in behavior modfication. New York:
Holt, Rinehart and Winston.
Ulman, J. D., & Sulzer-Azaroff, B. (1973, August). Multielement baseline design in applied
behavior analysis. Symposium conducted at the annual meeting of the American Psychological
Association, Montreal.
Ulman, J. D., & Sulzer-Azaroff, B. (1975). Multielement baseline design in educational research.
In E. Ramp &
G. Semb (Eds.), Behavior analysis: Areas of research and application (pp.
377-391). Englewood Cliffs, NJ: Prentice-Hall, 1975.
Underwood, B. J. (1957). Psychological research. New York: Appleton-Century-Crofts.
VanBierliet, A., Spangler, P. P., & Marshall, A. M. (1981). An ecobehavioral examination of a
simple strategy for increasing mealtime language in residential facilities. Journal of Applied
Van Hasselt, V. B., & Hersen, M. (1981). Applications of single-case designs to research with
visually impaired individuals. Journal of Visual Impairment and Blindness, 75, 359-362.
Van Hasselt, V. B., Hersen, M., Kazdin, A. E., Simon, J., & Mastantuono, A. K. (1983). Social
skills training for blind adolescents. Journal of Visual Impairment and Blindness, 75, 199-203.
Van Houten, R., Nau, P A., MacKenzie-Keating, S. E., Sameoto, D., & Colavecchia, B. (1982).
An analysis of some variables influencing the effectiveness of reprimands. Journal of Applied
Varni, J. W., Russo, D. C, & Cataldo, M. E (1978). Assessment and modification of delusional
speech in an 1
1
-year-old child: A comparative analysis of behavior therapy and stimulant drug
effects. Journal of Behavior Therapy and Experimental Psychiatry, 9, 377-380.
Veenstra, M. (1971). Behavior modification in the home with the mother as the experimenter: The
effect of differential reinforcement on sibling negative response rates. Child Development, 42,
2079-2083.
Venables, P. H., fe Christie, M. H. (1973). Mechanism, instrumentation, recording techniques
and quantification of responses. In W. F. Prokasy & D. C. Raskin (Eds.), Ectodermal activity

in psychological research (pp. 1-124). New York: Academic Press.
Venables, P. H., & Martin, I. (1967). A manual of psychophysiological methods. Amsterdam:
North-Holland.
Vermilyea, J. & Barlow, D. H. (in press). Rachman and Hodgson (1974) a decade
A., Boice, R.,
later: How do desynchronous response systems relate to the treatment of agoraphobia?
Behaviour Research and Therapy.

Vukelich, R., fe Hake, D. F. (1971). Reduction of dangerously aggressive behavior in a severely
retarded resident through a continuation of positive reinforcement procedures. Journal of
Wade, T. C, Baker, T. B., & Hartmann, D. P. (1979). Behavior therapists' self-reported viewsand
practices. Behavior Therapist, 2, 3-6.
Wahler, R. G. (1968, April). Behavior therapy for oppositional children: Love is not enough.
Paper presented at the meeting of the Eastern Psychological Association, Washington, DC.
Wahler, R. G. (1969a). Oppositional children: A quest for parental reinforcement control.
Wahler, R. G. (1969b). Setting generality: Some specific and general effects of child behavior
therapy. Journal of Applied Behavior Analysis, 2, 239-246.
Wahler, R. G., Berland, R. M., & Coe,D. (1979). Generalization processes in child behavior
T.
change. In B. B. Lahey & A. E. Kazdin (Eds.), Advances in clinical child psychology (pp.
36-72). New York: Plenum.
Wahler, R. G., & Pollio, H. R. (1968). Behavior and insight: A case study in behavior therapy.
Journal of Experimental Research in Personality, 3, 45-56.
Wahler, R. G., Sperling, K. A., Thomas, M. R., Teeter, N. C, & Luper, H. L. (1970).
Modification of childhood stuttering: Some response-response relationships. Journal of Ex-
perimental Child Psychology, 9, 411-428.
Wahler, R. G., Winkel, G. H., Peterson, R. E, & Morrison, D. C. (1965). Mothers as behavior
therapists for their own children. Behaviour Research and Therapy, 3, 113-124.
Waite, W.W, & Osborne, J. G. (1972). Sustained behavioral contrast in children. Journal of the
Experimental A nalysis of Behavior, 18, 113-117.
Walker, H. M., & Buckely, N. K. (1968). The use of positive reinforcement in conditioning
attending behavior. Journal of Applied Behavior Analysis, 1, 245-250.
Walker, H. M., & Lev, J. (1953). Statistical inference. New York: Holt, Rinehart and Winston.
Wallace, C. J., Boone, S. E., Donahoe, C. P., & Foy, D. W (in press). Chronic mental disabilities.
In D. H. Barlow (Ed,), Behavioral treatment of adult disorders. New York: Guilford Press.
Wallace, C. J. (1982). The social skills training project of the Mental Health Clinical Research
Center for the Study of Schizophrenia. In J. P. Curran & P. M. Monti (Eds.), Social skills
training (pp. 57-89). New York: Guilford Press.
Wallace, C. J., & Elder, J. P. (1980). Statistics to evaluate measurement accuracy and treatment
effects in single subject research designs. In M. Hersen, R. M. Eisler, & P. M. Monti (Eds.),
Progress in behavior modification, (Vol. 10, pp. 40-82). New York: Academic Press.
Wampold, B. E,, & Furlong, M. J. (1981a). The heuristics of visual inference. Behavioral
Wampold, B. E., & Furlong, M. J. (1981b). Randomization tests in single-subject designs:
Illustrative examples. Journal of Behavioral Assessment, 3, 329-341.
Ward, M. H., & Baker, B. L. (1968). Reinforcement therapy in the classroom. Journal of Applied
Warren, V. L., & Cairns, R. B. (1972). Social reinforcement satiation: An outcome of frequency
or ambiguity. Journal of Experimental Child Psychology, 13, 249-260.
Watson, J. B., & Rayner, R. (1920). Conditioned emotional reactions. Journal of Experimental
Watson, P. J., & Workman, E. A. (1981). The non-concurrent multiple baseline across-individu-
als design: An extension of the traditional multiple baseline design. Journal of Behavior
Therapy and Experimental Psychiatry, 12, 257-259.
Webb, E. J., Campbell, D. T, Schwartz, R. D., & Sechrest, L. (1966). Unobtrusive measures:
Nonreactive research in the social sciences. Chicago: Rand McNally.
Webb, E. J., Campbell, D. T, Schwartz, R. D., Sechrest, L., & Grove, J. B. (1981). Nonreactive
measures in the social sciences, (2nd ed.). Boston: Houghton Mifflin.
References 403
Weick, K. E. (1968). Systematic observational methods. In G. Lindzey & E. Aronson (Eds.)., The
handbook of social psychology, (Vol. 2, 2nd ed.). (pp. 357-451). Menlo Park, CA: Addison-
Wesley.
Weinrott, M. R., Garrett, B., & Todd, N. (1978). The influence of observer presence on classroom
behavior. Behavior Therapy P, 900-911.
Weinrott, M. R., Jones, R. R., & Boler, G. R. (1981). Convergent and discriminant validity of
five classroom observation systems: A secondary analysis. Journal of Educational Psychology,
73, 671-679.
Wells, K. C, Hersen, M., Bellack, A. S., & Himmelhock, J. M., (1979). Social skills training in
unipolar nonpsychotic depression. American Journal of Psychiatry, 136, 1331-1332.

Werner, J. S., Minkin, N., Minkin, B. L., Fixsen, D. L., Phillips, E. L., & Wolf, M. M. (1975).
"Intervention package": An analysis to prepare juvenile delinquents for encounters with police
officers. Criminal Justice and Behavior, 2, 55-83.
Whang, R L., Fletcher, R. K., & Fawcett, S. B. (1982). Training counseHng skills: An experimen-
tal analysis and social validation. Journal of Applied Behavior Analysis, 15, 325-334.
Wheeler, A. J., & Sulzer, B. (1970). Operant training and generalization of a verbal response form
in a speech-deficient child. Journal of Applied Behavior Analysis, 3, 139-147.
White, O. R. (1971). glossary of behavioral terminology Champaign, IL: Research Press.
/I
White, O. R. (1972), manual for the calculation and use of the median slope: A technique of
A
progress estimation and prediction in the single case. Eugene, OR: University of Oregon,
Regional Resource Center for Handicapped Children.
White, O. R. (1974). The "split middle": A "quickie" method of trend estimation. Seattle, WA:
University of Washington, Experimental Education Unit, Child Development and Mental
Retardation Center.
Wildman, B. G., & Erickson, M. T. (1977). Methodological problems in behavioral observation.
In J. D. Cone & R. P. Hawkins (Eds.), Behavior assessment: New directions in clinical
psychology (pp. 255-273). New York: Brunner/Mazel.
Williams, C. D. (1959). Case report: The elimination of tantrum behavior by extinction proce-
dures. Journal of Abnormal and Social Psychology, 59, 269.
Williams, J. G., Barlow, D. H., & Agras, W
S. (1972). Behavioral measurement of severe
depression. Archives of General Psychiatry, 27, 330-334.
Wilson, C. W, & Hopkins, B. L. (1973). The effects of contingent music on the intensity of noise
in junior high home economics classes. Journal of Applied Behavior Analysis, 6, 269-275.
Wilson, G. T, & Rachman, S. J. (1983). Meta-analysis and the evaluation of psychotherapy

outcome limitations and liabilities. Journal of Consulting and Clinical Psychology, 51, 54-64.
Wincze, J. P. (1982). Assessment of sexual disorders. Behavioral Assessment, 4, 257-271.
Wincze, J. P., & Lange, J. D. (1981). Assessment of sexual behavior. In D. H. Barlow (Ed.),
Behavioral assessment of adult disorders (pp. 301-329). New York: Guilford Press.
Wincze, J. P., Leitenberg, H., & Agras, W. S. (1972). The effects of token reinforcement and
feedback on the delusional verbal behavior of chronic paranoid schizophrenics. Journal of
Winkler, R, C. (1977). What types of sex-role behavior should behavior modifiers promote?
Winett, R. A., & Winkler, R. C. (1972). Current behavior modification in the classroom: Be still,
be quiet, be docile. Journal of Applied Behavior Analysis, 5, 499-504.

Wittlieb, E., Eifert, G., Wilson, E E., & Evans, I. M. (1979). Target behavior selection in recent
child case reports in behavior therapy. Behavior Therapist, 1, 15-16.
Wolery, M. & Billingsley, F. F. (1982). The application of Revusky's R^ test to slope and level
changes. Behavioral Assessment, 4, 93-103.
Wolf, M. M. (1978). Social validity: The case for subjective measurement or how applied
behavior analysis is finding its heart. Journal of Applied Behavioral A nslysis, 11, 203-215.
Wolf, M. M., Brinbrauer, J. S., Williams, T, & Lawler, J. (1965). A note on apparent extinction
of the vomiting behavior of a retarded child. In L. P. Ullmann & L. Krasner (Eds.). Case
studies in behavior modification (pp. 364-366). New York: Holt, Rinehart and Winston.
Wolf, M. M., & Risley, T. R. (1971). Reinforcement: Applied research. In R. Glaser (Ed.), The
nature of reinforcement (pp. 310-325). New York: Academic Press.
Wolfe, J. L., & Fodor, I. G. (1977). Modifying assertive behavior in women: A comparison of
three approaches. Behavior Therapy, 8, 567-574.
Wolpe, J. (1958). Psychotherapy by reciprocal inhibition. Stanford: Stanford University Press.
Wolpe, J. (1976). Theme and variations: A behavior therapy casebook. Elmsford, New York:
Pergamon Press.
Wolstein, B. (1954). Transference: Its meaning and function in psychoanalytic therapy. New
York: Grune & Stratton.
Wong, S. E., Gaydos, G. R., & Fuqua, R. W. (1982). Operant control of pedophilia. Behavior
Wood, D. D., Callahan, E. J., Alevizos, R N., & Teigen, J. R. (1979). Inpatient behavioral
assessment with a problem-oriented psychiatric logbook. Journal of Behavior Therapy and
Experimental Psychiatry, 10, 229-235.
Wood, L. E, & Jacobson, N. S. (in press). Marital disorders. In D. H. Barlow (Ed.), Behavioral
treatment of adult disorders. New York: Guilford Press.
Wright, H. E (1960). Observational child study. In R Mussen (Ed.), Handbook of research
methods in child development (pp. 71-139). New York: Wiley.
Wright, J., Clayton, J., & Edgar, C. L. (1970). Behavior modification with low-level mental
retardates. Psychological Record, 20, 465-471.
Yarrow, M. R., & Waxier, C. Z. (1979). Dimensions and correlates of prosocial behavior in young
children. Child Development, 47, 118-125.
Yates, A. J. (1970). Behavior therapy New York: Wiley.
Yates, A. J. (1975). Theory and practice in behavior therapy. New York: Wiley.
Yawkey, T. D. (1971). Conditioning independent work behavior in reading with seven-year-old
children in a regular early childhood classroom. Child Study Journal, 2, 23-34.
Yelton, A. R., Wildman, B. G., & Erickson, M. T. (1977). A probability-based formula for
calculating interobserver agreement. Journal of Applied Behavior Analysis, 10, 127-131.
Zeilberger, J., Sampen, S. E., & Sloane, H. N. (1968). Modification of a child's problem
behaviors in the home with the mother as therapist. Journal of Applied Behavior Analysis, 1,
47-53.
Zilbergeld, B., & Evans, M. B. (1980). The inadequacy of Masters and Johnson. Psychology
Today, 14, 28-43.
Zimmerman, E. H., & Zimmerman, J. (1962). The alteration of behavior in a special classroom
Journal of the Experimental Analysis of Behavior, 5, 59-60.
situation.
Zimmerman, J. Overpeck, C, Eisenberg, H., & Garlick, B. (1969). Operant conditioning in a
sheltered workshop. Rehabilitation Literature, 30, 326-334.
Subject Index
Actuarial issues, 62-63 Changing Criterion Design, 175,

Agoraphobic disorder, 55, 59, 326, 205-208, 319
329-330, 366 Classification, 26
Alcoholism, 145, 165, 170-171 Clinical significance, 35, 36, 45, 48,
Alternating Treatments Design, 65, 69, 282, 285, 320, 333, 369
95,99,210,211, 252-283,302, Component analysis, 110
319, 338, 344 Concurrent Schedule Design. See
Analysis of variance, 7, 56, 59, 60, 193, Simultaneous Treatment Design
287-290, 294 Confound, 19, 20, 142, 253, 256, 275
Anorexia Nervosa, 45-46, 69, 82, Control groups, 14, 56, 59, 60, 61, 143,
197-201, 343 226, 269
Anxiety, 34, 87, 136, 145, 241, 273 Correlation, 6, 17, 19, 28, 38, 45, 127
Assessment, 107-139 Correlogram, 288
direct, 108 Counterbalancing, 259, 260, 262, 263,
See also Repeated measurement 264, 269, 273, 274, 280, 284
Autism, 226-228, 232-233, 292, Criterion Reference Tests, 109, 110
354-355, 362, 366, 368-369 Critical Ratio Test, 6
Autocorrelation, 288, 293, 294, 295,
296, 299, 301, 302 Demand effects, 70, 134, 203
Averaging of results, 14-15, 16, 23, 54, Dependent variables, 10, 12, 17, 33, 35,
55, 60, 61, 66, 226 37, 39, 142, 236, 302
Depression, 15, 34, 35, 36, 54, 57-58,
Baseline, 39-45, 71-79 61, 64, 100, 109, 145, 146, 147,
Behavioral observation measures, 109, 154, 155, 156, 274, 275, 278, 366
110, 131, 146, 182 Deterioration, 16, 17, 36, 37, 44, 55,
behavioral products, 131-132 59, 64, 65, 74, 77, 88, 94, 104, 150,
codes, 125, 126, 130 152, 153, 154, 163, 228, 233, 328,
observers, 113, 115, 117, 118-122, 343
124-129, 130, 132, 282 Diagnostic category, 37
procedures, 113, 115, 116-118, 120, Differential attention, 347-362, 365, 366
129, 130 Direct observation. See Behavioral
settings, 109, 110, 112-114 observation
Between Series Design, 254 Drug evaluation, 28, 87-88, 100, 101,
Bidirectionality, 206 170, 183-192, 209, 249-251, 264
Blocking, 45
Enuresis, 98, 230-232
Carry-over effects, 96, 99-101 Error, 3, 5, 6, 26, 33
Case study, 1, 8-13, 17, 19, 22, 23, Equivalent Time Series Design, 28,
24-25, 56, 140-142, 351 157-166
Celeration line, 313-315, 316, 317 Ethics, 14, 74, 90, 96, 98, 100, 153,
Central tendency, 5 209, 249
405
1
Expectancy effects, 42, 184, 189, 219 classical conditioning, 39, 40,90
Experimental analysis of behavior, 8, operant conditioning, 8, 30, 99
29-31 Logical generalization, 253, 333, 369
Experimental criterion, 285, 286
Experimental psychology, 1, 2-5, 6, 14, Maintenance, 68, 105-106, 144, 230,
30, 35 236, 239, 248, 250
Matching, 15, 54, 68, 213, 214
Factor analysis, 6 Merit Method, 6
Factorial Design. See Analysis of Mixed Schedule Design, 255
variance Multi-Element Baseline Design, 254,
Field testing, 365, 367 255, 299, 319
Follow-up, 44, 89, 110, 145, 150, 151, Multi-Element Experimental Designs, 30
234, 236, 247, 248 Multiple Baseline Design, 9, 64, 66, 88,
Functional manipulation, 260 95, 101, 102, 106, 164, 209-251,
275,281, 308, 309, 311, 321, 333
Generality of findings, 2, 4, 7, 8, 14, across behaviors, 215-230, 247, 344
16, 25,28, 32, 33,49-66, 84, 112, across individuals, 244
113, 127, 130, 150, 153, 154, 162, across settings, 238-244, 247, 249
204, 205, 211, 216, 226, 232, 239, across subjects, 230-238, 249, 251,
241, 247, 252, 257, 260, 272, 325, 278, 343
325-371 Multiple Probe Technique, 245-248
Group comparison. See Group design Multiple Schedule Design, 254, 255
Group Comparison Design, 1, 2, 3, Multiple treatment interference, 143,
5-8, 11-13, 14, 15, 16, 17, 18, 19, 153, 179, 205, 256-263, 272, 273,
20, 21, 22, 23, 28, 29, 30, 31, 33, 281
35, 36, 51-66, 99, 108, 167, 178,
179, 191, 193, 205, 226, 238, 252, Naturalistic studies, 2, 17, 18-20, 21
259, 286, 287, 291, 320, 321, 365, Nonconcurrent Multiple Baseline
370 Design, 244, 248
Norm Reference Tests. See Criterion
Habituation, 138 reference tests
Headache, 135, 136, 161-162 Normal distribution, 3, 5, 305
Homosexuality, 10, 39-42, 70, 86,
103-105, 147, 334-339 Obsessive Compulsive Disorder, 15, 16
Operational definition, 1 1
Independent variables, 9, 10, 17, 18, 27,
28, 29, 30, 33, 34, 35, 39, 48, 67, Paranoid delusions, 26
154 Patient Uniformity Myth, 16
Independent verification, 259 Percent of success, 12, 17, 19, 56
Individual differences, 5, 6, 7 Period treatment design, 175, 206
Instrumentation, 108 Phase, 26, 67, 72, 93, 95-101, 154, 162,
Intensive Design, 28 165, 280, 286, 292, 295, 299, 301,
Interaction effects, 193-205, 249, 272 302, 316, 319
Intelligence, 5 Phobia, 53, 82, 195-197, 201, 216-219,
tests, 6 273, 284, 333, 343, 346, 347
Intrasubject averaging, 45-48 Physiological measures, 108, 131,
Introspection, 3-4 135-138, 150
Irreversible procedures, 101-105 Physiological psychology, 1, 2-5, 8, 23
Placebo effects, 39, 60-61, 75, 78, 87,
Law of Initial Values, 138 101, 104, 105, 141, 183, 184, 185,
Learning Theory, 4, 6, 30, 31 186, 187, 188, 189, 190, 191, 192,
Subject Index 407
209, 249, 251, 255, 330, 331, 333, Scientist-Practitioner Split, 21-22
335 Self-report measures, 70, 108, 109 131,
Population, 8, 16, 305 132-135, 136, 150, 218-219, 284
Post Traumatic Stress Disorder, 241 behavioral, 133
Probe measures, 241 questionnaires, 108, 109, 133-134
Process research, 2, 17, 20-21, 23, 25, êlf-monitoring, 108, 109, 133,
26, 27, 38 134-135, 203, 239
structured interviews, 108, 109
Quasi-Experimental Designs, 27-28, 71, Serial dependency, 287-290, 295, 296,
142, 143, 186, 206, 249 299, 300, 301, 302, 305-306
Questionnaires, 29 Sexual disorders, 86, 194, 220-222, 367
Simultaneous Replication Design, 226,
Random assignment, 15, 18, 19, 287 254
Random sampling, 52-54, 55, 65, 305 Simultaneous Treatment Design, 255,
Randomization Design, 254, 255 282-284, 319
Randomization Tests, 302-308, 319, 320 Social psychology, 30
Reactivity, 118, 120, 130, 135, 143, 245, Social validation procedures
247, 282 social comparison, 109, 110
Regression techniques, 110 subjective evaluation, 109, 110
Reliability, 68, 109, 114, 118, 122, Splitmiddle technique, 312-319, 321
124-129, 134, 158, 239, 286, 290, Spontaneous remission, 12, 19, 42
293, 308, 322, 325, 326, 327, 333, Statistical analysis, 3, 5, 6, 22, 28, 34,
338, 341, 346, 364 36, 126, 128, 129, 255, 257, 281,
Repeated measurement, 3, 4, 20, 21, 26, 282, 321
27, 30, 32, 37-38, 39,41,42,43, descriptive statistics, 3, 6, 22, 319,
44, 48, 64, 65, 67, 68-71, 72, 108, 321
110, 142, 179, 245, 287 inferential statistics, 1, 7-8, 16, 53,
Replication, 5, 11, 25, 26, 33, 51, 60, 65, 252, 318, 319, 321
56-62, 111, 143, 153, 154, 156, single-case, 285-324
162, 165, 179, 193, 196, 200, 204, Statistical significance, 35, 36, 48, 58,
205, 212, 225, 226, 232, 241, 244, 294, 302, 303-304, 308, 309, 313,
253, 260, 264, 286, 325-371 316, 318, 319-320
clinical, 325, 366-369 Structuralism, 4
direct, 50, 58, 61, 325, 326-347, 351,
364, 365
Target behaviors, 107, 108, 109-112,
systematic, 56, 59, 61, 62, 63, 101,
126, 129, 131, 134, 142, 145, 146,
325, 334, 339, 343, 344, 346,
156, 158, 187, 212, 228, 251, 309
347-354, 363-366
Term Series Design, 27
Representative case, 25-26
Therapeutic criterion, 285, 286
Response dimensions, 114-116
Time sampling, 70, 222, 224
Response guided experimentation, 38
Time Series Analysis, 71, 142, 288,
Response specificity, 138
296-302, 308, 319, 321, 353
Reversal design, 9, 30, 67, 88-95, 101,
Trend, 37, 38, 45, 73, 77
209, 210
Trend analysis, 28
Rn Test of Ranks, 308-312, 320
Triple response system, 108, 132
Sample, 8, 15, 16, 107

Sampling theory, 1, 8 Validity, 109, 129-131, 134, 135, 137
Schizophrenic Disorder, 15, 52, 80, 87, construct, 130
91, 167-168, 187, 205, 339-343, content, 130
366 convergent. 111
ecological, 114 272, 326-327, 339. 344. 346.

external, 28, 57, 143, 252, 260 362
internal, 24, 28, 57, 141, 143, 154, relationship. 17, 18, 19, 20
252 therapist, 17, 18, 19, 20, 25, 329, 361
See also generality of findings uncontrolled, 34. 35, 141
Variability, 5. 6, 7, 32-50, 72-73, 77, Visual inspection, 290-292, 293, 297,
100, 125, 129, 130, 157, 206, 225, 321. 322
262, 292, 301, 322
intersubject, 6, 8, 14, 17, 36, 39, 41, Withdrawal Design, 9, 26, 28, 30, 45,
50, 58,60,61.252. 271,272, 59, 66, 67, 74, 79, 88-95, 97, 98.
292, 338, 346, 369 99. 100, 101, 102, 106, 140-208,
intrasubject, 3, 38, 39, 48, 50. 63, 64, 209, 210, 212, 239, 241, 243, 249,
65, 205, 292 250, 269, 272, 280, 332-333, 340,
Variables, environmental, 33, 35, 59, 344
112, 137 Within Series Design, 254
patient, 17, 18. 19, 20, 120, 204-205, Within Subject Designs, 66, 179, 183
Name Index
Abel, G. G., 45, 46, 47, 198, 199, 201, Bailey, J. S., 175
262 Bakeman, R., 114, 116
Adams, H. E., 135, 139, 359 Baker, B. L., 357, 365
Agras, W. S., 30, 39, 40, 41, 42, 43, 45, Ban, T, 100
46,47, 56, 69, 71, 80, 81, 82, 85, Bandura, A., 73, 99, 101, 153
86, 102, 103, 104, 136, 137, 138, Barker, R. G., 114
147, 150, 154, 155, 156, 166, 174, Barlow, D. H., 9, 15, 22, 24, 25, 30,
175, 176, 183, 188, 189, 194, 195, 35, 39,40,41,42,43,45,46,47,
197, 201, 205, 255, 259, 273, 274, 61, 67, 69, 70, 71, 73, 74, 79, 80,
278, 282, 327, 329, 330, 332, 336, 82, 86, 88, 95, 96, 102, 103, 104,
337, 341, 342, 352 133, 136, 137, 138, 140, 141, 142,
Aikins, D. A., 247 143, 150, 151, 152, 153, 158, 164,
Alevizos, P. N., 115 166, 167, 184, 185, 194, 196, 198,
Alden, S. E., 355, 359 199, 201, 207, 209, 212, 253, 254,
Alford, G. S., 71, 147, 154, 155, 156, 255, 256, 257, 261, 263, 268, 270,
175, 214, 220, 221, 352 274, 280, 281, 282, 327, 329, 330,
Allen, K. E., 89, 90, 94, 354, 356, 358 332, 333, 336, 337, 347, 352, 366,
Allison, M. G., 214 367, 369, 370, 371
Allport, G. D., 24, 62 Barmann, B. C, 214, 230
Altmann, J., 116, 117 Barnes, K. E., 360
Anderson, R. L., 322 Barnett, J. T, 213
Andrasik, 191, 192 Baroff, G. S., 355
Angell, M. J., 215 Barrera, R. D., 268
Armstrong, M., 357 Barrett, R. P., 265, 267, 270, 271. 272,
Arnold, G. R., 357 282
Arrington, R. E., 114, 118, 122 Barrios, B. A., 132
Ashem, R., 141 Barton, E. S., 222, 223, 224, 275
Atiqullah, M., 287 Bates, P, 214, 224, 225, 236
Ault, 99, 108 Baum, C. G., 120
Austin, J. B., 150, 152 Beauchamp, K. L., 214, 230
Axelrod, S., 132 Beck, A. T, 146, 275
Ayllon, T, 64, 70, 166, 167, 168, 170, Beck, S. J., 8, 235
214, 348, 349, 351 Becker, D., 153, 154, 235
Azrin, N. H., 64, 70, 106, 122, 166, Becker, R., 69
167, 168, 170, 265, 266, 349 Becker, W. ., 357, 358
Bellack, A. S., 28, 68, 87, 133, 139,
Baker, T. B., 108 183, 191, 192, 214, 215, 217, 218,
Baer, D. M., 62, 63, 71, 88, 94, 102, 247, 248, 347
114, 116, 128, 138, 139,209,210, Bemis, K. M., 72
212, 214, 222, 223, 245, 246, 247, Berberich, J. P, 368
266, 286, 290, 322, 323, 356, 357, Berger, L., 17
358, 360 Bergin, A. E., 15, 16, 19, 21, 22, 23,
409
25. 33, 35, 36,41, 51, 54, 55,61, Bruce, C., 358
63, 74, 366, 370 Brunswick, E., 53
Berk, R. A., 126, 127 Bryan, K. S., 247
Berler, E. S., 214 Bryant, L. E., 214
Bernal, M. E., 112 Bucher, B., 268
Bernard, M. E., 175, 202 Buckley, N. K., 156, 157, 352
Bickman, L., 113 Budd, K. S., 214, 360
Bijou, S. W., 95, 99, 108, 117, 118, Buell, J. S., 89, 90, 354, 356, 357
356, 357 Bugle, C., 106
Billingsley, E E, 308, 318, 323 Burgio, L. D., 214
Birkimer, J. C, 129 Burke, M., 162, 163, 360, 369
Birney, R. C, 6 Butcher, J. N., 19, 31
280
Bittle, R., 255, 266, Buys, C. J., 359
Blackburn, B. L., 215
Blanchard, E. B., 71, 136, 263, 352 Cairns, R. B., 363
Blewitt, E., 171, 172 Calhoun, K. S., 139
Blough, P. M., 258 Callahan, E. J., 102, 104, 115
Blunden, R., 171, 172 Campbell, D. T, 27, 28, 45, 57, 71,
Boer, A. P., 118 111, 121, 126, 132, 138, 140, 142,
Boler, G. R., 131 143, 153, 157, 244, 252, 256
Bolger, H., 9 Capparell, H. V, 191, 192
Bolstad, O. D., 120, 121, 125, 129, 131, Carey, R. G., 268
139 Carkhuff, R. R., 167, 168, 169
Boone, S. E., 52 Carlson, C. S., 357
Bootzin, R. R., 94, 99 Carmody, T. B., 135
Borakove, L. S., 110 Carr, A., 243
Boring, E. G., 3, 4, 6 Carter, V, 358
Bornstein, M. R., 214, 215, 217, 218, Carver, R. P, 110
347 Cataldo, M. E, 360
Bornstein, PH., 108, 113, 133, 135 Catania, A. C, 212
Bower, S. M., 295 Celso-Goyos, A., 267
Bowdler, C. M., 25 Chai, H., 320
Box, G. E. P, 300, 301, 306 Chapin, H. N., 198, 199, 201, 275
Boyer, E. G., 139 Chapin, J. P, 45, 46, 47
Boykin, R. A., 139 Christian, W. P, 214, 231, 232
Bradley, L. A., 136 Christie, M. H., 137
Bradlyn, A. S., 147, 149 Chassan, J. B., 15, 16, 20, 28, 35, 36,
Brady, J. P, 352 55, 87, 95, 99, 100, 183, 184, 185
Brawley, E. R., 357, 358 Ciminero, A. R., 139
Breuer, J., 9 Clairborne, M., 343, 345
Breuning, S. E., 214, 249, 250, 251 Clark, R., 358
Bridgwater, C. A., 108 Clayton, J., 359
Brill, A. A., 10 Coates, T. J., 136
Brinbauer, J. S., 209, 265, 352, 366 Cohen, D. C, 22, 70
Broden, M., 355, 356, 357, 358 Cohen, J., 127
Brody, G. H., 287 Cohen, S., 293
Brookshire, R. H., 352 Colavecchia, B., 268
Brouwer, R., 215 Coleman, R. A., 175
Brown, J. H., 129 Coles, E. M., 138
Brown, R. A., 355, 359 Conderman, L., 358
Browning, R. M., 142, 256, 283 Cone, J. D., 108, 109, 115, 118, 122,
Name Index 411
124, 125, 127, 130, 131, 139 Dunlap, G., 5, 214

Conger, J. C, 127, 358 Dyer, K., 214, 231, 232, 233
Connis, R. T, 106 D'Zurilla, T. J., 110
Conover, W. J., 304, 306, 307
Conrin, J., 175, 180, 182 Edelberg, R., 137
Cook, T. D., 45, 121, 142, 143, 153, Edgar, C. L., 359
252 Edgington, E. S., 38, 52, 53, 55, 65, 71
Cormier, W. H., 355, 358 253, 254, 255, 282, 302, 306, 319,
Cornell, J. E., 254, 263 323, 324, 328
Corriveau, E. P., 350 Edwards, A. L., 66, 179, 330, 343
Corte, H. E., 266, 355, 359 Egel, A. L., 214
Cossairt, A., 360 Eifert, G., 109
Costello, G. C, 29 Eisenberg, H., 265, 266
Cranston, S. S., 213 Eisler, R. M., 69, 71, 82, 85, 102, 147,
Creer, T. L., 320 148, 154, 155, 156, 165, 166, 170,
Cristler, C, 213 171
Cronbach, L. J., 116, 124, 130, 134 Elashoff, J. D., 296, 301
Cummings, L. T, 106 Elkin, T. E., 69, 80
Curran, J. P., 350 D. P, 357
Ellis,
Cuvo, A. J., 110, 213, 214 Emerson, M., 159, 300
Emery, G., 275
Dalton, K., 101 Emmelkamp, P M., 135, 327, 328
Daneman, D., 235 Epstein, L. H., 71, 144, 161, 214, 235,
Davidson, P O., 29 236, 355, 359
Davis, C. M., 87, 137 Erikson, M. T, 118, 129
Davis, E, 300 Esveldt-Dawson, K., 266, 272
Davis, J., 187, 191 Evans, I. M., 109, 134, 153, 154, 358,
Davis, K. v., 183, 186 367
Davis, X, 159 Everett, P B., 292
Davis, V. J., 249 Ewalt, J. 183
Davison, G. C, 24, 141, 355, 356 Eyberg, S. M., 110
Dawson, J. H., 214, 239, 240, 268 Eysenck, H. J., 10, 11, 21
DeProspero, A., 293 Ezekiel. M., 322
deSantamaria, M. C, 215, 236, 237
Dignam, P J., 349 Fabry, B. D., 128
Dillon, A., 362, 363 Fairbank, J. A., 214, 241, 242
Dinoff, M., 349 Farkas, G., 235
Dobb, L. W., Ill Fawcett, S. B., 215
Dobes, R. W., Ill Ferguson, D. B., 214, 250, 251
Dobson, W. R., 350 Feuer stein, M., 135
Doke, L. A., 122, 256, 266 Fialkov, M. J., 202, 204
Dollard, J., Ill Fisher, E. B., 7, 267
Domash, M. A., 213, 214, 243 Fisher, R. A., 255
Donahoe, C. P, 52 Fiske, D. W., Ill
Dotson, V. A., 127 Fjellstedt, N., 115
Drabman, R. S., 139, 214, 215 Flanagan, B., 263
Dredge, M., 358 Fleiss, J. H., 126, 127
Dressel, M. E., 120 Fleming, I. R., 116
Dukes, W. E, 24, 56 Fleming, R. S., 358
Dulaney, S., 82, 83 Fletcher, R. K., 215
duMas, E M., 8 Foa. E. B., 334
Fodor, I. G., 134 Gregory, P. R., 215, 239, 240

Forehand, R. L., 110, 120, 133, 139, Greenspoon, J., 352
348, 362, 363 Greenwald, A. G., 259
Forman, J. B. W., 52 Greenwood, C. R., 122
Foster, S. L., 115, 118, 122, 125, 127, Grinspoon, L., 183
130, 131 Gross, A. M., 214
Fox, R., 159, 300, 322 Grove, J. B., 138
Foy, D. W., 52 Guess, D., 222, 223, 247
Frank, J. D., 10 Gullick, E. L., 352
Freitas, L., 368
Freud, S., 9 Hadley, S. W., 36
Frick, T, 126, 127 Hake, D. E, 255, 259, 266, 280, 359
Fuqua, R. W., 215 Hall, C., 214, 299
Furlong, M. J., 293, 302 Hall, R. v., 132, 158, 159, 174, 175,
206, 207, 213, 300, 355, 356, 357,
Garcia, E., 222 359, 360
Gardner, W, 129 Hallahan, D. P., 267
Garfield, S. L., 35 Halle, J. W., 214
Garlick, B., 265, 266 Hamilton, S. B., 135
Garrett, B., 267, 280 Hammer, D., 139, 215
Garton, K. L., 116, 181 Haney, J. L., 113, 214,233,234
Gaydos, G. R., 215 Harbert, T. L., 102, 150, 151, 152, 352
Geer, J. H., 133 Harris, E R., 89, 90, 94, 354, 356, 357,
Geesey, S., 266, 278, 282 358
Gelder, M. G., 210 Hart, B. M., 89, 90, 354, 356, 357
Gelfand, D. M., 350 Hartmann, D. R, 107, 108, 109, 111,
Gelfand, S., Ill, 112, 116, 118, 121, 112, 116, 117, 118, 120, 121, 122,
138, 350 123, 125, 126, 127, 129, 130, 132,
Geller, E. S., 292 138, 139, 175, 206, 254, 296, 299,
Gendlin, E. T, 20 301, 302
Gentile, J. R., 295 Hatzenbuehler, L. C., 362
Gerwitz, J. L., 356 Haughton, E., 349
Gilman, A., 209 Hawkins, R. R, 95, 99, 107, 109, 110.
Glass, G. S., 6, 101 Ill, 118, 127, 128, 130, 132, 134,
Glass, G. v., 287, 293, 296, 299, 300 138, 356
Glazeski, R. C., 120 Hay, L. R., 214, 302
Goetz, E. M., 116 Hayes, S. C., 9, 71, 95, 108, 110, 131,
Goldiamond, I., 122 175, 206, 207, 208, 253, 255, 256,
Goldfried, M. R., 110, 130 257, 262, 268, 274, 276, 277, 280
Goldsmith, L., 159, 300 Haynes, S. N., 108, 113, 114, 120, 121,
Goldstein, M. K., 353 122, 130, 132, 133, 135, 136, 137,
Goodlet, G. R., 358 138, 139
Goodlet, M. M., 358 Hasazi, J. E., 360
Goodman, L. A., 209 Hasazi, S. E., 360
Gorsuch, R. L., 299 Hendrickson, J. M., 159, 160
Gotestam, K. G., 215 Heninger, G. R., 101
Gottman, J. M., 139, 142, 213, 293, Henke, L. B., 356
296, 299, 302 Henson, K., 265, 266
Grayson, J. B., 334 Hemphill, D. R, 161
Green, J. D., 268, 360 Herbert E. W., 351, 360, 361, 362, 363,
Greenfield, N. A., 137 364
Name Index 413
Herman, S. H., 39, 40, 41, 42, 43, 334, Johnson, M., 110, 120, 121, 125,
S.
336, 337, 338 129, 131,139,266
Hernstein, R. J., 212 Johnston, J. M., 31, 37, 72, 90, 94, 95,
Hersen, M., 25, 35, 61, 67, 68, 69, 70, 96, 100, 111, 128, 132, 175, 182,
71, 73, 74, 79, 80, 82, 85, 86, 88, 291, 347, 354
94, 95, 96, 102, 105, 133, 137, 139, Johnston, M. K., 356, 357
140, 142, 144, 146, 148, 150, 152, Johnstone, G., 267
153, 154, 155, 156, 158, 161, 164, Jones, R. R., 125, 131, 290, 293, 296,
165, 166, 167, 170, 171, 175, 183, 297, 299, 301
184, 185, 191, 192, 209, 212, 214, Jones, R. T, 214, 233, 234
215, 217, 218, 228, 229, 247, 248,
347, 352, 366 Kanowitz, J., 121
Hickey, J. S., 108 Katz R. C., 214, 230
Hilgard, J. R., 213 Kaufman, K. E, 175
Himmelhock, J. M., 68, 347 Kazdin, A. E., 9, 19, 24, 25, 30, 31, 53,
Hinson, J. M., 258 56, 59, 60, 67, 88, 94, 95, 99, 101,
House, A. E., 126 102, 105, 106, 109, 110, 112, 113,
House, B. J., 126 115, 118, 120, 121, 130, 132, 139,
Horner, R. D., 245, 246, 349 141, 142, 153, 162, 202, 204, 206,
Home, G. P., 299, 302 209, 211, 212,214,215, 216, 223,
Hopkins, B. L., 116, 175, 179, 355, 228, 229, 234, 235, 247, 254, 256,
358, 360 260, 261, 266, 267, 278, 279, 282,
Honing, W. K., 38, 212 286, 290, 291, 292, 307, 318
Homer, A. L., 138 Kane, M., 214, 241, 242
Holz, W., 122 Keefauver, L. W., 175, 202
Holtzman, W. H., 322 Kelley, C. S., 354, 356
Holmes, D. S., 356 Kelly, J. A., 149, 214, 226, 343, 345
Hollon, S. D., 72 Kelly, M.
G., Ill, 115, 117, 126, 147
Holmberg, M., 114 Kendall. P C., 19, 31, 116
Holm, R. A., 115 Kennedy, R. E., 215, 301
Hollenbeck, A. R., 121, 127 Kent, R. N., 118, 121
Hollandsworth, J. G., 120 Kernberg, O. E, 18
Hoffman, A., 320 Kessel, L., 10
Hodgson, R. J., 333, 334 Kiernan, J., 110
Hocking, N., 355, 357 Kiesler, D. J., 16, 17, 18, 20, 49, 55, 60
Hoch, P. H., 17, 20 Kirby, E D., 360
Hubert, L. J., 127, 302 Kircher, A. S., 266
Huitema, B. E., 299 Kirchner, R. E., 215, 243
Hundert, J., 214 Kirk, R. E., 307
Hutt, C, 111, 112 Kistner, J., 215
Hutt, S. J., Ill, 112 Klein, R. D., 295
Hyman, R., 10, 17 Knapp, T. J., 293
Kneedler, R. D., 267
Inglis, J., 29 Koegel, R. L., 106, 214, 215, 226, 227,
Iwata, B. A., 267-268 368, 369
Kopel, S. A., 209, 211, 212, 216
Jackson, D., 355, 357 Kraemer, H. C, 55, 117
Jacobson, N. S., 353, 363 Krasner, L., 30, 57, 94, 99, 141
Jarrett, R. B., 268, 274, 276, 277 Kratchowill, T. R., 31, 67, 142, 175,
Jayaratne, S., 31 202, 287, 296, 301, 324
Jenkins, G. M., 301 Kulp, S., 117
Kuypers, D. S., 357 Lyman, R. D., 132

Kwee, K. G., 135
MacDonough, I. S., 53, 71, 72
Lacey, J. I., 138 Mackenzie-Keating, S. E., 268

Lambert, M. J., 36 Madsen, C. H., 357
Lang, P. J., 108 Maisto, S., 27
Lange, J. D., 69 Makohoniuk, G., 121
Larson, L., 243 Malaby, J., 175
Last, C. G., 268 Malan, D. H., 18
Laughlin, C, 343, 345 Malone, J. C, 258
Lawler, J., 352 Malow, R., 134
Laws, D. R., 355, 359 Mandell, M. R, 101
Lawson, D. M., 145 Mandell, R. M., 101
Lazarus, A. A., 24, 141 Mann, R. A., 166, 174, 175, 176, 266
Leaf, R. B., 110 Manning, P. J., 215
LeBlanc, J. M., 360 Mansell, J., 244
Leitenberg, H., 25, 29, 30, 45, 46, 47, Marascuilo, L. A., 302
59,69, 80, 81, 86, 89,90, 95, 101, Margolin, G., 353, 363
102, 103, 104, 115, 137, 138, 151, Marks, I. M., 15, 210, 215, 329
166, 174, 175, 176, 189, 194, 195, Marshall, A. M., 215
196, 197, 198, 199, 201, 205, 211, Marshall, K. J., 267
215, 255, 274, 327, 329, 330, 341, Martin, G. L., 266, 267
342, 352 Martin, L, 137
Lentz, R. J., 114, 117, 123, 138, 350, Martin, R J., 132
351, 363 Martindale, A., 117
Lev, J., 6 Mash, E. J., 107, 109, 110, 121, 131,
Levin, J. R., 302, 324 133, 139
Levy, R. L., 31, 67 Mastantuono, A. K., 215, 228, 229
Lewin, K., 7 Matherne, R M., 102, 352
Lewinsohn, P. M., 133, 275 Matson, J. L., 202, 204, 215, 266, 272
Libet, J., 133 Mavissakalian, M. R., 70, 136, 196,
Liberman, R. R, 87, 100, 183, 187, 247, 327, 330, 332
188, 190, 191,216,219,350,353, Max, L. W., 10
360 May, R R. A., 18
Lick, J. R., 134 Mayer, R. G., 117, 348
Light, F. J., 127 McCallister, L. W., 358
Lillisand,D. B., 133 McCleary, R., 302
Lind, D. L., 352 McCoy, D., 266
Lindsey, C. J., 132 McCullough, J. D., 254, 259, 263, 269
Lindsley, O. R., 183 McDaniel, M. H., 254, 263
Linehan, M. M., 130 McDonald, M. L., 133
Lloyd, J. W., 267 McFall, R. M., 72, 133, 213
Lobitz, G. K., 266 McFarlain, R. A., 183
Locke, B. J., 266, 355, 359 McGonigle, J. J., 260, 261, 267
Long, J. D., 368 McKnight, R L., 268, 274, 276, 277,
Lovaas, O. I, 169, 368, 369 282
Lowenstein, L. M., 112 McLaughlin, T. F, 175
Luborsky, L., 20, 54 McLean, A. R, 258
Luce, S. C., 214, 231,232 McNamara, J. R., 53, 71, 72, 116, 132
Lund, D., 355, 357 McNees, M. R, 243
Luper, H. L., 358 Melin, L., 215
Name Index 415
Mendelsohn, M., 146 Osborne, J. G., 258, 259

Metcalfe, M., 43 Overpeck, ., 265, 266
Meuller, R. K., 254, 263 Owen, M., 159, 300
Michael, J., 286, 290, 348, 351
Miller, P. M., 69, 97, 98, 111, 158, 165, Page, T. J., 267
166, 170, 171 Palotta-Cornick, A., 267
Mills, H. L., 330, 332, 369, 371 Panyan, M., 357
Mills, J. R., 330, 332 Paris, S. G., 363
Minkin, N., 109 Parsonson, B. S., 286
Mischel, W., 134, 275 Patterson, G. R., 125, 130, 139, 215,
Mitchell, S. K., 125, 126, 127, 299, 358 343, 345, 348, 363
Mock, J., 146 Paul, G. L., 9, 10, 20, 53, 55, 56, 57,
Montague, J. D., 138 60, 114, 117, 118, 121, 123, 131,
Monti, P. M., 350 138, 350, 351, 362
Moon, W., 87, 187, 191 Pavlov, I. R, 4
Moore, J., 187, 191 Pear, J. J., 266
Moore, R. C, 87, 104 Peckham, R D., 287
Morrison, D. C, 355, 356 Pendergrass, V. E., 82, 84, 174
Moses, L. E., 306 Pennypacker, H. S., 31, 37, 100, 111,
Moss, G. R., 166 132, 138, 175, 182, 291, 347
Mowrer, O. H., Ill Perloff, B. E, 368
Mulick, J. A., 266 Pertschuk, M. J., 343
Munford, P R., 360 Peterson, C. R, 109
Peterson, L., 95, 99, 108, 138
Neale, J. M., 52 Peterson, R. E., 355, 356
Nee, J., 52 Peterson, R. E, 94, 95, 357, 358
Neef, N. A., 267 Pettigrew, E., 355, 359
Nelson, R. O., 9, 108, 110, 114, 131, Phillips, E. L., 175
135, 139, 214, 268, 274, 276, 277, Pinkston, E. M., 360
368 Roche, ., 215
Neucherlein, K. H., 350 Poling, A. D., 249
Nathan, R E., 112 Pollio, H. R., 357
Nau, R A., 268 Pomerleau, O. E, 343
Nay, W. R., 113, 117, 118, 121, 123, Poole, A. D., 116
130, 132, 133, 135, 136 Porcia, E., 300
Nielson, T. J., 357 Porterfield, J., 171, 172
Nietzel,M. T, 136, 137 Powell, J., 117, 259
Nordquist, V. M., 359 Power, C. T, 117
Nunnally, J., 124 Prokop, C. K., 136
O'Brien, E, 106, 214, 230, 265, 266, Rabon, D., 357

282 Rachlin, H., 258
O'Brien, J. T, 268 Rachman, S. J., 6, 333, 334
O'Leary, K. D., 121, 129, 153, 154, Ramp, E., 82, 83
175, 193, 353, 358, 361 Rapport, M. D., 202, 203, 204
Ollendick, H., 215, 238, 265, 266,
T. Rast, J., 175, 182
270, 271, 272, 273, 278, 282, 338 Ravenette, A. T, 26, 28
Olson, D. G., 67 Ray, W. J., 137, 138
Oltmanns, X, 52 Rayner, R., 9
O'Neill, M. J., 214, 250, 251 Rees, L., 101
Orne, M. T, 70 Reese, N. M., 360
Redd, W. H., 265, 266, 352 Schweid, E., 95, 356

Redfield, J. P., 121 Schwitzgebel, R. L., 139
Reid, J. B., 117, 121, 123, 124, 125, Sears R. R., Ill
131, 299 Sechrest, L., 120, 132, 138
Revusky, S. H., 308, 311, 312 Semmel, M. I., 126, 127
Reynolds, G. S., 212, 258, 356, 357 Shader, R., 183
Reynolds, N. J., 357 Shapiro, D. A., 6
Richard, H. C, 132 Shapiro, E. S., 260, 261, 264, 265, 270,
Richman, G. S., 214 271, 272, 280, 282
Rickard, H. C, 349 Shapiro, M. B., 26, 27, 28
Risley, T. R., 64, 71, 142, 143, 162, 212, Shaw, B. J., 275
265, 266, 285, 357 Sheldon-Wildgen, J., 214
Riva, M. T, 213, 214 Sherman, J. A., 214, 247
Roberts, M. W., 362 Shields, E, 360
Roden. A. H., 295 Shigetomi, C., 132
Rogers, C. R., 20 Shine, L. C, 295
Rogers- Warren, A., 114 Shontz, E C., 25, 56
Rojahn, J., 266 Shores, R. E., 159
Roper, B. L., 107, 121, 138 Shrout, R E., 127
Rosen J. C, 215 Shuiler, D. Y, 116
Rosenbaum, M. S., 139 Sidman, M., 5, 15, 30, 33, 49, 58, 72,
Rosenzweig, S., 8 77, 90, 100, 129, 212, 254, 255,
Ross, A. O., 348 259, 260, 262, 291, 325, 326, 329,
Rossi, A. M., 112 341, 347, 364, 365
Rothblum, E., 215 Simmons, J. Q., 162, 368
Roxburgh, P. A., 183, 184 Simon, A., 214
Rugh, J. E., 139 Simon, J., 139, 228, 229
Rush, A. J., 275 Simpson, M. J. A., 115
Rusch, E R., 105, 106, 122 Singh, N. N., 215, 239, 240, 268
Russell, M. B., 112 Skiba, E., 355, 359
Russo, D. C, 106, 215, 226, 227, 360 Skinner, B. E, 5, 30, 59
Rychtarik, R. G., 133, 135 Slavon, R. E., 215
Sackett, G. R, 114, 118 Sloane, H. N., 99, 357, 358
St. Lawrence, J. S., 147, 148 Smeets, R M., 358
Sajwaj, T, 162, 163, 164, 212, 360, Smith, C. M., 266
361, 362, 363 Smith, M. L., 6
Sameoto, D., 268 Smith, R C, 116
Sampen, S. E., 99, 358 Smith, v., 216, 219
Sanders, S. H., 214, 220, 221, 287 Solnick, J. v., 209
Sanson-Fisher, R. W., 116, 118 Solomon, R, 112
Saudargas, R. A., 153, 154, 358 Sohis, W. A 202, 203
Schnelle, J. E, 243 Sowers, J., 106
Schaeffer, B., 368 Spangler, R E, 215
Schleinen, S. J., 110 Sperling, K. A., 358
Scheffe, H., 287, 288 Spitzer, R. L., 52
Schindele, R., 31 Spradlin, J. E., 214
Schofield, L., 70 Sprague, R. L., 183
Schreibman, L., 369 Stachowiak, J. G., 358
Schroeder, S. R., 266 Stanley, J. ., 27, 28, 45, 57, 71, 140,
Schumaker, J., 247 142, 143, 157, 244, 252, 256
Schutte, R. C., 116, 181, 355, 358 Stravynski, A., 215
Schwartz, R. D., 132, 138 Steinman, W. M., 266
Name Index 417
Steketee, G., 334 Ulrich, R., 82, 83, 122

Stern, R. M., 137 Underwood, B. J., 8, 27, 49, 51, 56
Sternbach, R. A., 137 Urey, J. R., 215
Stilson, D. W., 5
Stoddard, P., 357 Van Biervliet, A., 215
Stokes, T. E, 139, 215 Van Hasselt, V B., 71, 88, 209, 215,
Stoline, M. R., 299 228, 229
Stover, D. O., 142, 283 Van Houton, R., 268
Strain, P. S., 159, 160 Varni, J. W., 360
Striefel, S., 247 Vaught, R. S., 290, 296, 299
Strupp, H. H., 14, 15, 16, 19, 20, 21, Veenstra, M., 355, 359
22,23, 25, 33, 36,41, 51, 54,61, Venables, P H., 137
63, 366, 370 Veraldi, D. M., 135
Stuart, R. B., 132 Vermilyea, J. A., 136, 139
Sulzer-Azaroff, B., 115, 117, 118, 175,
215, 236, 237, 254, 255, 257, 265, Wade, T C, 108
266, 268, 280, 348 Wahler, R. G., 112, 355, 356, 357, 358,
Sushinsky, L. W., 134 361, 363
Swan, G. E., 133 Waite, W. W., 258, 259
Swearinger, M., 215 Wallace, C. J., 52, 71, 126, 350, 367
Sweeney, T. M., 108 Walker, H. M., 6, 122, 156, 157
Wampold, B. E., 293, 302
Talan, K., 101 Ward, M. H., 146, 357, 365
Tate, B. B., 355 Ware, W. B., 299
Taylor, C. B., 136 Warren, V L., 114, 363
Teevan, R. C, 6 Watson, P J., 9, 244, 245
Teigen, J. R., 115 Watts, J. G., 170, 171
Terdal, L. G., 107, 109, 110, 131, 133, Waxier, C. Z., 123
139 Webb, E. J., 120, 132, 138
Thomas, D. R., 357, 358 Webster, J. S., 214, 220, 221
Thomas, J. D., 359 Weick, K. E., 118, 120, 121, 122
Thompson, L. E., 69, 80, 81, 114, 195, Weinrott, M. R., 131, 267, 280, 282,
274, 330, 352 296
Thoresen, C. E., 136, 296, 301 Weiss, D. J., 126, 139
Thome, E C, 30 Werner, J. S., 110
Tiao, G. C, 306 Werry, J. S., 183
Tinsley, H. E. A., 126, 139 Wetherby, B., 247
Titler, N. A., 112 Weyman, P., 110
Todd, N., 267, 280 Whang, D. L., 215
Traux, C. B., 20, 167, 168, 169, 352 Wheeler, A. J., 175
Tremblay, A., 159, 160 White, O. R., 258, 312, 313, 315. 316,
Tryon, W. W., 319 318
llicker, B., 213 Whitman, T L., 244
Turkat, I. D., 27 Wildman, B. G., 118, 129
Tbrkewitz, H., 353 Willard, D., 159, 300
Tbrner, S. M., 175, 191, 192, 214, 247, Williams, C. D., 354, 356
248, 347 Williams, J. G., 69, 71, 146, 352
TWardosz, S., 162, 163, 164, 212, 360 Willson, V L., 142
Wilson, C., 133, 135, 136, 137, 138,
Ulman, J. D., 254, 255, 257, 265, 266, 139
280 Wilson. E E., 109
Ullmann, L. R, 30, 57, 141 Wilson, G. T, 6, 56
Wincze, J. P., 69, 137, 174, 178 179, Wooton, M., 360
330, 339, 341, 342, 343, 366 Workman, E. A., 244, 245
Winett, R. A., 138, 292 Wright, D. E., 80, 81, 195
Winkel, G. H., 355, 356 Wright, H. E., 114, 116
Winkler, R. C, 138 Wright, J., 358
Winton, A. S., 268 Wysocki, T, 249
Wittlieb, E., 109
Wodarski, J. S., 215 Yang, M. C. K., 299
Wolery, M., 308, 318, 323 Yarrow, M. R., 123
Wolf, M. M., 64, 71, 89, 90, 110, 142, Yates, A. J., 29
143, 175, 212, 266, 286, 290, 352, Yawkey, T. D., 359
354, 355, 356, 359 Yelton, A. R., 129
Wolfe, J. L., 134, 215 Yule, W, 215
Wolstein, B., 37
Wonderlich, S. A., 138 Zegiob, L. E., 120
Wong, S. E., 215 Zeilberger, J., 99, 358
Wood, D. D., 122 Zilbergeld, B., 367
Wood, L. E, 108, 115, 116, 117, 123. Zimmerman, E. H., 356
125, 129, 353 Zimmerman, J., 265, 266, 356
Wood, S.. 360 Zubin, J., 17, 20
About the Authors
DAVID H. BARLOW received his Ph.D from the University of Vermont in
1969 and has pubHshed over 150 articles and chapters and seven books,
mostly in the areas of anxiety disorders, sexual problems, and clinical re-
search methodology. He is formerly Professor of Psychiatry at the University

of Mississippi Medical Center and Professor of Psychiatry and Psychology at
Brown University, and founded clinical psychology internships in both set-
tings. Currently he is Professor in the Department of Psychology at the State

University of New York at Albany and has been a consultant to the National
Institute of Mental Health and the National Institutes of Health since 1973.
He is Past President of the Association for Advancement of Behavior
Therapy, past Associate Editor of the Journal of Consulting and Clinical
Psychology^ past Editor of the Journal of Applied Behavior Analysis and y
At the present he is also Director of the

currently Editor of Behavior Therapy.
Phobia and Anxiety Disorders Clinic and the SexuaHty Research Program at
SUNY at Albany. He is a Diplomate in Clinical Psychology of the American
Board of Professional Psychology and maintains a private practice.
MICHEL HERSEN (Ph.D., State University of New York at Buffalo,

1966) is Professor of Psychiatry and Psychology at the University of
Pittsburgh. He is the Past President of the Association for Advancement of
Behavior Therapy. He has co-authored and co-edited 33 books including:
Single-Case Experimental Designs: Strategies for Studying Behavior Change
(1st edition), Behavior Therapy in the Psychiatric Setting , Behavior Modifica-
tion: An Introductory Textbook , Introduction t\ ^linical Psychology In-
ternational Handbook of Behavior Modificatioi ^ Therapy, Outpatient
Behavior Therapy: A Clinical Guide, Issues in ^ ho therapy Research,
Handbook of Child Psychopathology, The Psychology Handbook,
Clinical
and Adult Psychopathology and Diagnosis. With Alan S. Bellack, he is editor
and founder of Behavior Modification and Clinical Psychology Review. He is
Associate Editor of Addictive Behaviors and Editor of Progress in Behavior
Modification. Dr. Hersen is the recipient of several grants from the National
Institute of Mental Health, the National Institute of Handicapped Research,
and the March of Dimes Birth Defects Foundation.
419
m
l?Tti
u
T
On the first edition:

"It ishard to imagine a more skillful blending of discourse and example
for a most difficult subject matter ... a model of scholarly acumen . .
beautifully written. .will undoubtably become a classic."

.
The American Journal of Mental Deficiency

"Recommended reading for all behavior therapists."
Betiavior Modifica tion
Barlow and Hersen present a thorough revision of a book which has
become a classic. The second edition has a completely new invited
chapter by Donald P. Hartmann on behavioral assessment, in addition to
Alan E. Kazdin's chapter on statistical analysis. A special feature of the
new edition is expanded material on clinical replication.
About the authors:

David H. Barlow has published over 150 articles and chapters and seven
books, including The Scientist Practitioner: Research and Accountability
in Clinical and Educational Settings. Currently, he is Professor of
Psychology at the State University of New York at Albany. He is Past
President of the Association for Advancement of Behavior Therapy and
current editor of Behavior Therapy. Professor Barlow is Director of the
Phobia and Anxiety Disorders Clinic and the Sexuality Research Pro-
gram at the State University of New York at Albany.
Michel Hersen has co-authored and co-edited 33 books, including
Behavioral Assessment: A Practical Handbook and The Clinical
Psychology Handbook. He is currently Professor of Psychiatry and
Psychology at the University of Pittsburgh, as well as the ^^î;esident
of the Association for the Advancement of Behavior Ther^^
S. Bellack, he is editor and founder of Behavior Modificai
Psychology Review. /a-:F p>?PFPTM!N! Al
M
if-
m i
iillil:!
\V

Single Case Experimental Designs: Strategies For Studying Behavior Change (Barlow & Hersen)

Uploaded by

Copyright:

Available Formats

Single Case Experimental Designs: Strategies For Studying Behavior Change (Barlow & Hersen)

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

Single Case Experimental Designs: Strategies For Studying Behavior Change (Barlow & Hersen)

Uploaded by

Copyright:

Available Formats

~

Barlow/Hayes/Nelson THE SCIENTIST PRACTITIONER: Research

Free specimen copies available upon request.

Single Case Experimental

With invited chapters by

SAO PAULO SYDNEY TOKYO TORONTO

JAPAN Pergamon Press, 5th Floor, Matsuoka Central Building,

CANADA Pergamon Press Canada Ltd., Suite No 271,

Copyright () 1984 Pergamon Press, Inc.

Library of Congress Cataloging in Publication Data

Single case experimental designs, 2nd ed.

(Pergamon general psychology series)

A// Rights reserved.No part of this publication may be

Printing: 456789 Year: 1234567890

Printed in the United States of America

1. The Single-case in Basic and Applied Research: An Historical

1.2. Beginnings in Experimental Physiology and Psychology 2

2. General Issues in a Single-case Approach 32

3. General Procedures in Single-case Research 67

4. Assessment Strategies 107

5. Basic A-B-A Withdrawal Designs 140

5.1. Introduction 140

6. Extensions of the A-B-A Design, Uses in Drug Evaluation, and

6.1. Extensions and Variations of the A-B-A Withdrawal

7. Multiple Baseline Designs 209

7.1. Introduction 209

8. Alternating Treatments Design 252

8.1. Introduction 252

9. Statistical Analyses for Single-case Experimental Designs 285

9.1. Introduction 285

10. Beyond the Individual: Replication Procedures 325

10.1. Introduction 325

Hiawatha Designs an Experiment 372

Subject Index 405

Name Index 409

About the Authors 419

In the preface to the first edition of this book we said:

We do not expect this book to be the final statement on single-case designs. We

At the time, seemed a reasonable statement to make, but we think that

Cohen made a significant contribution in searching out the seemingly endless

Conversation between Tolman and Allport

TOLMAN: "I know I should be more idiographic in my research, but I

just don't know how to be."

The Single-case in Basic and Applied

1.2. BEGINNINGS IN EXPERIMENTAL PHYSIOLOGY

the clinical extension of the newly developed experimental methodology

in the demonstration that important findings with wide generality were

most famous for developing measures of sensation through several psy-

modalities. What is common to these methods is the repeated measurement

To Wundt, the subject matter of psychology was immediate experience,

emotional experiences where scientific calm and objectivity might be dis-

1.3. ORIGINS OF THE GROUP COMPARISON

Important research in experimental psychology and physiology using single

Gallon and Pearson expanded the study of individual differences at che

sitate a comparison among individuals and a concern for a description of a

in 1932, while reviewing research in experimental psychology, Dunlap pointed

The influence of inferential statistics

During the 1930s, the work of R. A. Fisher, which subsequently exerted

The work of Fisher was not limited to the invention of sophisticated