Clifiicol Psychology Review, Vol. 10, pp. 649-668, 1990 0272-7358/90 $3.00 + .

oo
Printed in the USA. All rights reserved. Copyright 0 1990 Pergamon Press plc

FUNCTIONAL ANALYSIS IN BEHAVIOR

THERAPY

Stephen N. Haynes

University of Hawaii

William H. O’Brien

Illinois Institute of Technology

ABSTRACT. The functional analysis occupies a central role in behavioral construct

systemsand in behavior therapy. However, examination of the literature suggests that it is a
conceptually amorphous term with multiple definitions. Such imprecision has resulted in
inconsistency and restricted applications in clinical case conceptualizations and clinical
decision-making. In order to clatify the nature of the functional analysis, this paper examines
its eptitemological bases and important dimensions of functional relationships. A definition of
functional analysis is proposed that integrates the important characteristics of functional
relationships with the tenets of behavioral construct systems. This definition can accommodate
complex causal models which more accurately depict the behavioral interrelationships
presented by most clients, Methods for deriving the functional analysis are then examined. A
review of the literature suggests that pretreatment functional analyses are infrequently
conducted. The construction of functional analytic causal models is proposed as a method of
systematizing and organizing the functional analysis and facilitating its clinical impact.

One of the defining characteristics of behavior therapy is an emphasis on the

idiographic assessment of objectively defined target behaviors and the situational,
cognitive, and behavioral factors that control some aspect (e.g., frequency, inten-
sity, duration) of its topography (e.g., Haynes, 1978; Nelson 8c Hayes, 1986). The
data derived from this assessment are then used to design an intervention which

The authors would like to thank Linda Gannon for her comments on an earlier draft of this
manuscript.
Correspondence should be addressed to Stephen N. Haynes, Department of Psychology,
University of Hawaii at Manoa, 2430 Campus Road, Honolulu, HI 96822.

649
650 S. N. Haynes and W. H. O’Brien

introduces new controlling variables, or modifies the original variables presumed

to control the targeted behavior problems. The term functional analysis has been
loosely affixed to this assessment/intervention process, and, as a result, it occupies
a central role in behavioral construct systems (Bootzin, 1975; Craighead, Kazdin,
& Mahoney, 1981; Goldfried & Pomeranz, 1968).
As noted by numerous behavior analysts, an erroneous functional analysis can
lead to ineffective interventions because it often dictates the type of intervention
implemented with a client’ as well as the variables upon which they focus (Bootzin,
1975; Emmelkamp, 1986; Haynes, 1986a; Lanyon & Lanyon, 1976; Wolpe, 1977).
Despite the importance of the functional analysis in treatment formulation, there
continues to be considerable divergence among behavior analysts2 in how it is
defined and implemented. Furthermore, the conceptual bases, domain of utility,
and methods used in functional analyses have not been well articulated.
The purposes of this article are to: (a) present the epistemological bases for a
focus on functional and causal relationships in behavior therapy; (b) propose a
definition of functional analysis; (c) briefly present the methods of identifying
functional relationships; (d) examine the degree to which the functional analysis
affects intervention decisions; (e) examine causal relationships and their impact on
the functional analysis; (f) delineate the areas of needed research on functional
analysis; and (g) integrate causal modeling concepts with functional analysis in
clinical applications.

THE EPlSTEMOLOGlCAL FOUNDATIONS OF THE FUNCTIONAL ANALYSIS

An emphasis on functional relationships in behavior therapy has two interdepen-
dent origins: (a) a rejection of a structuralist approach to understanding behavior
problems and (b) an avoidance of some of the metaphysical issues associated with
a focus on “causal” relationships. The early proponents of functionalism (e.g.,
Angell, Carr, Cattell, Dewey, Thorndike, Titchner, Woodworth), who were heavily
influenced by Darwin, emphasized the inadequacies of an epistemology for
behavioral science based only on description of structure or topography. They
reasoned that in order to effectively predict behavior and understand the condi-
tions under which it occurs, behavioral scientists must consider its “utility” and
context, in addition to its form (Boring, 1957; Rachlin, 1970).
The conflicts between structuralists and functionalists have continued for almost
a century and are currently manifested in dialogues regarding the Diagnostic and
Statistical Manual for Mental Disorders (DSM-III-R: American Psychiatric Associa-
tion, 1987). The DSM-III-R is a taxonomy of behavior disorders adhering to a
structuralist approach. Symptoms are generally clustered according to topograph-
ical covariation, which is taken as evidence that some common unmeasured
“underlying” variable is operational. In contrast, a functional approach focuses on
the covariation between topography and the putative controlling variables. Topo-
graphical covariation, per se, is considered meaningful only to the extent that it
assists with identification of these controlling variables (“Behavioral,” 1988).

‘Client is a generic term referring to the object of behavioral assessment or therapy. A client
may include a family, classroom, inpatient adult, institution, teacher, etc.
‘Behuavior analyst is a generic term referring to a behavioral scientist interested in the
understanding of human behavior and the treatment of behavior disorders; it is not
restricted to individuals with any specific theoretical orientation.
 FunctionalAnalysis 651

Functionalism is also an accommodation to more than 2000 years of debate

concerning the nature of causality.3 Arguments have centered around the kinds of
causes (e.g., formal, efficient, material, final, necessary, external, internal), the
conditions necessary for inferring causality (e.g., constant conjunction), definitions
of causality, symmetry and directionality of causal relationships, levels of causality
(e.g., macro vs. micro), and termporal constraints on causal inferences (Aristotle,
1947; Blalock, 1964; Bunge, 1963; Cook SC Campbell, 1979; Grunbaum, 1966;
Hume, 1911; Hyland, 1981; James, Mulaik, & Brett, 1982; Kenny, 1979; Russell,
1948; Simon, 1957). To avoid the semantic and conceptual ambiguities associated
with the concept of “causality,” behavioral and other scientists began to stress
“functional” relationships in their discourses about the phenomena of their
respective disciplines.

BASIC CHARACTERISTICS OF FUNCTIONAL RELATIONSHIPS

A functional relationship exists when two or more variables have shared variance:
Some parameter (e.g., rate, magnitude, length, age) of one variable is associated
with some parameter of another. In an alternative language, variables are func-
tionally related when they demonstrate a mathematical relationship. A simple
functional relationship is described by the functional equation Y = f(aX, + bX, +
e) in which the dependent (criterion) variable “Y” is a function of two independent
(predictor, functional) variables, “Xi, Xs” each one with a specific weight (“u” or
“V’) and an error term “e” which indicates that Y is also a function of unmeasured
variables and measurement error.
A focus on functional relationships avoids most of the controversies surround-
ing casual relationships, primarily because functional relationships need not imply
causality. The familiar functional equation E = mc* suggests a functional relation-
ship between energy, “E,” and the speed of light, “c,” without implying that the
speed of light “causes” energy. Also, the temporal and directional relationships
among functionally related variables need not be specified (Bunge, 1963).
Because the functional analysis in behavior therapy is based on identification of
functional relationships, it is affected by the following characteristics:

1. Functional relationships imply only covariance among variables; there are

different apes of functional relationships (i.e., forms of functional relation-
ships) and functional variables. Some functional relationships are cuusal
while others are correlational but not causal. In addition, some functional
variables are controllable (e.g., social initiation skills) while others are not
(e.g., gender) and some functional relationships are important (i.e., account
for a relatively large proportion of variance) while others are trivial.

‘A “causal” variable is a functional variable whose modification leads (or would lead) to a
change in a parameter (e.g., probability, rate, magnitude) of a designated dependent
variable. Causal variables may be original or maintaining and need be neither necessary,
sufficient, exclusive, nor modifiable. Thus, all causal variables are functional variables but
not all functional variables are causal. Extended discussions of the concept of causality can
be found in Asher (1976), Blalock (1971), Bunge (1963), Cook and Campbell (1979),
McCormick (1937), Russell (1948), Sellitz, Wrightsman, and Cook (1976), and Simon
(1977).
652 S. N. Haynes and W. H. O’Brien

2. Because there are always unmeasured functional variables and measure-

ment error, functional relationships are always probabilistic rather than
exactly deterministic (Blalock, 1964; Cook 8c Campbell, 1979; Kaplan,
1964). Therefore, predictive efficacy is highest when predicting average
values and there is a positive relationship between predictive efficacy
and the number of cases or episodes to which a functional equation is
applied. This applies to the number of persons in a nomothetic application
(e.g., the functional relationship between eating patterns and social contexts
across a number of persons) or to the number of occurrences of an event in
an idiographic application (e.g., the functional relationships between eating
patterns and social contexts for one person across a number of episodes).
3. Functional relationships are nonexclusory: A functional relationship between
a dependent variable and a specified set of independent variables does not
preclude important functional relationships for that dependent variable
involving other independent variables (James et al., 1982; McCormick,
1937). For example, a demonstrated functional relationship between de-
pression and neurotransmitters does not preclude a functional relationship
between depression and cognitive or behavioral events (Beck & Young,
1985).
4. Functional relationships can vary over time and must be considered as
transient (Gadamer, 1975; James et al., 1982): The independent variables
that are functionally related to a particular dependent variable and the
strength of their relationships (i.e., the weight of the variable in a functional
equation) may vary across time. For example, pain responses of a client
initially may be a function of organic factors, but later, social-environmental
factors may predominate (Fordyce, 1976).
5. An independent variable in a functional relationship may be necessary (the
dependent variable never changes unless the independent variable has
changed first), sufficient (the dependent variable always changes whenever
the independent variable has changed), necessary and sufficient (the depen-
dent variable never changes without a change first occurring in the indepen-
dent variable and always changes whenever the independent variable has
changed), or neither necessary nor sufficient (e.g., correlational) to ac-
count for variance in the dependent variable, Blalock, 1964).
6. All functional relationships are subject to mathematical description (James et
al., 1982). These mathematical descriptions are usually referred to as
functional (or structural) equations. However, because the mathematical
symbols in functional equations have assigned meanings, they are not always
amenable to the same manipulations as true mathematical equations (where
mathematical symbols are not imbued with extraneous definitions). For
example, a functional equation which accurately predicts the degree of
cardiovascular reactivity to a stressor, Y, as a function of amount of prestress
caffeine ingestion, X,, and aerobic conditioning, X,, such that Y = 2X, -
X,, cannot be transposed to accurately predict caffeine intake as a function
of cardiovascular reactivity and aerobic conditioning such that X, =
.5Y + .5x,.
7. Functional relationships have boundaries or domains of operation (Johnston &
Pennypacker, 1980; McCormick, 1937). Therefore, functional relationships
are conditional, and their boundary conditions must always be specified. For
 Functional Analysis 653

example, the functional relationship between decreased self-efficacy and

posttreatment relapse for substance abuse may be valid only in an environ-
ment in which the substance is available (Marlatt, 1985).
8. Functional variables vary in their level (Dubin, 1969) and should be ex-
pressed in the level most relevant for their intended use. For example,
marital distress may be functionally related to macrolevel variables, such as
ethnicity, as well as to microlevel variables, such as the number of interrup-
tions in a communication exercise. In this example, the macrolevel func-
tional variable may be of interest to the sociologist and the microlevel
functional variable may be of interest to the behavioral marital therapist.
Although microlevel variables usually have greater clinical utility in behavior
therapy, both microlevel and macrolevel variables can be valid and useful,
depending on their intended applications.
9. Functional relationships that are causal can be reciprocal or bidirectional
(Cook & Campbell, 1979). For example, negative marital interactions may
lead to subsequent depressed behaviors and/or depressed behaviors may
lead to subsequent negative marital interactions.
10. Causal functional relationships require precedence of the causal variable
(Sellitz, Wrightsman, 8c Cook, 1976; Simon, 1971). For an event to be causal,
it is necessary, but not sufficient, for the event to precede that which it
causes.

The characteristics of the functional analysis are delineated in the subsequent

sections of this paper, but it should be pointed out here that behavior analysts are
most interested in a subset of functional relationships - those involving impor-
tant, controllable variables whose manipulation will result in a clinically significant
change in a target behavior for an individual client (Baer, 1986; Ferster, 1965).4
Therefore, despite an historical tie to functionalism, the functional relationships
targeted in functional analysis are those that conform to the more restrictive
concept of causality.

A PROPOSED DEFINITION OF THE FUNCTIONAL ANALYSIS

IN BEHAVIOR THERAPY

There are frequent connotative and denotative inconsistencies in use of the term
“functional analysis” in the behavior therapy literature. These are primarily
attributable to differences in its definitional components and in the degree to
which the definitions reflect the principles of functional relationships outlined
above. Definitions of “functional analysis” have included: (a) a specification of
target behaviors (e.g., Craighead et al., 198 l), (b) demonstrations of “control”
through the manipulation of hypothesized controlling (causal) variables, usually

4Sometimes behavior analysts focus on promoting positive alternatives to problem behav-

iors. In these cases, intervention strategies are not strictly based on the manipulation of
causal variables for the targeted problem. However, a causal analysis is still invoked in that
the behavior analysts presumes that the manipulation of certain variables (e.g., self-
instructions, social behaviors, autonomically-mediated arousal) will result in particular
behavior changes, even if those variables do not account for the current parameters of the
targeted behavior problems.
654 S. N. Haynes and W. H. O’Brien

within a single-subject design (e.g., Nelson, 1983; Nelson & Hayes, 1986; Peterson,
Homer, & Wonderlich, 1982), (c) specification of controlling factors for a class of
behavior problems rather than for an individual case (e.g., Boyd & Levis, 1980),
(d) identification of situational or setting factors (Bootzin, 1975; Maisto, 1985), (e)
stimulus-response or response-response relationships (Hawkins, 1986), (f) motiva-
tional and developmental factors (Kanfer & Philips, 1970), (g) identification of
@~&al functional relationships alternative to those operating for a client (Hawk-
ins, 1986), (h) predictions about a client’s behavior (Turkat & Maisto, 1985), (i)
specification of response components (Bernstein, Borkovec, 8c Coles, 1986), and (j)
an overall conceptual integration of behavior problems, causal and mediating
variables, resources, and so on (Correa & Sutker, 1986). Additionally, a functional
analysis has been defined as a process by some (e.g., Hake & Olvera, 1978) and as
a product by others (e.g., Ferster, 1965).
Semantic imprecision has also resulted from interchangeable use of the terms
“functional analysis, ” “behavioral analysis” (Lanyon & Lanyon, 1976), “functional
behavioral analysis” (Wincze, 1982), “behavioral assessment” (Kalish, 1981), and
“behavioral case formulation” (Wolpe & Turkat, 1985). Other authors seem to
have restricted the domain of functional relationships to those involving response
contingencies (Voeltz & Evans, 1982). Within-author inconsistency in definition of
the term “functional analysis” has also occurred (e.g., Keefe, Kopel, & Gordon,
1978).
In sum, there has been disagreement about the definition of functional analysis,
its underlying assumptions, methods of derivation, relevant components, and
domain of utility. These inconsistencies hinder communication among behavior
analysts about the characteristics of the functional analysis and its role in behavior
therapy. Additionally, many definitions do not incorporate the important princi-
ples of functional relationships they are meant to reflect. Consequently, the
elegance, communicability, cohesiveness, and amenability to empirical refinements
of behavioral construct systems are compromised.
Perhaps more importantly, a number of the definitions of functional analysis
include restraints on the methods considered appropriate for its derivation (e.g.,
functional analysis as “manipulation”), or on the types of variables (e.g., response
contingencies) considered appropriate for inclusion. These constraints limit the
power of our methods of identifying functional relationships and of the conse-
quent causal models of behavior disorders. Indeed, identification of important
functional relationships requires diverse methods, and many classes of variables
may have important causal functions.
To address some of these methodological and conceptual inconsistencies, a
definition of functional analysis is presented which reflects the axioms of func-
tional relationships presented earlier. It is also congruent with the tenets of
behavioral construct systems, is clinically useful, and avoids unnecessary restraints
on the methods for its derivation or on the types of variables that may be included.
Functional analysis is defined as:

The identification of important, controllable, causal functional relationships applica-

ble to a specified set of target behaviors for an individual client.

There are several inferences implicit in this apparently simple definition. First,
only some variables functionally related to a target behavior will be causal,
controllable, and important. For example, a history of exposure to a major
 Functional Analysis 655

environmental trauma might be considered an important, causal, but uncontrolla-

ble, functional variable for the occurrence of nightmares in a client with a
posttraumatic stress disorder (PTSD) diagnosis. There may also be statistically
significant but unimportant functional relationships between PTSD and job
satisfaction. Therefore, the functional analysis identifies only a subset of func-
tional variables for a target behavior - those that are likely to be important
controllable and causal and, consequently, primary targets of behavioral interven-
tion. Examples of causal variables that can be important and controllable include
parent-delivered response contingencies, ruminations about an unpleasant inter-
personal altercation, alcohol ingestion, and positive assertion skills.
It is important to acknowledge that many variables may be salient and of interest
to behavioral scientists but deemphasized in a functional analysis. For example, a
history of childhood sexual abuse may be an important determinant of distressed
adult interpersonal relationships (Harter, Alexander, 8c Neimeyer, 1988) and
certainly meritorious of empirical and social attention. However, a functional
analysis of an adult with severe social skill deficits, who was sexually abused as a
child, is more likely to focus on variables that are controllable, such as social
initiation and reception behaviors, outcome and self-efficacy expectations, rumi-
nations, self-instructions, labels, and/or psychophysiological responses. A func-
tional analysis, therefore, does not always “explain” behavior in the sense of
identifying all important causal variables. Rather, it identifies important causal
variables that may be manipulated or under the control of the client or behavior
analyst.
Second, because the definition centers on functional relationships in reference
to the target behavior, a functional analysis may also include effects associated with
changes in the targeted behavior problem that occur as a result of treatment. For example,
increasing the rate of affectionate interactions through behavioral marital therapy
may also positively affect job satisfaction and the level of alcohol intake for one of
the spouses. Thus, the target behavior (in this case, positive marital exchanges),
although considered the primary dependent variable in the functional analysis,
may act as an important causal variable for other behaviors. Such extended
functional relationships are important components of the functional analysis and
reflect a social systems emphasis (Karoly, 1988; O’Leary, 1984; Russo, Hamada, &
Marques, 1988; Wahler & Fox, 1981).
Third, the functional analysis also includes relationships among causal variables.
For example, the depressive behaviors of a client may be a function of aversive
marital interactions and low self-efficacy perceptions. Additionally, marital inter-
actions may also function as a causal variable for, or be correlated with, self-
efficacy perceptions. As noted later, the strength and direction of the relationships
among causal variables is an important determinant of which variables are selected
for intervention (Haynes, 1986a).
Fourth, the functional analysis can be differentiated from behavioral assessment. The
functional analysis is only one of many goals of behavioral assessment (e.g., the
identification of target behaviors, the design of intervention program: see reviews
in Ciminero, Calhoun, 8c Adams, 1986; Haynes, 1978, 1989, 1990; Nelson 8c
Hayes, 1986). Behavioral assessment also includes many methods, such as direct
observation, self-monitoring and interviews. A functional analysis is not a “method.”
It is but one possible product of the application of behavioral assessment methods.
Fifth, as noted earlier, functional relationships can vary in level. Consequently, a
functional analysis should emphasize the level most relevant to its intended
656 S. N. Haynes and W. H. O’Brien

application. For example, a functional analysis emphasizing pharmacological

factors will emphasize more micro-level variables than those examining behavioral
or social systems factors.
Sixth, functional analyses can be constructed for a behavior goal as well as a
behavior problem. For example, a functional analysis may be constructed to address
acquisition of social skills by a client or to account for the social skills deficits of that
client (Carr & Durand, 1985; Tofte-Tipps, Mendonca, & Peach, 1982). Further-
more, these two functional analyses may include different causal variables and
forms of causal relationships.
Seventh, the functional analysis is relevant to all behavior problems that are a
function of controllable variables. As has been advocated by others (Bornstein,
Bornstein, & Dawson, 1984; Evans, 1985; Kanfer & Grimm, 1980; Kantor, 1970;
McFall, 1986; Ollendick & Hersen, 1984; Ulrich, 1975) the current definition also
implies no proscriptions about the Qpe of functional variables targeted (e.g.,
physiological, social-environmental).
Lastly, functional analyses are idiographic (addressing causal relationships for
behavior problems of individual clients) rather than nomothetic (addressing causal
relationships for a behavior problem, across clients). Although there is conceptual
and methodological overlap between the two, the focus of this manuscript is upon
idiographic functional analyses - functional analyses of the behaivor problems of
individual clients. As discussed later in this paper, nomothetic analyses are
constructed from a number of individual cases or exemplars of the model, each of
which can have unique causal pathways. For example, a nomothetic causal model
of juvenile antisocial behavior positing that inept parental discipline and child
coercion are key causal variables (Patterson, 1986), will not always be applicable to
individual cases where other important factors (e.g., chronic illness) are opera-
tional (cf. Snyder, 1987).
Between-person differences in functional analyses involving the same behavior
problems should be expected and are congruent with the axiom that there are
often important individual differences in the causes of a behavior disorder
(Hawkins, 1986; Khouri, 8c Akiskal, 1986; Marlatt, 1985). Nomothetic models are
based on multiple measures and large samples of subjects (Wickens, 1982) and
provide the behavior analyst only with an array of possible causal variables that may
or may not be operational for a particular client. Such models are helpful in
suggesting which assessment foci are most likely to generate a valid functional
analysis for a client with a particular behavior problem but do not negate the
necessity for an individualized preintervention functional analysis.
In summary, a definition of functional analysis is proposed that incorporates
important concepts of functional relationships in a manner congruent with the
characteristics and goals of behavioral construct systems and behavioral interven-
tions. A functional analysis identifies functional relationships that are important,
causal, and controllable. It also includes functional relationships among causal
variables and the changes in other variables that may accompany the modification
of target behaviors. Functional analyses can vary in level, can be contructed for
behavior goals as well as target behaviors, and are idiographic in focus.

METHODS OF IDENTIFYING CAUSAL RELATIONSHIPS

IN CLINICAL SITUATIONS

Deriving information about possible causal relationships for a functional analysis

 Functional Analysis 657

of a client is an important goal of preintervention assessment. Inferences about

causal relationships require information about: (a) significantly elevated conditional
probabilities (Patterson, 1982; Schlundt, 1985); (b) $.Irecedence of the causal variable;
and (c) exclusion of alternative explanations for the observed functional relation-
ships (Asher, 1976; Simon, 1971). Failure to satisfy any of these conditions
precludes differentiation of causal from simply correlational relationships (Bla-
lock, 1964; Cook & Campbell, 1979; James et al., 1982). Methods of satisying these
conditions have been the topic of hundreds of books on assessment, research
design, and statistics. The goal of this section is only to briefly reiterate the
principles involved, while specifically referencing their application to the func-
tional analysis.
To satisfy the conditional probability requirement it must be demonstrated that
if a variable, Y, is causally related to another variable, X, the probability of change
in Y, given a change in X (i.e., its conditional probability), must be significantly
different from the overall or base-rate probability of change in Y (i.e., its
unconditional probability). For example, the paranoid behaviors of a client cannot
be causally related to social isolation (Haynes, 1986b) unless it can be shown that
probability of paranoid behaviors differs when that person is under conditions of
high than under and low social isolation. Examination of conditional probabilities
underlies such diverse statistical and methodological procedures as Markov models
and sequential analyses, chi square, logistic multiple regressions, and some Bayes-
ean statistical procedures. Furthermore, they underlie many assessment proce-
dures such as self-monitoring, questionnaires and interviews (e.g., “What do you
do when David hits another child?“).
A criterion for inferring causal relationships that is more problematic for the
behavior analyst is precedence. To validate an hypothesized causal relationship, it
must be shown that change in the hypothesized causal variable precedes change in
the dependent variable. However, a functional relationship between two variables
whose occurrence is reliably separated in time does not always imply a causal
relationship between the two (Dubin, 1969; McCormick, 1937). The demonstra-
tion of a consistent temporal relationship between two variables during preinter-
vention assessment with a client is difficult because it often requires either
manipulation or multiple measures across a long period of time (Kratochwill,
1978).
The causal role of a variable also depends on the specific parameters of a behavior
problem being addressed. For example, presleep worry cannot be considered a
cause of the onset of a client’s insomnia unless it preceded those difficulties.
However, presleep worry may have been triggered by sleep-onset difficulties and
may subsequently serve as a causal variable for their maintenance. Congruent with
this notion, Barnett and Gotlib (1988) suggested that learned helplessness may not
serve as a causal variable for onset of depressive episodes, but may serve as a causal
variable affecting their magnitude or duration.
There are several methods used by behavior analysts to derive hypotheses about
causal relationships for a functional analysis of a client. The most common is
through several contemporaneously administered traditional behavioral assess-
ment instruments (e.g., questionnaires, analogue observation). Although the re-
sultant data can then be examined using sophisticated statistical techniques, such
as path analysis and multiple regression in research applications, the clinician must
rely on less objective analytic procedures such as visual inspection of the data, zero
order correlations, or comparison with norms. Furthermore, regardless of which
 S. N. Haps and W. H. O’Brien

analytic procedure is used, causal inferences must be held more tentatively because
temporal precedence is difficult to establish. Causal inferences can be tesded more
powerfully through longitudinal designs such as those used in time-series or
sequential analysis. Finally, systematic manipulation of hypothesized controlling
variables (e.g., ABAB designs), can be used to test causal relationships by demon-
strating reliable control of a dependent variable.
These methods differ dramatically in the confidence engendered in the result-
ing causal inferences. It is important to note here that those methods that are the
most poweful (e.g., manipulation, time-series) tend to also be the most costly and
the most cumbersome in clinical situations. Consequently, most functional analyses
and intervention decisions tend to be based on the less objective analytic proce-
dures resulting many times in a specious understanding of the relationships
between the hypothesized independent and dependent variables (Nelson, 1988).
Most traditional behavioral assessment methods are based on unvalidated
assumptions that the data derived serve as valid marker variables for causal
relationships that occur in the natural environment.5 For example, we often
presume that interview-derived parent reports of how they respond to their child’s
tantrums accurately reflect (i.e., serve as a marker for) responses to their child’s
tantrums in the natural environment. However, validation of marker variables
requires extensive investigation of the degree to which they correlate with
behaviors they are intended to predict - an empirical process seldom pursued.
It should be reemphasized that development of a functional analysis is not tied
to any specific method. Imposed methodological limitations for establishing causal
relationships can also impose constraints on the adequacy of our functional
analyses. If, for example, the functional analysis is methodologically defined as the
manipulation of temporally contiguous response contingencies, then behavior
analysts must confine their search for important causal relationships to this class of
variables (e.g., Baer, 1978). Any construct system which includes methodologically
mandated conceptual restraints will be significantly limited in power, applicability
across populations and behavior problems, and evolution, because only a very
limited set of causal variables will be subject to inclusion. Hence, the explanatory
power of the construct system will be limited to a few persons, situations, behaviors
or variables.
In summary, there are several methods of identifying possible causal relation-
ships for a functional analysis of clients’ behavior disorders, and there is an inverse
relationship between the power of those methods and their clinical utility. The
most powerful, time-series analysis and manipulation, can be prohibitively time-
consuming. Contemporaneously administered behavioral assessment instruments
are a more efficient method of deriving causal hypotheses, but inferences must be
held more tentatively because causal and noncausal functional relationships cannot
be readily differentiated. The use of marker variables can increase the efficiency
and validity of measures of causal relationships but require extensive validation.

5A marker is an index (usually easily derived) of a complex or difftcult-to-measure

functional relationship. For example, an empirical determination of the causal roles of
psychosocial stressors for a client with migraine headaches would require months of daily
multimethod assessment. However, the correlation between psychosocial stressors and
headaches may be indicated by the magnitude of cardiovascular response to a brief
laboratory stressor or the rate of recovery to that stressor (Haynes, 1988).
 Functional Analysis 659

APPLICATIONS OF THE FUNCTIONAL ANALYSIS TO THE DESIGN OF

INTERVENTION PROGRAMS

One goal of this manuscript is to examine the role and impact of functional
analyses in the design of behavioral intervention programs. In order to estimate
the degree to which intervention designs for clients are based on preintervention
functional analyses in published case studies, a review was conducted of 156 case
studies, published in the years 1985-1988, appearing in Behavior Therupr (N = 21),
Journal of Applied Behavior Analysis (N = 78), Behavior Modification (N = 31), and
Behaviour Research and Therapy (N = 26). Each study was coded as to whether or
not the intervention was based on preintervention data collected on functional
relationships. For example, a study in which contingency management or self-
instruction was used to reduce classroom “hyperactivity” would be coded as “based
on a functional analysis” if preintervention data was collected which suggested a
causal role for those factors.
Intervention decisions (i.e., which independent variables were selected for
manipulation) were based on preintervention data collection on functional rela-
tionships in only 20% of the 156 cases. In contrast, intervention decisions in 80%
of the cases were based only on a presumed, but untested, functional relationship
between the designated independent and target variables. These results are
consistent with those reported by Haynes (1986a) and Lennox, Miltenberger,
Spengler, and Efanian (1988), in which preintervention assessment was found to
be infrequently used in the design of behavioral intervention programs. There-
fore, despite the emphasis in behavioral construct systems on individual differ-
ences in the causes of behavior problems and on individualized interventions,
preintervention functional analyses were infrequently conducted; intervention
decisions were usually based on a nomothetic model of the behavior problem.
Although the infrequent use of idiographic preintervention functional analyses
to design behavioral intervention programs may be a result of several factors (see
review in Haynes, 1986a), it is reasonable to suppose that the lack of efficient and
clinically useful methods of identifying, conceptualizing, and organizing func-
tional relationships is an important contributor. Before proposing a clinically
useful schema for constructing functional analyses, a discussion of complex causal
relationships and their impact on functional analysis strategies is warranted.

Areas of Needed Empirical Investigation

Areas of needed research on the functional analysis have been well articulated and
will not be reviewed in detail here. In brief, the functional analysis has been
infrequently subjected to empirical investigation. Sufficient data are unavailable
on: (a) its reliability across behavior analysts; (b) its criterion-related or treatment
validity; (c) the variables affecting its reliability and validity; (d) the assessment
instruments which are optimally suited for its derivation and under what condi-
tions they are useful; (e) the clincial decision-making processes involved in its
derivation; (f) markers of complex functional relationships; (g) the best methods
of testing hypothesized functional relationships; (h) its generalizability across
target behaviors and clients; and (i) its impact on intervention efficacy.
The empirical study of the functional analysis would be facilitated by the
adoption of a consistent definition and, especially, by the delineation of a
conceptual model for its construction. The last section of this paper attempts to
660 S. N. Haynes and W. H. O’Brien

integrate the conceptual and methodological issues into a conceptual schema for
the construction of functional analyses.

THE COMPLEXITY OF CAUSAL RELATIONSHIPS IN BEHAVIOR DISORDERS

As suggested earlier in this paper, the functional analysis of a client’s behavior

problems is closely linked to the extant conceptualizations of the causal variables
controlling specific target behavior(s) and of the form, strength and direction of
the causal relationships. Consequently, functional analyses reflect extant models of
causality. As noted by Haynes (1989) and Haynes and O’Brien, (1988), these
models of causality of behavior disorders are becoming increasingly complex.
Several basic characteristics of causal relationships (e.g., nonexclusiveness, do-
mains, bidirectionality) were discussed in the section “The Characteristics of
Functional Relationships,” but others are frequently incorporated into complex
models of behavior disorders. For example, there is increasing recognition that
behavor problems often involve multiple causes and multiple causal paths (e.g., there
are many means through which relaxation training may reduce chronic pain
reports), and that these causal variables and paths can vary across individuals
(Kanfer, 1985; Mirksy & Duncan, 1986; Turner, Beidel, 8c Costello, 1987).
Furthermore, the importance, or weights, of particular causal variables can vary
across persons. Paranoid behaviors, for example, may be a function of differing
permutations of attentional, attributional, parental modeling, concurrent social,
classical conditioning, life stressor, response contingency, and/or behavioral iatro-
genie factors (Haynes, 1986b).
Other characteristics of causal relationships (sometimes referred to as the
“form” of a causal model, in contrast to its “content”) are important but less
frequently reflected in current causal models of behavior disorders. First, causal
relationships may assume a quadratic, parabolic, log-linear, or biphasic, rather than
linear, form. Causal relationships may also exhibit ceiling or floor effects or
functional plateaus (Grove & Andreasen, 1986). For example, the effects of drugs
on behavior frequently assume a biphasic or nonlinear function (Marlatt, 1985)
and the effects of distraction on sexual arousal may assume a parabolic form
(Barlow, 1986).
Second, a causal relationship may be in effect only when a critical level of the
causal variable has been surpassed. For example, there may be a causal relation-
ship between life stressors and behavior problems only when the stressors exceed
a critical magnitude or frequency (Rosenthal & Rosenthal, 1985).
Third, causal relationships may vary in their temporal characteristics. The latency
and duration of impact of a causal variable may vary over time, across target
behaviors, and across developmental stages (Cook & Campbell, 1979; James et al.,
1982; Piaget, 1952; Scott, Stewart, & De Ghett, 1974).
Fourth, the causal relationship between two variables may be mediated by other
variables (Arnold, 1982). For example, sodium and potassium levels may mediate
the impact of psychosocial stressors on blood pressure (Gross & Strasser, 1983).
There are other characteristics of causal models of behavior disorders (for
additional discussions, see Haynes, 1988; Haynes & O’Brien, 1988), but the
dimensions outlined above illustrate their complexity. Such complexity does not
limit the utility of the functional analysis. However, it affects the methods used to
derive causal inferences and the form and content of the resulting causal model.
 Functional Analysis 661

Complexity mandates that a diversity of methods be used for identifying causal

relationships. For example, assumptions of multiple causes and causal paths
require a multivariate approach to assessment. Furthermore, research and stan-
dard statistical techniques, based on presumed linear relationships, may underes-
timate significantly the strength of nonlinear causal relationships. The delineation
of many causal relationships requires frequent sampling, across a range of
parameters of causal variables, within a time-series framework. James et al. (1982)
have also noted the importance of estimating the equilibrium-time of a functional
relationship: Sampling points must be programmed to correspond to the period
within which the effect of a causal variable is operational. For example, a discrete
stressor may cause a brief increase in blood pressure which dissipates after 5
minutes. To confirm the causal relationship, measurement of blood pressure must
occur within the 5-minute “window” of observable covariation. It should be
apparent that the methods for detecting causal relationships are prescribed by their
presumed characteristics.
More importantly, functional analyses must accurately reflect the complexity of
causal relationships. While causal models and functional analyses must strive for
parsimony, functional analyses which do not accurately reflect multiple causal
factors or temporal relationships, for example, are unlikely to engender optimally
efficient intervention strategies.

CAUSAL MODELING AS A STRUCTURE FOR FUNCTIONAL ANALYSIS:

FUNCTIONAL ANALYTIC CAUSAL MODELS

The central role of the functional analysis in the design of behavioral intervention
programs, their infrequent, informal and unsystematic construction by behavior
analysts, and the complexity of causal relationships which must be incorporated,
suggest that a more structured approach to the development of a functional
analysis in clinical settings is needed. A more systematic method of organizing the
behavior problems and hypothesized causal relationships of a client would facili-
tate selection of variables for intervention and design of intervention programs.
Causal modeling provides the conceptual basis for such an organizational scheme.
Causal modeling has been used for many years to derive and test hypotheses
about causal relationships among sets of variables for diverse targets, such as
coronary heart disease (Pepping 8c Vaitl, 1981), substance abuse (Huba & Bentler,
1982), political violence (Kritzer, 1977), and adolescent antisocial behavior (Patter-
son, 1986). Other discussions of causal model applications have been provided by
Evans (1985), Snyder (1987), and Taylor and Agras (1981). Basically, causal
modeling is a statistical technique for estimating patterns of relationships among a
set of variables, and the underlying organization of that pattern, given an
hypothesized causal structure. Usually, it is conducted with large samples, involves
multiple measures of each variable, and involves cross-sectional data collection.
Importantly, causal modeling provides estimates of shared variance among multi-
ple variables.
We do propose to review the theory or statistical basis of causal modeling here.
That has been done very well by many others (Asher, 1976; Bentler, 1980; Biddle
SCMarlin, 1987; Joreskog, 1978), and it is not the focus of this manuscript. Nor do
we suggest that the statistical techniques of causal modeling are adaptable to the
functional analysis of individual clients. We propose that some of the concepts and
662 S. N. Haynes and W. H. O’Brien

“5
Attribution - Strong Relationship

\\’ - Weak Relationship

&direction Causal

X = Causal variable Negative

z Relationship
Rumination
Y = Dependent variable t--, Covarying
Non-Causal
RelatIonship
R = Residual causes

FIGURE 1. Functional analytic causal model of client whose primary pre-

senting complaints were sleep onset insomnia and behavioral depression
(e.g., decreased social interaction, decreased recreational behaviors).

vector systems of causal modeling can help systematize the functional analysis of
individual clients.
Used for purposes of a functional analysis, functional analytic causal models are
vector diagrams (e.g., a conceptual schema) of all important, controllable, causal
variables associated with a particular clients’ target behaviors. These models also
include estimates of the strength and form of hypothesized causal relationships
and, therefore, conform to the definition of the functional analysis.
Figure 1 presents an illustrative functional analytic causal model of a client
whose major complaint was sleep maintenance isomnia. Note that the model
illustrates: (a) bidirectional causal relationships (e.g., between Y, and Ye; (b)
multiple “dependent” and “independent” variables and causal paths (although the
differentiation between independent and dependent variables is more conven-
tional than real when bidirectional relationships are postulated); (c) the relative
strengths (weights) of causal relationships; (d) mediated relationships (e.g., X,
---)X6--->Y2); (e) causal relationships between target behaviors; (f) functional
but noncausal relationships (X4<--->XS).
The structure of a causal model (or a subset of it) can be expressed more
specifically in functional equations (Duncan, 1975; Evans, 1985; James et al.,
1982). For example, for a subset of the model in Figure 1, the behavior analyst
may hypothesize that Y, = log(X,) + 2(X,)’ + R; indicating that the sleep
maintenance problem is a weak logarithmic function of causal attributions and a
stronger quadratic function of presleep ruminations (e.g., the relative impact
presleep ruminations would tend to rapidly increase as parameters (e.g., rate,
intensity) of both increased.
 Functional Analysis 663

Care must be exercised in applying functional equations with individual clients:

They are usually applied as explanatory formulae for nomothetic causal models,
imply more predictive precision than is possible with most case studies, and do not
possess true mshematical properties. However, when used in functional analysis,
they promote a systematic examination and delineation of the complex functional
relationships in a functional analysis. Perhaps most importantly, attempting to
express hypothesized causal relationships in a different medium can generate new hypotheses
about causal relationships and other possible causal variables and help clarify the hypothe-
sizes about the predicted relationships (Wickens, 1982).
Acknowledging that many other factors (e.g., the social significance of target
behaviors, the amenability of potential target behaviors to intervention) affect
clinical intervention decisions, intervention designs must be congruent with the
structure of a causal model for a particular client’s behavior problems. Unless an
alternative rationale can be provided, behavior problems and causal variables
selected for intervention should be those that account for the greatest proportion
of variance in the causal model: The primary guiding principle of selecting target
variables in intervention design is that of “shared variance” - selecting for modification
the variable or subset of variables that account for the greatest amount of variance
in the target behaviors. Thus, for the case illustrated in Figure 1, a treatment-
facilitated reduction in the parameters of many variables would positively affect
sleep disorders (the model would be characterized as “nonrecursive” and “overde-
termined” by causal model theorists). However, the model suggests that a reduc-
tion of aversive marital interactions may be the most powerful or cost-efficient
focus of intervention. Thus, despite multiple and bidirectional causal connections
among variables, the outcome, cost-efficiency, and generalizability of intervention
is significantly affected, by which causal variable is selected for modification -
functional analytic causal models guide that selection.
Functional analytic causal models should be expected to change during the
assessment-intervention process. First, they may change as new information on
problem behaviors, client goals, and causal variables is acquired during assess-
ment. Functional analytic causal models may also be modified when they are not
validated by the results of intervention. Functional analytic causal models should
be expected to evolve over the course of intervention, and the target variables of
choice may change as intervention progresses and the strength of specific causal
relationships are modified. Inspection of Figure 1 also suggests that successful
reduction in aversive marital interactions would change the relative weights of the
various causal paths in the model (e.g., residual sleeping difficulties would then be
a relatively stronger function of other causal variables).
As suggested previously, there is latitude in the permissible level of a causal
model. For example, “aversive marital interactions” could be broken down into
many microlevel behavioral and cognitive variables. Higher level causal models
(those with more abstract or macrolevel variables) are often conceptually and
statistically simpler but less useful clinically in that they pinpoint only an array of
possible intervention targets. For clients with behavior problems that are a
function of one or a small number of variables, the issue of level is not problematic
because variables can be reduced to a therapeutically useful level of specificity
without rendering the model needlessly complex. For clients necessitating more
complex causal models (consider “dual diagnosis” patients with multiple behavior
problems and causal factors) a step-wise approach to functional analysis seems
most appropriate: Variable sets in a higher level causal model are separately
664 S. N. Haynes and W. H. O’Brien

broken down into subsets of lower level causal models. Thus, for Figure 1, the
behavior analyst may construct a separate causal model for aversive marital
interactions which more specifically dictates the intervention target within this
higher order variable (e.g., more microlevel variables might include “criticism in
front of family members” or “sarcastic comments” during disagreements). At any
level, the goal of the functional analytic causal model is to provide a clinically
useful causal schema relevant to a client, and the most useful level may be
expected to vary across disciplines and across persons within disciplines.
Construction of functional analytic causal models is proposed as a method of
objectifying and systematizing case conceptualizations and clinical decision-making
processes, thereby rendering them more amenable to empirical investigation.
Causal model construction for individual clients can facilitate evaluations of the
following aspects of functional analysis: (a) the temporal stability and criterion
validity of hypothesized causal relationships and (b) between-behavior analyst
agreement, treatment utility and validity, social acceptability, and contribution of
various assessment methods in functional analyses. In short, increased objectifica-
tion of functional analyses in the form of functional analytic causal models may
help clarify an important clinical decision-making process, may help communicate
those processes to other professionals and students, and may open them to
empirical investigation and, therefore, refinement.
The authors have found preparation and discussions of functional analytic
causal models for clients to be an excellent vehicle for training graduate students
in many aspects of clinical decision-making. They promote a more organized
approach to understanding the behavior problems of individual clients and
designing intervention programs and help integrate the assessment and treatment
phases of behavior therapy. Furthermore, they are useful in consulting with
medical staff about patients. Functional analytic causal models provide a parsimo-
nious, graphic representation of the client which can usually be readily interpreted
by nonpsychologists.

SUMMARY

The functional analysis is one of the defining characteristics of behavior therapy.

However, it has been defined inconsistently and sometimes at variance with
principles of functional relationships. Characteristics of functional relationships
were reviewed, and a definition of functional analysis was proposed that is
consistent with characteristics of functional relationships and also with the pur-
poses and methods of behavior therapy. The definition proposed accommodates
complex relationships among causal and dependent variables.
Several methods of identifying causal relationships through identification of
conditional probabilities and precedence were outlined. An inverse relationship
between empirical power and clinical utility of those methods was noted. Develop-
ment of markers of causal relationships is a promising method combining power
and utility. Despite availability of several methods of identifying causal relation-
ships, a systematic functional analysis is seldom conducted prior to intervention in
published case studies.
Functional analyses are closely linked to causal models for behavior disorders,
and these conceptualizations are becoming increasingly complex. The types of
causal variables, relationships among functionally related variables, temporal
 Functional Analysis 665

parameters of causal relationships, and their domains were reviewed and their
impact on functional analyses noted.
Finally, functional analytic causal modeling was proposed as a method of
systematizing clinical intervention decisions involving complex causal relation-
ships. The generation of functional analytic causal models facilitates the system-
ization and empirical study of assessment and treatment decisions.

REFERENCES

American Psychiatric Association (1987). LXagnostk and statisticalmanual of mental disorders (3rd ed. rev.).
Washington, DC: Author.
Aristotle (1947). Metaphysics. Cambridge, MA: Harvard University Press.
Asher, H. B. (1976). Causal modeling. Beverly Hills: Sage.
Arnold, H. J. (1982). Moderator variables: A clarification of conceptual, analytic and psychometric
issues. Organizational Behavior and Human Performance, 29, 143-174.
Baer, D. M. (1978). On the relation between basic and applied research. In A. C. Catania & T. A.
Brigham (Eds.), Handbook of applied behavior analysis; social and instructional processes (pp. 11-16). New
York: Halsted Press.
Baer, D. M. (1986). Advances and gaps in a behavioral methodology of pediatric medicine. In N. A.
Krasnegor, J. D. Arasteh, & M. F. Cataldo (Eds.), Child health behavior: A behavioral pediatrics perspective
(pp. 54-69). New York: John Wiley & Sons.
Barlow, D. H. (1986). Causes of sexual dysfunction: The role of anxiety and cognitive interference.
Journal of Consulting and Clinical Psychology, 54, 140-148.
Barnett, P. A., & Gotlib, I. H. (1988). Psychosocial functioning and depression: distinguishing among
antecedents, concomitants, and consequences. Psychological Bulletin, 104, 97-126.
Beck, A. T., & Young, J. E. (1985). Depression. In D. Barlow (Ed.), Clinical handbook of psychological
disorders (pp. 206-244). New York: Guilford.
Behavioral Assessment and DSM-III-R [mini-series]. (1988). Behavioral Assessment, 10, 43-121.
Bentler, P. M. (1980). Multivariate analysis of latent variables: Causal modeling. Annual Rev& of
Psychology, 31,419-456.
Bernstein, D. A., Borkovec, T. D. & Coles, M. G. H. (1986). Assessment of anxiety. In A. R. Ciminero,
C. S. Calhoun, & H. E. Adams (Eds.), Handbook of Behavioral Assessment (pp. 353-403). New York:
John Wiley & Sons.
Biddle, B. J., & Martin, M. M. (1987). Causality, confirmation, credulity and structural equation
modeling. Child Development, 58, 4-17.
Blalock, H. M. (1964). Causal inferences in nonexperimental research. Chapel Hill, NC: The University of
North Carolina Press.
Bootzin, R. R. (1975). Behavior mod$ication and therapy, an introduction. Cambridge, MA: Winthrop
Publishers, Inc.
Boring, E. G. (1957). A history of experimental psycholology. New York: Appleton-Century Crofts.
Bornstein, P. H., Bornstein, M. T., & Dawson, D. (1984). Integrated assessment and treatment. In T.
Ollendick & M. Hersen (Eds.), Child behavioral assessment: Principles and procedures (pp. 223-243). New
York: Pergamon.
Boyd, T. L., & Levis, D. J. (1980). Depression. In R. J. Daitzman (Ed.), Clinical behavior therapy and
behavior modz$ation (pp. 301-350). New York: Garland STPM Press.
Bunge, M. (1963). Causality. Cambridge, MA: Harvard University Press.
Carr, E. G., & Durand, V. M. (1985). Reducing behavior problems through functional communication
training. Journal of Applied Behavior Analysis, 18, 111-126.
Ciminero, A. R., Calhoun, K. S., & Adams, H. E. (1986). Handbook of behavior assessment (2nd ed.). New
York: John Wiley & Sons.
Cook, T. D., & Campbell, D. T. (1979). Q u&-experimentation: Design and analysis issues for field settings.
Chicago: Rand McNally.
Correa, E. I., & Sutker, P.B. (1986). Assessment of alcohol and drug behaviors. In A. R. Ciminero, K.
S. Calhoun, & H. E. Adams (Eds.), Handbook of behavior assessment (2nd ed., pp. 446-495). New York:
John Wiley & Sons.
Craighead, W. E., Kazdin, A. E., & Mahoney, M. J. (1981). Behavior modification: Principles, issues and
applications. Boston: Houghton Mifflin Co.
666 S. N. Haynes and W. H. O’Brien

Dubin, R. (1969). Theory building. New York: Free Press.

Duncan, 0. D. (1975). Introduction to s&u&Ural equation models. New York: Academic Press.
Emmelkamp, P. M. (1986). Behavior therapy with adults. In S. L. Garfield & A. E. Bergin (Eds.),
Handbook of psychotherapy and behavior change (3rd ed.). New York: John Wiley & Sons.
Evans, I. M. (1985). Building systems models as a strategy for target behavior selection in clinical
assessment. Behavioral Assessment, 7, 21-32.
Ferster, C.B. (1965). Classification of behavioral pathology. In L. Krasner & L. P. Ulmann (Eds.),
Research in behavior modification (pp. 6-26). New York: Holt, Rinehart & Winston.
Fordyce, W. E. (1976). Behavioral methods for chronic pain and illness. St. Louis: Mosby.
Gadamer, H. G. (1975). Truth and method. New York: Seabury Press.
Goldfried, M. R., & Pomeranz, D. M. (1968). Role of assessment in behavior modification. Psychological
Reports, 23, 75-87.
Gross, F., 8c Strasser, T. (1983). Mild hypertension: Recent advances. New York: Raven.
Grove, W. M., & Andreasen, N. C. (1986). Multivariate statistical analysis in psychopathology. In T.
Millon & G. L. Klerman (Eds.), Contemporary directions in psychopathology, toward the DSM-IV (pp.
347-362). New York: Guilford Press.
Grunbaum, A. (1966). Causality and the science of human behavior. In R. Ulrich, T. Stachnik, & J.
Mabry (Eds.), Control of human behavior (pp. 3-10). Glenview, IL: Scott, Foresman and Co.
Hake, D. F., & Olvera, D. (1978). Cooperation, competition and related social phenomena. In A. C.
Catania & T. A. Brigham (Eds.), Handbook of applied behavior analysis: social and instructional processes
(pp. 208-245). New York: Halsted Press.
Harter, S., Alexander, P. C., & Neimeyer, R. A. (1988). Long-term effects of incestuous child abuse in
college women: Social adjustment, social cognition, and family characteristics. Journal of Consulting
and Clinical Psychology, 56, 5-8.
Hawkins, R. P. (1986). Selection of target behaviors. In R. 0. Nelson & S. C. Hayes (Eds.), Conceptual
foundations of behavioral assessment (pp. 31 l-385). New York: Guilford Press.
Haynes, S. N. (1978). Principles of behavioral assessment. New York: Gardner Press.
Haynes, S. N. (1986a). The design of intervention programs. In R. Nelson & S. C. Hayes (Eds.),
Conceptual foundations of behavioral assessment (pp. 386-429). New York: Guilford.
Haynes, S. N. (1986b). A behavioral model of paranoid behaviors. Behavior Therapy, 17, 266-287.
Haynes, S. N. (1989). Behavioral assessment of adult disorders. In M. Goldstein & M. Hersen (Eds.),
Handbook of psychological assessment. Elmsford, NY: Pergamon Press.
Haynes, S. N. (1990). Behavioral assessment. In M. Hersen, A. Kazdin, & A. S. Bellack Eds), The clinical
psychology handbook (pp. 435-460). New York: Pergamon Press.
Haynes, S. N., & O’Brien, W. (1988). The Gordian Knot of DSM-III-R Use: Integrating Principles of
Behavior Classification and Complex Causal Models. Behavioral Assessment, 10, 95-105.
Huba, G. J., & Bentler, P. M. (1982). A developmental theory of drug use: Derivation and assessment
of a causal modeling approach. In B. P. Bates & 0. G. Brim (Ed.), Life-span development and behavior
(pp. 147-203). New York: Academic Press.
Hume, D. (1911). A treatise of human nature. New York: Dutton. (Original work published 1740).
Hyland, M. (1981). Introduction to theoretical psychology. Baltimore, MD: University Park Press.
James, L. R., Mulaik, S. A., & Brett, J. M. (1982). Causal analysis: Assumptions, models and data. Beverly
Hills: Sage.
Johnston, J. M., & Pennypacker, H. S. (1980). Strategies and tactics of human behavioral research. Hillsdale,
NJ: Lawrence Erlbaum Associates, Publishers.
Joreskog, K. G. (1978). Structural analysis of covariance and correlation matrices. Psychomettika, 43,
443-477.
Kalish, H. I. (1981). From behavioral science to behavior mod@ation. New York: McGraw-Hill.
Kanfer, F. H. (1985). Target selection for clinical change programs. Behavioral Assessment, 7, 7-20.
Kanfer, F. H., & Grimm, L. G. (1980). Managing clinical change: A process model of therapy. Behavior
Modzjication, 4, 4 19444.
Kanfer, F. H., & Philips, J. (1970). b arning foundations of behavior therapy. New York: Wiley.
Kantor, J. R. (1970). An analysis of the experimental analysis of behavior (TEAB). Jountal of’the
Experimental Analysis of Behavior, 13, 10 I-l 08.
Kaplan, A. (1964). The conduct of inquiry: methodology for behavioral science. San Francisco: Chandler
Publishing Co.
Karoly, P. (1988). Handbook of child health asessment. New York: Wiley.
Keefe, F. J., Kopel, S. A., & Gordon, S. B. (1978). A practical gui& to behavioral a.ssessme7lt.New York:
Springer.
 Functional Analysis 667

Kenny, D. A, (1979). Cowelation and causality. New York: John Wiley.

Khouri, P. J., & Akiskal, H. S. (1986). The bipolar spectrum reconsidered. In T. Millon & G. L.
Klerman (Eds.), Conternporaly directions in psychopathology, toward the DSM-IV (pp. 457-47 1). New
York: Guilford Press.
Kratochwill, T. R. (1978). Single subject research: Strategies for evaluating change. New York: Academic
Press.
Kritzer, H. M. (1977). Political protest and political violence: A nonrecursive causal model. Social Forces,
5, 630-640.
Lanyon, R. I., & Lanyon, B. J. (1976). Behavioural assessment and decision-making: The design of
strategies for therapeutic behaviour change. In M. P. Feldman & A. Broadhurst (Eds.), Theoretical
and expzrimental bases of the behaviour therapies (pp. 289-329). London: John Wiley & Sons.
Lennox, D. B., Miltenberger, R. G., Spengler, P., & Erfanian, N. (1988). Decelerative treatment
practices with persons who have mental retardation: A review of five years of the literature. American
Journal of Mental Retardation, 92, 492-50 1.
Maisto, S. A. (1985). Behavioral formulation of cases involving alcohol abuse. In I. D. Turkat (Ed.),
Behavioral caseformulation (pp. 43-86). New York: Plenum Press.
Marlatt, G. A. (1985). Relapse prevention: Theoretical rationale and overview of the model. In Relapse
prevention: Maintenance strategies in the treatment of addictive behaviors (pp. 3-70). New York: Guilford.
McCormick, S. J. (1937). Scholastic metaphysics: Part I. Being, its division and causes. Chicago: Loyola
University Press.
McFall, R. M. (1986). Theory and method in assessment: The vital link. Behavioral Assessment, 8, 3-10.
Mirsky, A. F., & Duncan, C. C. (1986). Etiology and expression of schizophrenia: Neurobiological and
psychosocial factors. In M. R. Rosenzweig & L. W. Porter (Eds.), Annual review of psychology (Vol. 37,
pp. 291-319). Palo Alto, CA: Annual Reviews.
Nelson, R. 0. (1983). Behavioral assessment: Past, present, and future. Behavioral Assessment, 5,
195-206.
Nelson, R. 0. (1988). Relationship between assessment and treatment within a behavioral perspective.
Journal of Psychopathology and Behavioral Assessment, 10, 159-170.
Nelson, R. O., & Hayes, S. C. (1986). Conceptual foundations of behavioral assessment New York: Guilford.
O’Leary, K. D. (1984). Marital discord and children: Problems, strategies, methodologies, and results.
In A. Doyle, D. Gold, & D. S. Moskowitz (Eds.), Children in families under stress (pp. 35-46). San
Francisco: Jossey-Bass.
Ollendick, T. H., & Hersen, M. (1984). An overview of child behavioral assessment. In T. H. Ollendick
& M. Hersen (Eds.), Child behavioral assessment, principles and procedures (pp. 3-19). Elmsford, NY:
Pergamon Press.
Patterson, G. R. (1982). Coercive family processes. Eugene, OR: Castalia Publishing Co.
Patterson, G. R. (1986). Performance models for antisocial boys. American Psychologist, 41, 442-444.
Pepping, G., & Vaitl, D. (1981). Causal modeling: A tool in epidemiology. In T. M. Dembroski, T. H.
Schmidt, & G. Blumchen (Eds.), Biobehavioral bases of coronary heart disease (pp. 55-61). Basel: Karger.
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.
Piaget, J. (1952). The language and thought of the child. London: Routledge & Degan Paul.
Rachlin, H. (1970). Introdwtion to modern behaviorism. San Francisco: Freeman.
Rosenthal, T. L., & Rosenthal, R. H. (1985). Clinical stress management. In D. Barlow (Ed.), Clinical
handbook of psychological disorders (pp. 145-205). New York: Guilford.
Russell, B. (1948). Human knowledge: Its scope and limits. New York: Simon and Schuster.
Russo, D. C., Hamada, R. S., & Marques, D. (1988). Linking assessment and treatment in pediatric
health psychology. In P. Karoly (Ed.), Handbook of child health assessment (pp. 30-50). New York:
Wiley.
Schlundt, D. G. (1985). An observational methodology for functional analysis. BuL!etin of the Society of
Psychologists in Addictive Behaviors, 4, 234249.
Scott, J. P., Stewart, J. M., & De Ghett, V. J. (1974). Critical periods in organization of systems.
Developmental Psychobiology 7, 489-5 13.
Sellitz, C., Wrightsman, L. S., & Cook, S. W. (1976). Research methods in social relations. New York: Holt.
Simon, H. A. (1957). MO&~ of man. New York: Wiley & Sons.
Simon, H. A. (1971). Spurious correlation: A causal interpretation. In H. M. Blalock (Ed.), Causal
models in the social sciences (pp. 5-17). Chicago: Aldine Atherton.
Snyder, J. (1987). Behavioral analysis and treatment of poor diabetic self-care and antisocial behavior:
A single subject experimental study. Behavior Therapy, 18, 251-263.
668 S. N. Haynes and W. H. O’Brien

Taylor, C. B., & Agras, W. S. (1981). Assessment of phobia. In D. H. Barlow (Ed.). Behavioral assessment
ofudt& di.sorders (pp. 181-208). New York: Guilford Press.
Tofte-Tipps, S., Mendonca, R., & Peach, R. V. (1982). Training and generalization of social skills: A
study with two developmentally handicapped, socially isolated children. Behavior Modz$ation, 6,
45-71.
Turkat, I. D., & Maisto, S. A. (1985). Application of the experimental method to the formulation and
modification of personality disorders. In D. H. Barlow (Ed.), Clinical handbook of psychological disorders
(pp. 502-570). New York: Guilford.
Turner, S. M., Beidel, D. C., & Costello, A. (1987). Psychopathology in the offspring of anxiety
disorders patients. Journal of Consulting and Clinical Psychology, 55, 229-235.
Ulrich, R. (1975). Toward experimental living, Phase II: “Have you ever heard of a man named
Frazier, Sir?” In E. Ramp & G. Semb (Eds.), Behavior analysis: Areas of research and application (pp.
45-60). Engelwood Cliffs, NJ: Prentice-Hall.
Voeltz, L. M., & Evans, I. M. (1982). The assessment of behavioral interrelationships in child behavior
therapy. Behavioral Assessment, 4, 131-165.
Wahler, R. G., & Fox, J. J. (1981). Setting events in applied behavior analysis: Toward a conceptual and
methodological expansion. Journal of Applied Behavior Analysis, 14, 327-338.
Wickens, T. D. (1982). Models for behavior, Stochastic processes in psychology. San Francisco: W. H.
Freeman.
Wincze, J. P. (1982). Assessment of sexual disorders. Behavioral Assessment, 4, 257-27 1.
Wolpe, J. (1958). Psychothmae by reciprocal inhibition. Stanford, CA: Stanford University Press.
Wolpe, J. (1977). Inadequate behavioral analysis: The Achiles heel of outcome research in behavior
therapy. Journal of Behavior therapy and Exprimentul Psychiatry, 8, 1-3.
Wolpe, J., & Turkat, I. D. (1985). Behavioral Formulation of clinical cases. In I. Turkat (Ed.), Behavioral
case fomulution (pp. 213-144). New York: Plenum Press.

Received July 18, 1989

Accepted January 16, 1990