Mediators and Mechanisms of Change in Psychotherapy Research · Mediators and Mechanisms of Change...

ANRV307-CP03-01 ARI 20 February 2007 18:34

Mediators and Mechanismsof Change in PsychotherapyResearchAlan E. KazdinDepartment of Psychology, Yale University, New Haven, Connecticut 06520-8205;email: [email protected]

Annu. Rev. Clin. Psychol. 2007. 3:1–27

First published online as a Review inAdvance on October 11, 2006

The Annual Review of Clinical Psychology isonline at http://clinpsy.annualreviews.org

This article’s doi:10.1146/annurev.clinpsy.3.022806.091432

Copyright c© 2007 by Annual Reviews.All rights reserved

1548-5943/07/0427-0001$20.00

Key Words

mediators, moderators, mechanisms of psychotherapy, processesand outcomes of therapy, randomized controlled trials, mediationanalyses, treatment evaluation

AbstractThere has been enormous progress in psychotherapy research. Thishas culminated in recognition of several treatments that have strongevidence in their behalf. Even so, after decades of psychotherapy re-search, we cannot provide an evidence-based explanation for how orwhy even our most well studied interventions produce change, thatis, the mechanism(s) through which treatments operate. This chap-ter presents central requirements for demonstrating mediators andmechanisms of change and reviews current data-analytic and designsapproaches and why they fall short of meeting these requirements.The role of the therapeutic alliance in psychotherapy and cognitivechanges in cognitive therapy for depression are highlighted to illus-trate key issues. Promising lines of work to identify mediators andmechanisms, ways of bringing to bear multiple types of evidence,recommendations to make progress in understanding how therapyworks, and conceptual and research challenges in evaluating media-tors and mechanisms are also presented.

1

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


Contents

INTRODUCTION. . . . . . . . . . . . . . . . . 2CONCEPTUAL AND

DEFINITIONAL ISSUES . . . . . . . 3REASONS FOR STUDYING

MEDIATORS ANDMECHANISMS . . . . . . . . . . . . . . . . . 3

REQUIREMENTS FORDEMONSTRATINGMEDIATORS ANDMECHANISMS OF CHANGE . . 4Multiple Criteria . . . . . . . . . . . . . . . . . 4General Comments . . . . . . . . . . . . . . . 6

CURRENT STATUS OFRESEARCH ON MEDIATORSAND MECHANISMS . . . . . . . . . . . 6Examples Where Mediators and

Mechanisms are Discussed ButNot Well Established . . . . . . . . . . 6

General Comments . . . . . . . . . . . . . . . 8OVERVIEW OF METHODS FOR

STUDYING MEDIATORS ANDMECHANISMS INPSYCHOTHERAPY . . . . . . . . . . . . 8Statistical Techniques . . . . . . . . . . . . . 8Design Methods for Studying

Mediators and Mechanisms . . . . 10General Comments . . . . . . . . . . . . . . . 11

PATHS TO IDENTIFYING ANDELABORATING MEDIATORSAND MECHANISMS . . . . . . . . . . . 11

Meticulous Description . . . . . . . . . . . 12Moderators as a Path to Identifying

Mediators and Mechanisms . . . . 13Direct Intervention and

Manipulation. . . . . . . . . . . . . . . . . . 14Converging Lines of Work. . . . . . . . 15General Comments . . . . . . . . . . . . . . . 16

RECOMMENDATIONS FORRESEARCH . . . . . . . . . . . . . . . . . . . . . 16Use Theory as a Guide . . . . . . . . . . . 16Include Measures of Potential

Mediators in TreatmentStudies . . . . . . . . . . . . . . . . . . . . . . . . 17

Establish the Timeline of theProposed Mediator orMechanism and Outcome . . . . . . 17

Assess More than One Mediator orMechanism . . . . . . . . . . . . . . . . . . . 17

Use Designs that Can EvaluateMediators and Mechanisms . . . . 18

Examine Consistencies AcrossDifferent Types of Studies . . . . . 19

Intervene to Change the ProposedMediator or Mechanism . . . . . . . 20

SPECIAL CHALLENGES ANDOBSTACLES . . . . . . . . . . . . . . . . . . . . 21Mechanism-Outcome Relations . . . 21Everything in Moderation . . . . . . . . 22Measurement Development . . . . . . . 22

CONCLUSIONS. . . . . . . . . . . . . . . . . . . 23

INTRODUCTION

Several forms of psychotherapy for children,adolescents, and adults produce therapeuticchange, as demonstrated in scores of con-trolled treatment studies (Kazdin & Weisz2003, Lambert 2004, Nathan & Gorman2007). The changes can encompass social,emotional, cognitive, behavioral, educational,and physical spheres of functioning. We knowwell that therapy “works,” i.e., is responsiblefor change, but have little knowledge of whyor how it works.

Discussions and theory about why psy-chotherapy changes people are plentiful, butsupportive evidence is quite rare. The Amer-ican humorist and writer, Mark Twain (1835–1910) noted that, “everybody talks about theweather but nobody does anything about it.”1

Mechanism is the weather of psychotherapyresearch. The focus of this review is on whytreatment works, through what processes, and

1Mark Twain (Samuel L. Clemens) is credited with thisquote but some suggest that his coauthor A. Charles DudleyWarner wrote the statement (Twain & Warner 1874).

2 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


how change comes about, or more succinctly,on the mechanism(s) of therapeutic change.This review also discusses the importance ofstudying mediators and mechanisms of ther-apy, examines the limitations of existing dataevaluation and design strategies, and pro-vides recommendations for changes needed inresearch.

CONCEPTUAL ANDDEFINITIONAL ISSUES

Several related concepts are important to de-lineate in part because of their confusion butalso because they are relevant to elaboratingmechanisms (please see Table 1). It is useful tobegin with cause or causal relation. A random-ized controlled trial (RCT) may show thattreatment compared to no treatment leadsto therapeutic change. From the demonstra-tion, we can say that the treatment caused thechange, as the term “cause” is used in science.

Demonstrating a cause does not say whythe intervention led to change or how thechange came about. To evaluate how changecomes about, research often looks at me-diators. As noted in the table, mediator isa construct that shows important statisticalrelations between an intervention and out-come, but may not explain the precise pro-cess through which change comes about. Theterm “mechanism” refers to a greater level ofspecificity than does the term “mediator” andreflects the steps or processes through which

therapy (or some independent variable) actu-ally unfolds and produces the change. Mecha-nism explains how the intervention translatesinto events that lead to the outcome. Thisis easily confused with the notion of media-tion. For example, cognitions may be shownto mediate change in therapy, an importantlead perhaps. However, this does not explainspecifically how the change came about, i.e.,what are the intervening steps between cog-nitive change and reduced stress or anxiety. Inthis review, the primary focus is on mediatorsand mechanisms. The goal is to understandthe mechanisms of change; the study of medi-ators is often a first step, as is illustrated below.

Moderator refers to some characteristicthat influences the direction or magnitudeof the relation between the intervention andoutcome. If treatment outcome varies as afunction of characteristics of the patient ortherapist (e.g., sex, ethnicity, temperament),treatment delivery (e.g., individual versusgroup treatment), or cohort (e.g., so-calledgeneration X versus baby boomers), these lat-ter variables are moderators. I return to mod-erators below because they have importantbearing on mediators and mechanisms.

REASONS FOR STUDYINGMEDIATORS AND MECHANISMS

Evaluating mediators and mechanisms oftherapeutic change is important for sev-eral reasons. First, there is an embarrassing

Table 1 Key terms and concepts

Cause: a variable or intervention that leads to and is responsible for the outcome or change.Mediator: an intervening variable that may account (statistically) for the relationship between the

independent and dependent variable. Something that mediates change may not necessarily explain theprocesses of how change came about. Also, the mediator could be a proxy for one or more other variablesor be a general construct that is not necessarily intended to explain the mechanisms of change. A mediatormay be a guide that points to possible mechanisms but is not necessarily a mechanism.

Mechanism: the basis for the effect, i.e., the processes or events that are responsible for the change; thereasons why change occurred or how change came about.

Moderator: a characteristic that influences the direction or magnitude of the relationship between andindependent and dependent variable. If the relationship between variable x and y varies is different formales and females, sex is a moderator of the relation. Moderators are related to mediators and mechanismsbecause they suggest that different processes might be involved (e.g., for males or females).

www.annualreviews.org • Mediators and Mechanisms of Psychotherapy 3

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


wealth of treatments in use. For example, inthe context of child and adolescent therapyalone, 550+ psychotherapies can be delin-eated (Kazdin 2000). Some of these are knownto produce change; it is not likely that the dif-ferent treatments produce change for differ-ent reasons. Understanding the mechanismsof change can bring order and parsimony tothe current status of multiple interventions.

Second, therapy can have quite broad out-come effects, beyond the familiar benefitsof reducing social, emotional, and behavioralproblems (e.g., suicidal ideation, depression,and panic attacks). Therapy also alters physi-cal conditions (e.g., pain, blood pressure), im-proves recovery from surgery or illness, andincreases the quality of life (see Kazdin 2000).How do these effects come about? Elaborat-ing mechanisms of therapy will clarify theconnections between what is done (treatment)and the diverse outcomes.

Third, by understanding the processes thataccount for therapeutic change one ought tobe better able to optimize therapeutic change.Indeed, without understanding what is criticalto treatment and how it operates, we are at abit of a loss. Should we focus on more practice,catharsis, chatting, homework—what leads tochange and why? If we know how changescome about, perhaps we can direct better,stronger, different, or more strategies thattrigger the critical change process(es).

Fourth, extending treatments from re-search to clinic or “real world” settings willbe difficult without understanding how treat-ment works. We enter the clinical arena withone hand tied beyond our back if we applyan unspecified and possibly low dose of sometreatment that we do not understand. To opti-mize the generality of treatment effects fromresearch to practice we want to know whatis needed to make treatment work, what arethe optimal conditions, and what componentsmust not be diluted to achieve change.

Fifth, understanding how therapy workscan help identify moderators of treatment,i.e., variables on which the effectiveness ofa given treatment may depend. Understand-

ing the processes through which treatmentoperates can help sort through those facetsthat might be particularly influential in treat-ment outcome and permit better selection ofsuitable patients. For example, if changes incognitive processes account for therapeuticchange, this finding might draw attention tothe pretreatment status of related processes(abstract reasoning, problem-solving, attribu-tions) that might moderate who responds orfails to respond to treatment.

Finally, understanding the mechanismsthrough which change takes place is impor-tant beyond the context of psychotherapy.Many interventions or experiences in every-day life improve adjustment and adaptivefunctioning, ameliorate problems of mentaland physical health, help people manageand cope with stress and crises, and moregenerally navigate the shoals of life. As exam-ples, participating in religion, chatting withfriends, exercising, undergoing hypnosis,and writing about sources of stress all haveevidence in their behalf. Mechanisms thatelaborate how therapy works might have gen-erality for understanding human functioningbeyond the context of therapy. The otherside is also true. Mechanisms that explainhow other change methods work might wellinform therapy. Basic psychological processes(e.g., learning, memory, perception, persua-sion, social interaction) and their biologicalpathways (e.g., changes in neurotransmit-ters) may be common to many types ofinterventions, including psychotherapy.

REQUIREMENTS FORDEMONSTRATING MEDIATORSAND MECHANISMS OF CHANGE

Multiple Criteria

Establishing a mediator or mechanism hasseveral requirements. The requirements arehighlighted because they provide the back-ground for why changes are needed inresearch. I focus on mediation becausethis is an important interim step between

4 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


demonstrating a causal relation and under-standing concretely the mechanism of actionthrough which the effect occurs. Also, medi-ation is the primary focus of contemporaryresearch.

Strong association. Demonstration of astrong association between the psychothera-peutic (A) intervention and the hypothesizedmediator of change (B) is an initial require-ment. Then of course, there ought to be an as-sociation between the proposed mediator (B)and therapeutic change (C). Indeed, if thesethree variables are not related, the case for theoperation of a mediator is greatly weakened,if not eliminated.

Specificity. The second criterion refers tothe demonstration of the specificity of the as-sociation between the intervention, proposedmediator, and outcome. We would not wantmultiple mediators to account for the change,but rather show a more specific connection. Ademonstration that many plausible constructsdo not account for therapeutic change, withthe exception of one, strengthens the argu-ment that the proposed construct mediateschange.

Consistency. Replication of an observed re-sult across studies, samples, and conditions,i.e., consistency in the relation, contributesto inferences about mediators. We expect therelations among A, B, and C not to be sam-ple specific. Inconsistency across two or moredemonstrations does not necessarily meanthat the proposed mediator is not involved.The relation between a proposed mediatorand outcome might be perfectly consistent butmoderated by a variable we have not yet iden-tified. Yet, when consistency across studies isobtained, this greatly facilitates drawing in-ferences about whether a particular mediatormay be involved.

Experimental manipulation. Direct ma-nipulation through an experiment obviouslymakes a strong case between therapy and out-

come (A and C). This type of demonstra-tion (e.g., RCT) is common and demonstratescause. However, uncommon are experimentsthat manipulate the proposed mediator ormechanism (B) and show the impact on out-come (C). Experimental evidence strengthensthe case that a proposed mediator is responsi-ble for a change in the outcome of interest.

Timeline. A timeline must be established toinfer a causal relation or mediator of change.Causes and mediators must temporally pre-cede the effects and outcomes. Demonstrat-ing a timeline between cause and an effect, al-beit obvious, is the Achilles’ heel of treatmentstudies, as I elaborate below.

Gradient. Showing a gradient in whichstronger doses or greater activation of theproposed mediator is associated with greaterchange in the outcome can help make the casefor a particular mediator. A common analysisin medicine, epidemiology, and public healthis showing a dose-response relation. For ex-ample, there is a dose-response (and linear)relation between passive cigarette smoke (i.e.,exposure to secondhand smoke) and coronaryheart disease (He et al. 1999). Demonstratinga dose-response relation increases the plausi-bility of an agent being causally involved andmay point to likely mechanisms as well. Ofcourse, it is possible that there is no dose-response relation (e.g., a qualitative or on-off gradient) or that the relation is not linear.Such relations do not mean a particular con-struct is not causally related, but may makeinferences more difficult or require supple-mentary information.

Plausibility or coherence. Plausibility orcoherence of an explanation of how a me-diator or mechanism operates and integra-tion of findings with the broader scientificknowledge base contribute to the inferences.In medicine, pathophysiology often is invokedto meet this criterion. That is, in light of thefindings, is there a plausible, coherent, andreasonable process (e.g., buildup of plaque)


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


through which the disorder (e.g., atheroscle-rosis and heart attack) might be explained?An explanation is plausible because it invokesother information and steps in some process-outcome relation that are reasonable or sup-ported by other research.

The use of plausibility and coherenceto elaborate mechanisms is poignantly illus-trated in child abuse. Occasionally, parentsbring their very injured and pained child toan emergency room for treatment and tellthe physician that the child has been injured.Three examples from my own experience in-clude a child who allegedly fell off a bicycle,another who fell down the five front stairs ofa cement porch at home, and a child who gotinto a fistfight with a seven-year-old sibling.In each case, the physician (a different personfor each case) was suspicious because the in-juries consisted of large and deep bruise marksacross the back (with lines resembling a belt)and a mark that could resemble a belt buckleon the upper shoulder (child 1); three or possi-bly four round burn marks on the child’s backin the size of the end of a cigarette (child 2);and a black eye and open scalp wound un-der the hair (child 3). The physician in eachcase was suspicious primarily based on thecriterion of plausibility and coherence of the“mechanisms” or process involved leading tothese outcomes. In light of how a child is likelyto fall off a bicycle or down the stairs or tobe hit by young sibling, respectively, the in-juries were not very likely (plausible, coher-ent). However, the injuries were very plausibleby invoking another process or mechanism,namely, parent abuse of their children. (Onecould use the term “parsimonious” here, butI use “plausible” and “coherent” to focus ona greater level of specificity, namely, lookingat the operation of a mechanism and how itunfolds to produce an outcome.)

In relation to psychotherapy, plausibilityand coherence convey the importance of theo-retically based investigation of mediators andmechanisms of change. Here we need morethan a global construct that can be used to ex-plain onset of a clinical problem or therapeutic

change. We need a plausible account of howthe construct works and leads (in a testableway) to the outcomes.

General Comments

Drawing inferences about a mediator ofchange requires convergence of multiple cri-teria because they act in concert. Interpreta-tion of what accounts for or explains a par-ticular relation (mediator, mechanism) is notlikely to come from a single investigation. Bythe very nature of one of the criteria (consis-tency), replication is required. Yet, apart fromthat criterion, the case for a mediator is builtby a sequence of studies that may vary in theset of criteria they address and the clarity ofthe demonstration. After several studies, andwhen all or most of the criteria are met, onecan state that some intervening process ac-counts for change.

CURRENT STATUS OFRESEARCH ON MEDIATORSAND MECHANISMS

Mechanisms of treatment are increasingly dis-cussed, a likely precursor to more empiricalwork on the topic (e.g., Brent & Kolko 1998,Grawe 2004, Hofmann 2000, Kazdin 2006,Kazdin & Nock 2003, Weersing & Weisz2002). I believe this has fostered the viewthat we know about key processes leading tochange and are using suitable methodologi-cal, statistical, and design tools. Few empir-ical studies are available that meet even twoor three of the criteria mentioned previously.Consider briefly two therapy areas where me-diators and mechanisms of action are oftendiscussed.

Examples Where Mediators andMechanisms are Discussed But NotWell Established

Therapeutic alliance and treatment out-come. The therapeutic alliance refers to thecollaborative nature of the patient-therapist

6 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


interaction, their agreement on goals, andthe personal bond that emerges in treatment.A consistent finding is that the stronger thealliance the greater the therapeutic change(Horvath & Bedi 2002, Orlinsky et al. 2004).Studies that evaluate alliance during (e.g.,early, middle) treatment often show that al-liance predicts improvement in symptoms atthe end of treatment. Showing that alliancepredicts later symptom change by itself doesnot show that alliance plays a causal role,leaving aside the more specific matter of re-flecting a potential mediator. Merely becausesymptoms are not assessed in the middle oftreatment, does not mean they have not al-ready changed. Perhaps very early in treat-ment clients get a little better (some symptomimprovement) and as a result form a positivealliance with the therapist.

For example, a study of psychodynami-cally oriented supportive therapy showed thatchanges in alliance early in treatment pre-dicted symptom change at the end of treat-ment, in keeping with a large body of evidence(Barber et al. 2000). However, a critical ad-dition was included. Both symptom changeand alliance were assessed at multiple points.Symptom changes early in treatment pre-dicted alliance and that alliance also predictedfurther symptom change. Thus, the famil-iar alliance-outcome correlation in part re-flects the relation of early and later symptomchange, and the timeline is symptom changeto alliance as well as the reverse. Assessment ofboth symptom change and alliance were com-pleted at multiple points during the course oftreatment to identify these interesting rela-tions. Other studies with assessments at mul-tiple points have shown that a positive al-liance may follow improvements in symptoms(DeRubeis & Feeley 1990, Tang & DeRubeis1999).

From these examples, I do not wish to as-sert that alliance is invariably the effect ratherthan a cause. Indeed, the correlational evi-dence does not permit statements about causeor mediation. The reciprocal or bidirectionalrelations of symptoms and alliance are inter-

esting to pursue. The broader point is morepertinent here, to wit, in the vast majorityof studies the timeline between alliance andsymptom change has not been established.

Cognitions in cognitive therapy for de-pression. There are very few forms of psy-chotherapy as well established as cogni-tive therapy (CT) for unipolar depressionamong adults (American Psychiatric Associ-ation 2000, Hollon & Beck 2004). This treat-ment is evidence based, and then some, inlight of the range of trials. But why does CTwork, i.e., through what mediators or mech-anisms? In fact, little can be stated as to whytreatment works. In the development of thistreatment, the basis of therapeutic change wasthought to be changes in key cognitive pro-cesses (negative triad) that characterize manydepressed patients. CT is designed to changethese cognitions and in the process changedepression. The relation of cognitions andcognitive change in treatment to therapeu-tic change has been studied in different waysby assessing symptom change and cognitivechange at the end of treatment and show-ing that one shares variance with the other,or by evaluating whether cognitions assessedearly or in the middle of treatment corre-late with subsequent therapeutic change (e.g.,DeRubeis et al. 1990, Kwon & Oei 2003). Inboth of these methods, the timeline problemis unresolved, i.e., we do not know the order-ing of cognitive change and symptom change.This issue is similar to the concern raised inrelation to alliance, namely, in the vast ma-jority of studies, symptom change may havepreceded or occurred concurrently with cog-nitive changes. From research as currently de-signed and discussed, it is not possible to saythat cognitive processes serve as the mediatorsof therapeutic change.

Actually, unlike the research on alliance,perhaps one can say a bit more about me-diators and mechanisms of cognitive therapy.The research permits one to say more aboutwhat is not a likely mediator of the effectsof CT. Tests of mediation and evaluation of


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


therapeutic changes quite early in the courseof treatment suggest that improvements canreadily occur without changes in cognitions orin advance of implementing cognitive-changestrategies in treatment (e.g., Burns & Spangler2001, Tang & DeRubeis 1999). Challengesto the cognitive bases of change in CT fordepression are not new (Ilardi & Craighead1994, Whisman 1999). Perhaps we can statemore confidently now than before that what-ever may be the basis of changes with CT, itdoes not seem to be the cognitions as origi-nally proposed.

General Comments

I have highlighted two areas that are often dis-cussed as if we know the basis for the effect,i.e., the mechanism involved. In both allianceand cognitive therapy literatures, the timelineproblem is a methodological shroud that cov-ers most studies. Without clearly establishingthat the putative basis for the effect invariablycomes before symptom change, conclusionsabout mediation are in question. Of course,it follows that more specific statements aboutmechanisms are premature.

The two examples are intended to conveyhow a key criterion, establishing a timeline, isnot met in otherwise well-studied areas wheremechanisms are discussed. The examples wereused to illustrate this single point rather thanto review comprehensively the respective lit-eratures. Although I focused on the timelineproblem, other concerns in relation to theseliteratures could be illustrated by applying allof the criteria. Let me mention one to notethat this is not a vacuous claim.

We do not have a clear picture or set ofstudies that test how the putative mechanismunfolds in such a way as to alter symptoms. Inrelation to plausibility and coherence of themechanism-outcome relation, precisely whathappens that leads to symptom change? Forexample, through what process or sequence ofevents along any dimensions (cognitive pro-cesses, neurotransmitters, stress) does alliancelead to reductions in depression, anxiety, or

feelings that life is meaningless? The time se-quence problem is more basic, but how doesone get from “my therapist and I are bonding”to “my marriage, anxiety, and tics are better”?This is a leap with the intervening steps un-specified or untested, at least to my knowl-edge. The steps are not academic. If we couldidentify the steps, there may be other waysto activate them than through alliance alone.Also, we might identify novel moderators re-lated to the mechanisms that help us selectindividuals likely to vary in responsiveness tothe intervention.

OVERVIEW OF METHODS FORSTUDYING MEDIATORS ANDMECHANISMS INPSYCHOTHERAPY

Dominant methods of evaluating mediatorsin therapy research have limited what can beconcluded in large part because critical con-ditions mentioned previously are not met. Ihighlight current methods and discuss why wehave not been able to learn very much aboutmediators or mechanisms from them.

Statistical Techniques

Tests of mediation. Statistical evaluationcan play a central role in addressing whethera particular construct accounts for change.Multiple regression techniques, path analy-sis, structural equation modeling, and boot-strap methods are prominent options (Baron& Kenny 1986, Holmbeck 2002, Hoyle &Smith 1994, Kenny et al. 1998, MacKinnonet al. 2002, Shrout & Bolger 2002). Multipleregression analyses have been the most com-monly used techniques, and an overview of thelogic conveys the benefits as well as the prob-lems. Consider a hypothetical outcome studyin which we evaluate the following compo-nents:

� A = an intervention (the treatment)� B = a mediator or intervening variable� C = an outcome (therapeutic change)

8 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


The interrelations of A, B, and C, as eval-uated statistically, are used to infer whether Bcan explain why treatment works. In demon-strating mediation statistically, four condi-tions and tests are usually proposed:

� The treatment or intervention (A) mustbe related to therapeutic change (C);

� The treatment (A) must be related tothe proposed mediator (B);

� The proposed mediator (B) must be re-lated to therapeutic change (C); and

� The relation between the intervention(A) and therapeutic change (C) must bereduced after statistically controlling forthe proposed mediator (B).

The logic seems compelling because themany conditions, if met, suggest that the im-pact of treatment (A) on therapeutic change(C) really depends on some intervening pro-cesses (B). My simple description ignoresmany nuances. For example, the extent towhich the A-C relationship is reduced by con-trolling for B is a matter of degree, and thereis no cut point (beyond statistical significance[e.g., Sobel test]) to decide whether there isor is not support for a particular mediationalview. Also, the various statistical analyses arenot free from controversy or challenge (e.g.,Kraemer et al. 2001, 2002). I merely wish toconvey that the statistical analyses highlightedhere are used to evaluate mediators of changein therapy.

As is so often the case in statistical analyses,the concerns here are not about the statisticsper se but about their use and interpretation.A key interpretive limitation is the fact thatthe timeline between the mediator and theoutcome is not necessarily established. Mostpsychotherapy studies of mediation evaluatethe mediator and symptoms at pre and postor evaluate the mediator but not the symp-toms during treatment. Change in the media-tor is shown to correlate, predict, and accountfor variance in relation to the outcome. Thestatistical analysis alone cannot establish thatone influence preceded, and therefore possi-bly mediated, the other.

Even when the timeline is established,mediation does not necessarily suggest themechanism of action. If, for example, somecognitions are shown to come before symp-tom change and statistically explain theintervention-outcome (A-C) relation that byitself does not show the construct is themechanism. What precisely is the process ofchange, what are the steps from the constructto the change, and are other variables embed-ded in the measure? One might say that cog-nitions as a mediator might be a first step tomove to more fine-grained analyses, a defen-sible position. But one must also say that cog-nitions might not be the variable at all or atleast the cognitions of interest to the inves-tigator. Cognitions might be a proxy variablefor some other construct or be a global con-struct that includes multiple distinguishablecomponents (Kraemer et al. 2001).

In cross-sectional studies, the languageused to describe the data-analytic strategiesand the findings lends itself to misconceptionin relation to the timeline. For instance, re-gression analyses identify variables as “pre-dictors” or independent variables and othersas “outcomes” or dependent variables. Yet,the timeline is only established by the ex-perimental design. The distinction betweenantecedent (independent) and outcome (de-pendent) variables, from the standpoint of thesteps of the statistical analyses (and printouts)is arbitrary. Similarly, the accompanying dia-grams of the results with a flow chart of somekind to convey the authors’ views of mediation(e.g., a structural equation model with arrowspointing to the right) may lead the author orreader to conclude erroneously that there is atimeline.

Percentage of variance. Occasionally, re-searchers focus on the notion of percentageof variance accounted for, or “explained by,” avariable and consider this as proof that a criti-cal process or the critical explanation has beenidentified. If two variables are correlated (r),then one can identify the proportion of com-mon or shared influence (r2). For example,


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


therapeutic processes (e.g., alliance) “predict”therapeutic change. Researchers often notethat alliance accounts for a significant pro-portion of variance and sometimes even morevariance than other influences (e.g., treat-ment technique). Further interpretation is of-ten added to suggest this must mean that thealliance is why treatment leads to change oris the most significant/important influence intherapy.

Nothing in the measure of percentage ofvariance speaks to mediators or mechanisms.First, shared variance of alliance and outcomecould be huge, but that could be due to symp-tom change occurring before alliance. Sec-ond, the therapeutic alliance can “account for”treatment outcome variance but itself be ex-plained by one or more other variables, suchas common method variance in the allianceoutcome measures or even characteristics ofthe patients before they came to treatment(e.g., Kazdin & Whitley 2006, Zigler & Glick1986). In short, amount variance may or maynot point toward mediators or mechanisms.Whether the relation (any correlation) pro-vides meaningful leads will stem from the con-ditions required for establishing mediators, asenumerated above.

Biases in the data analysis. The way thedata analyses are completed occasionally canfoster the view that a critical influence or me-diator has been identified. An example in psy-chotherapy research pertains to integrity orfidelity of treatment, that is, the notion thattreatment was carried out as intended. In-vestigators evaluate whether clients who re-ceived the treatment as intended show greaterchange than those who did not (for a reviewsee Perepletchikova & Kazdin 2005). Obvi-ously, if critical procedures of treatment areresponsible for change, adherence to theseprocedures ought to make a difference in out-come. A measure of treatment integrity mayallow the investigator to delineate the ex-tent to which clients received the full doseor proper implementation of treatment. (Re-lated to this article, but not this section, many

studies that assess integrity do so at the endof treatment, at the same time that symptomchange is evaluated, raising some timeline is-sues we can forego here.) The investigatormay analyze the data with only those clientswho received the intervention as intended orwho received some minimal dose or by includ-ing all the data and showing a correlation be-tween how well treatment was implementedand the degree of therapeutic change. Receiv-ing the appropriate levels of treatment is notrandomly distributed and may well be con-founded with client or client x therapist char-acteristics. For example, getting more, better,or more carefully implemented treatment mayrelate to the personality of the client (or ther-apist), severity of his or her problems, matchof values and interests between the therapistand patient, and more.

The example is on treatment integrity butthe broader point is critical. In studying anyintervening process or construct of therapy,the investigator may wish to include in thedata analyses only those individuals for whomthe putative mediator was invoked or oc-curred. After random assignment of cases todifferent groups, keeping or using only caseswhere the mediator was effectively manipu-lated changes the equality of the sample andintroduces other constructs that are likely tobe confounded with the variable (mediator) ofinterest.

Design Methods for StudyingMediators and Mechanisms

Randomized controlled trials. RCTs re-main the primary method of demonstratinga causal relation between treatment and ther-apeutic change. The most common limitationof RCTs pertinent to this discussion is thefailure to establish a timeline between a pro-posed mediator or mechanism and outcome,as I illustrated above. Assessing the proposedmediator during treatment is necessary butnot sufficient to show the timeline betweenthe mediator and outcome. The assessment ofsymptom change is required during treatment

10 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


as well. Changes in assessment and design oftreatment trials, noted later in the chapter, canaddress this concern.

Component analyses of treatment tech-niques. One way that investigators attemptto get at mediators and mechanisms is by an-alyzing a treatment that is known to be effec-tive. In this context, treatment is considered a“package,” i.e., several distinguishable ingre-dients or components (e.g., x, y, and z). Dis-mantling studies provide all the componentsof the package to one group and variationswithout all of the components to one or moreother groups. The complementary approachis the constructive strategy in which one be-gins with one component and adds others (toother groups or individuals in a crossover de-sign) (see Kazdin 2003). The idea behind eachstrategy is to identify necessary, sufficient,and facilitative ingredients for treatment toachieve change. Under ideal circumstancesthere is a theory underlying the package with amore specific statement that some ingredient(e.g., focus on specific activities or exercises) isessential for therapeutic change. If a compo-nent is shown to contribute greatly to changeand little or no change occurs without thatcomponent, investigators often interpret thisas evidence for a mediator or mechanism.

Identifying a critical component of treat-ment, while valuable for many reasons (e.g.,for extending abbreviated versions of treat-ment to clinical work), does not provide directsupport for a mediator or mechanism. A com-ponent might achieve its effects for all sortsof reasons (processes) that must be assessed.Investigators might note that dismantling isa first step—true in principle. In practice,there are scores of very informative disman-tling studies that have not moved to the nextsteps to understand why a component mightbe important.

General Comments

In demonstrating causal relations and identi-fying candidates that might be mediators or

mechanisms of change, one ought to beginwith the criteria or requirements mentionedabove. Statistical analyses and experimentaldesigns (arrangements) are tools to addressthese requirements. One completes the sta-tistical analysis and then reverts to one ofthe criteria to ask, “Was this criterion met?”Whether the timeline of a supposed mediatoror mechanism of change or whether a con-struct is a plausible and coherent explanationof therapeutic change are not questions aboutstatistical analyses per se but about interpre-tation of those analyses.

PATHS TO IDENTIFYING ANDELABORATING MEDIATORSAND MECHANISMS

Understanding mediators and then mecha-nisms is not a matter of one study but is a mat-ter of creeping up on the process that drawson a series of projects often seemingly unre-lated or from different disciplines or types ofresearch. It is useful to think of the searchfor mechanisms as a chess game. Even thoughthere might be a final winning move (check-mate), the game is won on multiple fronts,an integrated sequence of actions, and con-verging moves that make checkmate possible.Critical to a chess game is that there is move-ment toward a goal; whether psychotherapyresearch shows this movement, at least in re-lation to understanding mechanisms, is not soclear. By movement, I refer to a progressionthat reflects depth or type of understanding ofthe factors that produce therapeutic change.

Consider the progression in understand-ing based on the concepts of correlate, risk fac-tor, and cause (Kraemer et al. 1997) in relationto the familiar example of cigarette smokingand lung cancer. The connection between thecharacteristic and outcome began as a corre-late (e.g., cross-sectional study findings thatpeople with lung cancer reported a higherrate of smoking) and moved to a risk factoror a correlate where one characteristic clearlyprecedes another (e.g., longitudinal studyfindings that those who smoked had higher


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


mortality rates for lung cancer). Then theconnection moved to being identified as acausal factor. Both quasi experiments withhumans and true experiments with animalsshowed that the amount of smoking alteredthe outcome. The dose-response relation ofthe findings as well as direct experimenta-tion supported the causal role of smoking anddisease. Once a causal role is demonstrated,one can ask more analytically how, or throughwhat mechanism, does the cause operate?

Much of research on treatment, but alsoon psychiatric disorders, identifies correlatesand risk factors (predictors) of the outcome.A difficulty is that the work rarely progressesto the next step that would provide a more in-depth evaluation of how the factor operates.Research on heart disease and cholesterol il-lustrates the progress and more in-depth eval-uation. Cholesterol has been known to be arisk factor and predictor of heart disease. Re-search progressed to evaluate whether choles-terol is causally involved, and indeed, it is.Changing (reducing) cholesterol, in fact, al-ters the subsequent risk for heart disease.

Psychological research often moves fromrisk factor to causal risk factor casually andwithout actual tests. In the usual instance, astudy identifies, let us say, two risk factors forsome deleterious outcome. The investigatorthen suggests that we ought to change, ad-dress, or in someway attend to these risk fac-tors to make people better. This is a non se-quitur. We do not know that the risk factorsbear any causal relation to the outcome. Thepractice of identifying a risk factor and thendiscussing an intervention to alter that riskfactor is so common that one looks for sys-tem issues that might foster it (e.g., journalsor funding agencies that insist on clinical im-plications on what might [merely] be a basiccritical finding).

Psychological research rarely moves theevaluation from correlate, to risk factor, tocausal agent. A usual reason given is thatone cannot experimentally manipulate crit-ical variables in humans (e.g., child-rearingpractices, attachment style), and therefore we

are confined to correlation. The problem iselsewhere, namely, little theory about keyconstructs (mediators) and how they couldbe studied, little effort to identify steps orprocesses (mechanisms) by which the con-struct leads to an outcome, and little useof convergent lines of inquiry that couldstrengthen inferences about causes, media-tors, and mechanisms. One does not need trueexperiment necessarily. One needs to buildthe case by meeting the requirements outlinedabove. There are many strategies to under-stand mechanisms or at least to move the ballforward significantly beyond correlation, as Iaddress below.

Meticulous Description

Understanding mechanisms is readily framedas an explanation of some phenomenon of in-terest. In research (and life) one can read-ily distinguish description (what is happen-ing) from explanation (why it is happeningor through what forces, processes, or mech-anisms). This is a helpful distinction to learnand to teach but for this discussion to blur. Inmany instances in science, one can conceiveof description and explanation as related andas opposite ends of a continuum. Dependingon the detail, level of analysis, and sequenceof moving from one to the other, descriptioncan become explanation. Let us continue theexample of cigarette smoking and lung cancer.

Spanning decades, cross-sectional and lon-gitudinal studies and research with humansand animals have established a causal rolebetween cigarette smoking and lung cancer,which is where we left off in the commentsabove. Establishing a causal relation does notautomatically explain the mechanisms, i.e.,the process(es) through which lung cancercomes about. The mechanism has been un-covered by describing what happens in a se-quence from smoking to mutation of cells intocancer (Denissenko et al. 1996). A chemical(benzo[a]pyrene) found in cigarette smoke in-duces genetic mutation at specific regions ofthe gene’s DNA that is identical to the damage

12 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


evident in lung cancer cells. This finding isconsidered to convey precisely how cigarettesmoking leads to cancer at the molecular level.This is an example of where the “what” (de-scription) can be sufficiently fine grained toconvey the “how.”

In therapy, proposed mechanisms mightencompass such constructs as the therapeuticrelationship. Research needs to go beyond thedemonstrated correlation and even the pre-dictive portion (i.e., on the assumption thatthe timeline can be firmly established). Oneway to move closer to understanding mecha-nisms would be to describe social interactionoutside of the context of therapy in relationto neurological or other biological indices(e.g., Adolphs 2003, Meyer-Lindenberg et al.2005). What changes take place in social in-teraction? There is still a huge leap betweenthese descriptions and explaining how a rela-tion in therapy leads to symptom change, butthis is a start and moves beyond where we aretoday in the therapy literature.

Moderators as a Path to IdentifyingMediators and Mechanisms

Moderators refer to characteristics that influ-ence the direction or strength of the relationbetween an intervention and outcome. For ex-ample, we know that childhood signs of an-tisocial behavior predict later delinquency inadolescence for boys but not for girls, i.e., sexmoderates the relationship (Tremblay et al.1992). This suggests that different mediatorsand mechanisms are likely to be involved inthe onset of delinquency for boys and girls.The finding is very useful indeed, because anysearch for mediators that combined boys andgirls might not find an effect; a clear effectfor boys might be diluted or nullified by theabsence of any effect among girls.

Moderators can play a more direct role inelaborating mediators and mechanisms of ac-tion, and these have yet to be exploited. Con-sider an example of the effect of experienceduring childhood on subsequent criminal be-havior, where a genetic characteristic is a mod-

erator. As is well known, children with a his-tory of physical abuse are at risk for later anti-social behavior. Most people who are abusedas children do not engage in antisocial be-havior. A genetic characteristic moderates therelationship. Abused children with a geneticpolymorphism (related to the metabolism ofserotonin) have much higher rates of antiso-cial behaviors than those without this poly-morphism (Caspi et al. 2002). Among boyswith the allele and maltreatment, 85% de-veloped some form of antisocial behavior(diagnosis of conduct disorder, personalityassessment of aggression, symptoms of adultpersonality disorder, or court conviction of vi-olent crime) by the age of 26. Individuals withthe combined allele and maltreatment consti-tuted only 12% of the sample, but accountedfor 44% of the cohort’s violent convictions.Further research has replicated and extendedthe finding by noting that parent neglect aswell as abuse in conjunction with the polymor-phism increase risk for conduct problems andviolence (Foley et al. 2004, Jaffee et al. 2005).

So far, this is a fascinating illustration ofmoderation. However, closer scrutiny is help-ful here because it hints at mechanism. Caspiand colleagues (2002) looked at the allele formonoamine oxidase A (MAO-A) because:

� The gene that encodes the MAO-A en-zyme that metabolizes neurotransmit-ters is linked with maltreatment victim-ization and aggressive behavior;

� A rare mutation causing a null allele atthe MAO-A locus in human males is as-sociated with increased aggression;

� Animal gene knockout studies show thatdeleting this gene increases aggression;and

� Restoring this gene expression de-creases aggression.

In one sense we have identified amoderator—the influence of an indepen-dent variable (abuse in the home) and out-come (antisocial behavior years later) is influ-enced by some other characteristic or variable(MAO-A allele). Clearly, we have much more


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


because the work and the results it gener-ated are beginning to point to the genetic andmolecular underpinnings. We do not knowhow the allele and abuse traverse specific stepsfrom a to z in which aggression emerges, butwe are getting close. For example, recent find-ings show the neural mechanisms throughwhich the genetic influence is likely to operate(Meyer-Lindenberg et al. 2006). The MAO-Aallele is associated with diminished brain cir-cuitry related to impulse control that wouldpromote aggression.

The type of moderator work illustratedhere has some characteristics uncommon inthe usual moderator research in relation totherapy. In the illustration, the moderator wasidentified based on considering mechanismsthat might be involved. Theory about poten-tial mechanisms, prior correlational evidence(abuse and victimization), and other studiesindirectly related served as background. Inmuch of treatment research and moderatorresearch in clinical psychology more gener-ally, moderators of convenience are used, suchas information routinely obtained and globalindices (e.g., socioeconomic class, ethnicity,comorbidity). There is little sound theory be-hind the research or predictions that derivefrom proposing precisely what facets of themoderator might be important in explain-ing the relation. Thus, there is a vast liter-ature with analyses showing boys and girls,younger versus older, and this ethnic groupversus that ethnic group differ. This is fine asa start, but much of the research never getspast the “start.” Moderation can lead to in-sights about mediation, as the example of ag-gression shows, but it requires tests of ideasabout what the mechanisms are or could be.

Direct Intervention andManipulation

Direct manipulation of a proposed mecha-nism is of course a powerful way to move ourunderstanding forward. Consider the work onfear conditioning and psychotherapy. Therehave been decades of research on Pavlovian

conditioning of fear in humans and animals.Conditioning as an explanation of fear acqui-sition and extinction as an explanation of fearreduction or elimination are useful paradigmsfor the processes that might be involved intreatment. Research has suggested that ex-tinction is not merely unlearning (elimina-tion of a previously established connection)because the connection is not erased or lost,but rather is actively suppressed through re-learning of an acquired inhibition (Myers &Davis 2002).

Understanding the neurological under-pinnings of extinction has moved to interven-tion research. Conditioning and extinction offear depend on a particular receptor in theamygdala (N-methyl-D-aspartate) (see Daviset al. 2006). Chemically blocking the recep-tor shortly before extinction training blocksextinction in animal research, a finding thatshows a dose-response relation. Blocking thereceptor after extinction training also blocksextinction, which suggests that the consoli-dation process can be interrupted. A com-pound (D-cycloserine) binds to the receptorand makes the receptor work better, i.e., en-hances extinction when given before or soonafter extinction training.

The laboratory research has moved totherapy trials where exposure therapy, basedon an extinction model, was evaluated to testwhether enhancing a mechanism of extinctionwould improve treatment outcome. An initialcontrolled trial was completed with individu-als who suffered acrophobia (fear of heights)(Ressler et al. 2004). Exposure therapy, oneof the most well demonstrated treatments foranxiety, was used as the treatment. The goalwas to extinguish fear; exposure to heights wasprovided in presentations via virtual reality.Presumably, activation of the critical receptor(with D-cycloserine) would improve theeffects of exposure therapy (i.e., augment ex-tinction). Indeed, that was found. Participantswho received the drug (oral administrationtwo to four hours before each session) showedgreater improvements than those who re-ceived a placebo. The results were reflected

14 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


on several measures of avoidance, anxiety,global improvement, and self-exposures toreal-world heights as well as skin conduc-tance, as a measure of anxiety during and aftertreatment. The effects were evident one weekand three months after treatment. The en-hanced outcome effects (with D-cycloserine)have been replicated for the treatment ofsocial anxiety (Hofmann et al. 2006).

The model of the research program, i.e.,movement from moderators and mediatorsto mechanisms and from basic to applied re-search, more than two outcome studies needsto be replicated. Understanding mechanismsof learning and extinction, but also memory,belief, persuasion, control, stress alleviation,anticipation, and so on are within empiricalreach in a similar way. Once such mechanismsare studied, potential targets can be identified,with a similar paradigm of manipulating themechanisms. Manipulation might be throughpsychological interventions as well as biolog-ical ones.

Converging Lines of Work

The prior examples emphasize key elementsof the demonstrations such as studying mod-erators or intervening directly on an intendedmechanism. Actually, the emphases are usefulbut the examples are part of a broader strat-egy. Multiple lines of evidence are likely tobe needed to converge on precisely what themechanism is. The examples I have providedfocus on moderators and mechanisms and un-derpinnings that are biological. This is nota coincidence; the technological advances forstudying biological processes are astoundingand in some cases, processes (e.g., neurotrans-mitter or synapse activity) can be observedin real time. Studying mediators and mecha-nisms and key theses of this review have noth-ing inherently to do with biology. The focuson mechanisms and the convergence of mul-tiple lines of work can be gleaned from study-ing psychological processes and human inter-action, as illustrated in research on parentingpractices in the homes of young children.

In the 1960s, Patterson and his colleaguesbegan an extensive research program designedto understand the emergence and mainte-nance of aggressive child behavior (Patterson1982, Patterson et al. 1992). The studies in-cluded directly observing child and parent in-teraction in the home in a detailed fashion(29 different behaviors and interactions oc-curring from moment to moment includingsuch behaviors as attending to and unwit-tingly reinforcing child deviant behavior, us-ing commands, delivering harsh punishment,and failing to attend to appropriate child be-havior). Among the many interaction pat-terns, those involving coercion have receivedthe greatest attention (Patterson et al. 1992,Snyder & Stoolmiller 2002). Coercion refersto a sequence of parent and child actionsand reactions that increase the frequency andamplitude of angry, hostile, and aggressivebehaviors. The sequence may begin with anargument over some action that has or hasnot been performed. This intensifies throughverbal statements (e.g., yelling, screaming) tomore intensive actions (e.g., hitting, shoving).Ultimately, a high-intensity action of one per-son (usually the child) ends the aversive be-havior of the other person (usually the par-ent). Thus through negative reinforcement(increase in likelihood of a behavior that ter-minates an aversive condition), children areinadvertently rewarded for their aggressiveinteractions. Their escalation of coercive be-havior is increased in the process, and childrenare likely to be more aggressive (more often,higher intensity) in the future. The parent be-haviors are part of the discipline practices thatsustain aggressive behavior. The interactiondoes not place a single-unidirectional causalrelation between the parent and child. Rather,a dynamic interaction exists in which aversivebehavior on the part of both parties escalatesand does so in a way that systematically pro-grams, fosters, and develops greater deviancein the child.

The parent-child interaction does not nec-essarily determine the next behavior but in-creases the probability that the behavior


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


would move in one direction and toward someend rather than another. Given x (behavior ofthe parent), y (behavior of the child) is muchmore likely to occur and so on in the sequence.Conditional probabilities of behaviors wereused to describe the interactions leading toaggression. I mentioned previously that thereis a way in which meticulous description canblend with and become an explanation. Muchof the sequence of interactions was of thistype, namely, showing that the interactionsfostered aggression and that the timeline wasclear.

The studies showing that specific ineptchild-rearing practices contributed to aggres-sive behavior served as a model for inter-vention (Reid et al. 2002). Several studieshave shown that changing parent-child in-teractions (via parent management training)significantly reduces aggressive behavior andrelated conduct problems (see Kazdin 2005,Reid et al. 2002 for reviews). Thus, a converg-ing set of studies showed a sequence of coer-cive parent-child interactions leading to esca-lation, support for a model that explains theinteraction (coercion theory and reinforce-ment leading to escalated aggression), and anintervention that changes putatively criticalparenting processes that controvert aggressivechild behavior. The unfolding of coercive be-havior and effective intervention go very farto suggest the mechanisms involved in onsetand elimination of aggression in the home, atleast for some children.

General Comments

The discussion highlights examples oftreatment-related research that moves from

Table 2 Recommendations for research

1. Use theory as a guide2. Include measures of potential mediators in treatment studies3. Establish the timeline of the proposed mediator or mechanism and

outcome4. Assess more than one mediator or mechanism5. Use designs that can evaluate mediators and mechanisms6. Examine consistencies across different types of studies7. Intervene to change the proposed mediator or mechanism

causal relations toward understanding me-diators and mechanisms. I have omittedstudies on mediation, which have becomerelatively common. The reason for omittingthese was explained in the discussion ofstatistical tests of mediation, namely, thestudies rarely establish the critical conditionsfor establishing a timeline and a mediatoris not necessarily a mechanism. When thetimeline is not established, it is even toomuch of a leap to imply there is a mediationrelation beyond a statistical connection inwhich the mediator and outcome could bereversed.

RECOMMENDATIONS FORRESEARCH

Psychotherapy research has a long history ofdiscussing processes of therapy, but little re-search has addressed the conditions neces-sary to establish mediators or mechanisms.In general, the investigation of mediators andmechanisms of therapy can be improved inseveral ways. Table 2 lists recommendationsto enhance our understanding of therapeuticchange.

Use Theory as a Guide

The guiding question for treatment researchis how does treatment achieve change? Theanswer may involve basic psychological pro-cesses (e.g., memory, learning, informationprocessing) or a broader theory (e.g., moti-vation). It is no longer sufficient to provideglobal conceptual views (e.g., psychodynamic,cognitive, or familial) that foster a treatmentapproach or orientation toward what to doin the sessions. Rather, to ensure progress,specific conceptual models are needed to ex-plain those processes that are responsible fortherapeutic change. What is needed further isgreater specificity in conceptualizing not onlythe critical construct but also how that oper-ates to produce symptom change. We needmore than tests of mediation to understandmechanisms. Mediation tests of plausible

16 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


constructs can provide a screening device ofsorts to identify potential avenues to be pur-sued in a fine-grained fashion.

It would be helpful for intervention re-search to identify “candidate mediators”and mechanisms or plausible constructs thatwould explain or account for (statistically)therapeutic change, manipulate the proposedmechanism, assess to ensure it has been ma-nipulated, and then evaluate change. For ex-ample, in relation to tobacco use amongteenagers, several mediators that may serve asuseful targets have been identified, includingcoping skills of the youth, peer influences, andavailability of tobacco, among others (Mac-Kinnon et al. 2002). The targets can be thefocus of intervention. If one of these targetsleads to change in tobacco use, this wouldserve as an excellent basis for further workto understand exactly how the influence pro-duces change. We need next-step researchthat begins with theory but tests directly howthe proposed mediator operates.

In research training, there is often a strongdemand of the investigator to begin with a the-ory or conceptual model. The study that fol-lows is a test of that theory. However, the goalof research is to end up with an understand-ing of how therapy works. This goal can beachieved by research that generates hypothe-ses and theories in addition to research thattests hypotheses. There is far too little re-search that focuses on generating hypothesesfrom careful observation and on building the-ory that can be tested (Kazdin 2003, McGuire1997).

Include Measures of PotentialMediators in Treatment Studies

The mediator or mechanism ought to be spec-ified so it can be measured. Studies occasion-ally include such measures (Hofmann 2000,Weersing & Weisz 2002), although their ad-ministration has not allowed evaluation oftimelines. Yet, measures are available. Morefine-grained analyses will be needed to studythe unfolding of processes over time and how

a change in some process results in symptomchange. As a prerequisite to understanding,assessments of potential mediators ought tobe included in treatment studies.

Establish the Timeline of theProposed Mediator or Mechanismand Outcome

It is important to establish that the proposedmediator is changing before the outcome.The timeline has two requirements: (a) theproposed mediator must be assessed beforethe proposed outcome, and (b) the “outcome”must also be assessed early to ensure the me-diator has in fact changed before the outcome.Even during the middle of treatment, longbefore the investigator may be interested intherapeutic change, it is quite possible thatimprovements occur in the client and theseimprovements come before change in the pu-tative mediator.

Assessment is the main change needed inresearch. Assessment on multiple occasionsduring treatment can provide information onthe timeline of mediators and mechanisms andoutcomes and the possibility of bidirectionalchanges, i.e., each one influences the otherin some way and at different points. Assess-ment on a session-by-session basis (i.e., everyoccasion over the course of treatment) per-mits evaluation of the mediator of change andsymptom reduction and considers individualdifferences in the course of these changes.

Assess More than One Mediator orMechanism

The accumulation of evidence would profitfrom the assessment of more than one media-tor in a given study. It is rare that one mediatoris studied, and hence there may be little valuein raising the bar even higher by recommend-ing the assessment of two or more mediators.Recommending the assessment of more thanone mediator during treatment means that theassessment battery (e.g., how many measures)will increase as each mediator is added to the


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


design. In laboratory (efficacy) studies of ther-apy, the addition of one or two measures dur-ing the course of treatment may not be par-ticularly onerous.

The assessment of multiple mediators ina given study has enormous benefits. If twoor more mediators are studied, one can iden-tify if one is more plausible or makes a greatercontribution to the outcome. In addition, theassessment of multiple potential mediatorswithin individual studies is economically ef-ficient, given the tremendous amount of timeand resources needed for any treatment inves-tigation. Across many studies, some mediatorsmay repeatedly emerge as possible contenderswhile others fall by the wayside.

Use Designs that Can EvaluateMediators and Mechanisms

Table 3 lists five designs that vary in the as-sessment of potential mediators or mecha-nisms of change and treatment outcome. As-sume all to be RCTs in which treatment iscompared with no treatment. The first andmost commonly used design variation omitsassessment of potential mediators. RCTs areexcellent in demonstrating a causal relationbetween the intervention and therapeuticchange. Yet, the designs that resemble the firstvariation can say nothing about mediators ormechanisms, even though we as authors oftendo. In the second design variation, symptomsand possible mediators are assessed at the

same time at pre- and post-treatment. Withthis variation, conclusions cannot be reachedabout whether improvements in symptoms in-fluenced the proposed mediator or vice versa,or whether both were altered by anothervariable.

The third design variation assesses symp-toms at pre- and post-treatment, but duringthe course of treatment (on one or more oc-casions) the proposed mediator is assessed.The data analyses then evaluate whether theprocess during treatment contributes to (pre-dicts, accounts for) treatment outcome. Thisresearch gives a strong but misleading im-pression that a timeline is established betweensome process (e.g., cognitions, alliance) andtherapeutic change. The failure to measuresymptoms at the same time or indeed beforethe mid-assessment of the supposed media-tor precludes conclusions about whether themediator comes before symptom change. Justbecause symptoms were not assessed in themiddle of treatment does not mean they didnot improve or indeed even improve beforethe putative process variable.

The fourth design variation improves onthe prior designs by including assessmentof the proposed mediator and the outcome(symptoms) during treatment. Ideally, therewill be more than one assessment occasionduring treatment. This variation can evaluatethe time sequence, i.e., whether changes inthe mediator preceded symptom change andwhether symptom change preceded change in

Table 3 Outcome study designs and evaluation of mediators or mechanisms

Mechanism assessment Outcome assessment

Design variation Pre During Post Pre During Post1. Usual outcome design N N N Y N Y2. Concurrent study of mechanisms and

outcomesY N Y Y N Y

3. Assessment of mechanism during treatment N Y N Y N Y4. Assessment of mechanisms and outcomes

during treatmentY Y Y Y Y Y

5. Assessment of mechanisms and outcomesall or most sessions

Y Y, Y, Y, . . . Y Y Y, Y, Y, . . . Y

Y = yes, assessment is conducted; N = no assessment.

18 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


the putative mediator (which may make themediator an effect rather than a cause). How-ever, if the assessment is only on one occasionduring treatment, it is possible that both theproposed mediator and symptom change oc-curred or appear to have occurred at the sametime. Their relation might not be easy to dis-cern and the possibility exists that a third vari-able led to both changes in the mediator andsymptoms.

A disadvantage of the fourth design vari-ation is that it presumes that the course ofchange for both the mediator and outcome iscaptured by measuring each of these at onlyone (or even two) fixed points during treat-ment. There could be great variation in whenthe change is made among patients receiv-ing the same treatment. Both the mediatorand symptoms may change at different pointsamong a set of patients. The fifth design varia-tion, an extension of the prior design, providesa more fine-grained analysis of change in me-diator and symptoms and overcomes this con-cern. Assessments are made so that one canexamine the course of change of the media-tor and symptoms and can take into accountindividual differences in when the changesoccur.

Examine Consistencies AcrossDifferent Types of Studies

Understanding mediators and mechanismsthrough which therapeutic change occurscould profit from different types of studies,beyond those that might be construed as ther-apy research. Conclusions from these studiesmay be consistent and converge in making aparticular process plausible.

Animal laboratory research. Granted,many of proposed mediators of therapymay not be amenable to mouse or zebrafish models. Yet, some of the mediators andmechanisms of therapy might be studied inthe lab, and we ought not to be shy aboutthem or shy away from them. I mentionedabove the work on fear conditioning and how

understanding key mechanisms of extinctionhas already improved the effectiveness ofextinction-based treatment (Davis et al.2005). Therapeutically relevant phenomena(e.g., attachment, separation, social support)can be studied in animal research to identifyprocesses (e.g., changes in the structure orfunction of the brain) and their consequencesin behavior. These in turn might direct re-search to plausible underpinnings to supporta conceptual view of the mechanism of ther-apeutic change. Such tests, far removed fromtherapy settings, provide important tests ofprinciple. For example, maternal caregivingbehaviors (e.g., nursing, licking, grooming)among rats influence the responsiveness tostress in the offspring; the effects can be seenin behavioral as well as from the neurologicaland endocrine responses of the offspring(e.g., Champagne et al. 2003, Pruessneret al. 2004). This might well be pertinentto understanding stress, coping with stress,and interventions designed to amelioratestress.

Naturalistic studies. If one is proposing amediator of change, is there a sample, popula-tion, or setting in which this mediator may beexpected to vary naturally, i.e., without inves-tigator intervention? For example, if changingparenting style is proposed to explain why aparent- or family-based treatment of a childclinical problem is effective, naturalistic stud-ies examining families with and without thesepractices and the short- and long-term childbehaviors with which these are associated arerelevant. Among naturally occurring instancesof the process or construct, is there a dose-response relation?

As an example and following up on theprior example of maternal caregiving amongrats, naturalistic studies of “normal” moth-ering have revealed that stress reactivity inhuman infants is influenced by maternalcaregiving (e.g., sensitivity, availability, lack ofintrusiveness) during routine activities (e.g.,feeding, meal preparation) very much in keep-ing with the animal research highlighted


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


above (Hume & Fox 2006). Low-qualitycaregiving was associated with greater stressreactivity of their infants (e.g., fearfulness,more right frontal brain asymmetry), an effectthat could not be explained by infant tempera-ment. Caregiving in relation to stress responseand reactivity “behaves” in a similar way acrossdifferent research paradigms and draws atten-tion to mediators or mechanisms that mightbe pertinent to therapy (e.g., trauma, stress,coping).

Naturalistic studies by themselves may notpermit strong causal conclusions. Yet, suchevidence can be enormously helpful. Manyadvances in understanding cancer, heart dis-ease, and stroke began by looking for variationin putative mediators (e.g., in health habits,diet) among individuals with varied outcomes(e.g., morbidity, mortality). Observing pro-cesses that may be operative in the naturalenvironment and their short- and long-termcorrelates can be very useful, for both gener-ating and testing hypotheses about mediatorsand mechanisms.

Qualitative research. Qualitative researchis an approach to the subject matter of hu-man experience and focuses on narrative ac-counts, description, interpretation, context,and meaning. Among the key characteris-tics is the in-depth study of the phenomenaof interest. Individual participants or casesare focused on intensely to examine pro-cesses, meaning, characteristics, and contexts.Qualitative research is a rigorous, verifiable,empirical, and replicable set of methodolo-gies that encompasses many different disci-plines and diverse design, assessment, anddata-analytic strategies (Berg 2001, Denzin &Lincoln 2005). In the context of the presentdiscussion, qualitative research might studythe process of therapy, how the patient andtherapist experience that process, and whatmight be critical actions or cognitions andhow they relate to improvements outside oftreatment. Qualitative research can provide afine-grained analysis by intensively evaluatingthe richness and details of the process, includ-

ing who changes and how change unfolds, andwho does not change and what might be op-erative there.

Laboratory studies of therapeutic pro-cesses. Such studies are viewed with ambiva-lence because they do not show whether treat-ment works in “real-life settings.” Controlledstudies of therapy in research rather than clin-ical settings are more important now than everbefore. The careful control afforded such re-search is precisely what is needed to identifymediators and mechanisms. Translational re-search without knowing what to translate willhave a checkered yield in clinical applicationsof treatment.

Intervene to Change the ProposedMediator or Mechanism

An excellent strategy is to conduct an exper-iment in which the proposed mediator is infact altered or varied across groups, as illus-trated in the treatment study cited above onextinction of fear (Davis et al. 2005). Groupsrandomly composed might be assigned to low,high, and medium levels of a proposed medi-ator (as a general concept) or mechanism (as amore specific set of steps expected to lead di-rectly to the outcome). Strong support wouldbe evident from findings that outcome variesdirectly as a function of levels of the manipu-lated dose.

Intervening to change a mediator is anexcellent strategy. Here, too, assessing morethan one mediator would be helpful in under-standing why change occurred. Intervening toalter the mediator and assessing the level ofthat mediator (as a check on the manipula-tion) and two or more plausible, other medi-ators that are not manipulated would be anelegant way to evaluate mechanisms. In sucha study, one can rule out or make implausiblesome mediators while providing evidence onbehalf of another mediator.

A variation of the intervention approachis worth distinguishing, and I refer to thishere as “therapy knockout studies.” The term

20 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


draws from genetic work (e.g., gene knockoutstudies with mice) where a particular gene isomitted or altered and the effects are evidenton behavior or some other facet suspectedto be controlled by the gene. The generalmodel of this research would be a wonder-ful extension to psychotherapy mechanisms.More specifically, if the investigator believesor theory predicts that a specific mechanismaccounts for change, it would be useful toprovide the therapy with an added interven-tion that is designed to “knock out” (inacti-vate) the mechanism. If role-play, practice, orwarm fuzzy relations are critical to the tech-nique, give two variations of the treatment:the original and the original with an effortto inactivate the mechanism. As with any sin-gle study, supportive evidence that treatmentworked wonders only when the mechanismwas allowed to operate could be explainedin multiple ways. Even so, this evidencewould be a superb addition to accumulatingevidence.

SPECIAL CHALLENGES ANDOBSTACLES

There are multiple challenges in consideringmediators and mechanisms that extend be-yond a few changes in designs or measure-ment strategies. Consider some of the keychallenges briefly.

Mechanism-Outcome Relations

The discussion has implied a simple model inwhich a single mechanism leads to a singleoutcome or the effects are strong, simple, andlinear. Yet multiple variants have implicationsfor conceptualizing, designing, and interpret-ing research.

Single agent (influence), multiple out-comes. One complexity occurs when a sin-gle influence produces multiple outcomes.For example, cigarette smoking leads to sev-eral physical and psychological conditions. Insome of these, we know there is a causal re-

lation and have identified the mechanism; inothers, we know of increased risk. The per-vasiveness of the influence of smoking on somany conditions can introduce complexitiesin the search for mechanisms because so manybiological systems are involved. There may bemultiple and different mechanisms for the sin-gle agent but different outcomes. On the otherhand, some common pathways may exist thathelp focus research.

Multiple influences, single outcomes.Similar outcomes may be reached throughmultiple paths. Thus, we do not expect tosee all people with a particular characteris-tic (high kindness, bipolar disorder) to havereached these delineations through the samepath. There are multiple paths. The paths mayreflect similar mechanisms activated by differ-ent experiences or different mechanisms. Forexample, low IQ could result from genetic,prenatal, cultural, and postnatal toxic (e.g.,lead) influences. This “single outcome” hasmany paths. Essential to work on mediatorsand mechanisms is distinguishing differentcourses or paths and moderating influences.Looking for one explanation or mechanismfor one group, one therapy, or one out-come may yield little. Conceptual work onpossible moderators and exploratory stud-ies (and yes, fishing expeditions) followed byconceptual work will be critical to look forsubgroups.

Nonlinear relations. We are trained tothink, love statistics that generate, and per-haps even are victims of a cognitive heuristic(I call it slippery-slope thinking) that attractsus to linear relations. Many critical relationsare nonlinear. For example, cholesterol andrisk for heart disease is positive and linear;higher cholesterol increases risk. Cholesteroland stroke are U shaped, i.e., nonlinear sothat low and high cholesterol increase risk.Nonlinear relations propose a challenge in thesense that dose-response relation (as a linearfunction) is one clue on the path toward mech-anisms, although it is not essential. Looking


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


at means of groups and using statistics thatevaluate and search for linear relations canspeed or delay progress. We may find a weakrelation or no relation between an agent andoutcome (e.g., cholesterol and stroke) for thesample as a whole. Analyses of subgroups andtests of nonlinear relations to identify reliablepatterns of mediator-outcome relations are astarting point.

Timing of Change. Assume for a momentthat 10 patients receive identical treatmentover the course of 20 sessions and that treat-ment works for each of them for the identicalreason, i.e., the same mechanism is responsi-ble for change. It is not likely that the processof change will follow the identical time courseso that by session 8, for example, the mecha-nism has changed in a critical way and symp-tom change is underway. Indeed, apparentlythe timeline of therapeutic change can be al-tered by what patients are told about the dura-tion of treatment prior to beginning treatment(Barkham et al. 1996).

Some patients may make rapid or suddengains at a particular point in treatment (e.g.,Tang & DeRubeis 1999). One could say thatat a given point, some have and some havenot made change in some qualitative or cate-gorical fashion. Alternatively, one could con-sider that the point of therapeutic change forall individuals is normally distributed with amean and standard deviation. In either sce-nario (sudden gains but not at the identicalpoint or normally distributed changes acrossseveral points), assessment of the mechanismis a challenge. Assessment of the mecha-nism at any one or two points in a studymay not capture when change in the mech-anism has occurred. A challenge for researchis ensuring that one can evaluate mechanismand change that may vary in course amongindividuals.

Everything in Moderation

The effects of an intervention may be moder-ated in ways that exert enormous impact. For

example, a “standard” dose of psychotropicmedication (e.g., for clinical depression) canbe an overdose or underdose for people ofdifferent ethnicities and countries (Lin et al.1993). Medication effects are moderated byethnicity. Among the intriguing issues, wouldmedication effects operate similarly if doseswere adjusted to each group, or is this notmerely a matter of dose? Either way, evalu-ating or analyzing data for an overall (main)effect of medication and ignoring ethnicitywould lead to a weak effect and not encouragepursuit of mechanisms.

It is possible that the mediator or mech-anism of change in psychotherapy varies as afunction of a moderator variable. Searchingfor moderators (a priori or post hoc), test-ing them (statistical power from dividing ofthe sample into subgroups), and interpret-ing them (e.g., is the moderator a proxy forsome other variable?) have their own specialchallenges. Rather than looking for main ef-fects of an intervention and a uniform mech-anism of change, we may need to identifyand characterize subgroups, very much inthe way that genetic researchers often profitfrom looking at special groups and individualoutliers.

Measurement Development

Mediators and mechanisms in therapy are of-ten discussed but validated measures of keyconstructs are not readily available. Many ad-vances in biotechnology as represented by thecontinued advances in neuroimaging have hadenormous impact on the search for neurolog-ical mechanisms, although these assessmentshave nontrivial interpretive challenges. If bymediators or mechanisms, we will be search-ing for psychological explanations of action,we will need valid measures. In the parenttraining literature, mentioned above, behav-ioral observations were used to show thatparent-child interactions unfolded in a se-quence leading to (conditional probabilities)child aggressive behavior. Among the manyvirtues of this work was the assessment of

22 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


observable actions and charting a sequencethat promotes aggression.

Presumably, many mediators of changebegin as broad constructs (e.g., changes incognition). We need valid measures of suchconstructs and then demonstration of howthe constructs operate. There are promisingleads. For example, break down of copingskills in high-risk situations is associated withcocaine and other substance abuse. Treatmentoften targets coping skills as the critical me-diator of change in reducing substance use.A role-play measure (Cocaine Risk ResponseTest) has been developed, evaluated, and in-tegrated into controlled treatment researchto evaluate the mediator (Carroll et al. 1999,2005). Assessment of mediators and mecha-nisms raises all of the usual questions in mea-surement development. The topic has notbeen accorded sufficient attention.

CONCLUSIONS

Enormous progress has been made in psy-chotherapy research. This has culminated inrecognition of several treatments that havestrong evidence in their behalf. Despite thisprogress, research advances are sorely neededin studying the mediators and mechanisms oftherapeutic change. It is remarkable that afterdecades of psychotherapy research, we can-not provide an evidence-based explanation forhow or why even our most well studied inter-ventions produce change.

Many researchers might regard the ratherlarge body of research on the therapeutic rela-tionship as a potential exception. Yet, the vastmajority of studies rarely rule out the possibil-ity that the relationship is the result of symp-tom change or some other variable rather thana mechanism responsible for it. I am not chal-lenging the importance of relationships—in

everyday life, I have tried one or two myself.This is a quarrel about the necessary assess-ment and design requirements that are in-frequently included in research. In addition,assuming the timeline were unequivocallyestablished, we need “next-step” researchthat clarifies how a relationship in ther-apy leads to symptom change, i.e., throughwhat specific steps. These steps need to beevaluated.

Prior research has provided importantgroundwork on which future studies couldbuild. For instance, an increasing numberof studies are including assessments duringthe course of treatment (e.g., Beauchaineet al. 2005, Eddy & Chamberlain 2000, Kolkoet al. 2000, Kwon & Oei 2003). The de-signs used in these investigations representa great improvement over prior studies andsignal progress in research on mechanisms ofchange. Yet, existing studies have attempted toevaluate only a handful of potential mediatorsand mechanisms of change.

The scientific study of mechanisms ofchange is certainly not an easy path onwhich to embark. A given treatment mightwork for multiple reasons. Just as there isno simple and single path to many diseases,disorders, or social, emotional, and behav-ioral problems (e.g., lung cancer, attention-deficit/hyperactivity disorder), there may beanalogous complexity in mechanisms for agiven treatment technique or therapeutic out-come. Two patients in the same treatmentconceivably could respond for different rea-sons. The complexities are critically impor-tant to understand because of a point madeabove, namely, the best patient care will comefrom ensuring that the optimal variation oftreatment is provided. Understanding mech-anisms of treatment is the path toward im-proved treatment.

ACKNOWLEDGMENT

Completion of this chapter was facilitated by support from the National Institute of MentalHealth (MH59029).


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


LITERATURE CITED

Adolphs R. 2003. Cognitive neuroscience of human social behavior. Nat. Rev. Neurosci. 4:165–178

American Psychiatric Association. 2000. Practice guideline for the treatment of patients withmajor depressive disorder (revision). Am. J. Psychiatry 157(Suppl. 4):1–45

Barber JP, Connolly MB, Crits-Christoph P, Gladis L, Siqueland L. 2000. Alliance predictspatients’ outcome beyond in-treatment change in symptoms. J. Consult. Clin. Psychol.68:1027–32

Barkham M, Rees A, Stiles WB, Shapiro DA, Hardy GE, Reynolds S. 1996. Dose-effectrelations in time-limited psychotherapy for depression. J. Consult. Clin. Psychol. 64:927–35

Baron RM, Kenny DA. 1986. The moderator-mediator variable distinction in social psycholog-ical research: conceptual, strategic, and statistical considerations. J. Personal. Soc. Psychol.51:1173–82

Beauchaine TP, Webster-Stratton C, Reid MJ. 2005. Mediators, moderators, and predictorsof 1-year outcomes among children treated for early-onset conduct problems: a latentgrowth curve analysis. J. Consult. Clin. Psychol. 73:371–88

Berg BL. 2001. Qualitative Research Methods for the Social Sciences. Needham Heights, MA: Allyn& Bacon. 4th ed.

Brent DA, Kolko DJ. 1998. Psychotherapy: definitions, mechanism of action, and relationshipto etiological models. J. Abnorm. Child Psychol. 26:17–25

Burns DD, Spangler DL. 2001. Do changes in dysfunctional attitudes mediate changes indepression and anxiety in cognitive behavioral therapy? Behav. Ther. 32:337–69

Carroll KM, Nich C, Ball SA. 2005. Practice makes progress? Homework assignments andoutcome in treatment of cocaine dependence. J. Consult. Clin. Psychol. 73:749–55

Carroll KM, Nich C, Frankforter TL, Bisighini RM. 1999. Do patients change in ways weintend? Assessing acquisition of coping skills among cocaine-dependent patients. Psychol.Assess. 11:77–85

Caspi A, McClay J, Moffitt TE, Mill J, Martin J, et al. 2002. Role of genotype in the cycle ofviolence in maltreated children. Science 297:851–54

Champagne FA, Francis DD, Mar A, Meaney M. 2003. Variations in maternal care in therat as a mediating influence for the effects of development on behavior. Physiol. Behav.79:359–71

Davis M, Myers KM, Chhatwal J, Ressler KJ. 2006. Pharmacological treatments that facilitateextinction of fear: relevance to psychotherapy. NeuroRx 3:82–96

Davis M, Myers KM, Ressler KJ, Rothbaum BO. 2005. Facilitation of extinction of conditionedfear by D-cycloserine. Curr. Direct. Psychol. Sci. 14:214–19

Denissenko MF, Pao A, Tang M, Pfeifer GP. 1996. Preferential formation of benzo[a]pyreneadducts at lung cancer mutational hotspots in P53. Science 274:430–32

Denzin NK, Lincoln YS, eds. 2005. The SAGE Handbook of Qualitative Research. ThousandOaks, CA: Sage. 3rd ed.

DeRubeis RJ, Evans MD, Hollon SD, Garvey MJ, Grove WM, Tuason VB. 1990. How doescognitive therapy work? Cognitive change and symptom change in cognitive therapy andpharmacotherapy for depression. J. Consult. Clin. Psychol. 58:862–69

DeRubeis RJ, Feeley M. 1990. Determinants of change in cognitive therapy for depression.Cogn. Ther. Res. 14:469–82

Eddy JM, Chamberlain P. 2000. Family management and deviant peer association as mediatorsof the impact of treatment condition on youth antisocial behavior. J. Consult. Clin. Psychol.68:857–63

24 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


Foley D, Wormley B, Silberg J, Maes H, Hewitt J, et al. 2004. Childhood adversity, MAOAgenotype, and risk for conduct disorder. Arch. Gen. Psychiatry 61:738–44

Grawe K. 2004. Psychological Therapy. Cambridge, MA: Hogrefe & HuberHe J, Vupputuri S, Allen K, Prerost MR, Hughes J, Whelton PK. 1999. Passive smoking and

the risk of coronary heart disease—a meta-analysis of epidemiologic studies. New Engl. J.Med. 340:920–26

Hofmann SG. 2000. Treatment of social phobia: potential mediators and moderators. Clin.Psychol. Sci. Pract. 7:3–16

Hofmann SG, Meuret AE, Smits JA, Simon NM, Pollack MH, et al. 2006. Augmentationof exposure therapy with D-cycloserine for social anxiety disorder. Arch. Gen. Psychiatry63:298–304

Hollon SD, Beck AT. 2004. Cognitive and cognitive behavioral therapies. In Bergin andGarfield’s Handbook of Psychotherapy and Behavior Change, ed. MJ Lambert, pp. 447–92.New York: Wiley. 5th ed.

Holmbeck GN. 2002. Post-hoc probing of significant moderational and mediational effects instudies of pediatric populations. J. Pediatr. Psychol. 27:87–96

Horvath AO, Bedi RP. 2002. The alliance. In Psychotherapy Relationships That Work: TherapistContributions and Responsiveness to Patients, ed. JC Norcross, pp. 37–69. New York: OxfordUniv. Press

Hoyle RH, Smith GT. 1994. Formulating clinical research hypotheses as structural equationmodels: a conceptual overview. J. Consult. Clin. Psychol. 62:429–40

Hume AA, Fox NA. 2006. Ordinary variations in maternal caregiving influence human infants’stress reactivity. Psychol. Sci. 17:550–56

Ilardi SS, Craighead WE. 1994. The role of nonspecific factors in cognitive-behavior therapyfor depression. Clin. Psychol. Sci. Pract. 1:138–56

Jaffee SR, Caspi A, Moffitt TE, Dodge K, Rutter M, et al. 2005. Nature x nurture: geneticvulnerabilities interact with physical maltreatment to promote behavior problems. Dev.Psychopathol. 17:67–84

Kazdin AE. 2000. Psychotherapy for Children and Adolescents: Directions for Research and Practice.New York: Oxford Univ. Press

Kazdin AE. 2003. Research Design in Clinical Psychology. Needham Heights, MA: Allyn & Bacon.4th ed.

Kazdin AE. 2005. Parent Management Training: Treatment for Oppositional, Aggressive, and An-tisocial Behavior in Children and Adolescents. New York: Oxford Univ. Press

Kazdin AE. 2006. Mechanisms of change in psychotherapy: advances, breakthroughs, andcutting-edge research (do not yet exist). In Strengthening Research Methodology: PsychologicalMeasurement and Evaluation. ed. RR Bootzin, PM McKnight, pp. 77–101. Washington,DC: Am. Psychol. Assoc.

Kazdin AE, Nock MK. 2003. Delineating mechanisms of change in child and adolescenttherapy: methodological issues and research recommendations. J. Child Psychol. Psychi-atry 44:1116–29

Kazdin AE, Weisz JR, eds. 2003. Evidence-Based Psychotherapies for Children and Adolescents. NewYork: Guilford

Kazdin AE, Whitley MK. 2006. Pretreatment social relations, therapeutic alliance, and im-provements in parenting practices in parent management training. J. Consult. Clin. Psychol.74:346–55

Kenny DA, Kashy DA, Bolger N. 1998. Data analysis in social psychology. In Handbook ofSocial Psychology, ed. D Gilbert, ST Fiske, G Lindzey, Vol. 1, pp. 233–65. New York:McGraw-Hill. 4th ed.


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


Kolko DJ, Brent DA, Baugher M, Bridge J, Birmaher B. 2000. Cognitive and family therapiesfor adolescent depression: treatment specificity, mediation, and moderation. J. Consult.Clin. Psychol. 68:603–14

Kraemer HC, Kazdin AE, Offord DR, Kessler RC, Jensen PS, Kupfer DJ. 1997. Coming toterms with the terms of risk. Arch. Gen. Psychiatry 54:337–43

Kraemer HC, Stice E, Kazdin AE, Offord DR, Kupfer DJ. 2001. How do risk factors worktogether? Mediators, moderators, independent, overlapping, and proxy-risk factors. Am.J. Psychiatry 158:848–56

Kraemer HC, Wilson GT, Fairburn CG, Agras WS. 2002. Mediators and moderators oftreatment effects in randomized clinical trials. Arch. Gen. Psychiatry 59:877–83

Kwon S, Oei TPS. 2003. Cognitive processes in a group cognitive behavior therapy of depres-sion. J. Behav. Ther. Exp. Psychiatry 34:73–85

Lambert MJ, ed. 2004. Bergin and Garfield’s Handbook of Psychotherapy and Behavior Change.New York: Wiley. 5th ed.

Lin K, Poland RE, Nagasaki G, eds. 1993. Psychopharmacology and Psychobiology of Ethnicity.Washington, DC: Am. Psychiatr. Press

MacKinnon DP, Lockwood CM, Hoffman JM, West SG, Sheets V. 2002. A comparison ofmethods to test mediation and other intervening variable effects. Psychol. Methods 7:83–104

MacKinnon DP, Taborga MP, Morgan-Lopez AA. 2002. Mediation designs in tobacco pre-vention research. Drug Alcohol Depend. 68:S69–83

McGuire WJ. 1997. Creative hypothesis generating in psychology: some useful heuristics.Annu. Rev. Psychol. 48:1–30

Meyer-Lindenberg A, Buckholtz JW, Kolachana B, Ahmad R, Hariri AR, et al. 2006. Neuralmechanisms of genetic risk for impulsivity and violence in humans. Proc. Natl. Acad. Sci.USA 103:6269–74

Meyer-Lindenberg A, Hariri AR, Munoz KE, Mervis CB, Mattay VS, et al. 2005. Neuralcorrelates of genetically abnormal social cognition in Williams syndrome. Nat. Neurosci.8:991–93

Myers KM, Davis M. 2002. Behavioral and neural analysis of extinction: a review. Neuron36:567–84

Nathan PE, Gorman JM, eds. 2007. Treatments That Work. New York: Oxford Univ. Press. 3rded. In press

Orlinsky DE, Rønnestad MH, Willutzki U. 2004. Fifty years of psychotherapy process-outcome research: continuity and change. In Bergin and Garfield’s Handbook of Psychotherapyand Behavior Change, ed. MJ Lambert, pp. 307–89. New York: Wiley. 5th ed.

Patterson GR. 1982. Coercive Family Process. Eugene, OR: CastaliaPatterson GR, Reid JB, Dishion TJ. 1992. Antisocial Boys. Eugene, OR: CastaliaPerepletchikova F, Kazdin AE. 2005. Treatment integrity and therapeutic change: issues and

research recommendations. Clin. Psychol. Sci. Pract. 12:365–83Pruessner JC, Champagne F, Meaney MJ, Dagher A. 2004. Dopamine release in response to

a psychological stress in humans and its relationship to early maternal care: a positronemission tomography study using [C]raclopride. J. Neurosci. 24:2825–31

Reid JB, Patterson GR, Snyder J, eds. 2002. Antisocial Behavior in Children and Adolescents: ADevelopmental Analysis and Model for Intervention. Washington, DC: Am. Psychol. Assoc.

Ressler KJ, Rothbaum BO, Tannenbaum L, Anderson P, Graap K, et al. 2004. Cognitiveenhancers as adjuncts to psychotherapy. Use of D-cycloserine in phobic individuals tofacilitate extinction of fear. Arch. Gen. Psychiatry 61:1136–44

Shrout PE, Bolger N. 2002. Mediation in experimental and nonexperimental studies: newprocedures and recommendations. Psychol. Methods 7:422–45

26 Kazdin

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.


Snyder J, Stoolmiller M. 2002. Reinforcement and coercion mechanisms in the developmentof antisocial behavior: the family. In Antisocial Behavior in Children and Adolescents: A De-velopmental Analysis and Model for Intervention, ed. JB Reid, GR Patterson, J Snyder, pp.65–100. Washington, DC: Am. Psychol. Assoc.

Tang TZ, DeRubeis RJ. 1999. Sudden gains and critical sessions in cognitive-behavioral therapyfor depression. J. Consult. Clin. Psychol. 67:894–904

Tremblay RE, Masse B, Perron D, LeBlanc M, Schwartzman AE, Ledingham JE. 1992. Earlydisruptive behavior, poor school achievement, delinquent behavior, and delinquent per-sonality: longitudinal analyses. J. Consult. Clin. Psychol. 60:64–72

Twain M, Warner ACD. 1874. The Gilded Age: A Tale of Today. Hartford, CT: Am. Publ.Weersing VR, Weisz JR. 2002. Mechanisms of action in youth psychotherapy. J. Child Psychol.

Psychiatry 43:3–29Whisman MA. 1999. The importance of the cognitive theory of change in cognitive therapy

of depression. Clin. Psychol. Sci. Pract. 6:300–4Zigler E, Glick MA. 1986. A Developmental Approach to Adult Psychopathology. New York: Wiley


Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.

AR307-FM ARI 2 March 2007 14:4

Annual Review ofClinical Psychology

Volume 3, 2007Contents

Mediators and Mechanisms of Change in Psychotherapy ResearchAlan E. Kazdin � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 1

Evidence-Based AssessmentJohn Hunsley and Eric J. Mash � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 29

Internet Methods for Delivering Behavioral and Health-RelatedInterventions (eHealth)Victor Strecher � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 53

Drug Abuse in African American and Hispanic Adolescents: Culture,Development, and BehaviorJose Szapocznik, Guillermo Prado, Ann Kathleen Burlew, Robert A. Williams,and Daniel A. Santisteban � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 77

Depression in MothersSherryl H. Goodman � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �107

Prevalence, Comorbidity, and Service Utilization for Mood Disordersin the United States at the Beginning of the Twenty-first CenturyRonald C. Kessler, Kathleen R. Merikangas, and Philip S. Wang � � � � � � � � � � � � � � � � � � � � �137

Stimulating the Development of Drug Treatments to ImproveCognition in SchizophreniaMichael F. Green � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �159

Dialectical Behavior Therapy for Borderline Personality DisorderThomas R. Lynch, William T. Trost, Nicholas Salsman,and Marsha M. Linehan � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �181

A Meta-Analytic Review of Eating Disorder Prevention Programs:Encouraging FindingsEric Stice, Heather Shaw, and C. Nathan Marti � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �207

Sexual Dysfunctions in WomenCindy M. Meston and Andrea Bradford � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �233

Relapse and Relapse PreventionThomas H. Brandon, Jennifer Irvin Vidrine, and Erika B. Litvin � � � � � � � � � � � � � � � � � � �257

vii

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.

AR307-FM ARI 2 March 2007 14:4

Marital and Family Processes in the Context of Alcohol Use andAlcohol DisordersKenneth E. Leonard and Rina D. Eiden � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �285

Unwarranted Assumptions about Children’s Testimonial AccuracyStephen J. Ceci, Sarah Kulkofsky, J. Zoe Klemfuss, Charlotte D. Sweeney,and Maggie Bruck � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �311

Expressed Emotion and Relapse of PsychopathologyJill M. Hooley � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �329

Sexual Orientation and Mental HealthGregory M. Herek and Linda D. Garnets � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �353

Coping Resources, Coping Processes, and Mental HealthShelley E. Taylor and Annette L. Stanton � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �377

Indexes

Cumulative Index of Contributing Authors, Volumes 1–3 � � � � � � � � � � � � � � � � � � � � � � � � � � �403

Cumulative Index of Chapter Titles, Volumes 1–3 � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �405

Errata

An online log of corrections to Annual Review of Clinical Psychology chapters (if any)may be found at http://clinpsy.AnnualReviews.org

viii Contents

Ann

u. R

ev. C

lin. P

sych

ol. 2

007.

3:1-

27. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by U

nive

rsity

of

Ari

zona

Lib

rary

on

08/2

2/09

. For

per

sona

l use

onl

y.

Mediators and Mechanisms of Change in Psychotherapy Research · Mediators and Mechanisms of Change...

Documents

Transcript of Mediators and Mechanisms of Change in Psychotherapy Research · Mediators and Mechanisms of Change...