Content uploaded by Sebastian Pintea
Author content
All content in this area was uploaded by Sebastian Pintea on Jul 15, 2017
Content may be subject to copyright.
Cognition, Brain, Behavior. An Interdisciplinary Journal
Copyright © 2015 ASCR Publishing House. All rights reserved.
ISSN: 1224-8398
Volume XIX, No. 4 (December), 299-311
MECHANISMS OF CHANGE IN PSYCHOTHERAPY:
METHODOLOGICAL AND STATISTICAL
CONSIDERATIONS
Ramona MOLDOVAN*, Sebastian PINTEA
Department of Psychology, Babes-Bolyai University, Cluj-Napoca, Romania
ABSTRACT
The ambition to understand why and how psychotherapy works has guided
theorists, researchers, and practitioners for decades. This has led to an
accumulation of literature, both theoretical and empirical, exploring the
mechanisms that facilitate change in the psychotherapeutic process. Over the past
decades there has been considerable advancement in the areas of investigating the
psychotherapy process (i.e., course of change, predictors of outcome, mediators
and moderators of change). The primary aim of our paper is to draw attention to
the importance of studying mechanisms of change and to delineate the most
important theoretical and methodological milestones for evaluating the processes
through which clinical change occurs. We first discuss prior work addressing
mechanisms of change and argue how mechanisms are linked to other related
concepts. We then outline several strategies in data analysis and briefly discuss
their role in investigating mechanisms of change. Finally, we suggest key
recommendations to be considered by researchers designing studies investigating
mechanisms of change in psychological treatments, as well as recommendations
for future research in this area.
KEYWORDS: psychotherapy, mechanism of change, mediation
From if to why psychotherapy leads to change
Patients come into therapy (as individuals, couples, or families) with various
behavioral, emotional, physiological and/or cognitive difficulties, and they seek
relief from these problems by the time therapy is completed. In most cases, their
needs are granted: psychotherapy works (Elkin, 1999; Lambert & Barley, 2002;
Lambert & Bergin, 1994; Lipsey & Wilson, 1993; Roth & Fonagy, 1996; Wampold
& Brown, 2005).
*
Corresponding author:
E-mail: ramonamoldovan@psychology.ro
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
300
The evidence-based movement within the psychological community has
been constantly striving both to improve the efficacy of psychological interventions
and to provide treatment guidelines for clients, professional providers, and third
parties alike (David & Montgomery, 2010). Up to this point, a vast amount of
research has clearly established the efficacy and effectiveness of a range of
psychological treatments (Murphy, Cooper, Hollon, & Fairburn, 2009). Meta-
analyses and qualitative reviews of controlled studies have indicated that many
forms of psychotherapy for children, adolescents, and adults lead to therapeutic
change (e.g., Kazdin & Weisz, 2003; Lambert & Ogles, 2004.). Research has
repeatedly demonstrated that individuals with various clinical problems will, on
average, benefit more from psychotherapy than from no treatment or a
psychological control treatment (Lambert & Ogles, 2004), mostly in terms of
emotional, behavioral, social, cognitive, educational, and physical functioning.
We now know well that psychotherapy works (i.e., it is responsible for
change) but still have rather little knowledge of for whom and under what
conditions psychotherapeutic treatments work, how they work, and why they work
(Kazdin, 2007) as most studies continue to focus on gathering empirical data to
support various (psycho)therapeutic packages while ignoring whether there is any
evidence to support the proposed theoretical underpinnings of these techniques
(David, 2004). The means through which these therapies exert their beneficial
effects are generally not well understood (Kazdin, 2009; Webb, 2010) as
investigations to date have yielded very few interpretable results (Doss, 2004). As a
matter of fact, it is quite remarkable that after decades of psychotherapy research,
with isolated exceptions, we cannot provide a clear-cut evidence-based explanation
for how and why even our most well studied interventions produce change (Kazdin,
2007). Certainly, all psychological interventions are based on theories that explain
why and how improvements supposedly occur (some of them have clearly
articulated mechanisms while others tend to be more focused on broad principles),
but these theoretical assumptions are rarely put to the test empirically (Johansson &
Høglend, 2007).
There are several reasons why a better understanding of mechanisms of
change is essential to the field of psychotherapy. First, because even the best
empirically supported treatment packages do not help all patients; it is essential to
validate the theoretically relevant mechanisms of change of efficacious treatments
as, undoubtedly, a better understanding of the mechanisms of change would provide
the best opportunity to further improve currently available treatments (Kraemer et
al., 2002; Connolly Gibbons et al., 2009). Second, identifying and understanding
the mechanisms of change in therapy can improve the understanding of clinical
disorders and the variables associated with their course. And third, collecting
information about mechanisms of change can help to distill the important
mechanisms of change that cut across different types of therapy and contribute to a
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
301
better understanding of psychological interventions in general (Kazdin & Nock,
2007). In this paper, we first develop on prior work defining mechanisms of
change and clarify how these mechanisms are linked to other related concepts such
as mediators (Kazdin & Nock, 2003; Nock, 2007). Next, we outline several
methodological criteria for investigating and testing for mechanisms of change.
Finally, we suggest key recommendations to be considered by researchers designing
studies aimed at investigating psychological treatments, as well as
recommendations for future research in this area.
Mechanisms of change versus other related concepts: theoretical delimitations
Given the inconsistencies that have been used when discussing mechanisms of
change, it is important to clarify key concepts as well as describe how they relate to
each other and how they fit into the broader scientific context (Nock, 2007). Several
interrelated and overlapping concepts are important to distinguish.
A mechanism of change refers to the process or series of events through
which one variable leads to and/or causes change in another variable. Mechanisms
of change reflect the processes through which some independent variable (i.e.,
therapy) actually produces the change and explain how the intervention eventually
leads to the outcome (Kazdin, 2007). This is easily confused with the notion of
mediation. For example, cognitions may be shown to mediate change in therapy.
However, this does not always explain specifically how the change came about (i.e.,
what are the intervening steps between cognitive change and reduced depression or
anxiety). The goal is to understand the mechanisms of change; the study of
mediators is often a first step.
A mediator is a construct that shows specific statistical relations between
an intervention and the outcome, but may not explain the precise process through
which change comes about. Mediators of treatment effects are variables which
account for, in a statistical sense, at least some of the effects of treatment on the
outcome (Baron & Kenny, 1986). Mediational analysis allows the clarification of
how treatments have effects and, particularly, what are the possible mechanisms
through which a treatment might achieve its effects (Kraemer et al., 2002). The
mediator is potentially a mechanism through which the change occurs (Johansson &
Hoglend, 2007). This suggests that treatment causes the mediator variable to
change, which then leads to the outcome. In psychotherapy, mediators are typically
processes within the patient (e.g., cognitions, abilities etc.)
A moderator refers to a characteristic that influences the direction or
magnitude of the relation between the intervention and the outcome. Generally
speaking, moderators clarify for whom or under what conditions an intervention
works (Baron & Kenny, 1986). If treatment outcome varies as a function of
different characteristics of the patient (e.g., age, symptoms, expectations), therapist
(e.g., sex, experience, self-efficacy) or treatment delivery (e.g., individual versus
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
302
group treatment), these variables are moderators (Kazdin, 2007). To show that a
variable is a moderator of treatment the variable must be a baseline or pre
randomization characteristic (in other words, it precedes treatment); second, the
variable must be uncorrelated with treatment; third, the variable has to be shown to
have an interaction effect with treatment on the outcome, that is ”explain”, in a
statistical sense, individual differences in the treatment effects (Kraemer et al.,
2002). Clearly, the mediator is proximal to the mechanism of change and also
necessary (though not sufficient) for demonstrating mechanisms of change; in the
following section we concentrate on several methodological and statistical aspects
related to mediation testing in randomized clinical trials.
Investigating mechanisms of change
Randomized clinical trials (RCTs) are widely regarded as the golden standard when
evaluating the efficacy and effectiveness in clinical research. There has been a
growing consensus that the RCT is one of the best methods for obtaining
convincing evidence for the efficacy of a psychological treatment (Haaga & Stiles,
2000). Acceptance of the RCT is so widespread that there are now precise
characteristics of a well-performed RCT. They include the following features (for
more details, see Kraemer et al., 2002): (1) A well-defined and justified population,
with a representative sample of sufficient size, to yield power to detect clinically
significant differences between treatments and to provide accurate estimates of the
effect sizes in that population (Borenstein, 1994; Jacobson & Truax, 1991; Kramer,
1993); (2) One or more control or comparison groups with clear treatment protocols
that could be replicated; (3) Randomization to treatment and control or comparison
groups in order to avoid confusing selection effects with treatment effects;
(4)Several justified outcome measures, selected in advance of the trial, obtained
either blinded to treatment group or having controlled measurement bias; (5)
Analysis performed by intention to treat (i.e., all randomized subjects are included
in the analysis of outcome); (6) A valid test for statistical significance and estimates
of effect sizes informative enough to guide consideration of clinical and policy
significance.
There is high consensus that there is much more that can be learned from a
successfully completed RCT than it is currently learned. RCT could also provide
information on possible mediators and moderators of treatment outcomes to guide
the next generation of studies and inform clinical applications (Kraemer et al.,
2002). RCTs provide an often-missed opportunity to investigate the mediators of
treatment effects; several guidelines have been therefore proposed (Kazdin, 2007;
Kraemer et al., 2002). First, the decision to perform a mediation analysis needs to
be made a priori as it will influence the choice and timing of measures used. Next,
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
303
hypotheses need to be derived from the theory underpinning the treatment and
formulated in terms of the putative mechanisms of change of the treatment
investigated. Then, the treatment, the putative mediators and the outcome need to be
clearly operationalized and appropriately assessed (Murphy, Cooper, Hollon, &
Fairburn, 2009).
There are two important features of the typical RCT design that can limit
researchers' ability to address questions of process and mechanisms of change
(Laurenceau et al., 2007). First,all measures are usually performed at pretreatment
and again at post treatment; this often used pre–post design is not effective for
adequately examining hypothesized mechanisms of change (Collins & Graham,
2002). Second, even when an outcome is measured at multiple points between the
beginning and end of treatment, rather few studies include measures of putative
mediators at multiple points between pre- and post-treatment; it is preferable to
obtain a minimum of three or more repeated measurements in order to adequately
evaluate a mediation model. Nevertheless, even with a pre–mid-post design, the
mediation effect can vary dramatically depending on the measurement interval used
to assess the putative mediator (Collins & Graham, 2002; Laurenceau et al., 2007)
(say, for instance, the middle assessment is far from the period when the treatment
has its strongest effect on the mediator).
Demonstrating and testing mediators
Over the past two decades, researchers have developed several methods for testing
whether a proposed mechanism can act as a mediator, which means to statistically
explain the relationship between an independent and a dependent variable.
Theoretically, to show that a variable is a mediator of a treatment, that
variable would have to measure an event or change occurring during treatment, and
it would have to correlate with treatment choice, hence to possibly be a result of
treatment, and have either a main or an interactive effect on the outcome (Kraemer
et al., 2002). According to the same authors, the directionality of mediation is
unambiguous since theoretical models are used in order to define putative mediators
while statistics are used to evaluate a presumed mediational model.
Essentially, in order to show such a relation, one must demonstrate that an
independent variable (A) is associated with a dependent variable (B); that A is
associated with the proposed mechanism (M); that M is associated with B; and
when A and M are both covaried with B, M continues to be associated with B while
the relationship between A and B is diminished. This pattern of relationships
provides evidence that A is associated with B through its relation with M (Nock,
2007; Baron & Kenny, 1986; Holmbeck, 1997; MacKinnon et al., 2002).
Statistical evaluation can play a central role in addressing whether a
particular construct accounts for change. A variety of procedural/statistical
solutions have been developed in order to assess whether a putative mediator meets
statistical criteria for mediation, each one with its own advantages and limits. We
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
304
next review the most frequently used methods, delineating their underpinning
theoretical principles and procedures. RCTs have the advantage of providing
longitudinal data therefore we are not explicitly addressing cross-sectional
mediation procedures, but suggest solutions for mediation with longitudinal data. A
number of comprehensive reviews detailing the limitations of cross-sectional
mediation procedures when applied to longitudinal data are available (MacKinnon,
2008; Cole & Maxwell, 2003,Gollob & Reichardt, 1985).
The difference scores solution. RCTs with two waves data (pretest-posttest
measures) offers the simplest case of longitudinal data. In such cases, one solution
to testmediation is using difference scores (or delta change scores). In other words,
the difference between the first and second measure of the mediator and the
outcome is calculated for each individual, and these difference scores are used in
the mediation equations (Baron & Kenny, 1986), along with the independent
variable, coded as a dummy variable.
One of the major advantages of this solution is that it includes information
about the dynamics of the mediator and the outcome. Even if in RCTs the timeline
between treatment and mediator or between treatment and outcome is clearly
established, the timeline between mediator and outcome is not always empirically
demonstrated. If the mediator and the outcome are measured simultaneously, the
statistical relationship between changes found in those variables does not show
which one has changed first. In order to overcome this limit, a growing body of
literature (Gelfand et al., 2009; DeRubeis & Feeley, 1990; Tenhave et al., 2007)
suggests that the temporal order between the mediator and the outcome should be
empirically investigated based on changes observed in these variables using non-
overlapping periods of time to reduce temporal ambiguity.
The residualized change solution. An alternative to the difference scores method is
the residualized change score. Briefly, in RCTs the residualized change score is the
difference between the after treatment score and the predicted after treatment score,
when the baseline measure is used to predict the after treatment score. The
residualized change scores obtained separately both for the mediator and the
outcome will be introduced in the mediation equations along with the treatment,
coded as a dummy variable, a very similar procedure to the difference score
approach. The advantage of this solution is that it adjusts for baseline differences
and avoids some of the problems with the reliability of the difference scores, but it
can still be susceptible to low reliability and its assumption about the regression to
the mean over time is sometimes inadequate (MacKinnon, 2008).
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
305
The ANCOVA solution. Another procedure that uses the baseline adjustment is the
analysis of covariance (ANCOVA). In a RCT, the mediator and outcome baseline
scores are included as covariates in the analysis. We indicate bellow the equations
for such a procedure, treated as regression (the treatment variable X is dummy
coded, the mediator is M and the outcome is Y). We have excluded the intercept
and the residuals from each of the following equations, in order to simplify the
presentation and make it more comprehensible.
Y2= c'X + b1M1 + b2M2 + s1Y1
M2= aX + s2M1
In these equations there are two estimators of the mediated effect: ab1 for
the longitudinal relations and ab2showing the relation across time for a and within
time for b2. Among the two estimators, ab1has more support as it reflects change
across time (Cole & Maxwell, 2003; McKinnon, 2008).
Two other models are frequently suggested when dealing with longitudinal
data: the autoregressive model and the Latent Growth Curve (LGC) model. Both
models show good potential for use in data obtained from RCTs. We will present
them briefly, concentrating more on their theoretical principles rather than their
technical aspects.
The autoregressive model solution. According to this model each variable is
predicted by the same variable at an earlier stage (i.e., regressed on itself). There are
three different ways of using the autoregressive model: (1) without including
contemporaneous relations among variables, (2) including contemporaneous
relations among variables, and (3) allowing for cross-lagged relations among
variables (the direction of the relations among X, M, and Y are all free to vary).
The first autoregression model is to some extent similar to the ANCOVA
model, except for the inclusion of contemporaneous relations among variables. As a
consequence, such an approach only allows for the longitudinal mediation effects to
be computed. We indicate bellow the equations for this model with data for a
potential RCT (the treatment variable X is dummy coded, the mediator is M and the
outcome is Y). With two waves data we only have one lag mediation effect, and its
estimator is ab1.
Y2= c'X + b1M1 + s1Y1
M2= aX + s2M1
We further indicate the procedure for three waves data. The terms in
brackets are not included in the first autoregressive model; they represent the
contemporaneous relations among variables. Variable X, the experimental
treatment, remains the same in all equations.
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
306
M2 = a1X + s2M1
M3 = a2X + s2M2
Y2 = b1M1 + c′1X + s3Y1 (+ b3M2)
Y3 = b2M2 + c′2X + s3Y2 + (b4M3)
With three waves data, there are several mediated effects, estimated by
a1b1 for the first lag, a2b2 for the second lag, and a1b2 reflecting the temporal
ordering of the mediated effect. It is also possible to consider two lag effects
(effects two waves apart).
The second autoregressive model, is depicted in the same equations from
above, but including the terms in brackets. By including the contemporaneous
relations among variables, this second model can compute autoregression and
longitudinal mediation effects (autoregressive mediated effects are estimated by
a1b1and a2b2, and the longitudinal mediated effect is estimated by a1b2) as well as
contemporaneous mediation relations (estimated by a1b3 at time 2 and a2b4 at time
3). The third autoregressive model violates the temporal precedence of X to M
to Y specified by the mediation model because paths in the reverse direction are
estimated as M to X and Y to M. In the context of a RCT, where the timeline
between the experimental treatment on the one hand and the mediator and the
outcome on the other is clear, the only cross-lagged relation that makes sense to be
tested is from the outcome to the mediator. As a consequence, the mediation
equations in such a case are as follows:
M2 = a1X + s2M1 + d3Y1
M3 = a2X + s2M2 + d3Y2
Y2 = c’1X + b1M1 + b3M2 + s3Y1
Y3 = c’2X + b2M2 + b4M3 + s3Y2
The mediation effects estimators in these equations, both contemporaneous
and longitudinal, are exactly the same as in the second autoregressive model.
Criticisms suggest that these models are not explicitly modeling change in
measures over time or individual differences in growth. Autoregressive models
focus more on the stability of the rank order of subjects on variables across time
rather than on trajectories of change across time (MacKinnon, 2008).
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
307
The Latent Growth Curve (LGC) Model solution. This model was developed in
order to overcome the limitations of the autoregressive models. In this case, the
mediation model examines whether the growth in the independent variable affects
the growth trajectory of the mediating variable which affects the growth trajectory
of the dependent variable (MacKinnon, 2008).
The solution of difference scores that we mentioned earlier for RCTs with
two data waves isin fact a particular case of the growth curve approach. For RCTs
with three or more data wavesthe mediation model examines whether the
experimental condition (i.e., treatment) affects the growth trajectory of the
mediating variable, which then affects the growth trajectory of the dependent
variable. The growth trajectory, for both the mediator and the outcome, is
represented by the slope factor which is specified in the model as a latent variable
(Cheong, MacKinnon, & Khoo, 2003).
The LGC model can be implemented in at least two versions: a single-stage
parallel process model and a two-stage piecewise parallel process model. The first
aims to demonstrate the relationships between treatment and changes in the
mediator and the outcome, without proving that a prior change in M is related to a
later change in Y. In the second version, the growth of the mediator and the
outcome process can be modeled separately for the earlier and for the later periods.
As a consequence, the mediated effects can be evaluated in different periods, as it is
more sensitive in estimating mediated effects when the trajectory shape changes
across time.
As one might expect, the LGC model is not perfect. The major criticism of
this model is that the measure itself may change over time, which may yield a
confusing representation of change over time.
With any of these mediational procedures in mind, it is important to note
that the mediated effects must be first tested for statistical significance; computing
the size of the mediated effect is also recommended. One way to test for the
significance of the mediated effect is to construct the confidence interval for the
mediated value and assess whether zero is included in the confidence interval.
Another way consists in calculating the standard error of the mediated effect (for
example using the formula derived by Sobel, 1982) and dividing the estimate of the
mediated effect by its standard error and by comparing this value to tabled values of
the normal distribution. Regarding the size of the mediated effect, there are three
categories of such measures: proportion or ratio measures, R squared measures and
standardized effect measures. Among them, one of the most common is the ratio
ab/c (where the ab is the mediated effect and c the total effect) which represents the
proportion of the total effect that is mediated. Such a measure has the advantage of
being easy to compute and also very intuitive.
Beyond the statistical procedures presented here, there are additional
methods to test for mediation such as the Autoregressive Latent Trajectory (ALT)
Model (Curran & Hussong, 2003; Bollen & Curran, 2004), differential equation
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
308
models (Boker & Nesselroade, 2002) and the Person-Centered Mediation Models
(Witteman et al., 1998). Yet as far as our goal is concerned getting into more details
might be redundant as there is a vast body of research addressing these very specific
issues.
Recommendations and strategic considerations for further research
Psychotherapy research has a long history, yet few studies have addressed the
theoretical and methodological milestones for evaluating the processes through
which psychotherapy exerts its effects. The investigation of mechanisms of change
via analyses of mediation is very informative yet it can be improved in a number of
ways. A growing body of literature has laid out recommendations for future
research in order to enhance our understanding of therapeutic change (Hinshaw,
2002; Holmbeck, 1997; Kazdin & Nock, 2003; Kraemer et al., 2002; Kazdin, 2007;
Johansson & Hoglend, 2007; Murphy et al., 2009). The general view is that in
addition to the minimum requirements for demonstrating mediation, there are a
number of ways to bring further evidence for mediation and possibly compensate
for common design flaws (Johansson & Hoglend, 2007).
First, the decision to perform a mediation analysis needs to be taken in
advance as it will influence the choice of measures as well as when their use. Next,
hypotheses need to be guided by the theory underpinning the treatment and the
findings of prior research and have to be formulated in terms of the likely
mechanisms of action of the treatment investigated (Murphy et al., 2009). Putative
mechanisms should make theoretical sense and/or be supported by other empirical
data; this way we can avoid arbitrary mediators that add little to our understanding
of psychotherapy (Johansson & Hoglend, 2007). Then, the treatment, the putative
mediators and the outcome need to be operationalized and integrated in a suitable
assessment protocol. To include more than one mediator not only makes economic
sense but it is also a wise decision from a methodological standpoint. Ruling out
seemingly plausible mediators strengthens the case for the remaining ones
(Johansson & Hoglend, 2007). Ideally, mediators are investigated in the context of
randomized clinical trials that include a control condition, as this can rule out the
possibility that what appears to mediate change is simply a general effect of
receiving treatment or a naturally occurring change rather than the specific effect of
the treatment under consideration (Murphy et al., 2009). Finally, future research
must deal with what has been called the Achilles’ heel of mediator studies (Kazdin
& Nock, 2003) - the timeline issue. Without demonstrating the temporal relation, it
is difficult to say whether the outcome caused the mediator or the other way around.
Clearly, one cannot talk about mechanisms of change without having sound
empirical (i.e., statistical) support. Thus, it is important to implement rigorous
statistical procedures when testing for mechanisms of change and mediation. As
previously discussed, each procedure/solution has its own advantages and
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
309
limitations. We strongly believe that statistical procedures used in testing for
mediation should remain as simple as possible while still being capable of offering
very intuitive results in identifying and testing significance as well as computing the
size of mediated effects. Nevertheless, more complicated statistical procedures
(such as the LGC Model) can clearly offer more detailed and accurate information
about mechanisms of change in psychotherapy. Having said that, a reasonable aim
seems to be finding an equilibrium between the quality of the research design and
the complexity of the statistical procedures used in order to bring intuitive yet solid
empirical data supporting not only the efficacy of the psychotherapeutic
intervention investigated but also the mechanisms of change explored.
REFERENCES
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations.
Journal of Personality and Social Psychology, 51, 1173-1182.
Boker, S. M., & Nesselroade, J. R. (2002). A method for modeling the intrinsic dynamics of
intraindividual variability: Recovering the parameters of simulated oscillators in
multi-wave panel data. Multivariate Behavioral Research, 37, 127–160.
Bollen, K. A., & Curran, P. J. (2004). Autoregressive latent trajectory (ALT) models: A
synthesis of two traditions. Sociological Methods and Research, 32, 336–383.
Borenstein, M. (1994). The case for confidence intervals in controlled clinical trials.
Controlled Clinical Trials, 15, 411-428.
Burr, J. A., & Nesselroade, J. R. (1990). Change measurement. In A. von Eye (Ed.),
Statistical methods in longitudinal research (Vol. I, pp. 3–34). New York:
Academic Press.
Cheong, J., MacKinnon, D. P., & Khoo, S. T. (2003). Investigation of mediational processes
using parallel process latent growth curve modeling. Structural Equation Modeling,
10, 238–262.
Cole, D. A., & Maxwell, S. E. (2003). Testing mediational models with longitudinal data:
Questions and tips in the use of structural equation modeling. Journal of Abnormal
Psychology, 112, 558–577.
Collins, L. M., & Graham, J. W. (2002). The effect of the timing and spacing of observations
in longitudinal studies of tobacco and other drug use: temporal design
considerations. Drug and Alcohol Dependence, 68, S85-S96.
Cronbach, L. J., & Furby, L. (1970). How should we measure “change” or should we?
Psychological Bulletin, 74, 68–80.
Curran, P. J., & Hussong, A. M. (2003). The use of latent trajectory models in
psychopathology research. Journal of Abnormal Psychology, 112, 526–544.
David, D., & Montgomery, G. (2011). The Scientific Status of Psychotherapies: A New
Evaluative Framework for Evidence-Based Psychosocial Interventions. Clinical
Psychology: Science and Practice, 18, 89-99.
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
310
David, D. (2004). Special issue on the cognitive revolution in clinical psychology: Beyond
the behavioral approach-Conclusions: Toward and evidence-based psychology and
psychotherapy. Journal of Clinical Psychology, 4, 447-451.
DeRubeis, R. J., & Feeley, M. (1990). Determinants of change in cognitive therapy for
depression. Cognitive Therapy and Research, 14, 469–482.
Doss, B. D. (2004). Changing the way we study change in psychotherapy. Clinical
Psychology-Science and Practice, 11, 368-386.
Elkin, I. (1999). A major dilemma in psychotherapy outcome research: Disentangling
therapists from therapies. Clinical Psychology-Science and Practice, 6, 10-32.
Gelfand, L. A., Mensinger, J. L., & Tenhave, T. (2009). Mediation analysis: a retrospective
snapshot of practice and more recent directions. The Journal General Psychology,
136, 153-76.
Gibbons, M. B. C., Crits-Christoph, P., Barber, J. P., Stirman, S. W., Gallop, R., Goldstein,
L. A., et al. (2009). Unique and common mechanisms of change across cognitive
and dynamic psychotherapies. Journal of Consulting and Clinical Psychology, 77,
801-813.
Gollob, H. F., & Reichardt, C. S. (1985). Building time lags into causal models of cross-
sectional data. Proceedings of the Social Statistics Section of the American
Statistical Association, 165–170.
Haaga, D. A. F., & Stiles, W. B. (2000). Randomized clinical trials in psychotherapy
research: Methodology, design, and evaluation. In Snyder, C. R., & Ingram, R. E.
(Eds.), Handbook of psychological change: Psychotherapy processes and practices
for the 21st Century (pp. 14−39). New York: Wiley.
Jacobson, N. S., & Truax, P. (1991). Clinical-significance - A statistical approach to
Defining meaningful change in psychotherapy-research. Journal of Consulting and
Clinical Psychology, 59, 12-19.
Judd, C. M., & Kenny, D. A. (1981). Estimating the effects of social interventions. New
York: Cambridge University Press.
Kazdin, A. E., & Nock, M. K. (2003). Delineating mechanisms of change in child and
adolescent therapy: methodological issues and research recommendations. Journal
of Child Psychology and Psychiatry, 44, 1116–1129.
Kazdin, A. E. (2007). Mediators and mechanisms of change in psychotherapy research.
Annual Review of Clinical Psychology, 3, 1-27.
Kazdin, A. E. (2009). Understanding how and why psychotherapy leads to change.
Psychotherapy Research, 19, 418-428.
Kazdin, A. E., & Weisz, J. R. (1998). Identifying and developing empirically supported child
and adolescent treatments. Journal of Consulting and Clinical Psychology, 66,
19-36.
Kraemer, H. C. (1993). Reporting the size of effects in research studies to facilitate
assessment of practical or clinical significance. Journal of Consulting and Clinical
Psychology, 17, 527-536.
Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002). Mediators and
moderators of treatment effects in randomized clinical trials. Archives of General
Psychiatry, 59, 877-883.
R. Moldovan, S. Pintea
Cognition, Brain, Behavior. An Interdisciplinary Journal
19 (2015) 299-311
311
Lambert, M. J., & Bergin, A. E. (1994). The effectiveness of psychotherapy. In Bergin, A. E.
& Garfield, S. L. (Eds.), Handbook of psychotherapy and behavior change (4th Ed.)
(pp. 143-189). New York: Wiley.
Lambert, M. J., & Barley, D. E. (2001). Research summary on the therapeutic relationship
and psychotherapy outcome. Psychotherapy, 38, 357-361.
Lambert, M. J., & Ogles, B. M. (2004). The efficacy and effectiveness of psychotherapy. In
M. J. Lambert (Ed.), Bergin and Garfield’s handbook of psychotherapy and
behavior change (5th Ed). New York: Wiley.
Laurenceau, J. P., Hayes, A. M., & Feldman, G. C. (2007). Some methodological and
statistical issues in the study of change processes in psychotherapy. Clinical
Psychology Review, 27, 682-695.
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis (Applied Social Research
Methods Series, Vol. 49). Thousand Oaks: Sage.
Murphy, R., Cooper, Z., Hollon, S. D., & Fairburn, C. G. (2009). How do psychological
treatments work? Investigating mediators of change. Behaviour Research and
Therapy, 47, 1-5.
Nock, M. K. (2007). Conceptual and design essentials for evaluating mechanisms of change.
Alcoholism-Clinical and Experimental Research, 31, 4S-12S.
Rogosa, D. R. (1988). Myths about longitudinal research. In K. W. Schaie, R. T. Campbell,
W. M. Meredith, & S. C. Rawlings (Eds.), Methodological issues in aging research
(pp. 171–209). New York: Springer.
Roth, A., & Fonagy, P. (1996). The relationship between outcome and therapist training,
experience and technique. In What Works for Whom? New York: Guilford.
Shadish, W. R., & Sweeney, R. B. (1991). Mediators and moderators in metaanalysis -
Theres a reason we dont let dodo birds tell us which psychotherapies should have
prizes. Journal of Consulting and Clinical Psychology, 59, 883-893.
Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in structural
equation models. Sociological Methodology, 13, 290–312.
Tenhave, T., Joffe, M., Lynch, K., Brown, G., Maistro, S., & Beck, A. (2007). Causal
mediation analyses with rank preserving models. Biometrics, 63, 926–934.
Wampold, B. E., & Brown, G. S. (2005). Estimating variability in outcomes attributable to
therapists: A naturalistic study of outcomes in managed care. Journal of Consulting
and Clinical Psychology, 73, 914-923.
Witteman, J. C. M., D’Agostino, R. B., Stijnen, T., Kannel, W. B., Cobb, J. C., de Ridder,
M. A. J., et al. (1998). G-estimation of causal effects: Isolated systolic
hypertension and cardiovascular death in the Framingham Heart Study. American
Journal of Epidemiology, 148, 390–401.
