Lorem, Ipsum

A Multivariate Meta-Analysis of Motivational Interviewing Process and Outcome

Brian T. Pace, Aaron Dembe, and Christina S. Soma University of Utah

Scott A. Baldwin Brigham Young Unversity

David C. Atkins University of Washington

Zac E. Imel University of Utah

Motivational interviewing (MI) theory proposes a process whereby a set of therapist behaviors has direct effects on client outcomes and indirect effects through in-session processes (e.g., client change talk). Despite clear empirical support for the efficacy of MI across settings, the results of studies evaluating proposed links between MI process and outcome have been less clear. In the present study, we used a series of multivariate meta-analyses to test whether there are differential relationships between specific MI-consistent and MI- inconsistent therapist behaviors, MI therapist global ratings, client change language, and clinical outcomes. Based on 19 primary studies (N � 2,614), we found a significant relationship between MI-consistent therapist behaviors and greater client change talk, as well as greater client sustain talk. Higher therapist global ratings (empathy and MI spirit) were significantly related to increased MI-consistent behaviors, decreased MI- inconsistent behaviors, increased client change talk, yet also increased client sustain talk. Therapist global ratings were not significantly related to clinical outcomes. Client sustain talk was a significant predictor of worse clinical outcomes, while client change talk was unrelated to outcome. Variability within the correlations indicated that MI-consistent and MI-inconsistent therapist behaviors were differentially related to therapist global ratings of empathy and MI spirit. Similar to past research, present findings provide equivocal support for hypothesized MI process outcome relationships. Clinical implications and future areas of MI mechanism research are discussed.

Keywords: motivational interviewing, process and outcome, meta-analysis, substance use treatment

Motivational interviewing (MI) is an empirically supported psy- chotherapy, originally developed for treating alcohol and sub- stance abuse and now applied to behavioral health problems (e.g.,

physical activity for obesity, medication management) across a variety of settings (e.g., mental health outpatient clinics, hospitals, and dental offices; see Armstrong et al., 2011; Lundahl, Kunz, Brownell, Tollefson, & Burke, 2010; Lundahl et al., 2013; Miller & Rose, 2009). MI theory proposes a linguistic process wherein specific therapist verbal behaviors (e.g., open questions, reflec- tions, closed questions, and confrontations) and general therapist skills (e.g., empathy, MI spirit) lead to client verbal behaviors (e.g., change talk and sustain talk), which in turn influence client outcomes (Figure 1; adapted from Miller & Rose, 2009).

Despite strong evidence that MI is effective in producing be- havior change (Lundahl et al., 2010, 2013), and a well-developed theory of how that change is produced, research support for many of the pathways shown in Figure 1 is inconclusive. Most MI process studies involve the use of trained human coders to rate therapist or client behaviors with measures such as the Motiva- tional Interviewing Skills Code (MISC; Miller, Moyers, Ernst, & Amrhein, 2008) and Motivational Interviewing Treatment Integ- rity Scale (MITI; Moyers, Manuel, & Ernst, 2014). Some studies have found that therapist use of MI specific behaviors or skills are related to treatment outcome (Baer et al., 2008; McCambridge, Day, Thomas, & Strang, 2011), but effects are inconsistent. For example, Tollison et al. (2008) found that complex reflections (but not simple reflections), correlated with better clinical outcomes. McCambridge et al. (2011) found that only therapist MI spirit and percentage of complex reflections correlated with cannabis use cessation, but no other MI behavior or skill was associated with

This article was published Online First June 22, 2017. Brian T. Pace, Aaron Dembe, and Christina S. Soma, Department of

Educational Psychology, University of Utah; Scott A. Baldwin, Department of Psychology, Brigham Young University; David C. Atkins, Department of Psychiatry and Behavioral Sciences, University of Washington; Zac E. Imel, Department of Educational Psychology, University of Utah.

Funding for the preparation of this article was provided by the National Institutes of Health/National Institute on Alcohol Abuse and Alcoholism (NIAAA) under award R01/AA018673 and National Institute on Drug Abuse under award R34/DA034860. In addition, David C. Atkins time was supported in part by K02 AA023814. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Also note both Zac E. Imel and David C. Atkins have minority equity stakes in a technology company, Behavioral Informatix (http://www.behavioralinformatix .com/), which is focused on developing computational models that quantify aspects of patient–provider interactions.

We presented preliminary findings from this project at the Association for Behavioral and Cognitive Therapies Conference in Philadelphia, Penn- sylvania in 2015.

Correspondence concerning this article should be addressed to Brian T. Pace, Department of Educational Psychology, University of Utah, 1721 Campus Center Drive, SAEC 3220, Salt Lake City, UT 84112–9255. E-mail: [email protected]

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Psychology of Addictive Behaviors © 2017 American Psychological Association 2017, Vol. 31, No. 5, 524 –533 0893-164X/17/$12.00 http://dx.doi.org/10.1037/adb0000280

524

outcomes (for similar patterns of results, see also Baer et al., 2008; Gaume, Bertholet, Faouzi, Gmel, & Daeppen, 2010, 2013).

Several hypotheses could explain such inconsistent findings. First, behavioral coding is extremely labor intensive, and sample sizes in studies of MI process and outcome studies are typically limited (see Vader, Walters, Prabhu, Houck, & Field, 2010, n � 30; Baer et al., 2008, n � 52), though there are exceptions (Apodaca, Magill, Longabaugh, Jackson, & Monti, 2013, n � 157; Gaume et al., 2014, n � 208). As a result, inconsistent correlations between process measures and clinical outcomes across studies might simply reflect sampling variation associated with small sample sizes. In addition, many studies do not report associations between all links in the MI causal chain, reporting only those that provided significant results (e.g., Hodgins, Ching, & McEwen, 2009; McCambridge et al., 2011). Selective reporting may lead to biased estimates of how MI processes and outcomes are related.

To summarize the extant literature on MI process-outcome associations, Magill et al. (2014) conducted a meta-analysis of 12 studies (n � 1,004 clients) that assessed the relationship between MI therapist behaviors, client language, and clinical outcomes. For analysis, specific therapist behaviors were aggregated into MI- consistent (e.g., open questions, complex reflections) and MI- inconsistent (e.g., confront, warn) categories. As predicted by MI theory, there was a significant correlation between MI-consistent behaviors and client language toward behavior change (i.e., “change talk”; r � .26, 95% confidence interval, CI [.16, .35], k � 7 studies). However, more MI-consistent behaviors also correlated weakly, but positively, with client language away from behavior change (i.e., “sustain talk”; r � .10, 95% CI [�.02, .22], k � 8). Greater MI-inconsistent behaviors were related to less client change talk (r � �.17, 95% CI [�.26, �.07], k � 6), and a smaller but still significant association, with client sustain talk (r � .07, 95% CI [.02, .13], k � 6). There was no significant relationship between change talk and clinical outcomes (r � .06, 95% CI [�.09, .21], k � 7), but greater client sustain talk was associated with worse clinical outcomes (r � �.24, 95% CI [�.36, �.11], k � 9). These findings provided only partial support for the standard theoretical model of MI.

There are two primary limitations in the Magill et al. (2014) meta-analysis. First, the study was an examination of the technical hypothesis of MI, focusing on behavioral codes only, and did not assess the relational factors theorized to predict positive change, that is, the “global” ratings of therapist empathy and MI spirit (see Figure 1; Miller & Rose, 2009). These relational factors are core

theoretical components of MI, but the extent to which they are associated with change processes or outcomes remains unexplored in meta-analytic work. While early reports provided strong asso- ciations between skills like empathy and treatment outcome (r � .82; Miller, Taylor, & West, 1980), recent work has suggested that these correlations may be less dramatic (Moyers, Houck, Rice, Longabaugh, & Miller, 2016) and more similar to smaller effects observed outside of MI and substance abuse research (r � .31; Elliott, Bohart, Watson, & Greenberg, 2011). Second, as is typical in many MI process studies, MI therapist behaviors were combined into broad categories of MI-consistent and MI-inconsistent behav- iors. While this approach reduces complexity, and provides a broad test of MI process, it also limits our ability to observe differences between specific MI behaviors (e.g., simple vs. com- plex reflections) in relation to therapist global ratings, client lan- guage, or clinical outcomes.

More recently, Romano and Peters (2015) examined an MI process and outcome causal model among nonsubstance use dis- order populations (e.g., anxiety, mood, eating, and psychotic dis- orders). Among 20 studies that compared either an MI stand-alone intervention or MI as an adjunct treatment to a minimal control group, results showed that MI led to increased patient motivation for all target problems except eating disorders, increased rates of attendance, and increased in-session engagement. In contrast, the authors failed to find support for increased patient confidence and increased therapist empathy or MI spirit in the MI treatment conditions. However, there were several limitations. First, the number of observations among each link of an MI causal model varied and some were as limited as one study. The authors tested a model separate from Miller and Rose’s (2009) version (see Apodaca & Longabaugh, 2009). Second, MI was often assessed as an adjunct treatment (16 of the 20 studies); thus, it is impossible to understand the unique contribution of MI to these associations. Third, data was often collected in each individual study via patient self-report and only one study used a bona-fide MI fidelity mea- sure (e.g., MISC or MITI), providing no information about specific therapist behaviors that occurred in session.

The present meta-analysis addresses several limitations of prior meta-analyses. First, we restricted our analyses to MI-only condi- tions and studies that used observational MI fidelity measures. Second, to explore relational hypotheses about how MI works, we included therapist global ratings. Third, rather than rely solely on associations between composite MI-consistent and MI- inconsistent scores, we used multivariate meta-analysis methods to

Figure 1. Hypothesized pathways that facilitate behavior change in motivational interviewing (MI). Figure adapted from “Toward a Theory of Motivational Interviewing,” by W. R. Miller and G. S. Rose, 2009, American Psychologist, 64, pp. 527–537. Copyright 2009 by the American Psychological Association.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

525MI META-ANALYSIS

understand how the individual therapist behaviors that underlie these composites are associated with client change language, ther- apist global ratings, and clinical outcomes. This is an attempt to answer a call to MI mechanism research by Miller and Rose (2009) stating a need to “look under the hood” at MI process and outcome to improve clinical training and treatment deliver (p. 527).

Based on the MI theoretical framework outlined by Miller and Rose (2009) and a previous meta-analysis (Magill et al., 2014), we examined the following hypotheses:

Hypothesis 1: We predicted that higher therapist global ratings (empathy and MI spirit) would correlate with greater MI- consistent behavior and client change talk and less MI- inconsistent behaviors and client sustain talk.

Hypothesis 2: We predicted that increased MI-consistent be- haviors would correlate with increased client change talk and that increased MI-inconsistent behaviors would correlate with increased client sustain talk.

Hypothesis 3: We expected the relationship between client change language and outcomes would replicate the Magill et al. meta-analysis, that is, client change talk would be unrelated to clinical outcomes, and client sustain talk would be related to worse clinical outcomes.

Hypothesis 4: We hypothesized that an overall increase in MI-consistent behaviors would correlate with slightly better clinical outcomes.

Hypothesis 5: We expected significant heterogeneity of the individual components of MI-consistent and MI-inconsistent with client change language and therapist global ratings. We did not have any specific hypotheses regarding which specific individual therapist behaviors would be differentially related to client change language or therapist global ratings, as this analysis was exploratory.

Method

Selection of Studies

We conducted a literature search to identify potential studies of MI adherence and outcome. Primary search terms included: mo- tivational interviewing, MI, and motivational enhancement. Sec- ondary search terms included: outcome, competence, adherence, fidelity, spirit, change talk, MISC, MITI, and Yale Adherence Scale. The search used the following databases: PsycINFO, Psy- cArticles, ERIC, Academic Search Premier, and PubMed. We also reviewed previously published meta-analysis on MI for studies that met the inclusion criteria.

We included studies that had the following components: (a) an association between some aspect of therapist and client behavior and/or outcome in an MI context, (b) statistical data sufficient to calculate correlational effect sizes, (c) publication in a peer- reviewed journal, (d) examination of adult and/or adolescent pop- ulations, (e) group or individual treatment format, and (f) assess- ment of adherence to MI through trained raters (see Figure 2). For studies that did not report data in a format necessary to calculate

correlational effect sizes, or studies that did not report all effects, we contacted primary authors and collected correlation matrices.

Variable Selection

Therapist behaviors. Analyses included individual adher- ence variables that comprised MI-consistent and MI-inconsistent summary scores. Unlike prior work that relies on these summary scores alone, we estimated correlations of therapist behavior vari- ables using the individual codes that make up the MI-consistent and MI-inconsistent broad categories. For MI-consistent, the vari- ables included: advise with permission, affirm, complex reflec- tions, simple reflections, emphasize control, open questions, raise concern with permission, reframe, and support. MI-inconsistent variables included: advise without permission, confront, direct, raise concern without permission, warn, and closed questions. Retaining the raw correlation allowed for the examination of variability within the aggregate category. The raw therapist behav- iors that comprised these correlations are listed in Table 1.

Global ratings. We included two primary therapist global ratings: therapist empathy and MI spirit. Empathy ratings measure how well the clinician attempted to understand and take the per- spective of the client. MI spirit is a gestalt rating of the clinician with three facets: collaboration (avoiding an authoritative stance), evocation (drawing out the client’s perspective rather than giving advice), and autonomy (accepting that clients have the choice of whether to change).

Figure 2. A primary study inclusion flowchart. � The two most frequent reasons for exclusion were that studies either did not test a link in the motivational interviewing (MI) process outcome causal model or studies failed to report statistical data sufficient to calculate correlational effect sizes.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

526 PACE ET AL.

Client change language. We coded two client change lan- guage variables. Client change talk is client utterances indicating movement toward the target behavior change, for instance, reduc- ing substance use. Client sustain talk indicated movement away from the target behavior change.

Clinical outcomes. We selected client outcomes that evalu- ated the primary target of behavior change. These mostly consisted of substance use outcomes, plus one HIV medication adherence study and one gambling study. When multiple outcomes were reported (e.g., drinking days per week and a consequences of drinking measure) these outcomes were aggregated across com- parison variables. We recorded the most recent time point after the intervention.

Meta-Analytic Approach

We conducted two series of six multivariate meta-analytic mod- els to quantify and test the relationship and variability between and among different MI therapist and client behaviors, therapist global ratings, and client outcomes. The six models tested associations of: (a) MI therapist behaviors (consistent/inconsistent) and client change language (change talk/sustain talk), (b) MI therapist be- haviors and clinical outcomes, (c) client change language and clinical outcomes, (d) MI therapist behaviors and MI therapist

global ratings (empathy/MI spirit), (e) client change language and MI therapist global ratings, and (f) MI therapist global ratings and clinical outcomes.

The first series of models (Models 1A–F) served as baseline models with a fixed effect for the type of correlation and a random effect for study. For example, in the first meta-analysis (Model 1A) of MI therapist behaviors and client change language, the fixed effects provided four correlation estimates, including the relationship between (a) therapist MI-consistent behaviors with client change talk, (b) therapist MI-consistent behaviors with client sustain talk, (c) therapist MI-inconsistent behaviors with client change talk, and (d) therapist MI-inconsistent behaviors with client sustain talk. Thus, this model provides both a fixed effect estimate of these four correlations and between study variability in the one overall aggregate correlation (i.e., MI-behavior and change language). Thus, across the six meta-analyses, the fixed effects yielded 18 unique correlation types, including associations between MI-consistent and MI- inconsistent behaviors, clinical outcomes, as well as change talk, sustain talk, and therapist empathy and MI spirit global ratings (see Figure 3).

The second series of six meta-analytic models (Models 2A–F) included a stratified random effect for the correlation type. This random effect allowed the fixed effect for correlation type to vary across studies, and provides separate variability estimates by type of correlation. Accordingly, this model tests whether the estimates of a specific correlation varied across studies, but also indexed the amount of variability within a particular correlation type. For example, with Model 2A, there are now separate estimates of variability of the correlation between (a) MI-consistent behavior and change talk, (b) MI-consistent behavior and sustain talk, (c) MI-inconsistent behavior and change talk, and (d) MI-inconsistent behavior and sustain talk. These variance estimates capture the associations of individual codes within a type (e.g., complex reflection and sustain talk).

A likelihood ratio test was used to determine if including ran- dom effects improved the fit of the models (Models 2A–F). A significant likelihood ratio test would indicate that there is signif- icant variability in raw correlations within correlation types. When this was true, we examined these specific correlations (via the fixed effects) included in the aggregate correlation to determine

Table 1 MI-Consistent (MICO) and MI-Inconsistent (MIIN) Variable Composition

MICO MIIN

Advise with permission Advise without permission Affirm Confront Complex reflections Closed questions Emphasize control Direct Open questions Raise concern without permission Raise concern with permission Warn Reframe Simple reflections Support

Note. MI � motivational interviewing. The variable selection was based on examination of MISC and MITI manuals.

Figure 3. Structure of the six separate motivational interviewing (MI) comparisons and the 18 correlation types. Models 1A–F including a fixed effect at the correlation type level and a random effect at the study level. Models 2A–F including an additional random effect at the correlation type level. MICO � MI-consistent behavior; MIIN � MI-inconsistent behavior; CT � client change language; ST � client sustain language; OUT � outcome; EMP � therapist empathy; MIS � therapist MI spirit.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

527MI META-ANALYSIS

which correlations may be larger than others (e.g., the relationship between therapist empathy and complex reflections, a variable within the MICO composite).

We aggregated duplicate effect sizes (i.e., estimates of the same raw correlation in the same study) within-study using the MAc package (Del Re & Hoyt, 2010), using Hunter and Schmidt’s (2004) aggregation approach and assuming a .50 within-study correlation (Wampold et al., 1997). We fit all models using the metafor package (Viechtbauer, 2010, 2014) with the R statistical software (R Development Core Team, 2010). For the fixed effects, interpretation of the magnitude of the correlation coefficients are .10 for small effects, .30 for medium effects, and .50 for large effects (Cohen, 1992).

Results

Reports from 19 primary studies (N � 2,614, M � 137.6, range � 30 –372) met the inclusion criteria and examined at least one link in the MI causal chain (see Figure 1). The number of effects (total n � 604, range per study � 1–73, mean per study � 31.8) included 81 different raw correlation types. Sample sizes for each meta-analysis were: (a) MI behavior and client change lan- guage (n � 214 correlations, k � 9 studies), (b) MI behavior and outcome (n � 94, k � 13), (c) client change language and outcome (n � 15, k � 8), (d) global ratings and MI behavior (n � 229, k � 14), (e) global ratings and client change language (n � 30, k � 8), and (f) global ratings and outcome (n � 22, k � 11).

Correlations of MI Processes and Outcome

Figure 4 provides estimates of each of the 18 correlation types within the six multivariate fixed effect models. For MI behavior and client change language correlations (Model A), there was a significant relationship between therapist MI-consistent behaviors

and client change talk (r � .17, 95% CI [.11, .23]), but not between MI-inconsistent behaviors and client change talk (r � .02, 95% CI [�.04, .08]). There was a small and significant correlation between MI-consistent and increased sustain talk (r � .10, 95% CI [.04, .16]), and a nonsignificant correlation between MI-inconsistent therapist behaviors and sustain talk (r � .06, 95% CI [�.001, .12]). MI-consistent therapist behaviors correlated with worse clin- ical outcomes (Model B; r �.04, 95% CI [�.07, �.005]) and MI-inconsistent was unrelated to clinical outcomes (r �.01, 95% CI [�.05, .02]). In regards to change language, client sustain talk significantly, negatively correlated with clinical outcome (Model C; r � �.23, 95% CI [�.35, �.12]), indicating that an increase in sustain talk was associated with worse clinical outcomes. The relationship between change talk and clinical outcomes was near zero and not significant (r � �.05, 95% CI [�.16, .06]).

Both therapist global ratings, empathy and MI spirit, correlated with increased MI-consistent behaviors (Model C; empathy, r � .16, 95% CI [.10, .22]; MI spirit, r � .15, 95% CI [.08, .20]), decreased MI-inconsistent behaviors (empathy, r � �.13, 95% CI [�.20, �.06]; MI spirit, r � �.19, 95% CI [�.28, �.11]), in- creased client change talk (Model E; empathy, r � .25, 95% CI [.11, .38]; MI spirit, r � .25, 95% CI [.11, .39]), but also with increased sustain talk (empathy, r � .18, 95% CI [.05, .31]; MI spirit, r � .13, 95% CI [�.005, .27]). However, the association between MI spirit and sustain talk was not significant (p � .06). The relationship of therapist global ratings and clinical outcomes was small and not significant (Model F; empathy, r � .03, 95% CI [�.05, .10]; MI spirit, r � .04, 95% CI [�.04, .12]).

Consistency of Correlations Within MI Categories

A particular strength of the multivariate mixed effects models is the ability to assess the variability of correlations within the 18 correlation types, addressing the extent to which all specific ther-

Figure 4. Effect sizes in the motivational interviewing (MI) causal model. emp � therapist empathy; mis � MI spirit. � p � .05. �� p � .01.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

528 PACE ET AL.

apist variables that comprise MI behaviors correlate similarly with the comparison variable (e.g., therapist empathy or MI spirit rat- ings). As noted above, we tested this by including random effects for the correlation type in the model, which provided tests of variability between the levels of each type of correlations.

In only one set of models was there evidence of variability within correlation types. As indicated by a significant improve- ment in fit between Models 1D and 2D, there was significant variability within correlation types for MI-consistent and MI- inconsistent variables and therapist global ratings of empathy and MI spirit (Model D; �2(10) � 89.3, p � �.001). This was the largest sample size of any submodel and had the largest power relative to the five other comparisons to detect differences. For the other models, none of the other likelihood estimates were signif- icant, indicating a lack of significant variability within the corre- lation types.

MI therapist behaviors and therapist global ratings. To explore variability within the significant correlation types, we reran Model 2D with the fixed effect at the level of the raw correlation type to examine differences at the individual therapist behavior level. We compared individual therapist behaviors among both MI-consistent and MI-inconsistent categories across two ther- apist global ratings: empathy and MI spirit (see Figure 5). MI- consistent behaviors generally trended in the hypothesized direc- tion wherein increased counts of these behaviors correlated with higher empathy and MI spirit ratings. The complex reflection variable had the largest effects with empathy and MI spirit (r � .38 and r � .35, respectively). In contrast, correlations between sup- port (empathy r � .20 and MI spirit r � .13) and simple reflections (empathy r � .17 and MI spirit r � .16) were smaller.

We found a similar pattern regarding MI-inconsistent behaviors but in the opposite direction. All MI-inconsistent behaviors corre- lated with lower empathy and MI spirit scores (range � �.34 to �.02). The variables confront (empathy r � �33, MI spirit r � �.34) and advise without permission (empathy r � �.20, MI spirit r � �.33) had the largest predictive weight of lower thera- pist global ratings. These data suggest that MI-inconsistent behav- iors, particularly confronting a client during session or giving advice without their permission, are related to lower empathy and MI spirit global ratings.

Discussion

The present study is the first aggregate examination of MI fidelity measures and outcome to include both technical elements (i.e., MI therapist behaviors in relation to client change language, including the dismantling of MI-consistent and MI-inconsistent therapist behaviors), and relational elements (i.e., therapist global ratings). Our analysis found support for some parts of the MI model but not for others. The ultimate goal of process research is to understand what in-session behaviors are related to client im- provement (or worsening). Our results suggest that the causal model proposed by Miller and Rose (2009) is clearly identifying important aspects of therapy process, but simultaneously does not capture all of the important variance in client outcomes.

Therapist Global Ratings Correlations

MI-consistent and MI-inconsistent. Therapist global ratings of empathy and MI spirit generally correlated with MI-consistent and

Figure 5. Correlations between individual motivational interviewing (MI) therapist behaviors and empathy and MI spirit ratings. The point estimated shows the aggregated correlation and error bars represent 95% confidence intervals. The plots depict MI-consistent (left) and MI-inconsistent (right) therapist behaviors.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

529MI META-ANALYSIS

MI-inconsistent behaviors in the hypothesized directions: Higher global ratings with more consistent behaviors and lower global ratings with more inconsistent behaviors. Of the 14 studies that examined MI-consistent behaviors in relation to empathy, 13 of the 14 were positive correlations, ranging from small to large in magnitude, with one small negative correlation (r � �.06; Gaume et al., 2013). We found a similar trend for the 13 studies that examined MI-consistent behaviors in relation to MI spirit, with nearly all positive correlations except for one small negative correlation (Gaume et al., 2013). The reverse was true for MI-inconsistent behaviors, a majority correlated negatively with empathy and MI spirit except a few small positive correlations (Catley et al., 2006; Tollison et al., 2008). These findings support the concept that an empathic therapist demonstrating MI spirit uses more MI-consistent behaviors and fewer MI-inconsistent behav- iors.

Client change language. As hypothesized, therapist global ratings also correlated significantly with increased change talk, yet contrary to MI theory also correlated with increased sustain talk (albeit to a lesser extent than change talk, with significant results only in the case of empathy). Of the eight studies that examined this link in the model, all but one study (Apodaca et al., 2016) found small positive correlations between therapist empathy and MI spirit with client change and sustain talk. The finding that increased therapist global scores are associated with increased sustain talk points to two possible explanations. First, more em- pathic and reflective listening may simply elicit more clinically relevant client talk in general, both change and sustain (Fischer & Moyers, 2014). This may not represent a deficit in the model, but rather highlight the importance of the MI adherent therapist’s role to continually work to reinforce change talk. Second, it is possible that there is a component of mutual influence present where therapists may be working with more difficult clients and in return working harder on MI principles. Specifically, therapists may work harder to collaborate and empathize when clients are strug- gling (Imel, Baer, Martino, Ball, & Carroll, 2011).

Clinical outcomes. We found small and nonsignificant correla- tions between therapist global ratings and clinical outcomes, failing to support both MI theory and general findings on the relationship between empathy and clinical outcomes (see Elliott et al., 2011). One possible explanation for this null finding is limited variability and high mean empathy across providers due to a selection of previously competent therapists who are highly motivated, receive training, and then are supervised weekly. Another possible explanation is that some studies were prevention oriented (rather than intervention) studies that sought to reduce or prevent problematic behaviors in a future event (e.g., drinking on one’s 21st birthday; Neighbors et al., 2012). Pre- vention studies have smaller effects than intervention studies, in which individuals are selected into the study for having problems in need of intervention. Moreover, some studies included in the current analyses were null trials, reporting an overall null effect of the MI intervention (e.g., Apodaca et al., 2016; Baer et al., 2008). Both of these are examples of restricted range that can reduce the power of process-focused analyses.

MI Therapist Behaviors Correlations

Client change language. We found small, significant aggre- gate correlations between MI-consistent behaviors and both client change and sustain language. Of the nine studies that examined

these links in the MI model, all but one showed positive correla- tions between MI-consistent behaviors and change and sustain talk (Gaume et al., 2014). These correlations ranged from small to medium, and the single negative correlation between MI- consistent and sustain talk was near zero. This is an important component in MI theory, as therapists who more often engage in MI-consistent behaviors are expected to elicit greater amounts of change talk. However, the results also suggest that similar behav- iors are also likely to elicit increased client sustain talk statements. These results parallel the above findings that increased therapist empathy and MI spirit correlate with both increased change and sustain talk. Again, it seems likely that more open-ended probing will elicit more substance abuse relevant client talk in general (Fischer & Moyers, 2014). If a primary goal of MI is to reduce sustain talk, it would seem up to the therapist to work to contin- ually reinforce the change talk and avoid unintentionally reinforc- ing the sustain talk during these sessions. More recent implemen- tations of MI coding systems are now focusing on the extent to which therapist are asking questions about or reflecting specific types of change language (Moyers, Martin, Houck, Christopher, & Tonigan, 2009), although there is not yet sufficient data to conduct systematic reviews.

Clinical outcomes. The aggregate correlations between MI- consistent and MI-inconsistent behaviors and clinical outcomes were near zero. Yet surprisingly, increased MI-consistent behav- iors correlated significantly with worse clinical outcomes (r � �.04). Two studies showed medium positive correlations between MI-consistent behaviors and outcome (Thrasher et al., 2006 and McCambridge et al., 2011); however, four studies with sample sizes more than four times larger showed small and neg- ative results (Gaume et al., 2014; Krupski et al., 2012; Lee et al., 2014; Neighbors et al., 2012). Again, one possible explanation is range restriction and reduced power in outcome assessment in prevention studies (vs. intervention) and studies that exhibited an overall null MI effect. This finding fails to support the link in the MI causal model that increased MI-consistent behaviors in session is related with better clinical outcomes.

Client Change Language and Outcome

An important component of the MI model is that more client change language is thought to predict clinical outcomes (Miller & Rose, 2009). There was no evidence that increased counts of change talk during session led to better outcomes. All of the eight studies that assessed the relationship between change talk to outcome showed correlations that were near zero, except one outlier with a medium negative correlation (Vader et al., 2010). In contrast to change talk findings, our results provided evidence for increased sustain talk leading to worse outcomes with all seven studies showing negative correlations. Thus, there is growing research support for the assertion that clients actively engaging in sustain talk statements are more likely to have worse clinical outcomes.

These findings have important implications for how we think about client language as an indicator of underlying ambivalence and a predictor of change. First, it may be that session-level counts of behavioral codes cannot sufficiently capture the nuanced rela- tionship between change language as an indicator of ambivalence and outcomes. It is possible that client language may shift through- out a session, and that certain statements, provided in more mean-

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

530 PACE ET AL.

ingful therapeutic moments, are better indicators of change. Se- quential analyses and more advanced statistical models may better be able to explore this relationship between these in-session be- haviors and their relation to clinical outcomes (Bertholet, Faouzi, Gmel, Gaume, & Daeppen, 2010; Moyers et al., 2009). However, there are very few studies in this area, and identifying these particular critical moments is likely to be challenging as they may be infrequent, and idiographic.

Second, change talk utterances may vary in their valence from weak to strong. A therapist may easily elicit change talk by asking about the negative impact alcohol, for example, but this change talk may be fleeting and not as significant as other forms of change talk, possibly those self-elicited by the client. The research on attempting to code strength of change language has been mixed (Amrhein, Miller, Yahne, Palmer, & Fulcher, 2003), and some replicate the present findings that even when assessing strength, only sustain talk significantly predicts posttreatment outcomes (Gaume et al., 2016). Many studies focus on general change or sustain categories because of interrater reliability problems asso- ciated with discriminating among different types of change talk. We may need to reevaluate how we code strength of change language, including possibly exploring the assessment of the un- derlying emotion during moments of change or sustain talk.

It is also possible that the technology of MI has made change talk less of a meaningful indicator of an underlying client state because the actual words of change are easily manipulated via therapist intervention. However, the underlying attitude may be more rigid—in contrast to what might be predicted by self- perception theory (Bem, 1972). It is unlikely that MI therapists are intentionally increasing sustain talk—and thus, observed sustain talk may be a more valid indicator of the actual underlying state of the client. Future work examining the nuanced relationship be- tween client change language and outcomes will benefit from both questioning the theoretical tenets on which this link in the model operates as well as utilizing advanced statistical methodologies to parse out this relationship further and extend on initial findings (Bertholet et al., 2010; Moyers et al., 2009).

Assessing Variability Across Comparisons

A particular strength of the present study is the assessment of variability within the correlation types, that is, looking “under the hood” at specific MI behaviors (Miller & Rose, 2009). It is common for researchers to aggregate MI therapist behaviors into broad “consistent” and “inconsistent” behaviors. Until the present study, there has not been a test of these composite correlations into their constituent parts. Given the selective reporting of behaviors across studies, this is a vital test to understand possible differential effects of individual therapist behaviors. We hypothesized that there would be significant variability within the MI-consistent and MI-inconsistent composite correlations in relation to client change language and therapist global ratings. Our results showed signifi- cant variability in the latter, but not the former.

We observed a significant amount of variability between sepa- rate MI therapist behaviors in relation to empathy and MI spirit. We found that MI-consistent behaviors generally correlated with empathy and MI spirit, and complex reflections had the largest effect size. Thus, when a therapist provided reflective statements above and beyond what the client stated, they were more likely to

be rated as empathic and engaging in MI spirit. In contrast, a number of therapist behaviors had negative correlations with ther- apist global ratings, including confront, advise without permission, raise concern without permission, and warn statements. Indepen- dent raters found therapists engaging in these behaviors to be less empathic and less engaged in MI spirit approach.

The lack of variability within the other composite MI-consistent and MI-inconsistent correlations in relation to client change lan- guage is indicative of a pattern as well. The aggregate correlations (shown in Figure 4) revealed a lack of differentiation between MI-consistent and client change and sustain talk. Correlations between both were small and near zero, yet significant, and sug- gest that MI-consistent therapist behaviors are not differentially related to the amount of change or sustain talk they predict. Correlations between MI-inconsistent and both change and sustain talk were near zero and nonsignificant. Thus, there does not seem to be evidence of the overall differential relationship between MI-consistent and change or sustain talk as well as individual therapist behaviors that create the composite variables.

Limitations and Directions for Further Research

This study was not without its limitations. The first limitation was the lack of data across all comparisons. As studies tended to selectively report significant relationships, or report effects that are not translatable to a correlation coefficient, it proved difficult to assess effects equally across the six broad comparisons models (shown in Figure 3). A second limitation is that all correlations are based on session-level counts of behavioral codes. This fails to assess the sequential processes at the talk-turn level, which is, examining whether specific therapist behaviors are more or less likely to elicit client change or sustain talk in the following talk turns. A third limitation is that this study did not assess additional MI variables found in the MISC manual, including a client global rating of self-exploration or MISC summary scores such as the ratio of reflections to questions or the percentage of open questions (Miller et al., 2008). We instead focused on counts of behavioral codes and therapist global ratings. An examination of other MI quality metrics and summary scores may lead to different results. Fourth, across all links in the MI causal model (see Figure 4) the aggregate effects were small or near zero and mostly homogenous (range � �.23–.25). We believe this is related to range restriction in therapist global ratings and low base rates in specific MI- consistent and MI-inconsistent behaviors where using well-trained and adherent MI therapists may attenuate the results. Another possible explanation is that MI fidelity measures may not adequately discrim- inate between therapists and examining sources of variability and therapist and client is a critical area of research going forward (Bo- swell et al., 2013; DeRubeis, Brotman, & Gibbons, 2005; Imel et al., 2011). Fifth, inherent in behavioral coding is addressing the issue of occasional poor interrater reliability. We decided to not restrict inclu- sion on this criterion due to limiting the number of observations in the meta-analysis and our desire to run multivariate models.1

1 We ran a subset of the data without the 68 effects with poor reliability. The effect sizes were nearly identical to those reported in the manuscript except for the link between MI-inconsistent and sustain talk. Although the overall effect was unchanged and small (r � .06), the p-value changed from .0504 to .0486.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

531MI META-ANALYSIS

Efforts to assess MI process outcome research are limited by the inherent constraints of labor-intensive behavioral coding, resulting in small sample sizes. Despite the increased sample size in the present meta-analysis, we are still only examining a fraction of the data from all published MI outcome trials. The present study included a total of 2,614 coded sessions across 19 primary studies. This figure may seem large in comparison the Magill et al. (2014) meta-analysis of 1,004 coded sessions across 12 studies, but pales in comparison to MI efficacy study research. A meta-analysis by Lundahl and colleagues (2010) included 204,415 sessions across 119 MI clinical trials. One way for researchers in our field to address this limitation is by attempting to scale up efforts to assess MI process and outcome. Recent work suggests technological advances in assessing MI adherence using automated speech rec- ognition and natural language processing (for a review see Pace et al., 2016) may increase our ability to collect and analyze data on MI process and outcome. Researchers have demonstrated the abil- ity to use machine learning procedures to analyze large unstruc- tured text corpora to identify semantic content (Steyvers & Grif- fiths, 2007), discriminate between different psychotherapies (Imel, Steyvers, & Atkins, 2014), automatically assign MI behavioral codes (Tanana, Hallgren, Imel, Atkins, & Srikumar, 2016), and assess MI empathy ratings with reliability on par with observer ratings (Xiao, Imel, Georgiou, Atkins, & Narayanan, 2015). These advances will allow us to scale up our ability to collect critical data, improve our ability to understand MI process and outcome, and answer key questions about how MI works.

References

References marked with an asterisk indicate studies included in the meta-analysis.

Amrhein, P. C., Miller, W. R., Yahne, C. E., Palmer, M., & Fulcher, L. (2003). Client commitment language during motivational interviewing predicts drug use outcomes. Journal of Consulting and Clinical Psy- chology, 71, 862– 878. http://dx.doi.org/10.1037/0022-006X.71.5.862

�Apodaca, T. R., Jackson, K. M., Borsari, B., Magill, M., Longabaugh, R., Mastroleo, N. R., & Barnett, N. P. (2016). Which individual therapist behaviors elicit client change talk and sustain talk in motivational interviewing? Journal of Substance Abuse Treatment, 61, 60 – 65. http:// dx.doi.org/10.1016/j.jsat.2015.09.001

Apodaca, T. R., & Longabaugh, R. (2009). Mechanisms of change in motivational interviewing: A review and preliminary evaluation of the evidence. Addiction, 104, 705–715. http://dx.doi.org/10.1111/j.1360- 0443.2009.02527.x

�Apodaca, T. R., Magill, M., Longabaugh, R., Jackson, K. M., & Monti, P. M. (2013). Effect of a significant other on client change talk in motivational interviewing. Journal of Consulting and Clinical Psychol- ogy, 81, 35– 46. http://dx.doi.org/10.1037/a0030881

Armstrong, M. J., Mottershead, T. A., Ronksley, P. E., Sigal, R. J., Campbell, T. S., & Hemmelgarn, B. R. (2011). Motivational interview- ing to improve weight loss in overweight and/or obese patients: A systematic review and meta-analysis of randomized controlled trials. Obesity Reviews, 12, 709 –723. http://dx.doi.org/10.1111/j.1467-789X .2011.00892.x

�Baer, J. S., Beadnell, B., Garrett, S. B., Hartzler, B., Wells, E. A., & Peterson, P. L. (2008). Adolescent change language within a brief motivational intervention and substance use outcomes. Psychology of Addictive Behaviors, 22, 570 –575. http://dx.doi.org/10.1037/a0013022

�Baer, J. S., Wells, E. A., Rosengren, D. B., Hartzler, B., Beadnell, B., & Dunn, C. (2009). Agency context and tailored training in technology

transfer: A pilot evaluation of motivational interviewing training for community counselors. Journal of Substance Abuse Treatment, 37, 191–202. http://dx.doi.org/10.1016/j.jsat.2009.01.003

Bem, D. J. (1972). Self-perception theory. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 6, pp. 1– 62). New York, NY: Academic Press.

�Bertholet, N., Faouzi, M., Gmel, G., Gaume, J., & Daeppen, J. B. (2010). Change talk sequence during brief motivational intervention, towards or away from drinking. Addiction, 105, 2106 –2112. http://dx.doi.org/10 .1111/j.1360-0443.2010.03081.x

Boswell, J. F., Gallagher, M. W., Sauer-Zavala, S. E., Bullis, J., Gorman, J. M., Shear, M. K., . . . Barlow, D. H. (2013). Patient characteristics and variability in adherence and competence in cognitive-behavioral therapy for panic disorder. Journal of Consulting and Clinical Psychology, 81, 443– 454. http://dx.doi.org/10.1037/a0031437

�Catley, D., Harris, K. J., Mayo, M. S., Hall, S., Okuyemi, K. S., Board- man, T., & Ahluwalia, J. S. (2006). Adherence to principles of motiva- tional interviewing and client within-session behavior. Behavioural and Cognitive Psychotherapy, 34, 43–56. http://dx.doi.org/10.1017/ S1352465805002432

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155–159. http://dx.doi.org/10.1037/0033-2909.112.1.155

Del Re, A. C., & Hoyt, W. T. (2010). MAc: Meta-analysis with correla- tions. R package version 1.1. Retrieved from http://CRAN.R-project .org/package�MAc

DeRubeis, R. J., Brotman, M. A., & Gibbons, C. J. (2005). A conceptual and methodological analysis of the nonspecifics argument. Clinical Psychology: Science and Practice, 12, 174 –183. http://dx.doi.org/10 .1093/clipsy.bpi022

Elliott, R., Bohart, A. C., Watson, J. C., & Greenberg, L. S. (2011). Empathy. Psychotherapy: Theory, Research, & Practice, 48, 43– 49. http://dx.doi.org/10.1037/a0022187

�Fischer, D. J., & Moyers, T. B. (2014). Is there an association between empathic speech and change talk in motivational interviewing sessions? Alcoholism Treatment Quarterly, 32, 3–18. http://dx.doi.org/10.1080/ 07347324.2014.856225

�Gaume, J., Bertholet, N., Faouzi, M., Gmel, G., & Daeppen, J. B. (2010). Counselor motivational interviewing skills and young adult change talk articulation during brief motivational interventions. Journal of Sub- stance Abuse Treatment, 39, 272–281. http://dx.doi.org/10.1016/j.jsat .2010.06.010

�Gaume, J., Bertholet, N., Faouzi, M., Gmel, G., & Daeppen, J. B. (2013). Does change talk during brief motivational interventions with young men predict change in alcohol use? Journal of Substance Abuse Treat- ment, 44, 177–185. http://dx.doi.org/10.1016/j.jsat.2012.04.005

�Gaume, J., Gmel, G., Faouzi, M., & Daeppen, J.-B. (2008). Counsellor behaviours and patient language during brief motivational interventions: A sequential analysis of speech. Addiction, 103, 1793–1800. http://dx .doi.org/10.1111/j.1360-0443.2008.02337.x

�Gaume, J., Magill, M., Longabaugh, R., Bertholet, N., Gmel, G., & Daeppen, J. B. (2014). Influence of counselor characteristics and behav- iors on the efficacy of a brief motivational intervention for heavy drinking in young men—A randomized controlled trial. Alcoholism: Clinical and Experimental Research, 38, 2138 –2147. http://dx.doi.org/ 10.1111/acer.12469

Gaume, J., Magill, M., Mastroleo, N. R., Longabaugh, R., Bertholet, N., Gmel, G., & Daeppen, J.-B. (2016). Change talk during brief motiva- tional intervention with young adult males: Strength matters. Journal of Substance Abuse Treatment, 65, 58 – 65. http://dx.doi.org/10.1016/j.jsat .2016.01.005

�Hodgins, D. C., Ching, L. E., & McEwen, J. (2009). Strength of commit- ment language in motivational interviewing and gambling outcomes. Psychology of Addictive Behaviors, 23, 122–130. http://dx.doi.org/10 .1037/a0013010

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

532 PACE ET AL.

Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis (2nd ed.). Thousand Oaks, CA: Sage. http://dx.doi.org/10.4135/9781412985031

Imel, Z. E., Baer, J. S., Martino, S., Ball, S. A., & Carroll, K. M. (2011). Mutual influence in therapist competence and adherence to motivational enhancement therapy. Drug and Alcohol Dependence, 115, 229 –236. http://dx.doi.org/10.1016/j.drugalcdep.2010.11.010

Imel, Z. E., Steyvers, M., & Atkins, D. C. (2014). Computational Psycho- therapy research: Scaling up the evaluation of patient–provider interac- tions. Psychotherapy, 52, 19 –30. http://dx.doi.org/10.1037/a0036841

�Krupski, A., Joesch, J. M., Dunn, C., Donovan, D., Bumgardner, K., Lord, S. P., . . . Roy-Byrne, P. (2012). Testing the effects of brief intervention in primary care for problem drug use in a randomized controlled trial: Rationale, design, and methods. Addiction Science & Clinical Practice, 7, 27. http://dx.doi.org/10.1186/1940-0640-7-27

�Lee, C. M., Neighbors, C., Lewis, M. A., Kaysen, D., Mittmann, A., Geisner, I. M., . . . Larimer, M. E. (2014). Randomized controlled trial of a Spring Break intervention to reduce high-risk drinking. Journal of Consulting and Clinical Psychology, 82, 189 –201. http://dx.doi.org/10 .1037/a0035743

Lundahl, B. W., Kunz, C., Brownell, C., Tollefson, D., & Burke, B. L. (2010). A meta-analysis of motivational interviewing: Twenty-five years of empirical studies. Research on Social Work Practice, 20, 137–160. http://dx.doi.org/10.1177/1049731509347850

Lundahl, B., Moleni, T., Burke, B. L., Butters, R., Tollefson, D., Butler, C., & Rollnick, S. (2013). Motivational interviewing in medical care set- tings: A systematic review and meta-analysis of randomized controlled trials. Patient Education and Counseling, 93, 157–168. http://dx.doi.org/ 10.1016/j.pec.2013.07.012

Magill, M., Gaume, J., Apodaca, T. R., Walthers, J., Mastroleo, N. R., Borsari, B., & Longabaugh, R. (2014). The technical hypothesis of motivational interviewing: A meta-analysis of MI’s Key Causal Model. Journal of Consulting and Clinical Psychology, 82, 973–983. http://dx .doi.org/10.1037/a0036833

�McCambridge, J., Day, M., Thomas, B. A., & Strang, J. (2011). Fidelity to Motivational Interviewing and subsequent cannabis cessation among adolescents. Addictive Behaviors, 36, 749 –754. http://dx.doi.org/10 .1016/j.addbeh.2011.03.002

Miller, W. R., Moyers, T. B., Ernst, D., & Amrhein, P. (2008). Manual for the Motivational Interviewing Skill Code Version 2.1 (MISC). Unpub- lished manual. University of New Mexico.

Miller, W. R., & Rose, G. S. (2009). Toward a theory of motivational interviewing. American Psychologist, 64, 527–537. http://dx.doi.org/10 .1037/a0016830

Miller, W. R., Taylor, C. A., & West, J. (1980). Focused versus broad- spectrum behavior therapy for problem drinkers. Journal of Consulting and Clinical Psychology, 48, 590 – 601. http://dx.doi.org/10.1037/0022- 006X.48.5.590

Moyers, T. B., Houck, J., Rice, S. L., Longabaugh, R., & Miller, W. R. (2016). Therapist empathy, combined behavioral intervention, and alco- hol outcomes in the COMBINE research project. Journal of Consulting and Clinical Psychology, 84, 221–229. http://dx.doi.org/10.1037/ ccp0000074

Moyers, T. B., Manuel, J. K., & Ernst, D. (2014). Motivational Interview- ing Treatment Integrity Coding Manual 4.1. Unpublished manual.

�Moyers, T. B., Martin, T., Christopher, P. J., Houck, J. M., Tonigan, J. S., & Amrhein, P. C. (2007). Client language as a mediator of motivational interviewing efficacy: Where is the evidence? Alcoholism: Clinical and Experimental Research, 31(Suppl.), 40s– 47s. http://dx.doi.org/10.1111/ j.1530-0277.2007.00492.x

�Moyers, T. B., Martin, T., Houck, J. M., Christopher, P. J., & Tonigan, J. S. (2009). From in-session behaviors to drinking outcomes: A causal chain for motivational interviewing. Journal of Consulting and Clinical Psychology, 77, 1113–1124. http://dx.doi.org/10.1037/a0017189

�Neighbors, C., Lee, C. M., Atkins, D. C., Lewis, M. A., Kaysen, D., Mittmann, A., . . . Larimer, M. E. (2012). A randomized controlled trial of event-specific prevention strategies for reducing problematic drinking associated with 21st birthday celebrations. Journal of Consulting and Clinical Psychology, 80, 850 – 862. http://dx.doi.org/10.1037/a0029480

Pace, B. T., Tanana, M., Xiao, B., Dembe, A., Soma, C. S., Steyvers, M., . . . Imel, Z. E. (2016). What about the words? Natural language processing in psychotherapy. Psychotherapy Bulletin, 51, 17–18.

R Core Development Team. (2011). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from http://www.R-project.org/

Romano, M., & Peters, L. (2015). Evaluating the mechanisms of change in motivational interviewing in the treatment of mental health problems: A review and meta-analysis. Clinical Psychology Review, 38, 1–12. http:// dx.doi.org/10.1016/j.cpr.2015.02.008

Steyvers, M., & Griffiths, T. (2007). Probabilistic topic models. Handbook of Latent Semantic Analysis, 427, 424 – 440.

Tanana, M., Hallgren, K. A., Imel, Z. E., Atkins, D. C., & Srikumar, V. (2016). A comparison of natural language processing methods for au- tomated coding of motivational interviewing. Journal of Substance Abuse Treatment, 65, 43–50. http://dx.doi.org/10.1016/j.jsat.2016.01 .006

�Thrasher, A. D., Golin, C. E., Earp, J. A. L., Tien, H., Porter, C., & Howie, L. (2006). Motivational interviewing to support antiretroviral therapy adherence: The role of quality counseling. Patient Education and Coun- seling, 62, 64 –71. http://dx.doi.org/10.1016/j.pec.2005.06.003

�Tollison, S. J., Lee, C. M., Neighbors, C., Neil, T. A., Olson, N. D., & Larimer, M. E. (2008). Questions and reflections: The use of motiva- tional interviewing microskills in a peer-led brief alcohol intervention for college students. Behavior Therapy, 39, 183–194. http://dx.doi.org/ 10.1016/j.beth.2007.07.001

�Vader, A. M., Walters, S. T., Prabhu, G. C., Houck, J. M., & Field, C. A. (2010). The language of motivational interviewing and feedback: Coun- selor language, client language, and client drinking outcomes. Psychol- ogy of Addictive Behaviors, 24, 190 –197. http://dx.doi.org/10.1037/ a0018749

Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36, 1– 48. http://dx.doi.org/10 .18637/jss.v036.i03

Viechtbauer, W. (2014). metafor: Meta-Analysis Package for R. R package version 1.9 – 4. Retrieved from http://CRAN.R-project.org/ package�metafor

Wampold, B. E., Mondin, G. W., Moody, M., Stich, F., Benson, K., & Ahn, H. (1997). A meta-analysis of outcome studies comparing bona fide psychotherapies: Empirically, “All must have prizes.” Psychologi- cal Bulletin, 122, 203–215. http://dx.doi.org/10.1037/0033-2909.122.3 .203

Xiao, B., Imel, Z. E., Georgiou, P. G., Atkins, D. C., & Narayanan, S. S. (2015). “Rate My Therapist”: Automated detection of empathy in drug and alcohol counseling via speech and language processing. PLoS ONE, 10, e0143055. http://dx.doi.org/10.1371/journal.pone.0143055

Received November 30, 2016 Revision received March 31, 2017

Accepted March 31, 2017 �

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

533MI META-ANALYSIS