The Efficacy of Motivational Interviewing: A Meta-Analysis of Controlled Clinical Trials Brian L. Burke Fort Lewis College Hal Arkowitz and Marisa Menchola University of Arizona A meta-analysis was conducted on controlled clinical trials investigating adaptations of motivational interviewing (AMIs), a promising approach to treating problem behaviors. AMIs were equivalent to other active treatments and yielded moderate effects (from .25 to .57) compared with no treatment and/or placebo for problems involving alcohol, drugs, and diet and exercise. Results did not support the efficacy of AMIs for smoking or HIV-risk behaviors. AMIs showed clinical impact, with 51% improvement rates, a 56% reduction in client drinking, and moderate effect sizes on social impact measures (d 0.47). Potential moderators (comparative dose, AMI format, and problem area) were identified using both homogeneity analyses and exploratory multiple regression. Results are compared with other review results and suggestions for future research are offered. Motivational interviewing is a relatively new and promising therapeutic approach that integrates the relationship-building prin- ciples of humanistic therapy (Rogers, 1951) with more active cognitive��� behavioral strategies targeted to the client���s stage of change (Prochaska, DiClemente, & Norcross, 1992). It has been defined as a client-centered yet directive method for enhancing intrinsic motivation to change by exploring and resolving client ambivalence (Miller & Rollnick, 2002). Since publication of the first edition of the motivational interviewing book (Miller & Rollnick, 1991), empirical research has advanced on approaches related to motivational interviewing for a variety of clinical prob- lems. In the past decade, three exemplary studies that eliminated almost all threats to internal validity (Miller, Benefield, & Toni- gan, 1993 Project MATCH, 1997 Stephens, Roffman, & Curtin, 2000) have provided strong support for the efficacy of these approaches in the areas of alcohol and drug addiction. The present article reviews this research domain meta-analytically, focusing on controlled clinical trials of individually delivered interventions that incorporated the four basic principles of motivational interviewing discussed in turn in the following text: (a) expressing empathy, (b) developing discrepancy, (c) rolling with resistance, and (d) sup- porting self-efficacy (Miller & Rollnick, 2002). Although expressing empathy is fundamental to virtually all psychotherapies, in motivational interviewing it takes the specific form of reflective listening (or accurate empathy) as described by Carl Rogers (1951). Underlying this principle of empathy is a client-centered attitude of acceptance, wherein client ambivalence or reluctance to change is viewed as a normal part of the human experience rather than as pathology or pernicious defensiveness. Developing discrepancy, the second principle of motivational in- terviewing, is where it begins to depart from classic client-centered therapy. A key goal in motivational interviewing is to increase the importance of change from the client���s perspective. This is accom- plished using specific types of questions, along with selective reflections, that direct the client toward the discrepancy between his or her problem behavior and broader personal values. Although motivational interviewing is intentionally directive, the therapist is careful not to explicitly advocate for change it is the client who presents the reasons for change. Accordingly, when a client expresses resistance to change, it is a signal for the interviewer to respond differently. Resistance is conceptualized as an interpersonal variable, and the third basic principle of motiva- tional interviewing is not to oppose the client���s resistance actively but rather to accept and flow with it, again using reflective listen- ing skills. Finally, a client���s readiness for change is hypothesized to stem from two main factors: the importance of the change for the client (as discussed above) and the confidence the client has about successfully making the change. This confidence, often termed self-efficacy, is an essential element in motivation and a good predictor of treatment outcome (Bandura, 1997). The fourth guiding principle of motivational interviewing, therefore, is to enhance the client���s confidence in his or her own capability to cope with obstacles and to succeed in changing. In the research literature, the most widely used approach related to motivational interviewing has been one in which the client (often alcohol or drug addicted) is given feedback based on indi- vidual results from standardized assessment measures, such as the Drinker���s Check-Up (Miller, Sovereign, & Krege, 1988) or a Brian L. Burke, Department of Psychology, Fort Lewis College Hal Arkowitz and Marisa Menchola, Department of Psychology, University of Arizona. This article is based on Brian L. Burke���s dissertation, which was submitted in partial fulfillment of the requirements for a doctoral degree in psychology at the University of Arizona. Special thanks are given to Varda Shoham, Michael Rohrbaugh, Julie Feldman, and Chris Segrin for assis- tance with the article to Bob Rosenthal, Drew Westen, and David Wilson for their generous consultation and to Leslie Goldstein for assistance in every other possible way. Correspondence concerning this article should be addressed to Brian L. Burke, Department of Psychology, Fort Lewis College, Durango, Colorado 81301-3999. E-mail: burke_b@fortlewis.edu Journal of Consulting and Clinical Psychology Copyright 2003 by the American Psychological Association, Inc. 2003, Vol. 71, No. 5, 843��� 861 0022-006X/03/$12.00 DOI: 10.1037/0022-006X.71.5.843 843

modification of it. This feedback, which concerns the client���s level of severity on the target symptom compared with norms, is deliv- ered in a motivational interviewing ���style,��� wherein possibilities for change are elicited from the client in a nonthreatening manner. Discussion of the problem may extend to one or more sessions that continue to embody the fundamental spirit and methods of moti- vational interviewing. We, along with Miller (W. R. Miller, per- sonal communication, March 2001), consider this feedback-based approach to constitute an adaptation of motivational interviewing (AMI) because it is defined by the presence of the feedback component and not solely by the use of motivational interviewing per se. More broadly, we also apply the term AMI to interventions that incorporate additional nonmotivational interviewing tech- niques while retaining motivational interviewing principles as the core of treatment as well as to interventions that have been spe- cifically adapted for use by nonspecialists (Rollnick, Heather, & Bell, 1992). To date, virtually all of the empirical studies in this area (and therefore in this review) have dealt with the efficacy of AMIs, and no studies have addressed the efficacy of motivational interviewing in its relatively pure form. Three previous reviews of approaches related to motivational interviewing have been published. Noonan and Moyers (1997) reviewed the 11 clinical trials of AMIs available at that time (9 with problem drinkers and 2 with drug abusers) and concluded that 9 of these studies supported the efficacy of AMIs for addictive behaviors. Dunn, DeRoo, and Rivara (2001) performed a system- atic review of 29 randomized trials of brief interventions that claimed to use the principles and techniques of motivational in- terviewing (or what we have called AMIs) to change behavior in four areas: (a) substance abuse, (b) smoking, (c) HIV-risk reduc- tion, and (d) diet and exercise. Data on methodological features were tabled, as were calculations of effect sizes and their 95% confidence intervals, although the authors chose not to combine or compare data meta-analytically. The strongest evidence for effi- cacy was found in the alcohol and drug abuse areas, in which AMIs appeared to work well for problem drinkers and improved the rate of entry into and retention in intensive substance abuse treatment. AMI effects did not appear to diminish over time, and the effect sizes for AMIs as preludes to other treatments (e.g., inpatient care) were roughly equivalent to those for AMIs as stand-alone interventions. More recently, Burke, Arkowitz, and Dunn (2002) qualitatively reviewed 26 studies that met their specified inclusion criteria. The authors concluded that the research supported the efficacy of AMIs for alcohol problems, drug addiction, hypertension, and bulimia as well as its efficacy for encouraging compliance in patients with diabetes. Mixed support was found for AMIs in the areas of reducing cigarette smoking, increasing physical activity, and en- hancing dietary adherence in patients with hyperlipidemia. No support was found for AMIs in the reduction of HIV-risk behav- iors (e.g., needle-sharing). In general, the AMIs reviewed were superior to no-treatment control groups and less credible alterna- tive treatments and equal to active comparison treatments. After examining evidence regarding the mechanism of AMIs, Burke et al. (2002) reported that the research literature failed to shed light on how the treatment actually works. For instance, no direct support was found for the idea that AMIs exert their clinical effects by enhancing the client���s motivation to change. In addition, the authors found virtually no data to indicate for whom these treat- ments were optimal, as most clinical trials of AMIs that looked for aptitude by treatment interactions (Shoham & Rohrbaugh, 1995) were unable to find them. Thus, three previous reviews have examined the AMI literature using largely qualitative methodol- ogy, with no quantitative synthesis of this research domain to date. For the current article, we decided to review the AMI literature meta-analytically for three main reasons: (a) traditional methods of reviewing may suffer a considerable loss of power relative to meta-analytic methods, hence inflating the probability of Type II errors (Rosenthal, 1991) (b) meta-analytic reviews are likely to lead to summary statements of greater precision and objectivity (Kaplan, 1964) and (c) several controlled clinical trials of AMIs could be included that were not yet available for inclusion in any of the previous reviews. To our knowledge, the current review is the first meta-analytic examination of the motivational interviewing literature. This meta- analysis is multidimensional (Westen & Morrison, 2001), provid- ing a range of statistics bearing on outcome in addition to the usual effect sizes. Our review has five main objectives. First, we present the basic characteristics (e.g., problem types, settings, treatment lengths, comparison groups) of the controlled clinical trials of AMIs. Second, we evaluate the evidence for the efficacy of AMIs across clinical problem areas compared with control procedures (e.g., no treatment or weak alternatives) and with other active treatments. Third, these studies are examined for evidence of sustained efficacy, the ability of the AMIs to produce lasting symptomatic changes rather than an initial response solely (Westen & Morrison, 2001). Fourth, we explore evidence for the clinical impact of AMIs���the practical value or importance of an intervention to clients or to others with whom clients interact (Kazdin, 1999). Fifth, we focus on the identification of moderator variables, factors associated with variations in the outcome of controlled clinical trials of AMIs that may shed light on why different studies produced different results (Rosenthal & DiMat- teo, 2001). Method Study Selection For this review, we searched through the reference sections of all three prior reviews and the motivational interviewing website (www .motivationalinterview.org). We also conducted a database search (PsycINFO) using motivational interviewing as a key phrase, and finally, we sent out an electronic message to all members of the Motivational Interviewing Network of Trainers asking for any published or unpublished studies relevant to our purposes. This meta-analysis follows the general guidelines commonly used in reviews of the efficacy of various psychotherapies (e.g., Kazdin, 1992). For this reason, studies had to satisfy the following criteria to be included in this review: (a) the intervention under study consisted primarily of imple- menting the motivational interviewing principles discussed above rather than principles of some other approach (such as cognitive��� behavioral therapy), (b) the intervention was delivered on an individual (i.e., not group) and face-to-face (i.e., not telephone) basis, and (c) the study design met our criteria for a controlled clinical trial. In our definition, a controlled clinical trial must use the following: (a) random assignment to groups or an alternative way of equating groups of clients before treatment (e.g., se- quential assignment), (b) at least one comparison group, and (c) adequate measurement targeting pertinent problem areas. Although the controlled clinical trial has recently come under criticism (e.g., Borkovec & Caston- 844 BURKE, ARKOWITZ, AND MENCHOLA

guay, 1998), it remains the gold standard for evaluating treatment outcome (Stanton & Shadish, 1997). Statistical Analyses Our general data analytic approach for this review is outlined in Table 1. The specific strategies and procedures used in implementing this approach are discussed in detail in the following text. What Kind of Controlled Trials Have Been Done With AMIs? The first goal of this meta-analysis was to characterize the types of questions that have been investigated in controlled trials of AMIs. Accord- ingly, we coded descriptive characteristics relating to important method- ological and substantive features of each study. Substantive features in- cluded problem area as well as treatment setting, format, and dose, whereas methodological features included sample size, study design (i.e., type of control and/or comparison groups used), dependent measures used, and follow-up lengths and rates. What Is the Comparative Efficacy of AMIs? For each study reporting sufficient information, we calculated effect sizes and confidence intervals for the main behavioral and health outcomes at all reported follow-up times. When necessary, authors were contacted for group means and standard deviations not reported in the original article (Gentilello et al., 1999 Juarez, �� 2001 Martino, Carroll, O���Malley, & Rounsaville, 2000). All author inquiries yielded useful data for effect size calculations. For each follow-up interval of each treatment comparison involving an AMI, a unit-free effect size, g, was calculated by subtracting the control group mean from the experimental (AMI) group mean and dividing the result by the pooled standard deviation according to the following formula (Hedges & Olkin, 1985, pp. 78 ���79): Effect size g (YE YC)/s and s [(nE 1)(sE)2 (nC 1)(sC)2]/[nE nC 2] (1) where YE and YC are the experimental and control group means posttreat- ment, s is the pooled standard deviation, sE and sC are the experimental and control group standard deviations, and nE and nC are the experimental and control group sample sizes. In all cases, an unbiased estimate of the population effect size was then obtained by correcting for the bias in g (Hedges & Olkin, 1985, p. 81). When means were not available, the effect size was estimated directly from significance tests (t, F, or chi-square) according to the requisite procedures (for more details, see Rosenthal, 1991, pp. 18 ���20). For all effect sizes, 95% confidence intervals were derived from their variance, which was estimated according to the following formula (Hedges & Olkin, 1985, p. 86): 2 d) [(nE nC)/nEnC] [d2/2(nE nC)] (2) where nE and nC are the experimental and control group sample sizes respectively. How Do AMIs Compare With No-Treatment or Placebo Groups and With Other Active Treatments Across Different Problem Areas? Prior to generating combined effect size estimates, effect sizes were grouped into theoretically meaningful subcategories according to the fol- lowing two variables: (a) clinical problem area (alcohol, smoking, drug addiction, HIV-risk behaviors, or diet and exercise) and (b) design type (no-treatment and/or placebo control or active treatment comparison group). A group of independent effect size estimates was then generated within each category (e.g., AMI vs. no-treatment and/or placebo control for alcohol problems). For studies using multiple follow-up points, the first posttreatment effect size was selected. For studies using multiple outcome measures, the effect size associated with the best target measure was selected a priori according to psychometric properties and common usage (e.g., standard drinks/week as a measure of alcohol consumption). When possible, combined effect sizes were also generated separately for second- ary target measures, such as peak blood alcohol concentration for alcohol problems. The combined effect size (dc), was computed by weighting each indi- vidual effect size according to the inverse of its variance. In this way, each study contributed to the combined estimate according to the precision of its own effect size estimates (i.e., studies with larger sample sizes contributed more heavily to the combined effect size). For each combined effect size, 95% confidence intervals were derived from the variance of dc (Hedges & Olkin, 1985, p. 113). Table 1 General Data Analytic Approach for Review Key question Data analytic strategy 1. What kind of controlled trials have been done with adaptations of motivational interviewing (AMIs)? Code descriptive characteristics of all 30 studies 2. What is the comparative efficacy of AMIs? Compute individual effect sizes for each study (a) How do AMIs compare with no-treatment or placebo groups across different problem areas? Compute combined effect sizes for AMIs versus no-treatment/ placebo comparison groups by problem area (b) How do AMIs compare with other active treatment across different problem areas? compute combined effect sizes for AMIs versus active treatment comparison groups by problem area 3. What is the sustained efficacy of AMIs? Compare posttreatment and follow-up effect sizes of AMIs 4. What is the clinical impact of AMIs? For AMI studies, whenever available, combine percentage improved data compute and combine within-AMI effect sizes combine drinking frequency data from alcohol studies compute combined effect sizes for social impact measures 5. What factors might account for any observed differences in effect sizes across these studies? If Q tests indicate significant heterogeneity, subdivide effect size groupings further using categorical moderators code and analyze potential moderators by multiple regression 845 MOTIVATIONAL INTERVIEWING META-ANALYSIS