Johathan Shedler

Page 3 of 8 Previous page Next Page

The Efficacy of Psychodynamic Psychotherapy

Jonathan Shedler
University of Colorado Denver School of Medicine

February–March 2010 American Psychologist
© 2010 American Psychological Association 0003-066X/10/$12.00
Vol. 65, No. 2, 98–109 DOI: 10.1037/a0018378

How Effective Is Psychotherapy in General?

In psychology and in medicine more generally, meta-analysis is a widely accepted method for summarizing and synthesizing the findings of independent studies (Lipsey & Wilson, 2001; Rosenthal, 1991; Rosenthal & DiMatteo, 2001). Meta-analysis makes the results of different studies comparable by converting findings into a common metric, allowing findings to be aggregated or pooled across studies. A widely used metric is effect size, which is the difference between treatment and control groups, expressed in standard deviation units.2 An effect size of 1.0 means that the average treated patient is one standard deviation healthier on the normal distribution or bell curve than the average untreated patient. An effect size of 0.8 is considered a large effect in psychological and medical research, an effect size of 0.5 is considered a moderate effect, and an effect size of 0.2 is considered a small effect (Cohen, 1988).

The first major meta-analysis of psychotherapy outcome studies included 475 studies and yielded an overall effect size (various diagnoses and treatments) of 0.85 for patients who received psychotherapy compared with untreated controls (Smith, Glass, & Miller, 1980). Subsequent meta-analyses have similarly supported the efficacy of psychotherapy. The influential review by Lipsey and Wilson (1993) tabulated results for 18 meta-analyses concerned with general psychotherapy outcomes, which had a median effect size of 0.75. It also tabulated results for 23 metaanalyses concerned with outcomes in CBT and behavior modification, which had a median effect size of 0.62. A meta-analysis by Robinson, Berman, and Neimeyer (1990) summarized the findings of 37 psychotherapy studies concerned specifically with outcomes in the treatment of depression, which had an overall effect size of 0.73. These are relatively large effects. (For a review of psychotherapy efficacy and effectiveness research, see Lambert & Ogles, 2004).

To provide some points of reference, it is instructive to consider effect sizes for antidepressant medications. An analysis of U.S. Food and Drug Adminstration (FDA) databases (published and unpublished studies) reported in the New England Journal of Medicine found effect sizes of 0.26 for fluoxetine (Prozac), 0.26 for sertraline (Zoloft), 0.24 for citalopram (Celexa), 0.31 for escitalopram (Lexapro), and 0.30 for duloxetine (Cymbalta). The overall mean effect size for antidepressant medications approved by the FDA between 1987 and 2004 was 0.31 (Turner, Matthews, Linardatos, Tell, & Rosenthal, 2008).3 A meta-analysis reported in the prestigious Cochrane Library (Moncrieff, Wessely, & Hardy, 2004) found an effect size of 0.17 for tricyclic antidepressants compared with active placebo (an active placebo mimics the side effects of an antidepressant drug but is not itself an antidepressant).4 These are relatively small effects. Methodological differences between medication trials and psychotherapy trials are sufficiently great that effect sizes may not be directly comparable, and the findings should not be interpreted as conclusive evidence that psychotherapy is more effective. Effect sizes for antidepressant medications are reported to provide reference points that will be familiar to many readers (for more comprehensive listings of effect size reference points, see, e.g., Lipsey & Wilson, 1993; Meyer et al., 2001).

2 This score, known as the standardized mean difference, is used to summarize the findings of randomized control trials. More broadly, the concept effect size may refer to any measure that expresses the magnitude of a research finding (Rosenthal & Rosnow, 2008).

3 The measure of effect size in this study was Hedges’ g (Hedges, 1982) rather than Cohen’s d (Cohen, 1988), which is more commonly reported. The two measures are based on slightly different computational formulas, but in this case the choice of formula would have made no difference: “Because of the large sample size (over 12,000), there is no change in going from g to d;bothvaluesare.31totwo decimal places” (R. Rosenthal, personal communication to Marc Diener, January 2008).

4 Although antidepressant trials are intended to be double-blind, the blind is easily penetrated because the adverse side effects of antidepressant medications are physically discernible and widely known. Study participants and their doctors can therefore figure out whether they are receiving medication or placebo, and effects attributed to medication may be inflated by expectancy and demand effects. Use of “active” placebos better protects the blind, and the resulting effect sizes are approximately half as large as those otherwise reported.