How to interpret correlational process-outcome effect sizes in psychotherapy: a meta-analytic benchmark study.

Abstract

Objective: This study aims at developing empirically driven criteria for correlational effect size interpretations grounded on the actual psychotherapy process-outcome effect size distributions.

Method: We performed a meta-analysis on PubMed and PsycINFO databases searching for meta-analyses reporting correlational process-outcome associations.

Results: Forty-three meta-analyses met inclusion criteria, reporting 1,637 effect sizes from 859 unique studies. A four-level meta-analytic model resulted in an estimated mean effect size of r = .24 (z = .24, 95% CI[.22,.27]), significantly different from Cohen's proposed value for moderate effects (i.e.,.30), z = -.04, SE = 0.01, t(1628) =  -4.17, p < .001). Percentiles derived from the models showed that Cohen's criteria were too conservative, with the 25th percentile = .12, 50^th percentile = .26, and 75^th percentile = .39. Based on these findings, we suggest the benchmarks .10, .25, and .40, for small, moderate, and large effect sizes. Even when using these less restrictive criteria, the majority of the correlation analyses from primary studies (81.8%) were underpowered to identify at least a moderate effect size.

Conclusion: The current findings might help to enhance effect size interpretations and power calculations in psychotherapy process-outcome research. Further replications are necessary to extend these benchmarks to other areas of clinical psychology.