Moderators in psychotherapy meta-analysis

Psychotherapy meta-analyses sometimes generate heterogeneous results, partially due to key methodological characteristics which vary between studies (e.g., psychotherapy conditions are contrasted with structurally different control conditions). Examining these potential moderator variables can help explain heterogeneous results within and between psychotherapy meta-analyses. The present manuscript provides an overview of moderators that are highly relevant to test the generalizability of effects across psychotherapy trials. These moderators mainly fall into one of the following groups: (a) structural equivalence of interventions, (b) preferences/allegiances, (c) therapist effects, and (d) sample representativeness. Individual moderators include: Bona fide psychotherapy, proximity to psychological interventions, psychotherapy orientation, pre-training of therapists, supervision, caseload of therapists, dosage, homework, patient preferences, researcher and therapist allegiance, therapist effects in nested designs, aspects of sample representativeness, multiple outcomes, and time of assessment. Our analysis of 15 psychotherapy meta-analyses published in 2016 suggests that the structural equivalence of psychotherapeutic conditions, patient and therapist preferences/allegiances, therapist effects and nested data structures as well as sample representativeness were often neglected and little-discussed as potential moderators. The manuscript describes further conceptual and methodological challenges when conducting moderator analyses such as the categorization of psychological treatments and the importance of interrater coding. We encourage meta-analysts to consider moderators which have previously shown utility in explaining heterogeneous results in the psychotherapy literature. Clinical or methodological significance of this article: Relevant moderator variables help explain heterogeneous results in psychotherapy meta-analyses. Though these variables are often overlooked, they should be regularly incorporated in meta-analyses.