Moderator Search in Meta-Analysis: A Review and Cautionary Note on Existing Approaches

Abstract

Meta-analytic searches for moderators or boundary conditions of relationships between variables have increased over the years. Multiple regression and subgroup analysis are the two most common strategies used to search for moderators in a meta-analysis. This short note develops the argument that estimates of the moderator effect may vary depending on the approach (regression or subgroup) chosen by the meta-analyst. The difference results from the ambiguity in assigning the variance shared by correlated moderators when that shared variance is also shared with the effect size. Formulae are presented to estimate the magnitude of the difference, and illustrative examples of tables for different levels of moderator effects and moderator intercorrelations are provided.