Testing moderators

When heterogeneity is non-negligible, the next step is to test moderators (or predictors): factors assumed to affect the magnitude of the effect sizes across the studies in which they are present. While the overall effect size answers the review's main research question, moderator analyses explain why some studies found the effect and others didn't.

• Subgroup analysis, for categorical moderators: a meta-analysis is run separately for each level (a good practice is at least three studies per level), and the levels are then compared for significant differences.
• Meta-regression, for continuous (or dummy-coded categorical) moderators: conceptually similar to a regression analysis run within a primary study.

Moderators worth testing are usually already coded during the coding step — study or sample characteristics such as the geographic region a study was conducted in, the mean age of participants, publication year, or the quality assessment itself. A study can contribute to more than one moderator analysis at once, since each tests a different candidate explanation for the same heterogeneity.

The result of this step is, for each moderator tested, whether it significantly explains part of the heterogeneity in effect sizes — narrowing down which study or sample characteristics actually drive the differences observed.