Testing Categorical Moderators in Mixed-Effects Meta-analysis in the Presence of Heteroscedasticity

AbstractMixed-effects models can be used to examine the association between a categorical moderator and the magnitude of the effect size. Two approaches are available to estimate the residual between-studies variance, τres2—namely, separate estimation within each category of the moderator versus pooled estimation across all categories. We examine, by means of a Monte Carlo simulation study, both approaches for τres2 estimation in combination with two methods, the Wald-type χ2 and F tests, to test the statistical significance of the moderator. Results suggest that the F test using a pooled estimate of τres2 across categories is the best option in most conditions, although the F test using separate estimates of τres2 is preferable if the residual heterogeneity variances are heteroscedastic.