Testing continuous moderators in meta‐analysis: A comparison of procedures

Using Monte Carlo simulation, we compare the statistical power and Type I error rates of three procedures commonly applied in meta‐analysis to detect the influence of a continuous moderator variable on the estimated effect sizes from a set of independent primary studies. We use the conventional T test (TTEST), the Hedges & Olkin (1985) z HO test (H&O), and the Rosenthal & Rubin (1982) z RR test (R&R). The H&O and R&R procedures weight every estimated effect size by its inverse‐variance, whereas in the conventional T test the individual effect sizes are unweighted. Results show an adequate control of the Type I error rate for the three procedures in all of the conditions. The statistical power is very similar in the H&O and R&R procedures, and systematically lower in the conventional T test. Both H&O and R&R procedures were similarly the most appropriate and powerful to detect a quantitative moderator variable on effect sizes, supporting the suitability of applying weighting schemes in meta‐analysis.