ON THE EFFICACY OF CREDIBILITY INTERVALS AS INDICATORS OF MODERATOR EFFECTS IN METAANALYTIC RESEARCH

Researchers have suggested the use of credibility intervals for identifying moderator effects in meta-analysis. In the present study, the accuracy of decisions derived from credibility intervals was determined from Monte Carlo generated correlations based on two series of meta-analyses, one containing 20 studies and the other, 100 studies. Within each study, the responses of 100 subjects were simulated. Although in some marginal cases the accuracy measures (Type I error rate and power) from the credibility intervals were better than those derived from the SH-75 per cent and U rules, the former technique was generally found to be a poor indicator of heterogeneity or homogeneity. Finally, the concept of a large interval was discussed and a rule of thumb for using it to detect moderators was suggested.