The Q test is regularly used in meta-analysis to examine variation in effect sizes. However, the assumptions of Q are unlikely to be satisfied in practice prompting methodological researchers to conduct computer simulation studies examining its statistical properties. Narrative summaries of this literature are available but a quantitative synthesis of study findings for using the Q test has not appeared. We quantitatively synthesized estimated Type I error rates and power values of a sample of computer simulation studies of the Q test. The results suggest that Q should not be used for standardized mean difference effect sizes like Hedges’ g unless the number of studies and primary study sample sizes are at least 40. Use of the Fisher’s r-to-z transformed effect size, on the other hand, resulted in Q performing well in almost all conditions studied. We summarize our findings in a table that provides guidelines for using this important test.
Maeda, Yukiko and Harwell, Michael R.
"Guidelines for Using the Q Test in Meta-Analysis,"
Mid-Western Educational Researcher: Vol. 28:
1, Article 4.
Available at: https://scholarworks.bgsu.edu/mwer/vol28/iss1/4