THE SMALL SAMPLE PERFORMANCE OF 4 TESTS OF THE DIFFERENCE BETWEEN PAIRS OF META-ANALYTICALLY DERIVED EFFECT SIZES

Four tests for the difference between pooled estimators of effect size from separate meta-analyses are discussed, and small sample performance compared via Monte Carlo simulation. In term of Type I error and power, a z-test based on the difference between pooled estimators appears most accurate, while a confidence interval approach modified to permit testing between two mean effect sizes and Hedges and Olkin's (1985) chi-square test with one degree of freedom appear roughly equivalent, and each of these is superior to a z-test based on an r-transformation. The modified confidence interval approach has particular promise for graphical examination of presumed moderators in meta-analyses.