Testing the significance of a common risk difference in meta-analysis. (Q1575216)

Using the Monte Carlo simulation, we estimated the statistical power and Type I error rates of five procedures for testing the significance of a common risk difference in a set of independent \(2\times 2\) tables. It was found that the unweighted procedure for testing the significance of a common risk difference showed Type I error rates systematically larger than the nominal significance level, and that its power was lower than that of the other procedures. The conditional weighted procedure showed the worst performance, with remarkably anomalous results under many of the conditions. Cochran's, Mantel-Haenszel's, and Yusuf's unconditional weighted procedures showed very similar results, with the best performance in both Type I error values and power values.