Key points are not available for this paper at this time.
Five procedures for estimating a common risk difference in a set of independent 2 × 2 tables were assessed via Monte Carlo simulation in terms of their bias, efficiency, confidence level adjustment, and statistical power. The maximum likelihood estimator showed the best performance, very closely followed by Cochran’s and Mantel-Haenszel’s procedures. The conditional weighted estimator, d CW, showed an irregular performace. The unweighted estimator, d U, showed less efficiency and statistical power than that of the other procedures. As a consequence, the use of the d CW and d U estimators is not recommended. The implications of the results in the practice of meta-analysis are discussed.
Sánchez‐Meca et al. (2001) studied this question.