Small sample properties of the Mantel-Haenszel test

Abstract

SUMMARY For the analysis of combinations of 2×2 contingency tables, Bennett & Kaneshiro (1974) suggested that the use of a continuity correction in the Mantel-Haenszel test is unnecessary for small samples. Their calculations also implied that the significance level of the resulting test is ‘essentially independent’ of the block effect. We demonstrate that such claims are misleading.