An Empirical Assessment of the Mantel-Haenszel for Studying Differential Item Performance

Abstract
The Mantel-Haenszel statistic has been receiving considerable attention lately as a technique for assessing differential item performance. An empirical study was carried out to determine the effect of the number of score groups and the inclusion or exclusion of the studied item in forming score groups on estimating as. In both White-Black and White-Hispanic comparisons, four or more score groups appear to provide stable a estimates for a 40-item vocabulary test. The inclusion of the studied item seems to result in fewer items with significant chi squares than the exclusion of the studied item in forming score groups.