AN UNSIGNED MANTEL-HAENSZEL STATISTIC

This paper introduces a new method for detecting differential item functioning (DIF), the unsigned Mantel-Haenszel (UMH) statistic, and compares this method with two other chi-square methods, the Mantel-Haenszel (MH) and the absolute mean deviation (AMD) statistics, in terms of power and agreement between expected and actual false positive rates. Three hundred datasets included items with uniform DIF; another 300 datasets included items with nonuniform DIF; and the other 300 datasets included items with both uniform and nonuniform DIF. All methods produced higher false positive rates than the theoretically expected false positive rates after application of a purification procedure. The second step of the purification procedure produced more false positives for the MH and UMH methods than the first step, but it reduced false positives for the AMID method. The two-step purification procedure also reduced power in most conditions for all three methods. (Contains 11 tables and 30 references.) (SLD) * Reproductions supplied by EDRS are the best that can be made * * from the original document.