Effects of Unequal Ability Variances on the Performance of Logistic Regression, Mantel-Haenszel, SIBTEST IRT, and IRT Likelihood Ratio for DIF Detection.

The article presents a study which evaluated the effects of unequal ability variances between reference and focal groups on the performance of the four commonly used differential item functioning (DIF) detection methods. Detection methods include logistic regression modeling, the Mantel-Haenszel chi-square test (MH test), simultaneous item bias test (SIBTEST) and likelihood ratio test. The importance of the DIF items in test fairness and equity in large-scale assessments is stressed which may threaten the validity of test scores for subgroups. Simulated data are used in manipulating factors, including sample size, mean difference in ability and difference in ability variance, and dichotomous item responses were generated for each item and examined.