The performance of the Mantel-Haenszel and logistic regression DIF identification procedures with real data.

Abstract

Numerous statistical methods have been proposed for detecting differential item functioning (DIF). Among them, methods based on item response theory (IRT) are theoretically preferred but very complicated and expensive to implement. As an alternative, the Mantel-Haenszel (MH) procedure has emerged as one of the most popular procedures because of its ease of implementation, relatively small sample size requirement, and associated test of significance. In addition, it provides a measure of the amount and direction of DIF. However, the MH procedure is not designed for and therefore not very effective in detecting nonuniform DIF. As an extension of the MH procedure, a more general DIF detection method, a logistic regression procedure (LR) has been shown to be powerful in detecting both uniform and nonuniform DIF. The purpose of this study is to examine the consistency of the MH and LR procedures and their agreement in the identification of DIF across sample size and criterion when using real examinee data. (Abstract shortened by UMI.)