The MIMIC Model as a Method for Detecting DIF: Comparison With Mantel-Haenszel, SIBTEST, and the IRT Likelihood Ratio

Abstract

This study compares the ability of the multiple indicators, multiple causes (MIMIC) confirmatory factor analysis model to correctly identify cases of differential item functioning (DIF) with more established methods. Although the MIMIC model might have application in identifying DIF for multiple grouping variables, there has been little examination of how well the technique works in terms of correct and incorrect identification of DIF. A Monte Carlo methodology is used in this study, with manipulation of the number of items, number of examinees, differences between the mean abilities of the reference and focal groups, level of DIF contamination of the anchor items, and amount of DIF in the target item. Results indicate that the MIMIC model is effective for DIF identification for 50 items or when the two-parameter logistic model underlies the data but has a very high rate of incorrect DIF identification for 20 items with three-parameter logistic data.