A Comparison of Six Differential Item Functioning Detection Methods

• In Event: Poster Session for Quantitative Methods and Statistical Theory

Abstract

This study compared the effectiveness of six DIF detection methods: The Mantel-Haenszel procedure, the Logistic Regression, the MIMIC model, the IRT-LR test, Lord’s Wald test and a Randomization Test. A simulation study was conducted in which the factors manipulated were the percentage of DIF items, sample size, test length, DIF type and magnitude, and the mean ability difference between groups. Results showed that the MIMIC model had the greatest power with longer tests. The logistic regression and randomization test efficiently detect uniform DIF, but the randomization test only applies when the two testing groups have identical mean ability. The IRT methods are better at detecting nonuniform DIF. Most methods are better with large magnitude DIF and large sample sizes.

Author

• Xueying Gao, University of Central Florida