DIF Detection with Zero-Inflation under the Mixture Modeling Framework

• In Event: Division D Graduate Student in Progress Research Gala

Abstract

This study aims to compare two approaches in detecting DIF items when the zero¬-inflation occurs: the Wald test and the mixture IRT approach with MIMIC. With zero¬-inflation, the standard DIF detections such as the Wald test, possibly fail to detect items exhibiting DIF, and compromise the test validity. The simulation was conducted, with the manipulation of the following research design factors: sample size, size of DIF effect, and proportions of zero-inflated class. The preliminary simulation identified that the mixture IRT outperformed the standard DIF test when there is domination of zero responses.

Author

• Sooyong Lee, The University of Texas at Austin