Interactive DIF detection by HLM: does interacted DIF matter?

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Differential Item Functioning (DIF) is a test bias that describes the phenomena which measures the same performance level differently because they are from different groups. A fair test should eliminate items that contain DIF. Traditional DIF detections analyze DIF between two groups, ignoring the possible interaction of the factors. In this study, a 3-level HLM model is used to determine if the existence of interactive DIF factor effects the regular DIF detection. Results show that the interactive DIF does not influence the regular DIF detection, regardless of the coefficient of interactive DIF, the overall DIF coefficient of the interacted subgroup, and the sample size. The finding indicates that researchers are not concerned about the interactions between factors, and the regular Rasch HLM DIF detection is robust.