Differential item functioning in diagnostic classification models

Assessment of differential item functioning (DIF) in diagnostic classification models (DCMs) has begun to attract research attention. In previous studies, authors found that DIF detection in DCMs appeared to be very powerful even when most or all the items on the studied test had DIF and no scale purification was necessary. This surprisingly good result was built on studies that made the unrealistic assumption of equality of the model and the Q-matrix across groups. The present study clarifies these weaknesses in previous studies, identifies various types of DIF, and proposes new DIF detection methods that are powerful in detecting DIF in DCMs. An illustrative simulation study was conducted to demonstrate the feasibility and advantages of the new methods. Finally, conclusions and suggestions for future studies are provided.