Log-Linear Item Response Theory Model for Person-by-Item Interactions

• In Event: Advances in Item Response Theory Estimation

Abstract

In this paper, we propose a log-linear item response theory (IRT) model with person-by-item interactions, and use the obtained log-linear model for detecting differential item functioning (DIF). The model is developed by incorporating item properties and person properties and the person-to-item interactions in an IRT model, and deriving the manifest probability in the form of a log-linear model. Pseudolikelihood estimation is used to fit the model and pseudolikelihood tests are proposed for testing the significance of the interaction term (the DIF effect). The effectiveness of the developed model and the pseudolikelihood tests is demonstrated by simulation studies and the application to real data.

Author

• Zhushan Mandy Li, Boston College