Detection of latent differential item functioning (DIF) using mixture 2PL IRT model with covariate
Ya Zhang · D-Scholarship@Pitt (University of Pittsburgh) · 2017
Mixture IRT models have been shown to improve the identification of latent group structure and facilitate the estimation of model parameters when covariates are incorporated or the Bayesian estimation method is employed. However, the efficiency of mixture IRT models in DIF analysis has not been systematically studied due to the challenges of identifying DIF with a relatively complex model. The present dissertation aims to explore the effect of covariate and estimation method on the detection of latent DIF under the mixture IRT framework. A Monte Carlo simulation study was performed by manipulating the magnitude of DIF, type of DIF, proportion of DIF items, group impact, and relationship between the covariate and the latent group membership. The generated response data were analyzed using the mixture 2PL IRT model by manipulating the inclusion of covariate and the estimation method. The estimation results were evaluated in terms of the recovery of the latent group structure, recovery of the model parameters, and detection of DIF. The goal is to provide insights and suggestions on the use of mixture IRT models in the analysis of DIF.