Differential Item Functioning Simulation Study Based On Mixed Irt Model

Differential item functioning is an important measurement index for evaluating fairness of assessment tool and mixed IRT models provide a new way to assess differential item functioning.But researchers ignore the limitation of DIF numbers,which contradict with the evidences of empirical studies.This study explored the accuracy of mixed IRT models using computer simulation and manipulating sample size,latent class difference,number of DIF and number of invariant items in contexts where less than 30% items showed DIF.This simulation research evaluated the accuracy of the recovery of classes,estimates of class mean differences,item parameter estimates,and DIF effects using the maximum likelihood estimation method based on EM algorithm.Overall,the results revealed that regardless of uniform DIF or non-uniform DIF,the performance of mixed IRT models was not ideal.Decreasing of sample size and DIF items was associated with more biased estimates and larger standard errors.But different trait distribution for two classes resulted in better model estimation.The limitation of the DIF item numbers between focus group and reference group may be the main cause of the poor model estimation.