The Effects Of Q-matrix Misspecification On Parameter Estimates And Classification Accurancy In RRUM

Based on the reduced reparameterized unified model(rRUM)in the cognitive diagnosis model,this paper discusses the estimation of the model parameters and the accuracy of the respondent classification considering the Q matrix with correct calibration or misspecification.Comparison between the using results on EM estimation method and on MCMC method of Gibbs sampling algorithm is used to enhance the accuracy of the estimation and classification.First,we introduce the rRUM,and two methods that we mentioned respectively.Then the correct calibration of Q matrix is used to judge whether the two estimates can get similar results,and for the Q matrix with misspecification the effects of different misspecification types and different misspecification rates on parameter estimation and classification accuracy are studied.The results show that when the Q matrix is correctly calibrated the two algorithms get the similar conclusion,and the EM algorithm gets slightly better parameter estimation with the number of the attributes increasing and small sample size.When the Q matrix exists underfitting type of misspecification,EM algorithm is invalid,while the MCMC algorithm can still keep good results.When When the Q matrix exists overfitting type of misspecification,both of the algorithms get smaller item parameter estimation,and MCMC algorithm obtains higher classification accuracy.