TWO MCMC-METHODS FOR GRADED RESPONSE MODEL WITH MISSING RESPONSES
Jinhui Wang · Journal of Beijing Normal University · 2011
Item Response Theory(IRT) plays an important role in educational measurement.How to estimate item parameters with missing data in IRT is an interesting issue.Zeng Li et al.(2009) proposed two MCMC-methods to solve the problem for 2PL model.This article extends their methods to Graded Response Model(GRM),a polytomous IRT in tests.In simulation study,the results of item parameter estimates of two MCMC-methods were compared with that of Multilog(missing responses are seemed as wrong responses),a widely used software for parameter estimation.The comparison was made individually under different conditions: three types of missing mechanism,two kinds of parameter priors,two sample sizes and three missing data proportions.The study provides a reference for item parameters estimation of GRM in practical analysis.