Empirical evaluation of fully Bayesian information criteria for mixture IRT models using NUTS

Abstract This study is to evaluate the performance of fully Bayesian information criteria, namely, LOO, WAIC and WBIC in terms accuracy determining number latent classes a mixture IRT model while comparing it conventional via non-random walk MCMC algorithms further compare their with criteria including AIC, BIC, CAIC, SABIC, DIC. Monte Carlo simulations were carried out these under different situations. The results indicate that related CAIC SABIC tend select simpler are not recommended when actual data involve multiple classes. For three measures, can be used for detecting tests at least 30 items, LOO suggested together effective parameters choosing correct