A Comparison Study Among General-Purpose Bayesian Implementations for Item Response Theory Models

• In Event: Innovative Approaches to Item Response Theory

Abstract

This study compares the point estimates and SEs, convergence rates and running times among two general-purpose Bayesian implementations, Stan and OpenBUGS and some specific-purpose Bayesian programs for the item response theory (IRT) models. To fully illustrate their distinction, four real world datasets with dichotomous and polytomous IRT models were run via three programs. These examples include the four-parameter normal ogive model and the multidimensional 2-parameter logistic model for dichotomous responses, and the generalized partial credit model and the generalized graded unfolding model for polytomous responses. With these examples, we encourage researchers to use Stan to estimate IRT models for its computational efficiency.

Authors

• Chanjin Zheng, East China Normal University

• Shaoyang Guo, East China Normal University