A Bayesian Beta-Mixture Model for Nonparametric Item Response Theory

• In Event: 58.028 - Applications of Item Response Theory

Abstract

Item response models typically assume that the item characteristic (step) curves follow a logistic or normal cumulative distribution function, which can be overly-restrictive for real item response data sets. The monotone homogeneity Item Response Theory (IRT) model relaxes this assumption by only requiring that these functions be monotone-increasing over the person ability domain. We propose a simple Bayesian model for nonparametric IRT, which constructs monotone item characteristic (step) curves by a finite mixture of beta distributions. We illustrate our IRT model through the analysis of data from a dichotomous item test of reading performance, and the analysis of data from a rating scale questionnaire of teacher preparation.

Authors

• Ethan Arenson

• George Karabatsos, University of Illinois at Chicago