A Bayesian Robust Item Response Theory Outlier-Detection Model

• In Event: 54.069 - Advances in Rasch Model

Abstract

In the practice of Item Response Theory (IRT) modeling, item response outliers can bias estimates of person ability and item parameters in a standard IRT model. We propose and illustrate a Bayesian IRT model that includes person and latent item-response outlier parameters, in addition to person ability parameters and item parameters. The outlier parameters, not only identify the presence of outliers at the model level, but also helps the model provide robust estimates of the person and item parameters. Hence, under this IRT model, it is not necessary to remove outlying items or persons from the data analysis. We illustrate our model through the analysis of dichotomous item-response data.

Authors

• Nicole K. Ozturk, University of Illinois at Chicago

• George Karabatsos, University of Illinois at Chicago