Bayesian Estimation of Graded Response Models

• In Event: Item Response Theory: Applications and Extensions

Abstract

Markov chain Monte Carlo (MCMC) methods enable a fully Bayesian approach to parameter estimation of item response models. In this simulation study, we compared the recovery of graded response model parameters using marginalized maximum likelihood (MML) and Bayesian estimation under various latent trait distributions, test lengths, and sample sizes.
Although there was little difference between MML and Bayesian estimation in item parameter recovery in samples with more than 500 respondents, Bayesian estimation recovered item threshold parameters better in samples with fewer than 300 respondents. Person parameters were recovered considerably better by Bayesian estimation for all sample sizes and test lengths.

Authors

• Vincent Kieftenbeld, Southern Illinois University - Edwardsville

• Prathiba Natesan, University of North Texas