Exploring Hamiltonian Monte Carlo in Estimation of the Four-Parameter Item Response Theory Model

• In Event: Investigations in Measurement and Statistics

Abstract

Recent research interest in the four-parameter IRT model (4PM) has been revived after several decades of neglect, and studies of estimation of the four-parameter IRT model (4PM) has investigated the standard Marginal Maximum Likelihood and Bayesian methods via analytics and Markov chain Monte Carlo (MCMC) with Gibbs sampling. This paper adds to the existing literature on estimation of this relatively new unidimensional model for binary response data with an investigation of the Hamiltonian Monte Carlo (HMC) sampler. Calibration of the 4PL model is performed across three conditions of test length sample size (one, five and ten thousand simulated respondents) by the Stan program via the rstan interface package. At present, literature review for the paper is done and R codes for HMC simulation are being verified. It is anticipated that the simulation will be completed by the time proposal review result is available and the full paper will be finished one month before the conference.

Author

• Hoan Do, Ohio University