Bayesian Analysis of IRT (Item Response Theory) Parameters and Amount of Information

• In Event: Roundtable Session 117
  In Roundtable Session: Item Response Theory

Abstract

This analytical study is conducted following a Bayesian approach to Item Response Theory (IRT), a measurement tool widely used to synthesize observations of students’ performances in an assessment. Markov Chain Monte Carlo (MCMC) estimation was employed using the WinBUGS computer program. The same data set was analyzed under several conditions, namely, the number of observations, strength of prior distributions, chain length and sampling procedures in MCMC estimation. The results showed that prior distributions influenced posterior means. The influence, however, depends on the strength of belief, likelihood, and the target IRT parameter. Further, when prior was diffused and data set was small, WinBUGS aborted the estimation of a-parameter after 350 cycles.

Authors

• Taslima Rahman, University of Maryland - College Park

• Robert J. Mislevy, University of Maryland