The Influence of the Prior Distribution on Ability Estimates in Bayesian Item Response Theory

Dr. Matthew Zeigenfuse

Bayesian Item Response Theory (IRT) models are a common statistical tool for measuring ability by means of psychological tests. Bayesian IRT models incorporate prior information about the distribution of abilities, allowing them to be used in situations where it is difficult or impossible to fit IRT models with standard estimation approaches. The price of this flexibility is that ability estimates in Bayesian IRT models depend upon one's choice of prior. In some situations, the influence of one's choice of prior can be substantial. This project will investigate using simulation how the choice of prior distribution affects ability estimates for different test lengths and numbers of individuals. Ideally, the student will have taken our Master's seminars offered in IRT and Bayesian statistics. The topic will be supervised mainly in English and it is recommended that the thesis be written in English.