Sensitivity of prior specification within testlet model

Abstract

There have been enormous statistical advances made in the analysis of standardized educational and psychological tests. Parallel with this, the practical advantages of the Bayesian approach were recognized in item response theory (IRT) and have been adopted to provide more detailed information about item parameters and an individual's underlying latent ability.

The purpose of this study is to examine the sensitivity of di erent prior distributions within the three-parameter logistic testlet (3PLT) model. First, the e cacy of the 3PLT model in the WinBUGS 1.4 program (Spiegelhalter, Thomas, Best, & Lunn, 2003) was compared to the 3PLT model in the SCORIGHT 3.0 (Wang, Bradlow, & Wainer, 2004) and the Gibbs (Du, 1998) programs, neither of which can manipulate prespeci ed prior distributions. Later, the impacts of di erent prior distributions in the 3PLT model will be discussed.