A1285
Title: Variable selection in dynamic item response theory models via Bayes factors with a single MCMC output
Authors: Jingyu Sun - University of Connecticut (China)
Liu Yang - University of Connecticut (China)
Xiaojing Wang - University of Connecticut (United States) [presenting]
Ming-Hui Chen - University of Connecticut (United States)
Abstract: Item response theory (IRT) models are essential analytic tools used in educational testing. The recent surge in computerized testing enables easier collection of longitudinal data in educational studies, but it brings new challenges for the appropriate analysis of educational testing data. To overcome the limitations of classic IRT models and to handle individually varying and irregularly-spaced longitudinal responses, a dynamic IRT model framework has been employed, and the dynamic changes of ability are further linked with individual characteristics in hierarchical modelling. The algorithm developed only needs one single Markov chain Monte Carlo runs to compute all possible Bayes factors for selecting individual characteristics in the proposed model. Further, the model selection consistency of Bayes factors is verified in both theory and simulations when the Zellner-Siow prior is used. In the end, computerized testing has been applied to illustrate the usage of the proposed model and computational strategies.