A0767
Title: Bayesian item response theory models with local dependence
Authors: Yu-Wei Chang - National Chengchi University (Taiwan) [presenting]
Abstract: Based on the responses to several items by each of the subjects, item response theory (IRT) models aim to make inferences on latent factors, item discrimination, item difficulties, and so on. Local independence is a model assumption required to form the likelihood function in classical and neat IRT models. However, the assumption has been found to be violated in many real applications, such as testlet items, mixture models, and time-limit tests. Therefore, local dependence (LD) modelings have gained much attention in recent years. The focus is on the LD model of a prior study, where they incorporated the Ising model into IRT models to capture LD among items and proposed some frequentist estimation procedures. In order to further allow those LD parameters to borrow information from each other, some Bayesian inference is proposed for the LD model. The main challenge of statistical inference is the intractable posterior distributions. The issue is solved using the pseudo-likelihood methods. Simulation studies are conducted to demonstrate the validation of the proposed estimation procedure under various conditions.
