Bayesian Decision Theory for Multi-Category Adaptive Testing

This work presents a method for item selection in adaptive tests based on Bayesian Decision Theory (BDT). Multiple categories of examinee's competence level are assumed. The method determines the probability an examinee belongs to each category using Bayesian statistics. Before starting a test, prior probabilities of an examinee are assumed. Then, each time an examinee responds to a single item, a new competence level is estimated "a-posteriori" using item response and prior probabilities values. A customized focus-of-attention vector of probabilities is estimated, which is used to draw the next item from the Item Bank. The latter vector considers both Personalized Cost and content balancing percentages of items.