Bias Reduction of Abilities for Adaptive Online IRT Testing Systems

Abstract

To evaluate the ability of examinees, the item response theory (IRT) gives us useful information. The IRT evaluates the examinee's ability and provides the item characteristics. Recently, we have configured the adaptive online testing using the IRT as one of the CBT (Computer Based Testing) methods, and we have been able to obtain the examinee's ability rating in short testing time. However, when the number of items is too small, we observed the bias of estimates for the ability parameter. In this study, we have performed simulation studies under various conditions, and have known such biases. To circumvent the biases due to the Bayes procedure, we have proposes a simple method to reduce the biases.