A NEW ITEM RESPONSE MODEL AS A MIXTURE OF TWO DIFFERENT PROCESSES
Shigemasu, K. and Hashimoto, T., The University of Tokyo, Japan
Sometimes subjects make correct responses by chance, but at the same time, they make careless mistakes for some test items, even when they have a skill and knowledge good enough to solve them. The accidental correct responses can be handled by one traditional item response model, namely, a three parameter logistic model. The model proposed in this study handles the accidental and careless mistakes by assuming two different psychological processes. Usually, the subjects solve test items carefully making use of their best knowledge and skill, but sometimes they fail to do so by chance. The first process is modeled by a three parameter logistic model and the second process is modeled by a uniform distribution over an entire interval of capacity. The proposed model assesses the plausibility of two different models by posterior probability. The probability of correct response is predicted by a mixture of two models weighted by the posterior probabilities. By stimulated data, we show the effectiveness of the proposed model.