A Mixture Rasch Model-Based Computerized Adaptive Test for Latent Class Identification
This paper applied computerized adaptive test algorithm for mixture model with two latent classes from two (large vs. small) ability distributions and two item pools of large or small difficulty difference. Four KL methods were conducted and three KL designs were carry out. The results illustrated that four methods shown trivial difference for four pools. Moreover, it was found that classification could be highly improved when difficulty difference is large. The performance on reversed KL had no difference from not reversed one.
Comments and Questions:
1) Results indicated that larger the difference of difficulty of pool, the larger the accuracy the classification would be. However, once difference on difficulty for the same item but for different examinees, it seems that the item is a DIF item. That means the accuracy could be improved by large DIF effect. That’s quite confused to me.
2) It is supposed that the related KL methods would have better estimated ability than random selection due to the fact that the study is based on no other constrains of content balance and exposure rate control. Why do results of the index of absolute bias show the opposite expectation?
3) It may worthy to consider mixture model in computerized classification test. To investigate the relationship between class classification and final categorical classification.