Investigation of Constraint-Weighted Item Selection Procedures in Polytomous CAT

To fulfill a large number of statistical and non-statistical constraints in computerized adaptive testing (CAT), the maximum priority index approaches can be used to handle many constraints simultaneously and efficiently for the construction of assessments. Many previous studies in CAT were conducted for dichotomously scored items; however, only few studies were conducted for polytomously scored items. In practice, because polytomous items are more informative, polytomous CAT tends to need fewer items than dichotomous CAT does. Many important issues in polytomous CAT need further attention. Therefore, the purpose of the study was to investigate constraint-weighted item selection procedures in polytomous CAT. The generalized partial credit model (GPCM) was considered in this study. It was found that the maximum priority index was implemented with the Fisher information, the interval information, and the posterior expected Kullback-Leibler information successfully in polytomous CAT. These three item information criteria had similar performance in terms of measurement precision, exposure control, and constraint management.