Abstract
Computerized adaptive testing (CAT) cannot only enable efficient and precise ability estimation but also increase security of testing materials. To meet a large number of statistical and nonstatistical constraints in CAT, the maximum priority index approaches can be used to handle these constraints simultaneously and efficiently for the construction of assessments. Many previous CAT studies were investigated for dominance items; however, only few CAT studies were investigated for unfolding items. In practice, an attitude measurement or personality test, such as the Minnesota Multiphasic Personality Inventory-2 or Cattell’s 16 Personality Factors Test, might fit better with the unfolding models than with the dominance ones. Besides, these tests commonly have hundreds of items from complex structures. Therefore, the purpose of this study was to investigate constraint-weighted item selection procedures in unfolding CAT. 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 unfolding CAT. These three item information criteria had similar performance in terms of measurement precision, exposure control, and constraint management. The generalized graded unfolding model and the two-parameter logistic model had similar performance in item selection.