Computerized classification testing under the higher-order polytomous IRT model

Latent traits in the human sciences may have a hierarchical structure, for example, a second-order latent trait “language proficiency” covers four first-order latent traits: listening, speaking, reading, and writing. The higher-order IRT model has been developed to account for tests measuring hierarchical latent traits. Being an IRT model, computerized classification testing (CCT) and computerized adaptive testing (CAT) under the higher-order IRT model can be developed. In this study, we focused on the development of CCT algorithms. Specifically, item responses on the first-order latent traits were assumed to follow the generalized partial credit model. The current-estimate approach was adopted for item selection and the ability-confidence-interval approach was adopted for classification. The results of simulation showed that the CCT algorithms were efficient; the higher correlation between the first-order latent traits, the more points each item had, the fewer categories for classification, the more accurate and efficient the classification would be.