The Choice of Item Selection Strategies and Termination Criteria in Grid Multidimensional Computerized Classification Test

• In Event: Modern Psychometric Methodologies and Applications

Abstract

The sequential probability ratio test (SPRT) was the most commonly used termination criterion in computerized classification test (CCT). The measurement efficiency of grid multidimensional CCT can be improved by applying the conditional latent trait distribution adjustment to SPRT (called SPRT_C; Liu et al., 2022). In this study, the adjustment was further applied to other two criteria: generalized likelihood ratio and stochastically curtailed SPRT, while separately combined with two item selection strategies: maximum Fisher information and expected likelihood ratio. The results indicated the generalized likelihood ratio combined with expected likelihood ratio can yield shortest or nearly shortest average test lengths in all conditions, while maintaining comparable PCCs as SPRT. The implications of our findings were elaborated and discussed.

Authors

• BoHsien Hu, National Sun Yat-sen University

• Tian-Hsiang Liu

• Ching-Lin Shih, National Sun Yat-Sen University