Online Calibration Designs of Multidimensional Computerized Adaptive Testing Using Kullback-Leibler Information

Fisher information depends heavily on provisional parameter estimates, which limits its efficiency in the early calibration stage.To address this limitation, this study introduces Kullback-Leibler (KL) information into online calibration for multidimensional computerized adaptive testing (MCAT) and proposes four KL-Downloaded from https://www.cambridge.org/core.03 Nov 2026 at 19:15:31, subject to the Cambridge Core terms of use.information-based designs: M-KI, M-PKI, M-SKI1, and M-SKI2, where M-PKI is a posterior-weighted extension of M-KI, and M-SKI1 and M-SKI2 are simplified approximations thereof.Simulation results showed that all four designs outperformed the Fisher-information-based M-D design in the early calibration stage, and that M-PKI further achieved the best final calibration precision, particularly in recovering the discrimination parameters.The simplified M-SKI1 and M-SKI2 substantially reduced computation time while preserving the early-stage advantage.In a semi-real data analysis, a hybrid design (M-SKI1&D), which applies M-SKI1 in the early stage before switching to M-D, yielded the best overall parameter recovery.Together, these findings establish KL-information-based designs as a promising alternative for MCAT online calibration, particularly when early-stage accuracy and the precise recovery of discrimination parameters are the primary concerns.