Maximum Information per Time Unit Designs for Calibrating Multidimensional Items Online

In multidimensional computerized adaptive testing (MCAT), the existing online calibration designs do not account for the time spent on items, potentially resulting in reduced calibration efficiency. To improve the calibration efficiency for MCAT, response time was used as auxiliary information for the online calibration of items. Drawing upon the D-optimal design for MCAT (M-D design) and A-optimal design for MCAT (M-A design), this study introduced the D-optimal per expectation time unit design (M-D-ET design) and A-optimal per expectation time unit design (M-A-ET design) for MCAT. These two new designs prioritized the selection of new items that yield the greatest increment in calibration information per time unit for examinees. Simulation studies were conducted to compare the performances of the M-D, M-A, M-D-ET, and M-A-ET designs across various scenarios. The results indicate that, prior to the calibration of approximately 70% of the new items, the M-D-ET/M-A-ET designs outperformed the M-D/M-A designs in terms of calibration information per time unit and average response time. In the continuous testing scenario, the M-D-ET/M-A-ET designs outperformed the M-D/M-A designs by generating more information per time unit, reducing response time, and maintaining approximately equivalent calibration accuracy. A study based on real response data yielded comparable findings.