Calibrating pretest items in multistage adaptive testing (MST) remains challenging, particularly under conditions of sparse response data, limited item exposure, and the need to preserve scale stability. To address these challenges, this study proposes a Simulation-Extrapolation Bayesian Logistic Regression (SBLR) approach and evaluates its performance relative to established methods, including test characteristic curve, information-weighted characteristic curve, fixed item parameter calibration, and Bayesian logistic regression. Using simulated MST data under varying ability distributions and sample sizes, performance was assessed using multiple item parameter recovery indices. Results indicate that SBLR – implemented via maximum a posteriori estimation with informative priors – consistently achieves higher precision and lower bias than alternative methods, particularly under sparse data conditions. These findings suggest that SBLR provides a robust and efficient solution for MST programs by reducing scale drift, improving calibration accuracy, and supporting long-term score comparability. Implications for future research and operational practice are discussed.
TsungHan Ho (Tue,) studied this question.