Enhancing Test-Taker Experience in Computerised Adaptive Testing: A Comparison of Constraint Methods for Regulating Item Difficulty Jumps

Computerised Adaptive Testing (CAT) leverages Item Response Theory (IRT) to provide efficient, individualised assessments. However, traditional CAT algorithms often produce abrupt "item difficulty jumps", sudden increases in difficulty between consecutive items, which can trigger test anxiety and frustration. This study proposes four methods to regulate these jumps: Fixed Start Item Design (FS), Constrained Item Difficulty Jumps (CDJ), Constraint on Difficulties Lower than Ability (CDLA), and Weighted Fisher Information (WFI).Through two simulation studies, one theoretical and one using empirical data from 13,701 students, the methods were evaluated for measurement accuracy, efficiency, and difficulty stability. Results indicate that while item difficulty converges in 1PL models, it frequently exhibits large, persistent jumps in 2PL models under traditional selection criteria. Among the proposed methods, CDJ effectively restricted jumps within a 0.5 logit threshold without significantly sacrificing measurement precision or increasing test length. CDLA demonstrated high efficiency when item banks contained a surplus of easy items, making it suitable for lower-ability or vulnerable populations. Conversely, WFI provided the strictest jump control but at a substantial cost to measurement efficiency at extreme ability levels. FS performed poorly, particularly for high-ability test-takers. The study concludes that CDJ is recommended as a robust default for 2PL-based assessments, while CDLA is preferable for item banks with predominantly easy content. Future research should experimentally validate the psychological impact of specific logit-scale difficulty jumps.