IRT Residual-Based Approach to Detecting Item Parameter Drift in CAT

This study proposes a residual-based item parameter drift (RIPD) detection framework for computerized adaptive testing (CAT). Introducing three statistics (RIPD R , RIPD S , and RIPD RS ), the framework detects drift by comparing focal group responses to synthetic drift-free reference responses generated without item recalibration. Three simulation studies evaluated the RIPD framework under curriculum-related drift and item compromise scenarios. Results indicated that RIPD R and RIPD RS consistently outperformed the pseudo-count D 2 method, demonstrating robust false positive control and detection power, particularly with sufficiently large reference groups and a longer test length. While RIPD S was critical for detecting nonuniform drift, it exhibited inflated false positive rates under severe contamination. Overall, the findings support RIPD as a practical and scalable method for post-administration IPD detection in CAT.