A new method for cheating detection during computerized adaptive testing

In the field of educational and psychological measurement, computerized adaptive testing (CAT) is flexible and convenient, but its reliance on repeatedly administered, pre-calibrated items makes it vulnerable to item exposure and pre-knowledge. We propose a method called CHeater Identification using Interim Person fit Statistic (CHIPS) and a slight modification of it, called Modified CHIPS (M-CHIPS), both designed to identify and limit cheaters during test administration. The methodological novelty lies in redefining a likelihood-based person-fit statistic for response times so that it becomes computable at each adaptive step. CHIPS replaces parameters that traditionally require full-test MCMC estimation with interim maximum-likelihood estimators of speed and expected log-response times, yielding a statistic (IPS) with an analytically tractable asymptotic $${{\upchi}}^{2}$$ distribution. This allows the IPS to be embedded as a constraint within the Shadow Test Approach, producing a dynamic item-selection algorithm that switches between databases based on real-time evidence of item pre-knowledge. M-CHIPS further introduces an early-stage speed-based intervention to improve detectability under extreme cheating scenarios. A simulation study evaluates estimation accuracy, error rates, and computational performance under varying pre-knowledge levels, ability–speed correlations, and test-length settings. Results show that the proposed methods substantially improve ability estimation for cheaters without affecting non-cheaters, demonstrating the statistical and algorithmic effectiveness of incorporating interim fit statistics into adaptive testing.