Bayesian Change Point Detection for Aberrant Behaviors in Computerized Testing

• In Event: Item Response Theory With Complex Data

Abstract

Changepoints are abrupt variations in a sequence of data in statistical inference. In educational and psychological assessments, it is pivotal to properly differentiate examinees aberrant behavior from solution behavior to ensure the test validity. In this paper, we propose a Bayesian changepoint detection algorithm to monitor the location of changepoint in the real time, meanwhile, further identify aberrant behaviors using responses and response times. Two simulation studies were conducted to investigate the performance of the proposed approach in which rapid guessing behavior, cheating behavior, warming-up effect and mixed behavior phenomenon occur more frequently at individual and group levels, respectively. Simulation results indicate that ability estimations could be improved based on solution behavior after removing the aberrant behaviors.

Author

• Jing Lu, Northeast Normal University