A hierarchical IRT model for identifying group-level aberrant growth to detect cheating

Abstract

As cheating on high-stakes tests continues to threaten the validity of score interpretations, approaches for detecting cheating proliferate. Most research focuses on individual scores, but recent events show group-level cheating is also occurring. The present Bayesian IRT simulation study extends the Bayesian Hierarchical Linear Model (BHLM) for detecting group-level aberrance. This preliminary study shows that the model reliably recovers individual ability as well as group-level increases. This model provides a valuable way for testing programs to analyze and detect potential cheating behaviors at the instructor, proctor, or administrator levels.

Publication
Conference on Test Security
Date