The detection of cheating in multiple choice examinations

TL;DR

This paper evaluates two methods for detecting collusion in multiple-choice exams, using real data and simulations to compare their effectiveness in identifying anomalous answer patterns.

Contribution

It introduces a simulation approach to validate empirical methods for detecting cheating, providing a comparative analysis of statistical techniques.

Findings

Empirical analysis aligns with traditional correlation methods.

Simulation confirms the validity of the empirical approach.

Bonferroni method is less reliable for this purpose.

Abstract

Cheating in examinations is acknowledged by an increasing number of organizations to be widespread. We examine two different approaches to assess their effectiveness at detecting anomalous results, suggestive of collusion, using data taken from a number of multiple-choice examinations organized by the UK Radio Communication Foundation. Analysis of student pair overlaps of correct answers is shown to give results consistent with more orthodox statistical correlations for which confidence limits as opposed to the less familiar "Bonferroni method" can be used. A simulation approach is also developed which confirms the interpretation of the empirical approach.