Introduction

Answer-copying and similarity statistics have been studied by researchers for several years. These statistics have been used both to confirm allegations of test-taker cheating and to detect potentital situations of test-taker cheating. Because test results with identical answers ("identical tests") are a special case of situations involving potential answer-copying or collusion, these have not been studied by researchers. However, a study of identical tests is important because these data are symptomatic of large-scale test security breaches which are not detectable by answer-copying and similarity statistics. The presented research demonstrates one approach for obtaining credible evidence that test security has been breached from the analysis of identical tests.

Objectives

This paper demonstrates probability computations that can determine whether a group of identical tests is anomalous or within expected variation.

Design/Methodology

Methodology is demonstrated for determining whether an anomalous group of identical tests is symptomatic of test fraud or exam content disclosure. Patterns of exam content disclosure are shown using scatter plots. After recognizing patterns of exam content disclosure, Bayesian classification analysis is presented for estimating probabilities and incident rates of individuals using disclosed exam content.

Results

The techniques in this paper are demonstrated using live data where test security breaches have been detected and confirmed using this methodology. The techniques also are demonstrated using simulated data.

Conclusions

The results of the paper are summarized to indicate when the presented technique could be useful for test security practitioners, measurement professionals, and testing program managers.