Aberrant Response Patterns as a Multidimensional Phenomenon: Using Factor-Analytic Model Comparison to Detect Cheating

Abstract

This dissertation proposes a new factor-analytic technique for detecting cheating on exams. Person-fit statistics have been developed to assess the extent to which examinees' response patterns are consistent with expectation, with expectation defined in the context of some model. Response patterns that are inconsistent with expectation are said to be aberrant. Many person-fit statistics have been developed, mostly in the context of classical test theory or item response theory. However, in the person-fit literature, most of these techniques rely on assessing person-fit for unidimensional measurement models. This dissertation proposes that cheating can be conceptualized as a multidimensional phenomenon. A new person-fit technique that involves comparing changes in person-fit across one-factor- and two-factor exploratory factor analysis models is investigated. A statistically-significant improvement in person-fit when adding a second factor to the model is taken as evidence of cheating. Results indicate that this new technique may be useful for detecting cheating when a small-to-moderate proportion of examinees are cheaters. Suggestions are offered for future research on this new technique.