Collusion among Accounting Students: Data Visualization and Topic Modeling of Student Interviews

Abstract

Although classroom cheating violates academic standards of behavior, it occurs frequently. Although the research on cheating is extensive, few researchers have interviewed students directly involved in cheating behaviors. We explore interview responses gathered from a cohort of graduate accounting students, some of whom colluded on an assignment, whereas others did not. We use Latent Dirichlet Allocation (LDA), a powerful text mining algorithm, as our primary tool to explore the underlying topical structure of the interviews and to demarcate subtle differences among students’ reactions to and explanations of their experience. Because LDA does not impose or require a priori theories, we use it to provide ideas for future research rather than to test extant theories about classroom collusion. We identify five primary topics that emerged from the accounting students’ reflections: (1) general course context (including honor code), (2) the rigor of the assignment, (3) student teams as support mechanisms, (4) the perceived repercussions of cheating (colluding), and (5) personality differences between the tax and audit track students. We find subtle language differences between colluders and noncolluders. Colluders considered the nature of the assignment and the difference between tax and audit majors more significant than noncolluders did. Additionally, the role of teams and the general institutional context were somewhat less relevant for colluders than for noncolluders. We conclude by exploring ethical and pedagogical implications of structuring courses as heavily team based for teaching and future research purposes.

Nachnutzung von Forschungsdaten für qualitative Forschungen

“…Daher sind computationelle Verarbeitungen von Interviewtranskripten, die nicht an das spezifische Korpus angepasst sind, problematisch, wann immer sie über einzelne Begriffe hinaus gehen. Dennoch wurden digitale Methoden bereits in einzelnen Studien erfolgreich angewendet (Sherin 2013;Karlgren et al 2020;Shrader et al 2021). Es ist also wichtig, dass bei der Anwendung computationeller Verfahren gewisse Besonderheiten berücksichtigt werden.…”