An Investigation of Answer Changing on a Large-Scale Computer-Based Educational Assessment

Abstract

Answer changing on tests has been studied for decades, however more recently answer-changing analysis has risen as an approach for exploring potential test fraud on high-stakes achievement tests. The purpose of this study was to document answer-changing patterns of students grades 3-11 on computer-based English language arts and mathematics mandated state achievement tests. Frequencies and distributions of answer-changing patterns as well as response times were reported. Additionally, relationships between student demographic characteristics and answer-changing variables were modeled using Poisson and negative binomial regression approaches. Results were consistent with prior research that has indicated large numbers of answer changes are rare occurrences that could warrant further exploration. Negative binomial regression was a better approach than Poisson regression due to overdispersion in the Poisson models. Student demographic variables were not useful in explaining answer-changing behaviors, for any of the independent variables examined. Results also add to the field’s understanding of answer-changing and response-time behaviors as constructs, as well as their utility as statistical means of detecting unusual patterns on achievement tests.