The effectiveness of learning analytics for identifying at-risk students in higher education

Abstract

In this article we investigate the effectiveness of learning analytics for identifying at-risk students in higher education institutions using data output from an in-situ learning analytics platform. Amongst other things, the platform generates ‘no-engagement’ alerts if students have not engaged with any of the data sources measured for 14 consecutive days. We tested the relationship between these alerts and student outcomes for two cohorts of first-year undergraduate students. We also compared the efficiency of using these alerts to identify students at risk of poorer outcomes with the efficiency of using demographic data, using widening participation status as a case study example. The no-engagement alerts were found to be more efficient at spotting students not progressing and not attaining than demographic data. In order to investigate the efficacy of learning analytics for addressing differential student outcomes for disadvantaged groups, the team also analysed the likelihood of students with widening participation status generating alerts compared with their non-widening participation counterparts. The odds of students with widening participation status generating an alert were on average 43% higher, demonstrating the potential of such a system to preferentially target support at disadvantaged groups without needing to target directly based on immutable factors such as their socio-economic background.