Predicting students’ academic performance by their online learning patterns in a blended course : To what extent is a theory-driven approach and a data-driven approach consistent?

One of the major objectives of precision education is to improve prediction of educational outcome. This study combined theory-driven and data-driven approaches to address the limitations of current practice of predicting learning outcomes only using a single approach. The study identified the online learning patterns by using students’ self-reported approaches and perceptions of online learning and by using the observational digital traces of the sequences of their online learning events in a blended course. The study examined predictions of the academic performance using the online learning patterns generated by the two approaches separately. It also investigated the extent to which the online learning patterns identified by the two approaches were associated with each other. The theory-driven approach adopted a hierarchical cluster analysis using the self-reported data and found a ‘deep’ and a ‘surface’ online learning patterns, which were related to differences in the academic performance. The data-driven approach used an agglomerative sequence clustering and detected four patterns of online learning, which not only differed by quantity (number of learning events), but also differed by quality (the proportions of types of learning events). A one-way ANOVA revealed that the online learning pattern which had the most learning events, and was characterized by high proportions of viewing course contents and of performing problem-solving exercises, had the highest academic performance. A cross-tabulation revealed significant association between the self-reported and observational online learning patterns, demonstrating consistency of the evidence by a theory-driven and a data-driven approach and triangulating the results of the two approaches.