Mastery data derived from formative assessments constitute a rich data set in the development of student performance prediction models. The dominance of formative assessment mastery data over use intensity data such as time on task or number of clicks was the outcome of previous research by the authors in a dispositional learning analytics context [1–3]. Practical implications of these findings are far reaching, contradicting current practices of developing (learning analytics based) student performance prediction models based on intensity data as central predictor variables. In this empirical follow-up study using data of 2011 students, we search for an explanation for time on task data being dominated by mastery data. We do so by investigating more general models, allowing for nonlinear, even non-monotonic, relationships between time on task and performance measures. Clustering students into subsamples, with different time on task characteristics, suggests heterogeneity of the sample to be an important cause of the nonlinear relationships with performance measures. Time on task data appear to be more sensitive to the effects of heterogeneity than mastery data, providing a further argument to prioritize formative assessment mastery data as predictor variables in the design of prediction models directed at the generation of learning feedback.

Blended learning, Computer assisted assessment, Dispositional learning analytics, E-tutorials, Formative assessment, Student profiles,
Rotterdam School of Management (RSM), Erasmus University

Tempelaar, D.T, Rienties, B, & Giesbers, B. (2015). Understanding the role of time on task in formative assessment: The case of mathematics learning. doi:10.1007/978-3-319-27704-2_12