Detecting distributed practice in trace data: A comparison of different operationalizations and their predictive value for lasting learning

Background: Spreading study sessions over time (i.e., distributed practice) is considered beneficial for lasting learning. Oftentimes, findings on distributed practice stem from laboratory settings in which learners’ study behaviors are manipulated. Trace data from digital learning platforms offer new ways of assessing distributed practice in naturalistic contexts. However, the fine-grained resolution of trace data allows for very different operationalizations of distributed practice. As of now, it is unclear which of these operationalizations are suitable. Aims: We developed and compared different operationalizations of distributed practice in a naturalistic online learning setting, using objective trace data.Sample: 331 medical students preparing for a high-stakes exam. Methods: The fine-grained longitudinal data allowed us to compare four different operationalizations of distributed practice in online learning environments: 1) distributing studying over more vs less days, 2) the growth in questions answered before the exam (via growth curve modeling), 3) the regularity in studying over time (via time-series clustering), and 4) the predictability in study trajectories (via time-series entropy). The resulting scores were then used to predict students’ exam performance, controlling for the total amount studied.Results: All four approaches could capture individual differences in distributed practice, but only distributing study activities over more days positively predicted exam scores over and above the total amount of studying done by students. Conclusions: Trace data offer various possibilities to capture distributed practice with great precision. To promote lasting learning, students should be helped to get into the habit of learning every day, at least a little.