Investigating SMART Models of Self-Regulation and their Impact on Learning

Self-regulated learning (SRL) is a critical 21^st-century skill. In this paper, we examine SRL through the lens of the searching, monitoring, assessing, rehearsing, and translating (SMART) schema for learning operations. We use microanalysis to measure SRL behaviors as students interact with a computer-based learning environment, Betty's Brain. We leverage interaction data, survey data, in situ student interviews, and supervised machine learning techniques to predict the proportion of time spent on each of the SMART schema facets, developing models with prediction accuracy ranging from rho = .19 for translating to rho = .66 for assembling. We examine key interactions between variables in our models and discuss the implications for future SRL research. Finally, we show that both ground truth and predicted values can be used to predict future learning in the system. In fact, the inferred models of SRL outperform the ground truth versions, demonstrating both their generalizability and their potential for using these models to improve adaptive scaffolding for students who are still developing SRL skills.