Developing methods to understand how learners’ motivation shapes engagement and achievement in undergraduate STEM courses has remained an important goal in efforts to meet increasing demand for STEM graduates. Researchers have documented strong associations between motivational beliefs, self-regulated learning (SRL) behaviors, and academic outcomes in STEM courses, yet researchers still lack analytic approaches within learning analytics contexts that fully integrate motivation with observable behavioral traces of SRL when developing prediction models of course performance. This limitation constrains efforts to identify how motivational differences among learners’ condition engagement with instructional resources and relate to course outcomes. Building on Expectancy-Value Theory and Achievement Goal Theory, I examined how configurations of goals, values, and costs provided context for interpreting and predicting relationships between SRL behaviors and course performance in a high-structure, active-learning biology course.Using digital trace data collected across multiple semesters, I examined how motivational profiles and learning behaviors related to predictions of course performance. Addressing three research questions, I first identified distinct motivational profiles using latent profile analysis and compared their demographic, academic, and behavioral characteristics. Then, I evaluated the performance of an omnibus motivation-behavior prediction model relative to profile-specific prediction models that relied on behavioral features as candidate predictors. Finally, I compared the behavioral predictors retained in the omnibus model with those retained across motivation profile-specific models to clarify how motivational conditions provided predictive utility to learners’ behavioral trace data.Results indicated that learners assigned to different motivational profiles varied in their engagement with course resources and levels of course performance. Profile-specific prediction models revealed differences in how behavioral indicators related to outcomes across motivational profiles, indicating that similar observable behaviors corresponded with course performance in differing ways across motivational profiles. Although profile-specific models improved within profile training set performance, omnibus modeling produced ultimately more stable and generalizable predictions when applied to a withheld test population. These findings demonstrate that incorporating motivational profiles into analyses of behavioral trace data can help instructors and researchers interpret patterns of learner engagement more accurately and identify groups of learners who may benefit from targeted instructional support in high-structure STEM courses.
Robert D. Plumley (Fri,) studied this question.