Predicting course performance from homework habits

Abstract

Homework exercises are a cornerstone of engineering education. In this work, we seek to understand how homework habits correlate with course performance. We conducted a study in which students in an undergraduate statics course used LivescribeTM smartpens to complete their coursework. Smartpens serve the same purpose as traditional ink pens, but also digitize the writing, producing a time-stamped record of the work. From this data, we compute a variety of quantitative features characterizing homework habits, such as the total amount of ink written and the time of the day at which the work is done. We use regression models to examine how these features correlate with the final course grade. We control for student ability using the students' scores on the Force Concept Inventory (FCI), which is completed during the first week of class. The FCI itself has some predictive ability, correlating with final course grade with R2 = 0.27. However, combining this score with the total ink written on the third homework assignment produced a much stronger correlation with R2 = 0.35. A regression model including the FCI score and four features computed from this assignment resulted in R2 = 0.43. Thus, by the end of the third week of the quarter, it is possible to explain a significant amount of the variance in final course grade by considering homework habits. This work provides a foundation for building early warning systems that examine homework activity to identify students at risk of performing poorly in a course. ©American Society for Engineering education, 2013.