Metrics for timely assessment of novice programmers

We describe a study of programming behaviors of three sections of an introductory computer programming course with the goal of finding metrics for determining, early in a course, which students may be at risk. Programming activity logs generated by an instructional programming environment were used to generate a dataset of variables such as session time, number of errors, number of file edits, and other basic data. Results from correlations and linear regressions identify two simple metrics that may identify at-risk students. We also present findings that provide a basis for these models, namely, the large number of edit/compile/interpret-errors cycles that are required for beginning programmers to gain competency.