Seeking Flow from Fine-Grained Log Data

Flow is the experience of deep absorption in a demanding, intrinsically-motivating task conducted with skill. We consider how to measure behavioural correlates of flow from fine-grained process data extracted from programming environments. Specifically, we propose measuring affective factors related to flow non-intrusively based on log data. Presently, such affective factors are typically measured intrusively (by self-report), which naturally will break the flow. We evaluate our approach in a pilot study, where we use log data and survey data collected from an introductory programming course. The log data is fine-grained, containing timestamped actions at the keystroke level from the process of solving programming assignments, while the survey data has been collected at the end of every completed assignment. The survey data in the pilot study comprises of Likert-like items measuring perceived educational value, perceived difficulty, and students' self-reported focus when solving the assignments. We study raw and derived log data metrics, by looking for relationships between the metrics and the survey data. We discuss the results of the pilot study and provide suggestions for future work related to non-intrusive measures of programmer affect.