Effect of emotion and workload on expertise in programming

Abstract

In recent times, few studies on program comprehension have incorporated the use of psycho-physiological devices and self-report tools to measure participants’ emotions and cognitive workload. The use of self-report scales is beneficial in settings where there is no access to biosensors or a need for a general understanding of programmers’ emotions and workload during learning or in an assessment of a large group. Hence, we designed a study that would use self-report tools — the Self-Assessment Manikin (SAM) test and NASA TLX (Task Load Index) to obtain the emotions and workload of the participants on two types of programming tasks (logical and conceptual). Our aim was to identify the relationship between emotions and expertise, and workload and expertise on conceptual and logical questions in timed and untimed conditions. Our findings indicate that participants showed Low Valence-High Arousal (LVHA) in timed conditions and High Valence-Low Arousal (HVLA) in untimed conditions. Findings further indicate that high performers showed High Valence-Low Arousal (HVLA) on conceptual questions and High Valence-High Arousal (HVHA) on logical questions, whereas low performers showed High Valence-Low Arousal (HVLA) on conceptual questions and Low Valence-High Arousal (LVHA) on logical questions. We found that participants had a higher workload in timed conditions than untimed conditions and had a higher workload on logical questions than conceptual questions. Furthermore, we employed machine learning algorithms to predict expertise using emotions and workload and the best classifier produced an accuracy of 71.4%, with 72% weighted precision, 71.5% weighted recall, and 71.11% weighted F-score.