Do feelings matter? On the correlation of affects and the self-assessed productivity in software engineering

Background: software engineering research (SE) lacks theory and methodologies for addressing human aspects in software development. Development tasks are undertaken through cognitive processing activities. Affects (emotions, moods, feelings) have a linkage to cognitive processing activities and the productivity of individuals. SE research needs to incorporate affect measurements to valorize human factors and to enhance management styles. Objective: analyze the affects dimensions of valence, arousal, and dominance of software developers and their real-time correlation with their self-assessed productivity (sPR). Method: repeated measurements design with 8 participants (4 students, 4 professionals), conveniently sampled and studied individually over 90 minutes of programming. The analysis was performed by fitting a linear mixed- effects (LME) model. Results: valence and dominance are positively correlated with the sPR. The model was able to express about 38% of deviance from the sPR. Many lessons were learned when employing psychological measurements in SE and for fitting LME. Conclusion: this article demonstrates the value of applying psychological tests in SE and echoes a call to valorize the human, individualized aspects of software developers. It reports a body of knowledge about affects, their classification, their measurement, and the best practices to perform psychological measurements in SE with LME models.

💡 Research Summary

The paper addresses a notable gap in software engineering (SE) research: the systematic incorporation of human affective states into the study of developer productivity. While SE has traditionally focused on technical artifacts and processes, the authors argue that cognitive activities underlying software development are intrinsically linked to emotions, moods, and feelings—collectively referred to as “affects.” Their objective is to examine how three affect dimensions—valence (pleasure), arousal (activation), and dominance (control)—correlate in real time with developers’ self‑assessed productivity (sPR).

To this end, the authors designed a repeated‑measures experiment involving eight participants, equally split between university students and professional developers. Each participant worked on a single programming task for ninety minutes while being prompted every ten minutes to report their current affective state using the Self‑Assessment Manikin (SAM), a pictorial 9‑point scale that captures valence, arousal, and dominance simultaneously. At the same intervals, participants rated their productivity on a five‑point Likert scale, answering the question “How productive do I feel right now?” This design yields nine paired observations per participant, providing a total of 72 data points.

Given the hierarchical nature of the data (multiple observations nested within individuals), the authors employed a linear mixed‑effects (LME) model. The fixed effects in the model are the three affect dimensions, while random intercepts and slopes for time are included to account for individual differences in baseline productivity and temporal trends. Model fitting was performed via restricted maximum likelihood, and multicollinearity among predictors was assessed using variance inflation factors.

The statistical analysis revealed two significant relationships. First, valence exhibited a positive coefficient (β ≈ 0.42, p < 0.01), indicating that higher reported pleasure is associated with higher self‑assessed productivity. Second, dominance also showed a positive effect (β ≈ 0.35, p < 0.05), suggesting that feelings of control or influence over the task boost perceived productivity. In contrast, arousal did not reach statistical significance (β ≈ 0.08, p = 0.34), implying that heightened activation alone does not translate into better productivity in this context. The model’s marginal R² (variance explained by fixed effects) was about 38 %, while the conditional R² (including random effects) rose to roughly 52 %, indicating that affect explains a substantial, though not exhaustive, portion of productivity variance.

Beyond the quantitative findings, the paper contributes methodological insights for SE researchers wishing to incorporate psychological measures. The authors discuss challenges such as translating SAM items into Korean while preserving semantic fidelity, designing unobtrusive data‑collection interfaces to avoid disrupting the programming flow, and handling convergence issues in LME fitting (e.g., choosing appropriate optimizer settings and simplifying random‑effects structures when necessary). They also note the importance of pilot testing to calibrate task difficulty and measurement intervals, thereby reducing fatigue and measurement noise.

The discussion situates the results within broader SE and psychology literature. Positive affect (high valence) is known to broaden attentional scope and foster creative problem solving, which aligns with the observed productivity boost. Dominance reflects a sense of agency and self‑efficacy; when developers feel they are in control, they are more likely to engage deeply with the code. The lack of a significant arousal effect may be explained by the Yerkes‑Dodson law: moderate arousal can enhance performance, but excessive activation may cause stress without productivity gains.

Practical implications are drawn for managers and team leads. Interventions that cultivate positive affect—such as constructive feedback, recognition, and a supportive work environment—could directly improve developers’ perceived output. Similarly, granting autonomy, clear task ownership, and opportunities for skill mastery may increase dominance feelings, further enhancing productivity. The authors caution, however, that self‑assessed productivity is a subjective metric; future work should triangulate with objective measures like code churn, defect density, or task completion time.

Limitations are acknowledged. The small, convenience‑sample size restricts external validity, and the homogeneous task may not capture the diversity of real‑world development activities. Moreover, the repeated‑measures design, while powerful for within‑subject analysis, cannot fully disentangle causality—high productivity might itself elevate mood. The authors propose larger‑scale studies, longitudinal designs, and experimental manipulations (e.g., mood‑induction protocols) to address these concerns.

In conclusion, the study demonstrates that affective states, particularly pleasure and perceived control, have measurable, positive associations with developers’ self‑reported productivity. By applying a rigorous LME framework, the authors provide a replicable analytical template for future SE research that seeks to integrate human factors. Their work underscores the value of treating developers as individuals with emotional lives, and it calls for SE methodologies that explicitly account for these dimensions to foster more humane and effective software development practices.