Exploring Indicators of Developers' Sentiment Perceptions in Student Software Projects

Communication is a crucial social factor in the success of software projects, as positively or negatively perceived statements can influence how recipients feel and affect team collaboration through emotional contagion. Whether a developer perceives a written message as positive, negative, or neutral is likely shaped by multiple factors. In this paper, we investigate how mood traits and states, life circumstances, project phases, and group dynamics relate to the perception of text-based messages in software development. We conducted a four-round survey study with 81 students in team-based software projects. Across rounds, participants reported these factors and labeled 30 decontextualized statements for sentiment, including meta-data on labeling rationale and uncertainty. Our results show: (1) Sentiment perception is only moderately stable within individuals, and label changes concentrate on ambiguity-prone statements; (2) Correlation-level signals are small and do not survive global multiple-testing correction; (3) In statement-level repeated-measures models (GEE), higher mood trait and reactivity are associated with more positive (and less neutral) labeling, while predictors of negative labeling are weaker and at most trend-level (e.g., task conflict); (4) We find no clear evidence of systematic project-phase effects. Overall, sentiment perception varies within persons and is strongly statement-dependent. Although our study was conducted in an academic setting, the observed variability and ambiguity effects suggest caution when interpreting sentiment analysis outputs and motivate future work with contextualized, in-project communication.

💡 Research Summary

This paper investigates which personal and contextual factors influence how software developers perceive the sentiment of written communication. The authors conducted a longitudinal survey with 81 computer‑science students who were working in 28 team‑based software projects as part of a university course. Over four measurement points (corresponding to different project phases) each participant labeled the same set of 30 de‑contextualized development‑related statements as positive, neutral, or negative. In addition to the sentiment labels, participants reported a range of psychological and situational variables: a long‑term mood trait measured with a 15‑item German Mood Survey (including a reactivity subscale), short‑term affect using the Positive and Negative Affect Schedule (PANAS), overall life satisfaction, and perceived team dynamics (task conflict and relationship conflict). The study also recorded the participants’ confidence and rationale for each label, allowing the authors to quantify labeling uncertainty.

The authors first examined the stability of sentiment perception within individuals. Using Cohen’s κ and generalized estimating equations (GEE), they found a moderate average agreement of κ ≈ 0.45 across the four rounds, indicating that developers’ sentiment judgments are only partially stable over time. Changes in labels were concentrated on statements that were rated as “ambiguous” in the meta‑data, suggesting that linguistic vagueness drives intra‑person variability.

Next, the authors explored simple correlations between each psychological variable and the proportion of positive, neutral, or negative labels. After correcting for multiple testing with the false discovery rate, virtually no correlations survived, implying that any single factor has only a weak direct association with sentiment perception.

To capture the joint influence of multiple variables while accounting for repeated measurements, the authors fitted a multinomial logistic GEE model with random effects for statements. The main findings from this model are:

• Mood trait and reactivity: Higher scores on the long‑term mood scale and on the reactivity subscale were significantly associated with an increased probability of assigning a positive label (β ≈ 0.12, p < 0.01) and a decreased probability of assigning a neutral label (β ≈ ‑0.08, p < 0.05). This suggests that developers who generally feel good or who experience frequent mood swings tend to interpret ambiguous messages more positively.

• PANAS affect: Positive affect showed a modest, non‑significant trend toward more positive labeling (β ≈ 0.05, p ≈ 0.07). Negative affect did not have a clear impact, and there were no significant interactions between PANAS scores and mood traits.

• Team conflict: Higher task conflict scores were linked to a slight increase in negative labeling (p ≈ 0.08), but this effect did not reach conventional significance. Relationship conflict showed no measurable effect.

• Project phase: The three phases (early, middle, deadline) did not differ significantly in the distribution of sentiment labels, nor did phase interact with any of the psychological predictors.

• Statement effects: Fixed effects for each of the 30 statements explained roughly 70 % of the variance in labeling outcomes, confirming that the linguistic properties of the statements (e.g., ambiguity, length, presence of technical jargon) dominate over individual traits.

The authors discuss several implications. First, the moderate intra‑individual stability and the concentration of label changes on ambiguous statements highlight that sentiment perception is a dynamic, context‑sensitive process. Consequently, static sentiment‑analysis tools that rely on a single, context‑free annotation may misrepresent the emotional climate of a development team. Second, while mood‑related traits have a detectable influence, their effect sizes are modest, and other factors such as task conflict only show exploratory trends. Third, the lack of systematic phase effects suggests that temporal pressure alone does not systematically bias sentiment perception in this student sample.

Limitations are acknowledged: the participant pool consists of university students rather than professional developers, the statements were stripped of surrounding conversation context, and all measures are self‑reported, which may introduce common‑method bias. Nevertheless, the study fills a gap in the literature by providing a longitudinal, within‑subject view of sentiment perception and by jointly modeling psychological and situational variables.

In conclusion, developers’ sentiment judgments are partially stable but heavily dependent on the wording of the message and on individual mood characteristics. Positive mood traits and higher emotional reactivity tend to shift interpretations toward positivity, whereas task conflict may nudge judgments toward negativity. The findings caution practitioners and researchers against over‑reliance on automated sentiment scores without considering the underlying variability and ambiguity of human perception. Future work should extend the design to professional settings, incorporate full conversational context, and explore more sophisticated multivariate models (e.g., structural equation modeling or Bayesian hierarchical approaches) to better capture the complex interplay of affect, team dynamics, and project circumstances.