Mouse Interactions as Digital Measures of Arousal and Pleasure at Work: Limited Between-Person Generalization and no Reliable Chronic-Stress Association
Journal
SSRN
Type
working paper
Date Issued
2025-09-08
Author(s)
Abstract
Background: Workplace stress harms health and productivity; unobtrusive digital biomarkers could trigger preventive interventions. Prior work suggests—though not uniformly—that computer, especially mouse, interactions change under stress, yet most evidence stems from lab studies of acute stress. We test whether mouse features indicate momentary arousal and pleasure, and whether long-term patterns relate to chronic stress.
Method: A 7-week field study with N=87. Twice daily Affective Slider (arousal, pleasure); TICS-SSCS at baseline and midpoint. Operating-system events yielded speed, acceleration, curvature, accuracy (path deviation), clicks, scrolling, pauses, and click-induced application window switches. Features from the prior 30 min classified arousal/pleasure (low/medium/high) with random forests (10-fold CV within participants; group-wise CV holding out participants). For chronic stress, person-level mouse/usage summaries were regressed on TICS-SSCS via OLS and principal-component regression.
Results: Within participants, mouse features moderately classified arousal and pleasure (accuracy 65%, weighted F1≈0.65); high arousal was best (F1=0.74), while low was often labeled medium. Across participants, performance fell below majority-class baseline (accuracy ≈47–48% vs. ≈56–57%; weighted F1≈0.44–0.45). Chronic stress showed no significant associations: OLS yielded R^2=0.31, adj. R^2=-0.11, p=.33; PCA regression R^2=0.09, adj. R^2=0.03, p=.07; no significant components.
Conclusions: Mouse dynamics signal momentary arousal/pleasure—especially high arousal—but generic models did not transfer across users, and usage statistics did not distinguish chronic stress. Person-specific baselines and context (e.g., physiology, task) appear necessary for workplace stress monitoring.
Method: A 7-week field study with N=87. Twice daily Affective Slider (arousal, pleasure); TICS-SSCS at baseline and midpoint. Operating-system events yielded speed, acceleration, curvature, accuracy (path deviation), clicks, scrolling, pauses, and click-induced application window switches. Features from the prior 30 min classified arousal/pleasure (low/medium/high) with random forests (10-fold CV within participants; group-wise CV holding out participants). For chronic stress, person-level mouse/usage summaries were regressed on TICS-SSCS via OLS and principal-component regression.
Results: Within participants, mouse features moderately classified arousal and pleasure (accuracy 65%, weighted F1≈0.65); high arousal was best (F1=0.74), while low was often labeled medium. Across participants, performance fell below majority-class baseline (accuracy ≈47–48% vs. ≈56–57%; weighted F1≈0.44–0.45). Chronic stress showed no significant associations: OLS yielded R^2=0.31, adj. R^2=-0.11, p=.33; PCA regression R^2=0.09, adj. R^2=0.03, p=.07; no significant components.
Conclusions: Mouse dynamics signal momentary arousal/pleasure—especially high arousal—but generic models did not transfer across users, and usage statistics did not distinguish chronic stress. Person-specific baselines and context (e.g., physiology, task) appear necessary for workplace stress monitoring.
Publisher
Elsevier BV