Task Context Shapes Ocular Associations with Stress, Cognitive Workload, and Fatigue: A Wearable Eye-Tracking Study

Background: Eye-tracking correlations with cognitive and emotional states vary across studies, likely because they depend on specific task demands. We investigated the consistency and specificity of such associations between wearable eye-tracking data and self-reported measures of stress, cognitive workload, and fatigue within assessment-centre-style tasks.Methods: Participants performed tasks involving social evaluation, reasoning, planning, comprehension, and presentation while wearing Meta Aria Gen 1 glasses. Data from 97 participants were analysed using linear mixed-effects models to examine associations between 104 wearable eye-tracking and head-movement metrics and each self-reported state, while accounting for task effects. Metric-by-task interactions assessed task dependence, and sensitivity analyses examined the effects of elapsed session time and shared variance among states.Results: After false-discovery-rate correction, significant metric-by-task interactions were observed in 53 stress, 61 workload, and 12 fatigue models, especially for saccade amplitude, speed, and duration. Across tasks, stress was linked to increased gaze variability, while workload and fatigue were linked to greater fixation angular distance. Fatigue associations weakened after adjusting for session time, and no fatigue metric-by-task interactions remained significant after adjustment for elapsed session time. After controlling for overlap among self-reported states, associations remained significant for stress, but not workload or fatigue.Conclusion: Ocular associations varied by task and by the shared variance across subjective states. Measures of fixation distribution appeared to capture a more general subjective-state component rather than exclusively distinguishing workload or fatigue. Several gaze and saccade associations with stress remained significant after adjustment for workload and fatigue. Predictive modelling should distinguish shared from state-specific information and be validated across diverse activities.