No Stress: Longitudinal Observation of Contributors to Workplace Stress
Paul G. Allen Center for Computer Science & Engineering, University of Washington September 2019 - Present
Stress has a significant impact on health and well-being as well as workplace productivity.
In order to understand stress and its relationship to personality, behavior, and expression, we present an analysis of a first-of-its-kind longitudinal dataset of workplace behavior, including passively sensed and actively reported measures.
We find that sensed facial expressions, email sentiment, and meeting duration are highly predictive of stress, and their accuracy is moderated by gender.
This dataset and analysis can be used to inform predictions and interventions for stress based on unobtrusive observed behavior in the workplace.
(1) How we can use every day sensing technology to detect stress, particularly across expressed affect and workplace demands?
(2) How do coping strategies and personality characteristics impact stress?
We used four datasets for the analysis in our study: user survey, expressed affect, email and calendar.
The user survey data was collected proactively by asking each participant to fill out surveys during or at the end of the study.
The expressed affect, email and calendar datasets were gathered passively by running the multi-modal Emotion Sensing Platform developed by Microsoft in the background of each participant’s workstation (a desktop or laptop).
(1) We present an analysis of contributors to stress using a first-of-its-kind longitudinal dataset of workplace behavior in situ, including passively sensed and actively reported measures.
(2) We find that sensed facial expressions, email sentiment, and meeting duration are highly predictive of stress, and their accuracy is moderated by gender.
(3) We highlight opportunities and discuss relative merits for interventions as well as implications for gender differences in managing and coping with stress.
• Pre-processed data, extracted and aggregated useful features and conducted through statistical analysis to gain insights of the dataset demographics
• Investigated correlations between features and generated hypotheses to test for
• In preparation of a paper that presents our results