Mental-healthMost people will experience being over-worked or work-related stress at some stage in their careers, which if left unaddressed, can lead to fatigue, poorer mental-health, and costly workplace errors. We know a lot about the general features of work that can undermine or support mental health and wellbeing, but we don't yet understand how work stress accumulates or how resilient individuals draw upon protective resources over time. To gain these insights, my research focuses on developing predictive models that can describe patterns of data collected from workers intensively over time. Ultimately, my approach will support the development of individually personalized workplace interventions, and methods that can proactively detect and prevent work stressors from accumulating to long-term health problems.
Principal investigatorMichael David Wilson firstname.lastname@example.org
Area of scienceHealth
Applications usedR, Docker, Singularity
My research aims to understand critical patterns of change in mental health and wellbeing over both short and longer periods of time. In the organizational sciences, individual change in mental health is typically studied with relatively few “snapshots” at specific times. Relying on sparse measurements over long timeframes limits understanding of the factors that underlie mental wellbeing, and reduces the effectiveness of evidence-based interventions. For example, current methods can miss large but transient dips in mental wellbeing that occur during vulnerable times and, therefore, fail to identify critically important workplace causes of poor wellbeing. Further, these methods often treat the workforce as homogenous, neglecting the importance of individuals’ unique trajectories of wellbeing. Critically, these methods prevent us from evaluating how acute stressors accumulate to eventually result in more chronic and severe mental health outcomes.
My data is analyzed with continuous-time dynamic modelling which is a new statistical method for analysing human data continuously over time. Until recently, these methods have rarely been applied in human sciences due to data limitations and computational constraints. Dynamic modelling approaches provide more complete and tractable solutions to questions of causality between measures.
These models take many days to completely run, due to the Bayesian sampling process. The models also cannot be distributed. Nimbus allows me to run these models without importance sampling, which provides huge gains in accuracy of generated models.