Positive mental health in work and private life: Extending modeling to a data-driven approach

Résumé 0

This research aims to extend the concept of positive mental health (PMH) as flourishing (Keyes, 2002) from a general context to specific work and non-work contexts. Conceptually, based on the integrated work-life balance framework (Sirgy & Lee, 2018), we explore the possibility that PMH in work life may be distinct from yet correlated with its counterpart in non-work life. Methodologically, we integrate several approaches. We analyze the multifaceted construct of PMH in such a way as to include and differentiate the general from the domain-specific expressions of PMH, using a person- and variable-centered approach, while accounting for the work and non-work contexts. Two different studies—respectively n=304 and n=1066—are used to explore the factor structure of this extended integrated construct of PMH. For each sample, a bifactor exploratory structural equation modeling (Bi-ESEM) provides the best fit to the data. Latent class analysis provides a means to explore different classes of PMH. A machine learning algorithm (classification trees) is used as an operational method to diagnose class assignment and to test classification predictive performance. Drawing on data from a sample of 1,066 French workers, our model reveals four classes of respondents, each displaying a particular profile of PMH in work-life contexts. Applying it to psychosocial functioning, we show that general and specific factors have significantly different associations with the reduction of psychosocial risks, and that there are significant differences between the profiles, wherein the full cross-context flourishers display the lowest level of absenteeism, presenteeism, turnover intentions, and work unhappiness, and the professional languishers the highest. We conclude that splitting PMH into work and non-work contexts highlights some significant facets and outcomes of the construct that are not available in its general expression alone. Machine learning proves to be an efficient and reliable way to diagnose and predict PMH with a high level of performance (accuracy=0.90, Kappa=0.86).