Unemployment, health and mortality: evidence from longitudinal register data
Publications of the Faculty of Social Sciences 211
84 pages. Helsinki, Unigrafia (2022)
It is a well-established finding that compared to the employed, the unemployed have poorer health and higher mortality. Even so, it is unclear whether unemployment causes poor health. The association may also be due to direct health-related selection if poor health increases the risk of becoming or remaining unemployed. The association could also be a product of indirect
selection if some other characteristics such as prior life experiences increase both the risk of unemployment and poor health. It is widely recognised that adequately controlling for selection is challenging. Additionally, the role of causality and selection in the association between unemployment and health may vary depending on which aspect of health is considered. While the causal
evidence is stronger for self-reported mental well-being, it is less consistent for objective mental health outcomes and physical health.
The purpose of this study was to contribute to our current understanding of the role of causality and selection in the health and mortality differences between the employed and the unemployed. Quasi-experimental designs that more strongly control for selection were utilised to explore various aspects of health among working-age Finnish men and women. The specific aims of the study were 1) to assess differences in cause-specific mortality between the unemployed and the employed using a workplace downsizing and closure design; 2) to evaluate the causes of mental health-related morbidity among the unemployed by adjusting for time-invariant confounding in a fixed effects regression; and 3) to quantify the contribution of direct selection in transitions from unemployment to employment. Two of the four sub-studies were based on large-scale, general population samples, and two on full population data. The longitudinal data combined individual-level information from various administrative registers. Health was measured using a wide range of objective health outcomes: reimbursement rights for drugs for specific health conditions, visits to specialised care, hospitalisations, and causes of death. Differences in pre-existing health conditions and other individual and contextual factors fully explained the small increase in non-alcoholic disease mortality associated with unemployment, and some of the increase in psychiatric, alcohol-related and external-cause mortality. Moreover, mental health and alcohol-related conditions may be particularly important predictors of both becoming unemployed and remaining there. Having an alcohol-related condition halved the chances of finding employment among the unemployed, while mental health conditions decreased the chances by 30%. And while currently unemployed men and women were approximately two percentage points more likely to have a mental health-related visit to specialised care than those in employment, 70% of that association was due to confounding from time-invariant characteristics such as past experiences and personality. Controlling for all time-invariant confounding also explained a substantial share of the poorer mental health of those exposed to repeated or extended periods of unemployment, particularly among women. Even with the extensive controls for unmeasured confounding with the quasi-experimental designs, unemployment was associated with certain aspects of health. It is therefore possible that these associations are causal. Most notably, men who are unemployed following workplace closures (that is, the unemployment estimate the least affected by confounding) had a 2.2-fold risk of alcohol-related mortality when compared to the employed. The risk was 1.5-fold among the respective women. Furthermore, unemployment was associated with an elevated mental health-related morbidity among both men and women as well as an increased risk in external-cause mortality among men. This study suggests that health-related selection is likely to contribute both to the health differentials between employed and unemployed people, and to the differences in labour-market attachments between those people with and without health problems. Controlling for both measured and unmeasured confounders using a strong study design, unemployment does not appear to be a risk factor for non-alcoholic disease mortality, but may be harmful to certain aspects of health, in particular mental health, and alcohol-related conditions.