Abstract and Introduction
Workplace downsizing and closure have been considered natural experiments that strengthen causal inference when assessing the association between unemployment and health. Selection into unemployment plays a lesser role among those exposed to severe workplace downsizing. This study compared mortality for individuals unemployed from stable, downsized, and closed workplaces with a reference group unexposed to unemployment. We examined nationally representative register data of residents of Finland aged 25–63 years in 1990–2009 (n = 275,738). Compared with the control group, the hazard ratio for substance use–related mortality among men unemployed from stable workplaces was 2.43 (95% confidence interval (CI): 2.22, 2.67), from downsized workplaces 1.85 (CI: 1.65, 2.08), and from closed workplaces 2.16 (CI: 1.84, 2.53). Among women, the corresponding estimates were 3.01 (CI: 2.42, 3.74), 2.39 (CI: 1.75, 3.27), and 1.47 (CI: 1.09, 1.99). Unemployment from stable workplaces was associated with mortality from psychiatric and self-harm–related conditions. However, mortality due to ischemic heart disease and other somatic diseases decreased for those unemployed following closure. The results indicate that selection mechanisms partially explain the excess mortality among the unemployed. However, substance-use outcomes among men and women, and fatal accidents and violence among men, might be causally associated with unemployment.
Unemployment is associated with negative health outcomes, such as increased health service use, decreased psychological well-being,[2–4] excess all-cause mortality,[1,5] and mortality due to causes such as cardiovascular disease and suicide.[2,6] However, the literature has yet to determine to what extent unemployment causally leads to poor health.[2–4] Suggested mechanisms through which unemployment might affect health are the loss of social (e.g., time-structures, social contacts, status) and material (income, benefits) resources, increased stress, and changes in health behaviors (poor coping styles). Alternatively, poor health might increase unemployment risk (direct selection).[1–4] Other preceding individual characteristics such as personality and health behaviors might also increase the probability of both unemployment and poor health (indirect selection).
A common strategy to address selection has been to adjust for preexisting health status and sociodemographic measures, while acknowledging the challenges of sufficiently controlling for residual confounding due to unmeasured characteristics. Information on workplace downsizing has also been used to strengthen causal inference. Workplace downsizings and closures can be considered natural experiments, given that they are independent of employee characteristics. Therefore, individual characteristics are less likely to affect unemployment the more heavily workplaces downsize. When a workplace closes, subsequent unemployment is more random, and selection into unemployment is reduced.[8–11] Nonetheless, recent large-scale longitudinal studies have focused less on the workplace contexts behind unemployment and how they affect the association between unemployment and health.
The early workplace closure studies were typically small, with homogeneous single-plant samples and short follow-up periods. More recently, 5 general population studies have utilized closure data. Salm examined 2 waves of biannual survey data and found that self-reported workplace closure had no association with self-rated health, limitations in daily living, longevity expectations, depression score, and self-reported depression diagnoses among 6,867 older working-age Americans. However, negative health changes were observed when a combination of these outcomes was examined among a very small subgroup of those unemployed during the second interview. The control group was stably employed.
Strully discovered an association between self-reported unemployment due to closure and fair/poor self-rated health, as well as a combination measure of self-reported stroke, hypertension, heart disease, heart attack, arthritis, diabetes, and mental health problems among 8,125 Americans. Because mental and physical outcomes were combined, the results might reflect mental health, physical health, or both. No association was detected for a combination measure of lung disease, cancer, or memory loss. The comparison group again were those stably employed.
Schmitz reported no association between self-reported unemployment due to closure and health satisfaction, a mental health score, and self-reported all-cause hospitalization among 23,734 Germans. The respondents were classified at each wave as employed (referent), unemployed due to closure, or other unemployed. A strength of the study was that it controlled for confounding from time-invariant individual characteristics using a fixed-effects design. A downside was that only those individuals with variation in the outcome contributed to these estimations.
Magnusson Hanson et al. compared the register data of 1,654,239 unemployed and employed individuals who were exposed to downsizing with unexposed employees over 2 years before and after the event. Antidepressant purchases increased slightly, especially close to the downsizing event among unemployed considered healthy at baseline.
Martikainen et al. compared 8-year mortality among unemployed persons and employed controls according to their workplace downsizing level, as well as during periods of high and low national unemployment (n = 159,736). They detected no excess all-cause mortality for those unemployed from severely downsized and closed workplaces. The association between unemployment and mortality was also weaker in high unemployment conditions, which has been interpreted to suggest that during recession times, the unemployed are a less select group.[6,10,11]
Previous studies have indicated that unemployment might be unassociated with all-cause hospitalizations and mortality. The evidence is conflicting for mental health[13,14] and lacking for mortality due to somatic conditions and external causes. We aimed to address these gaps by conducting an individual-level analysis with additional data on workplace downsizing and closure to obtain stronger causal inference on the relationship between unemployment and cause-specific mortality. Individuals who became unemployed from stable, downsizing, and closing workplaces were compared with a reference group that was unexposed to unemployment from a workplace.
The large data set, with over 90,000 unemployed individuals, enabled us to assess various physical and mental health outcomes at different levels of workforce reductions. We included several somatic health outcomes, whereas the literature has focused primarily on mental health.[3,11] Instead of merely analyzing the short-term association,[8,12,14] our maximum follow-up is 20 years. Register data have no self-report or recall bias, or the nonrandom attrition associated with survey studies.[8,12,13] Finally, we considered measures related to the workplace and regional context, individual health status, employment history, and sociodemographic characteristics prior to the onset of unemployment. While some suggest that closure designs already control for selection into unemployment,[9,13] we argue that these constitute an important addition. Although all employees lose their jobs in workplace closures, individual and regional characteristics might affect whether individuals find new employment or become unemployed. While our observational study is not an experimental study, which is unfeasible for our research question, we are contributing to the literature by combining workplace and individual-level data to reduce selection bias.
Am J Epidemiol. 2020;189(7):698-707. © 2020 Oxford University Press