What explains the negative effect of unemployment on health? An analysis accounting for reverse causality

Abstract

The unemployed are often in poorer health than their employed counterparts. This cross-sectional correlation is often attributed to a causal effect of unemployment on health. Recent research analyzing longitudinal data often supports alternative explanations, such as spurious correlation and/or selection of unhealthy workers into unemployment (i.e., reverse causality). In this paper, we apply a dynamic panel data estimator (system GMM) to account for both unobserved confounders and reverse causality. Despite some evidence for health selection, we still find strong support for the causality thesis. Furthermore, we show that the adverse health effect is partially explained by the loss of self-perceived social status due to unemployment but not by the loss of household income or social contacts.

Introduction

It is well documented that the unemployed are in worse health than their employed counterparts (Kroll, Müters, & Lampert, 2016). There are three distinct but not mutually exclusive explanations for this difference: social causation, indirect selection and direct (or health) selection. The first and long-standing social causation explanation assumes that becoming unemployed is damaging to an individual‘s health (for overviews see Brand, 2015; Wanberg, 2012). Unemployment is considered a stressful life event (cf., Pearlin, 1989) that creates distress and leads to health problems. In support of this explanation, a variety of mechanisms − again, none of them mutually exclusive − have been put forward. One of the most comprehensive discussions of this issue is provided by Jahoda (1981), who distinguished between the manifest and latent functions of work. In this framework, becoming unemployed leads to a loss of these functions and consequently to a deterioration of health. First and foremost, the unemployed are deprived of the manifest function of work, i.e., to provide the necessary financial means so that they can support themselves and their families. In addition, work fulfills a variety of latent functions. Work not only enables individuals to contribute to a collective purpose and structures their daily activities but also provides status and societal recognition and widens individuals’ social networks beyond family. Despite a long research tradition, the literature is often criticized for not testing the pathways or mechanisms underlying the social causation thesis (Bartley, 1994; Janlert & Hammarström, 2009). This lack of testing is unfortunate because, as argued by Goldthorpe (2001), for example, a test of the hypothetical mechanism underlying a causal relationship can improve the ability of statistical analysis to provide us with convincing evidence of the existence or absence of this causal relationship.

Testing the underlying mechanisms of social causation is even more important, given that social causation is, by far, not the only explanation for the unemployment-health relationship. A second explanation refers to “indirect selection”, i.e., the assumption that the relationship between health and unemployment might be mainly due to common causes. For example, workers with lower education have a higher risk of becoming unemployed (Mincer, 1991; Wolbers, 2000). At the same time, education is positively related to health (Conti, Heckman, & Urzua, 2010; Grossman, 1972). As a consequence, the common cause creates compositional differences such that the unemployed tend to be in poorer health than their employed counterparts. Accounting for indirect selection in quantitative research is a matter of controlling for all common causes, including those unobserved in the data, by suitable statistical methods.

Third, other scholars argue that there is a direct selection of unhealthy workers into unemployment. Sometimes, the term “reverse causality” is used to indicate that, compared to the social causation thesis, the direct selection explanation reverses the causal order between unemployment and health. The theoretical argument behind direct selection is that those workers whose health deteriorates become less productive and are more likely to be laid off (Bartley, 1988; Cook, 1985; West, 1991). Whereas the main bulk of the literature focuses on social causation, several authors have also argued for the importance of direct selection (Andreeva, Magnusson Hanson, Westerlund, Theorell, & Brenner, 2015; Heggebø, 2015; Korpi, 2001; McDonough & Amick, 2001 Stewart, 2001; Strully, 2009; Virtanen et al., 2006). Empirically, the social causation and direct selection explanations are not mutually exclusive. However, most research designs do not effectively distinguish empirically between the two. As a consequence, empirical results in support of social causation are often criticized by referring to the possibility of direct selection as an alternative explanation.

In the analysis below, we attempt to go beyond the current literature by making the following contributions. First, we aim to identify the net effect of social causation by explicitly ruling out both indirect and direct selection as alternative explanations. We do this by applying a dynamic panel data regression based on the generalized method of moments (GMM). Distinguishing between social causation and selection is important because the two explanations have very different implications for policymakers. In cases of social causation, it would make sense to take measures to avoid (prolonged) unemployment. In case of (direct) health selection, such measures will only induce costs and will be ineffective. Here, measures to foster the health of employed individuals would be more efficient. Second, we aim to further investigate the social causation effect by testing several causal mechanisms, namely those based on the reduction in financial resources, social network resources and self-perceived social status. Identifying the relevant mechanisms will help policy makers better focus potential measures against the adverse health effects of unemployment.

The article proceeds as follows: In the next section, we review the literature on social causation and derive our hypotheses. We then describe our data − the German Panel Study “Labour Market and Social Security” (PASS) − and the dynamic panel data estimator based on the generalized method of moments as our method of choice. After presenting our results, we conclude with a discussion.