Why are the unemployed in worse health? The causal effect of unemployment on health

Abstract

We analyse the effect of unemployment on health using information from the German Socio-Economic Panel of the years 1991–2008. To establish a causal effect we rely on fixed-effects methods and plant closures as exogenous entries into unemployment. Although unemployment is negatively correlated with health, we do not find a negative effect of unemployment due to plant closure on health across several health measures (health satisfaction, mental health, and hospital visits). For this subgroup of the unemployed, unemployment does not seem to be harmful and selection effects of ill individuals into unemployment are likely to contribute to the observed overall correlation between poor health and unemployment.

Introduction

The association between unemployment and health is well documented in the empirical literature. Various studies report a strong negative correlation between individual health and the experience of unemployment, or, more generally, between health and low income (see, e.g., Adams et al., 2003). However, the direction of causality is not yet well understood. There are at least three pathways that can lead to the observation of a less healthy stock of unemployed compared to the stock of employed. First, there is a selection of ill workers from work into unemployment. García-Gómez et al., 2010, Arrow, 1996, Riphahn, 1999, Lindholm et al., 2001 provide evidence that the likelihood of becoming unemployed is higher for ill workers. Second, poor health causes longer unemployment spells, as shown by Stewart (2001). Both points – selection of ill workers into unemployment and selection of healthy workers out of unemployment – increase the probability of observing an ill individual in the stock of unemployed and, thus, lead to a lower average health status of the stock of unemployed.

Third, unemployment itself might lead to a deterioration in health. The causal effect of unemployment on health is probably the most difficult of the three to show. There are most likely individual unobservable effects that both affect health and the probability of becoming unemployed, for instance a general frailty or other genetic factors. Usually, panel data help to control for this unobserved heterogeneity. Moreover, the health-related selection into unemployment needs to be considered. Since there might be reversed causality (e.g., a health shock that both decreases health and leads to unemployment), a causal effect can only be established if this selection effect is controlled for.

If unemployment indeed deteriorates health the individual and social costs of unemployment are higher than usually assumed and policymakers should try even harder to get the unemployed back into the labour market. An additional motivation to examine this third point is to find out more about the nature of unemployment. The life satisfaction literature concludes that unemployment is involuntary if it causally reduces life satisfaction (Winkelmann and Winkelmann, 1998). Similar arguments hold for health. It can be assumed that unemployment negatively affects health especially if it is involuntary.¹

There are only a few studies that analyse the effect of unemployment on health with German data. Romeu Gordo (2006) finds a negative effect of short-term unemployment on health satisfaction for men but no effect for women with SOEP data from 1984–2001. Moreover, long-term unemployment decreases health satisfaction of both men and women. However, although Romeu Gordo (2006) uses panel data and can therefore control for unobserved heterogeneity, the author cannot exclude that reversed causality may have biased the result. Huber et al. (2010) focus on welfare recipients (usually long-term unemployed) as a subgroup of unemployed and restrict their analysis to exit from unemployment. They find positive effects of returning to work on health, in particular the mental health of males.

The contribution of this paper is the following: First, in contrast to most of the international literature (Salm, 2009 being among the exceptions), we do not only use one health measure as an outcome variable but several different ones. In addition to health satisfaction (as in Romeu Gordo, 2006) these are the probability of a hospital visit within four years after the interview as a more objective health measure, and a measure of mental health. Moreover, we use the appropriate econometric methods by accounting for the ordered nature of the health satisfaction variable. Third, and most importantly, we extend the analysis by Romeu Gordo (2006) by accounting for the possible endogeneity of the entry into unemployment. In principle, this can be done by estimating a simultaneous equations model with health and the labour market status as endogenous variables (see Cai, 2010). Here, we rely on an alternative approach by using plant closures as an exogenous reason for unemployment. Doing this, reversed causality (from poor health to unemployment) is ruled out.

We find that the reason of unemployment entry has a strong impact on the results. Using data from the German Socio-Economic Panel for the years 1991–2008 and including all unemployed in the analysis, we find that unemployed are less healthy than employed according to all health measures. However, this is not causally due to unemployment, since there is no effect at all if unobserved heterogeneity is controlled for and exogenously unemployed are considered. In this latter group, unemployment does not deteriorate health. Thus, the worse health status of the unemployed might only be a selection effect into unemployment.

The results are in line with several international studies in the recent literature that either use plant closures or mass lay-offs to rule out a health-driven selection into unemployment. Browning et al. (2006) find no causal effect of job loss on the probability of entering a hospital due to symptoms caused by mental stress after four years with Danish register data. Salm (2009) finds no effect on several subjective and objective health measures with data from the HRS. Kuhn et al. (2009) do not find short-run effects of job loss on public health costs associated with health care utilisation. However, they do find that job loss increases hospitalisations for mental health reasons and prescriptions for antidepressants (both for males only). There are also studies that do find strong effects of involuntary job loss on subsequent mortality (e.g., Sullivan and von Wachter, 2009, Eliason and Storrie, 2009). A difference in our study is that we analyse the effect of actually being unemployed instead of the mere job loss on health. Also doing this, Böckerman and Ilmakunnas (2009) do not find negative effects of unemployment on self-assessed health for Finland, although they do not restrict their analysis to mass lay-offs or plant closures as reasons for unemployment.

The next section presents the data used in the analysis. Section 3 explains the econometric strategy while Section 4 reports the regression results. Section 5 presents robustness checks and Section 6 concludes.