The Effect of Depression on Labor Market Outcomes

Mental health problems are prevalent among U.S. adults and there is a well-documented cross-sectional association between mental health status and labor market outcomes in the literature. In general, the negative effects of mental health problems on the likelihood of employment are large and persistent across studies, ranging from 11 to 19 percentage points. However, results are mixed when it comes to the hours worked or wages (Ettner et al,1997; Alexandre and French,2001; Chatterji et al,2007). The main empirical challenge in estimation is that various mental health measures are highly endogenous in both a structural and statistical sense, and it is extremely difficult to find credible instruments for mental illness. Nonetheless, in cross-sectional studies, researchers have relied mainly on instrumental variables (IV) to identify the causal impact of mental illness on labor market outcomes. The drawback of this approach is that identification is driven by sometimes questionable exclusion restrictions. In this paper, we focus on the impact of depression and attempt to address the endogeneity with the correlated random effects (CRE) panel data model.  This allows us to control for unobserved time-invariant factors, such as productivity and access to care, that are potentially correlated with depression and labor market outcomes, without the need for an exclusion restriction. We estimate this model using the 2004-2009 Medical Expenditure Panel Survey, which contains a clinically validated measure of depression and rich data on labor market outcomes. We find that depression reduces the likelihood of employment in our CRE probit models. However, we don’t find any negative impact of depression on hourly wages or weekly hours worked in the conditional sample of workers. The magnitude of our estimates is substantially smaller than those from previous studies: depression reduces the probability of working by only 2.6 percentage points. In addition, we also examined the effect of depression on lost work days, which are more responsive to depression in the short run. To handle the excessive amount of zeros in this outcome, we adapted the zero-inflated ordered probit (ZIOP) to the CRE framework.  Our results indicate that depression increases work day loss by 1.4 days (33 percent), which implies that the aggregate productivity losses by U.S. employers due to absenteeism ranges from 700 million to 1.4 billion in 2009 USD.