Health Shocks, Education, and Labor Market Outcomes

A large literature in health economics tries to estimate the relationship between health and socioeconomic status (SES). Due to numerous measurement and endogeneity issues, however, it is difficult to identify the causal effect of health on SES. In this paper, we combine administrative data from Chile on monthly earnings and hospital discharges to estimate the effect of health shocks on earnings as a measure of SES. Our estimates have a causal interpretation because we use unpredictable health shocks such as injuries due to car accidents and we control for unobserved heterogeneity using individual fixed effects. Moreover, we answer the policy relevant question to what extent health insurance and education reduce the negative effects of health shocks on future earnings.

The Grossman (1972) model provides a theoretical background. Productivity and hence earnings depend on an individual's health stock. Random health shocks reduce the health stock, hence a health shock decreases earnings. Education increases marginal productivity conditional on health, so a higher level of education reduces the negative effect of health shocks on earnings. Finally, individuals can invest in their health to (partly) offset the effects of health shocks. Health insurance makes it cheaper to purchase health investment goods, hence individuals with health insurance should experience smaller decreases in earnings following a negative shock. We expect the effect of health shocks to be the strongest in the period of the shock itself. Depending on the severity of the shock, however, these effects can potentially spread into subsequent periods. To test the model predictions, we regress monthly earnings on health shocks and their lags. Unobserved individual-specific variables such as the underlying health stock and preferences for leisure and health affect all of these predictions. To control for them, we include individual fixed effects in all of our regressions.

Our earnings data come from the Chilean unemployment insurance (UI) system. We observe monthly earnings for the universe of Chilean workers who are enrolled in UI from 2003 to 2011. The earnings data also contain workers' education. We merge these data with the universe of hospital discharge records from 2004 to 2007. These records contain detailed diagnosis codes, so we know if a hospital stay is due to an unexpected health shock or a chronic condition. Moreover, we observe the specific cause of an injury, so that we are able to distinguish between work place and traffic accidents, for example. The severity of health shocks is approximated by the length of the hospital stay and whether any surgical procedures were performed. Finally, we observe patients' source of health insurance.

The main result shows a 20 percent decrease in monthly earnings in the month of the health shock. Although earnings eventually increase, they do not catch up over a five-year period with the earnings of those who never had a health shock. As predicted by our model, these negative effects are smaller for more highly educated workers and those with more generous health insurance.