Disparities in health and mortality across educational groups are striking and pervasive. It has also been established that lower mental health increases mortality and that the lower educated have more mental health problems. The association between education and mortality may, therefore, be partly explained by the increased incidence of mental problems of the low educated. An important issue is that education attainment, mental hospitalization and mortality may all depend on the same observed and unobserved individual factors. Such confounding factors render education and mental hospitalization endogenous in the mortality analysis. To obtain the causal impact of education, running through mental hospitalization, on mortality we account for both the selection into mental hospitalization, by using a correlated multistate model for the mental hospitalization process (both admittance an discharge) and mortality and, the selection into education, by using a re-weighting technique (inverse propensity weighting) based on the probability to attain higher education.

We use Swedish Military Conscription Data (1951-1983), linked to administrative Swedish registers. These data, comprising 1,7 mln men, allow us to distinguish five education levels and the timing of admittance and discharge from mental hospitals and the moment death of each men. We estimate the effect of moving up one educational level on the morality rate, both directly and running though mental hospitalization. Our empirical results indicate a strong educational gain on mortality, even after accounting for confounding factors. The (negative) impact of mental hospitalization on survival . We also found that ignoring confounding would overestimate the educational gain and the impact of mental hospitalization on mortality. Accounting for confounding in mental hospitalization is more important than accounting for selective educational attainment.