People quit the labor force for many different reasons, voluntarily or not, through various arrangements such as unemployment benefits, disability benefits or specially designed early retirement schemes. This paper complements the existing literature by considering a large, register-based sample including objective medical diagnosis codes. We estimate detailed hazard models of duration until retirement, controlling for unobserved heterogeneity and nonparametric baseline hazards, as well as observed heterogeneity through time-varying explanatory variables. These include diagnosis codes, along with a host of demographic, labor market and financial regressors. The panel structure of the data allows following individuals year by year from the age of 50 and precisely measure changes in objectively measured health and other regressors, as well as labor market status. We consider 12 broad, mutually exclusive and exhaustive categories of health diagnoses defined by aggregation across ICD codes. The use of objective medical diagnosis codes should eliminate the justification bias due to self-reports of health, and the large sample size obtained by using register rather than survey data should mitigate the e¤ect of any remaining mismeasurement of true work incapacity. Together, these improvements should help distinguish empirically important effects of health and economic variables on retirement. We distinguish a number of alternative exit routes, in particular, disability, early retirement, unemployment, and others (including out of the labor force and welfare). We estimate both single risk models, lumping all retirement states, and competing risk specifications, including all separate exit routes. Throughout, females are included in the estimations, and we present separate results by gender. We find sizeable differences in retirement behavior across marital status, gender, labor market attachment, occupation, income, and in particular health. We find that the disability retirement exit route that requires specific medical criteria to be met is different from the early retirement route. The latter shares similarities with private pension schemes in a number of countries, including the U.S., where benefits are tied to previous wages, and employers also contribute to this retirement scheme. These differences are pronounced within labor market attachment, income, and in particular health. Furthermore, unemployment followed by early retirement is different from unemployment followed by other programs regarding marital status, gender, income, and health. These comparisons hinge on the competing risk framework. Finally, even when using objective medical diagnosis measures we still find significant effects from health on retirement. Thus, not all health impact on retirement reported in earlier literature was due to justification bias.
