Sick of work or too sick to work? Evidence on self-reported health shocks and early retirement from the BHPS

We follow individuals as they retire using discrete-time hazard models applied to a stock sample from 12 waves of the British Household Panel Survey. We use a generalised ordered probit model to explore the existence of reporting bias and to purge the self-reported measure of health of this bias. This model takes into account the possibility that anticipated retirement may influence reporting of health. There is evidence that health shocks are a determinant of retirement age. This is the case for both men and women and is observed for both health limitations and a measure based on latent health status derived from the generalised ordered probit model. While the size of the health effect varies according to how we measure health, the relative effect is large compared to the other variables, and in particular when compared to the effects of private pensions. However numerical simulations show that this high hazard ratio interacts with the, relatively low, incidence of new health limitations among the age group of interest to generate only a modest number of excess early retirements. Further, our results do not show evidence that the health status of their partner affects the hazard of early retirement for both men and women. Having an employed partner does reduce the hazard of early retirement.