Flexible Modeling of Time to Failure in Risky Careers

Failure time models correcting for heterogeneity are used to explain the length of participation in a risky career. Using data from the National Football League, first we employ a class of techniques which ignore unobserved heterogeneity; hence these methods impose severe restrictions on the estimate hazard. We then examine a second class of techniques which correct for unobservables and thereby allow greater flexibility in the estimated hazard. Within this second class, we find that the estimated hazard using the Burr-12 density is much more accurate than densities in the first class, which include the exponential and Weibull. We expect that this density could be employed to successfully explain career duration in other high-risk, high-stress careers as well.