Answers are grounded in this paper and cite their sources.
In the authors' analysis of nonwork spells for Workers Compensation Insurance, they use the family of generalized gamma distributions for the structural model of failure time and both nonparametric and parametric controls for heterogeneity. These specifications allow for nested likelihood tests of not only the correct form for parametric heterogeneity, but for the structural distributions as well. The authors find in their data that while heterogeneity is important, both parametric and nonparametric types of control do about equally well. The "best" (in terms of statistical fit) structural distribution is the generalized gamma function; the exponential distribution fits especially poorly.
No takes yet. Share an insight, caveat, or question.
Butler et al. (1991) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: