A Physical Model Justifying Proportional Hazards.

The proportional hazards model has been applied successfully to a wide variety of biomedical data to explain the dependence of survival curves on covariates. It has also been applied to reliability problems, but has not been as successful in reliability. In this note we demonstrate that the proportional hazards model can be derived as a way of describing the effect of accidents on a population of uniformly degrading devices. Thus poor performance on accelerated life test modeling would be expected, but one might expect it to be a useful tool for modeling field data with covariates. We also develop a generalization of the proportional hazards model which results in an ANOVA type representation of survival curves.