Cox-McFadden Partial and Marginal Likelihoods for the Proportional Hazard Model with Random Effects

Abstract

In survival analysis, Cox's name is associated with the partial likelihood technique that allows consistent estimation of proportional hazard scale parameters without specifying a duration dependence baseline. In discrete choice analysis, McFadden's name is associated with the generalized extreme-value (GEV) class of logistic choice models that relax the independence of irrelevant alternatives assumption. This paper shows that the mixed class of proportional hazard specifications allowing consistent estimation of scale and mixing parameters using partial likelihood is isomorphic to the GEV class. Independent censoring is allowed and I discuss approximations to the partial likelihood in the presence of ties. Finally, the partial likelihood score vector can be used to construct log-rank tests that do not require the independence of observations involved.