Inference in Mixed Proportional Hazard Models with K Random Effects

Abstract

A general formulation of Mixed Proportional Hazard models with K random effects is provided. It enables to account for a population stratified at K different levels. We then show how to approximate the partial maximum likelihood estimator using an EM algorithm. In a Monte Carlo study, the behavior of the estimator is assessed and I provide an application to the ratification of ILO conventions. Compared to other procedures, the results indicate an important decrease in computing time, as well as improved convergence and stability.

Keywords: EM algorithm, penalized likelihood, partial likelihood, frailties

JEL Classification: C13, C14, C41