The Incidence of Monotone Likelihood in the Cox Model

Abstract

In estimating covariate coefficients for the Cox proportional-hazards model, there is a nonzero probability for any finite sample that the maximum likelihood estimate will be infinite. This implies that in Monte Carlo studies the variance is not a suitable estimator of precision. This also may necessitate a stratified analysis when multiple covariates are involved. A simple method is given for detecting and handling these infinite estimates.