| Abstract |
The mixed proportional hazards model generalizes the Cox model by incorporating a random effect. In the case of two samples, it is chiefly determined by a triple consisting of a number representing the treatment effect, the integrated base-line hazard, and the distribution of the unobserved random effect. If the latter has expectation 1, then this triple is known to be identifiable, and the treatment effect is known to have a consistent estimator. We prove that estimators, however, cannot be uniformly square root n consistent, neither in a local nor in a global sense.
|