The Paradoxical Nature of The Proportional Hazards Model of Random Censorship

Abstract

We show that the proportional hazards model of random censorship is too good to be frequently true as measured by mean squared errors: for estimating the underlying distribution function F(x) it is better to have a censored sample for a suitable expected censoring proportion than an uncensored full sample of the same size for any xbelow the 0.56-quantile of F.

AMS 1991 Subject Classifications::

• Primary: 62G05
• 62G20
• Secondary: 62N05

Keywords:

• Random censorship
• proportional
• hazards
• censored and uncensored samples