Seminars

General Departmental Seminar Series

Explained variation in proportional hazards regression

John O'Quigley, Professor of Mathematics, University of California at San Diego

Friday, September 20, 2002, 12:00 - 1:00 p.m.

G5/ 136-142, Clinical Sciences Center, 600 Highland Ave.

ABSTRACT

The basic idea of explained variation is very simple and readily expressed. For the multi-normal linear model it is widely known that a parameter of the model can be seen to be algebraically equivalent to the ratio of two variances; the residual or conditional variance of Y given X to the marginal variance of Y. Outside of the normal model, in particular for models commonly used in epidemiology such as the logistic or the proportional hazards model, there have been a variety of attempts at generalizing these concepts. Little consensus has been established, the various attempts often producing rival indices that may contradict one another. There does not seem to have been any attempts at a simple approach based on elementary concepts. The purpose of this talk is to examine this question and, in particular, consider the implications for constructing a suitable R2 measure of explained variation for proportional hazards models.

Back to General Departmental Seminar Series