Another question for the coxph gurus:
At the end of the print out of summary(coxph) there is a line that reads:
"Rsquare= 0.316 ( max possible= 0.97 )"
(Numbers vary, of course, depending on the model.)
I have looked up the code for this in summary.coxph, and I recognize that
these two numbers are:
(1-exp(2* average change in loglikelihood of fitted vs null model
per observation)), and
(1-exp(2* average change in loglikelihood of a saturated vs null
model per observation))
I recognize, too, that this is a measure of the goodness of fit of the
model. (Hence the nickname "Rsquare.") I want to report a minimal
multivariate model of baseline covariates for a clinical study, and I want
to give a sense of how much of the variability in outcomes is explained by
each additional variable as it is added to the final model. The above
Rsquare seems to be a good way to do this.
My problem is that the quantities above are obviously not real model R
squares, since the model is a maximum likelihood model rather than a least
squares model. I don't think that I can call this "Rsquare" in the paper.
Are there correct names for the above terms? What do other people call
these numbers in formal publications?
Many thanks in advance for your help on this issue.
Larry Hunsicker
|