> It seems to me that R returns the unpenalized log-likelihood for the
> ratio likelihood test when ridge regression Cox proportional model is
> implemented. Is this as expected?
It is easy to verify that this is correct:
> fit1 <- coxph(Surv(time, status) ~ ridge(age) + ph.ecog, lung)
> fit2 <- coxph(Surv(time, status) ~ age + ph.ecog, lung,
init=fit1$coef, iter=0) # A model with pre-fixed coefficients
> c(fit1$loglik[2], fit2$loglik[2])
[1] -735.0954 -735.0954
It this intentional? Yes. Is it optimal? Probably not. The dominating
design goal for the penalized Cox code was extensibility --- you can
actually add your own penalized methods. It also meant that individual
plugins such as frailty, pspline, and ridge have compromises. However,
I am not aware of anyone ever using this extensibility; given this
experience the design priorities would be different if I were to write
it anew. Hindsight is always 20/20.
BUT -- I would be quite happy to have someone write a better ridge()
plugin. Interested?
Terry Therneau
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
