On 05/06/2012 16:17, Terry Therneau wrote:
I was looking at how the model.frame method for lm works and comparing
it to my own for coxph.
The big difference is that I try to retain xlevels and predvars
information for a new model frame, and lm does not.
I use a call to model.frame in predict.coxph, which is whyI went that
route, but never noted the difference till now (preparing for my course
in Nashville).
Could someone shed light on the rationale for non-preservation?
T'other way round ... it would have needed a conscious decision to
preserve them: these all predate xlevels and predvars.
model.matrix.lm does make use of xlevels, and I think that explains the
difference as most lm() auxiliaries use the model matrix.
And I don't see predvars used in survival:::model.frame.coxph.
Terry T.
Simple example
> library(survival)
> lfit <- lm(time ~ factor(ph.ecog) + ns(age, 3), data=lung)
> ltemp <- model.frame(lfit, data=lung[1:2,])
> ltemp
time factor(ph.ecog) ns(age, 3).1 ns(age, 3).2 ns(age, 3).3
1 306 1 -0.1428571 0.4285714 0.7142857
2 455 0 0.0000000 0.0000000 0.0000000
> lfit$model[1:2,]
time factor(ph.ecog) ns(age, 3).1 ns(age, 3).2 ns(age, 3).3
1 306 1 0.4443546 0.3286161 0.1900511
2 455 0 0.5697239 0.3618440 -0.1297479
> levels(ltemp[[2]])
[1] "0" "1"
> levels(lfit$model[[2]])
[1] "0" "1" "2" "3"
> cfit <- coxph(Surv(time, status) ~ factor(ph.ecog) + ns(age,3), lung)
> model.frame(cfit, data= lung[1:2,])
Surv(time, status) factor(ph.ecog) ns(age, 3).1 ns(age, 3).2 ns(age, 3).3
1 306 1 0.4443546 0.3286161 0.1900511
2 455 0 0.5697239 0.3618440 -0.1297479
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
--
Brian D. Ripley, rip...@stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
