Hi -
1)in my origianl post, I negelcted to say:
Given a survReg model model<-survReg(Surv(time,cens)~covariate1*factor2,data)
(I wrote....)
I have just found (I think) that the request predict(model,type="terms"),
when plotted with the appropriate transformations for the model, provides
the same estimate that the request predict(model,type="response",p=.5)
gives, though this might only be true for a single covariate. Can anyone
tell me if this is wrong?
While the SPLUS manual suggests on page 336 that predict(model,
type="terms") are useful for examination of terms in the model that expand
to multiple dummy variables, such as factors...", would it be at all
sensible to show an interaction between covariates and or factors in a
survReg( ) model by plotting the 50 percentile (or any other) response
against the factor levels using, say, xyplot? e.g.
xyplot(fifty~covariate1|factor2,data=data)
And, if there is time or interest:
coxph can call survfit to plot survival curves for stratified variables,
controlling for other factors in a model. I have used the following code
to do a similar thing with the output of a survReg model...
plot(1-pweibull(((1:44)/45),3.97,.719))#intercept and scale for strata level 1
matpoints(1:44,1-pweibull(((1:44)/45),3.97,.817),pch="@")#intercept and
scale for level 2
where the only change is the scale parameter difference estimated for the
stratified variable in a model. Does this represent the same thing as in
the coxph call to plot(survfit) ?
------------------------------------------------------------------------------
Dr. Andrew Beckerman
Institute of Biological Science
University of Stirling, Stirling FK9 4LA, Scotland, UK
phone:+44 (0)1786 467808 fax: +44 (0)1786 464994
|