R help - Dec 2010 - problem with package rsm: running fit.mult.impute with cph

Hi all (and especially Frank),

I'm trying to use x=T, y=T in order to run a validated stepwise cox
regression in rsm, having multiply imputed using mice.  I'm coding

model.max<-fit.mult.impute(baseform,cph,miced2,dated.sexrisk2,x=T,y=T)

baseform is
 baseform<-Surv(si.age,si=="Yes")~ peer.press + copy.press +
excited +
worried + intimate.friend + am.pill.times + info.parents + info.teacher +
info.sch.nurse + info.friends + info.media + info.clinic + info.gp +
info.fpa + info.chemist + nearer.clinic + uti + thrush + herpes + love +
strat(gender) + Own.space + ordered(Chat.Mother) + ordered(Chat.Father) +
ordered(Prts.interested) + ordered(Someone.stands.up.4u) + Thought.runaway +
as.numeric(Prts.know.location)+ pubty + partner + arguments + mfqc.total +
esteem.total + events.total + bmi

This gives the error message


Error in  1:n.impute : NA/NaN argument
In addition : Warning message:
In 1:n.impute : numerical expression has 130 elements: only the first used
:>

It doesn't seem to be a singularity problem, as the simple equation

model.test<-fit.mult.impute(Surv(si.age,si=="Yes")~peer.press,cph,miced2,dated.sexrisk2,x=T,y=T)


gives an identical message, and the code runs perfectly in either without
x=T, y=T

PS the dataset is 827 rows x 130 variables.

	[[alternative HTML version deleted]]

rms and fit.mult.impute don't necessarily play well with mice.

Note also that you will need a special adjustment for variances for having
done stepwise regression.  One approach is to insert a zero for a
coefficient every time that variable is not 'selected', then compute the
sample variance of the beta hats and zeros [using the bootstrap].

Frank


-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
-- 
View this message in context:
http://r.789695.n4.nabble.com/problem-with-package-rsm-running-fit-mult-impute-with-cph-tp3069319p3070218.html
Sent from the R help mailing list archive at Nabble.com.