R help - Jul 2013 - multcomp on significant interaction in coxme model

Dear R community
I currently try to get post hoc multiple comparisons with the package 
multcomp from a cox mixed-effects model, where the survival is explained 
by two variables (cover with levels: nocover and cover; treatment with 
levels: tx, uv, meta), whose interaction is significant.
I read Hothorn, T. 2011: Additional multcomp Examples and there is an 
example involving a two-way ANOVA with significant interaction. Using 
this guideline I wanted to compare the levels of treatment in my data 
within the levels of cover.

My model is as follows:

m.lapf<-coxme(Surv(day,status)~cover*treatment+(1|r.brood)+(1|r.worker)+(1|species/brood),data=removal.lapf)

First I constructed a contrast matrix based on Tukey-contrasts as follows:

Tukey<-contrMat(table(removal.lapf$treatment), "Tukey")
Tukey
K1<-cbind(Tukey,matrix(0,nrow=nrow(Tukey),ncol=ncol(Tukey)))
K1
rownames(K1)<-paste(levels(removal.lapf$cover)[1],rownames(K1),sep=":")
K1
K2<-cbind(matrix(0,nrow=nrow(Tukey),ncol=ncol(Tukey)),Tukey)
K2
rownames(K2)<-paste(levels(removal.lapf$cover)[2],rownames(K1),sep=":")
K2
K<-rbind(K1,K2)
colnames(K)<-c(colnames(Tukey),colnames(Tukey))
K

This gives the following matrix
meta tx uv meta tx uv
cocooned:tx - meta              -1       1  0       0     0  0
cocooned:uv - meta             -1        0  1      0     0  0
cocooned:uv - tx                    0       -1  1      0     0 0
naked:cocooned:tx - meta    0        0  0     -1    1  0
naked:cocooned:uv - meta    0       0  0     -1     0  1
naked:cocooned:uv - tx         0        0  0      0    -1  1

then I constructed a secon matrix

tmp<-expand.grid(treatment=unique(removal.lapf$treatment),cover=unique(removal.lapf$cover))
X<-model.matrix(~cover*treatment,data=tmp)
X

which gives the following

   (Intercept) covernaked treatmenttx treatmentuv covernaked:treatmenttx 
covernaked:treatmentuv
1           1          1           1 0                      
1                      0
2           1          1           0 1                      
0                      1
3           1          1           0 0                      
0                      0
4           1          0           1 0                      
0                      0
5           1          0           0 1                      
0                      0
6           1          0           0 0                      
0                      0
attr(,"assign")
[1] 0 1 2 2 3 3
attr(,"contrasts")
attr(,"contrasts")$cover
[1] "contr.treatment"

attr(,"contrasts")$treatment
[1] "contr.treatment"


however when performing the tests via

summary(glht(m.lapf,linfct=K%*%X))

I get the following error message

Error in summary(glht(m.lapf, linfct = K %*% X)) :
   error in evaluating the argument 'object' in selecting a method for 
function 'summary': Error in glht.matrix(m.lapf, linfct = K %*% X) :
   'ncol(linfct)' is not equal to 'length(coef(model))'

Performing exactly the same routine with the same data on a logistic 
model with family=binomial does not give this error message.
So my question is, what am I missing here?

Thanks, for any possible input

-- 
Simon Tragust
Animal Ecology I
University Bayreuth
D-95440 Bayreuth

+49 921 552464


	[[alternative HTML version deleted]]

--- most of the message deleted....

Performing exactly the same routine with the same data on a logistic
model with family=binomial does not give this error message.
So my question is, what am I missing here?

Thanks, for any possible input

------------

I'm not a user of multcomp, so will follow with one question.  Results from
coxph and
coxme do not contain an intercept column.  Could this be the problem?

  A second possiblity is that coxme will in some cases return a bdsmatrix object
as the
variance matrix (to save space).  In such a case  "fit$var <-
as.matrix(fit$var)" might
correct the problem.

Terry Therneau