[R] warning message: coxme with package multcomp

[Simon Tragust]

Hi
I'm encountering an error/warning when doing multiple comparisons with 
the package multcomp on a coxme model.

My data:
I'm looking at the removal of brood from the nest according to three 
treatments I applied on the brood.
The brood and the workers caring about the brood in the nest, belonged 
to different colonies.

Factor: treatment (3 levels: tx,uv,meta)
Random effect 1: origin of brood
Random effect 2: origin of workers

modelling this with
pp.coxme<-coxme(Surv(day,status)~treatment+(1|r.brood)+(1|r.worker),data=p)

gives me first this first warning message:
Warning message:
In coxfitfun(x, y, strata = strata, offset = offset, init = ifixed,  :
   Loglik converged before variable  1 ; beta may be infinite.

but the output looks ok:
pp.coxme
Cox mixed-effects model fit by maximum likelihood
   Data: p;  Subset: (brood == "pupae")
   events, n = 3, 48
   Iterations= 2 42
                     NULL Integrated    Fitted
Log-likelihood -11.07857  -9.559384 -9.557399

                   Chisq df       p   AIC   BIC
Integrated loglik  3.04  4 0.55143 -4.96 -1.36
  Penalized loglik  3.04  2 0.21879 -0.96  0.84

Model:  Surv(day, status) ~ treatment + (1 | r.brood) + (1 | r.worker)
Fixed coefficients
                    coef    exp(coef) se(coef)     z    p
treatmenttx -22.1625984 2.370859e-10  0.00000  -Inf 0.00
treatmentuv  -0.6931777 4.999847e-01  1.22475 -0.57 0.57

Random effects
  Group    Variable  Std Dev      Variance
  r.brood  Intercept 0.0199992574 0.0003999703
  r.worker Intercept 0.0199992574 0.0003999703


to see if removal is overall dependend upon treatment i use the package 
multcomp as following with the second warning message:

pp.coxme.glht<-glht(pp.coxme,linfct=mcp(treatment="Tukey"))

summary(pp.coxme.glht,test=Chisqtest())

          General Linear Hypotheses

Multiple Comparisons of Means: Tukey Contrasts


Linear Hypotheses:
                Estimate
tx - meta == 0 -22.1626
uv - meta == 0  -0.6932
uv - tx == 0    21.4694

Global Test:
   Chisq DF Pr(>Chisq)
1 71.94  1  2.217e-17
Warning message:
In cov2cor(covm) :
   diag(.) had 0 or NA entries; non-finite result is doubtful

and all pairwise comparisons between treatments are then not possible:

summary(pp.coxme.glht,test=adjusted("Westfall"))
Error in if (!chkcorr(corr)) stop(sQuote("corr"), " is not a correlation 
matrix") :
   missing value where TRUE/FALSE needed
In addition: Warning messages:
1: In cov2cor(covm) :
   diag(.) had 0 or NA entries; non-finite result is doubtful
2: In cov2cor(covm) :
   diag(.) had 0 or NA entries; non-finite result is doubtful


I think this is because one of the levels of the factor treatment has 
only cencored data. Using the function coxph instead of coxme gives me 
the same first warning message on the model, but not the second warning 
/ error message doing multiple comparisons of the treatment levels.
My question is: Am I doing something fundamentally wrong or is it not 
possible to get multiple comparisons on a coxme model if one of the 
levels of a factor has only cencored data?.

Many thanks in advance!
Simon

-- 
Simon Tragust
Animal Ecology I
University of Bayreuth