[R] glht (multcomp): NA's for confidence intervals using univariate_calpha

[Ulrich Halekoh]

Hej,

Calculation of  confidence intervals for  means
based on a model fitted with lmer 

using the package multcomp

- yields results for  calpha=adjusted_calpha
- NA's for            calpha=univariate_calpha


Example:
library(lme4)
library(multcomp)
###  Generate data
set.seed(8)
d<-expand.grid(treat=1:2,block=1:3)
e<-rnorm(3)
names(e)<-1:3
d$y<-rnorm(nrow(d)) + e[d$block] 
d<-transform(d,treat=factor(treat),block=factor(block))
##### lmer fit 
Mod<-lmer(y~treat+ (1|block), data=d)
### estimate  treatment means
L<-cbind(c(1,0),c(0,1))
s<-glht(Mod,linfct=L)
## confidence intervals
confint(s,calpha=adjusted_calpha())
#produces NA's for the confidence limits
confint(s,calpha=univariate_calpha())

#for  models fitted with lm the problem does not occur
G<-lm(y~treat+ block, data=d)
L<-matrix( c(1,0,1/3,1/3,1,1,1/3,1/3),2,4,byrow=TRUE)
s<-glht(G,linfct=L)
confint(s,calpha=adjusted_calpha())
confint(s,calpha=univariate_calpha())



multcomp  version 1.2-7
R:platform       i386-pc-mingw32                             
version.string R version 2.13.1 Patched (2011-08-19 r56767)


Regards

Ulrich Halekoh
Department of Molecular Biology and Genetics,
Aarhus University, Denmark
Ulrich.Halekoh at agrsci.dk