Greetings,
I'm analyzing data with one fixed (TMT) and one random (BATCH) factor using
lme() and I have a basic question on results interpretation. Specifically, I'm
interested in assessing the difference between levels of TMT. I've looked at
the output from
summary(lme.object) and have used multicomp.default to do this. From the lme
summary (fixed effects), we see a significant difference in changing from TMT A
to TMT C (p = 0.0282) and between TMT A and TMT D (p = 0.0023), yet the
multicomp.default call
comparing all treatments to TMT A shows only 1 significant difference, D to A.
I thought I should get identical results with these two calls. Can anyone
point out why these are different? I'm running "Trial Professional Edition
Version 6.1.2 Release 1
for Microsoft Windows : 2002"
Thanks,
Sally
> options()$contrasts
[1] "contr.treatment" "contr.poly"
prot.lme_lme(Mean~TMT,random=~1|BATCH,data=Exp,method="ML")
Value Std.Error DF t-value p-value
(Intercept) 37.63795 3.211696 232 11.71903 <.0001
TMTB -2.94970 2.006021 232 -1.47042 0.1428
TMTC -4.42930 2.006021 232 -2.20800 0.0282
TMTD -6.19176 2.006021 232 -3.08659 0.0023
params <- fixef(prot.lme)
nparams <- length(params)
muest <- cbind(rep(1, nparams), contr.treatment(nparams))
muvec <- as.numeric(muest %*% params)
names(muvec) <- c("A","B","C","D")
vmat <- muest %*% prot.lme$varFix %*% t(muest)
prot.multi_multicomp.default(x=muvec, vmat=vmat, df.residual =
as.numeric(prot.lme$fixDF$X[1]),comparison="mcc",control=1,method="best",alpha=.05)
> prot.multi
95 % simultaneous confidence intervals for specified
linear combinations, by the simulation-based method
critical point: 2.3514
response variable:
simulation size= 12616
intervals excluding 0 are flagged by '****'
Estimate Std.Error Lower Bound Upper Bound
B-A -2.95 1.99 -7.63 1.730
C-A -4.43 1.99 -9.11 0.248
D-A -6.19 1.99 -10.90 -1.510 ****
>
|