[R-sig-ME] Confidence intervals of variance components

Julien Martin julien.martin2 at usherbrooke.ca
Thu Mar 26 18:29:59 CET 2009

I want to estimate standard error or confidence intervals for the 
estimate of varaince associated with a random effect.
With "lme", I could use "intervals" however, this does not work with "lmer"
I tried "pvals.fnc" from "languageR" package that should provide what I 
However, I find stange that the MCMCmean is quite different from the 
model estimate and that the model estimate is not even included in the 
95CI provided.
Is it normal? Do I misinterpret the output?
Is there another way to obtain CI or SE of varaince (or Std.Dev) of 
random effects with lmer?

Here an example
Estimate of Standard Deviance for Subject (Intercept) is not even close 
from the MCMCmedian
 >fm2 <- lmer(Reaction ~ Days + (1|Subject) + (0+Days|Subject), sleepstudy)
    Groups        Name Std.Dev. MCMCmedian MCMCmean HPD95lower HPD95upper
1  Subject (Intercept)  29.8219    15.4880  15.5692     8.6060    22.9435
2  Subject        Days   7.1285     6.4052   6.5464     4.1641     9.1933
3 Residual              30.4337    26.8669  26.9354    23.8227    30.1357


More information about the R-sig-mixed-models mailing list