Re: [R] bVar slot of lmer objects and standard errors

From: Spencer Graves <>
Date: Thu 29 Dec 2005 - 22:57:47 EST

          Have you received a satisfactory reply to this post? I haven't seen one. Unfortunately, I can't give a definitive answer, but I can offer an intelligent guess. With luck, this might encourage someone who knows more than I do to reply. If not, I hope these comments help you clarify the issue further, e.g., by reading the source or other references.

          I'm not not sure, but I believe that
lmertest1@bVar$schoolid[,,i] is the upper triangular part of the covariance matrix of the random effects for the i-th level of schoolid.   The lower triangle appears as 0, though the code (I believe) iterprets it as equal to the upper triangle. More precisely, I suspect it is created from something that is stored in a more compact form, i.e., keeping only a single copy of the off-diagonal elements of symmetric matrices. I don't seem to have access to your "nlmframe", so I can't comment further on those specifics. You might be able to clarify this by reading the source code. I've been sitting on this reply for several days without finding time to do more with it, so I think I should just offer what I suspect.

          The specifics of your question suggest to me that you want to produce something similar to Figure 1.12 in Pinheiro and Bates (2000) Mixed-Effects Models in S and S-Plus (Springer). That was produced from an "lmList" object, not an "lme" object, so we won't expect to get their exact answers. Instead, we would hope to get tighter answers available from pooling information using "lme"; the function "lmList" consideres each subject separately with no pooling. With luck, the answers should be close.

          I started by making a local copy of the data:

OrthoFem <- Orthodont[Orthodont$Sex=="Female",]

          Next, I believe to switch to "lme4", we need to quit R completely and restart. I did that. Then with the following sequence of commands I produced something that looked roughly similar to the confidence intervals produced with Figure 1.12:

fm1OrthF. <- lmer(distance~age+(age|Subject), data=OrthoFem)

fm1.s <-  coef(fm1OrthF.)$Subject
fm1.s.var <- fm1OrthF.@bVar$Subject
fm1.s0.s <- sqrt(fm1.s.var[1,1,])
fm1.s0.a <- sqrt(fm1.s.var[2,2,])
fm1.s[,1]+outer(fm1.s0.s, c(-2,0,2))
fm1.s[,2]+outer(fm1.s0.a, c(-2,0,2))

	  hope this helps.  	
	  Viel Glueck.
	  spencer graves

Ulrich Keller wrote:

> Hello,
> I am looking for a way to obtain standard errors for emprirical Bayes
estimates of a model fitted with lmer (like the ones plotted on page 14 of the document available at

Harold Doran mentioned
( that the posterior modes' variances can be found in the bVar slot of lmer objects. However, when I fit e.g. this model:
> lmertest1<-lmer(mathtot~1+(m_escs_c|schoolid),hlmframe)
> then lmertest1@bVar$schoolid is a three-dimensional array with dimensions (2,2,28).
The factor schoolid has 28 levels, and there are random effects for the intercept and m_escs_c, but what does the third dimension correspond to? In other words, what are the contents of bVar, and how can I use them to get standard errors?
> Thanks in advance for your answers and Merry Christmas,
> Uli Keller
> ______________________________________________
> mailing list
> PLEASE do read the posting guide!

Spencer Graves, PhD
Senior Development Engineer
PDF Solutions, Inc.
333 West San Carlos Street Suite 700
San Jose, CA 95110, USA <>
Tel:  408-938-4420
Fax: 408-280-7915

______________________________________________ mailing list
PLEASE do read the posting guide!
Received on Thu Dec 29 23:07:10 2005

This archive was generated by hypermail 2.1.8 : Fri 30 Dec 2005 - 00:00:05 EST