[R-sig-ME] [R] Extract Residual Variance from lmer using VarCorr

[Douglas Bates]

It is important when discussing the behavior of lmer and other
functions in the lme4 package to state the version of the package that
you are using.  The package changes as I experiment with the
computational methods.

Using version 0.999375-27 of the lme4 package you can see the following

> fm1 <- lmer(Reaction ~ Days + (Days|Subject), sleepstudy)
> str(VarCorr(fm1))
List of 1
 $ Subject: num [1:2, 1:2] 612.1 9.6 9.6 35.1
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : chr [1:2] "(Intercept)" "Days"
  .. ..$ : chr [1:2] "(Intercept)" "Days"
  ..- attr(*, "stddev")= Named num [1:2] 24.74 5.92
  .. ..- attr(*, "names")= chr [1:2] "(Intercept)" "Days"
  ..- attr(*, "correlation")= num [1:2, 1:2] 1 0.0655 0.0655 1
  .. ..- attr(*, "dimnames")=List of 2
  .. .. ..$ : chr [1:2] "(Intercept)" "Days"
  .. .. ..$ : chr [1:2] "(Intercept)" "Days"
 - attr(*, "sc")= Named num 25.6
  ..- attr(*, "names")= chr "sigmaREML"

so the value that you want for the residual standard deviation is

attr(VarCorr(fm1), "sc")

An alternative is to examine

> fm1 at deviance
          ML         REML         ldL2        ldRX2      sigmaML    sigmaREML
1.751986e+03 1.743628e+03 7.595997e+01 8.276128e+00 2.544924e+01 2.559182e+01
       pwrss         disc         usqr         wrss          dev         llik
1.165795e+05 9.888088e+04 1.769928e+04 9.888024e+04           NA           NA
     NULLdev
          NA

and notice that there are two elements called "sigmaML" and "sigmaREML".

Both these methods are sub-optimal in that they rely on a particular
version of the fitted model object.  I wrote an extractor named
"sigma" to obtain the estimate of the standard deviation but did not
export it because I am not convinced that this is a good name for such
an extractor.  Suggestions for the name of an extractor (or comments
that I have missed the obvious generic function) and what it should
return are welcome.

I have cc:'d the R-SIG-Mixed-Models mailing list as it is the more
likely list for help on such questions.


On Tue, Nov 4, 2008 at 1:14 PM, Benjamin Michael Kelcey
<bkelcey at umich.edu> wrote:
>
> I am trying to extract the residual variance from an lmer object. I have
> tried using VarCorr() which allows me to see the estimate but can not save
> it by itself.
>
> How can I get this from VarCorr() or is there another function available for
> this?
> Thanks,
> ben
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>