[R] how to fix the level-1 variances in lme()?

[Prof Brian Ripley]

On Thu, 29 Sep 2005, Mike Cheung wrote:

> Dear all,
>
> Edmond Ng (http://multilevel.ioe.ac.uk/softrev/reviewsplus.pdf) provides
> an example to fit the mixed effects meta-analysis in Splus 6.2. The
> syntax is:
> lme(fixed=d~wks, data=meta, random=~1|study, weights=varFixed(~Vofd),
> control=lmeControl(sigma=1))
> where d is the effect size, study is the study number, Vofd is the
> variance of the effect size and meta is the data frame.
>
> "sigma=1" is required to constrain the level 1 variance in applying
> mixed-effects models in meta-analysis.
>
> In Splus 6.1, I found that the help manual of nlme includes "sigma" as
> an optional argument in lmeControl() to fix the within-group standard
> error during the optimization.
>
> However, both Pinheiro and Bates (2000, p.476) and the help manual of
> the nlme package in R for Version 3.1-65
> (http://cran.r-project.org/doc/packages/nlme.pdf, p.172) do not include
> "sigma" as an argument for lmeControl().
>
> I would like to know how I could fix the level-1 variances as known
> values in lme().

Use S-PLUS and nlme >= 3.3 (as I recall).

R's and S-PLUS's nlme are not the same (they diverged from a common base 
around 1998).  It is a while (2 years?) since I needed this capability, 
but at the time it was not possible in R's lme.

Since you have access to S-PLUS, why not use it?

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595