[R] sampling weights for multilevel models

[Thomas Lumley]

On Mon, Jun 11, 2012 at 7:00 AM, Tamara <petrova.toma at gmail.com> wrote:
> Dear all,
>
> I am struggling with a problem which I have been reading on the forums about
> and it did not seem to me that there is a precise answer to my question.
> However, I still hope there is one.
>
> I am working with  http://timss.bc.edu/ PIRLS   data and trying to conduct
> multilevel analysis. There are different weights for each level of analysis
> in the PIRLS dataset (e.g.  there is a school weight, class weight, student
> weight).
> Is there a function in R which would let me use different weights for
> different levels of my model?
> If yes, which package contains it?
>

As far as I know there is no function that does what you want.   In
particular, lme() and lmer() don't work correctly with sampling
weights.

It does depend on why you want a multilevel model.  If you are
primarily interested in the mean model and the variance components are
just needed to get appropriate standard errors, then you can use the
svyglm() function in the survey package to fit a linear regression
with appropriate standard errors.   On the other hand, if you are
interested in estimating the variance components for their own sake,
you need some other software.

I do have longer-term plans to add multilevel modelling capabilities
to the survey package, but it's harder than it may appear.


   -thomas

-- 
Thomas Lumley
Professor of Biostatistics
University of Auckland