[R] Using GEE with sample weights

[Thomas Lumley]

On Mon, Apr 30, 2012 at 10:05 PM, Stefán Hrafn Jónsson <shj107 at gmail.com> wrote:
> Dear R community
>
> I am using the gee package to run logistics regression on paired cases from
> a panel sample.
>
> We are getting request from a reviewer to use sample weights to account for
> different level of attrition.
>
> After searching the documentation I am unable to find a way to incorporate
> sample weights into the gee formula. I found a way to incorporate precision
> weight but I understand that I cant use that function for sample weight.

For gee it doesn't matter whether you have sampling weights or
precision weights:
  -  the point estimate is the same (as usual for linear estimating functions)
  -  the standard errors are not model-based and they are also the
same (the standard error estimates are the same as the
Horvitz--Thompson + linearization estimates that you would use for
sampling)

If you wanted to use other features of a complex sampling design you'd
need different software, but not just for per-observation sampling
weights.

    -thomas

-- 
Thomas Lumley
Professor of Biostatistics
University of Auckland