Steven--
It strikes me as odd that the ref cited does not discuss the HLM
software package by Raudenbush and Bryk and Congdon, which can handle
a variety of weights at different levels, I believe. Reminds me of a
very interesting unanswered question from 2 years ago:
http://www.stata.com/statalist/archive/2005-10/msg00079.html
(sample weights on kids, and analytic weights on test scores).
If anyone has insight into that old question,
I'd be interested to hear it!
On 11/24/07, Steven Joel Hirsch Samuels <[email protected]> wrote:
> I believe that the questions of whether to include weights and of
> which weights to include, must be decided on a case-by-case basis. I
> might not use supplied weights in analyses of sub-populations or for
> certain analytic studies. I do not know that the reasons for my
> decisions not to weight would be considered prevalent concepts. In
> MLM, estimated variance components might be important, and I have not
> thought about weighting in connection with these. Certainly, a major
> reason that scientists do not consider whether to weight multilevel
> analyses is that their software cannot do so.
>
> The reference below states that the following MLM packages can
> incorporate sampling weights: MPLUS, LISREL, MLWIN, and the
> contributed Stata program GLLAMM. Construction of the weights for
> these programs is not straight-forward, because different factors
> apply at each level of a hierarchical model. The reference describes,
> and contains links to, SAS and Stata programs that construct
> composite weights for two-level analyses.
>
> Software to Compute Sampling Weights for Multilevel Analysis by Kim
> Chantala, Dan Blanchette, and C. M.Suchindran, 2006 ( http://
> www.cpc.unc.edu/restools/data_analysis/ml_sampling_weights/Compute%
> 20Weights%20for%20Multilevel%20Analysis.pdf )
>
>
> Steven Samuels
>
> > To Steven Samuel:
> >
> > First, let me apologize myself for "interfering" into
> > your fascinating conversation. I wonder if it is your
> > opinion that the same concepts are prevalent when some
> > scientists do not take into account sampling weights
> > when analyzing cluster samples arising from household
> > survey with multilevel regression techniques, or it is
> > simply that techniques for dealing with sampling
> > weights are not available for multilevel regression?
> >
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
