Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: Multi-level modeling, PCA and wealth scores


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: Multi-level modeling, PCA and wealth scores
Date   Mon, 22 Aug 2011 15:35:44 -0500

On Mon, Aug 22, 2011 at 2:28 PM, Nanlesta Pilgrim <nanlesta@gmail.com> wrote:
> A colleague of mine asked for my help on the following and would be
> most grateful for any recommendations you may offer.
>
> "I have a question regarding creation of wealth measures. We are
> trying to create a community level measure of wealth to be used in a
> multi-level model. I have used the pca command (with household assets)
> to create a pca score for each household. Now, I want to weight that
> score at the household level and then just determine the mean
> household factor load score for each county and save that value for
> every household by county. Is that possible to do?- Is that even the
> right thing to do for a multi-level model? Or do I just input the
> weighted household level value and the multilevel commands
> automatically generates a county-level score for this variable?"


I think you should be more explicit at how these scores be formed by
the software (see
http://www.newciv.org/pic/nl/artpic/10/1953/SidneyHarris_MiracleWeb.jpg).
If you DO know that the software will create the scores for you, then
the next step is to figure out if these are the right scores (see our
paper on this, doi:10.1111/j.1475-4991.2008.00309.x, and the earlier
expanded working paper versions of it). If you don't know, assume the
software won't create the scores. I'd suggest you run analysis
separately by country, as there will be mismatch in the available
variables between different countries. Then you could pull out the
scores, but then you have a problem of measurement error that would
bias the coefficients toward zero. The best solution would be to adopt
a multilevel structural equation modeling framework for this, which
means figuring out GLLAMM.


-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index