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 at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: using weights with -svy- commands


From   Jeph Herrin <stata@spandrel.net>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: using weights with -svy- commands
Date   Wed, 14 Mar 2012 13:56:09 -0400

Thanks. To be clear: I am summarizing the data at the respondent level into something like a daily average. The exact algorithm for summation is rather complex (see http://www.ncbi.nlm.nih.gov/pubmed/18714120) but I would like to give more weight to respondents that have more days reported.

thanks,
Jeph


On 3/14/2012 12:49 PM, Stas Kolenikov wrote:
[aweight] would have been called for if you summarized the 10k
functional data points into a single number, an aggregate indicator of
the level of physical activity. If you are analyzing all of the
available motion data, I don't see any need to specify any additional
weights. Depending on what exactly you are going to undertake in your
analysis, you may or may not need the [pweights] or -svy- settings
(and you can always compare the results using Hausman test, see
Pfeffermann 1993
http://www.citeulike.org/user/ctacmo/article/1036965).

On Wed, Mar 14, 2012 at 11:48 AM, Jeph Herrin<stata@spandrel.net>  wrote:
All,

This seems like a question that has been asked&  answered before, but I
can't find it in the archives.

I have data which was was collected using a stratified survey design
(NHANES, http://www.cdc.gov/nchs/nhanes.htm) which I have been analyzing
using -svy- commands.

However, for a random subset of survey respondents, additional data i
collected - up to 10080 motion counts collected each minute for up to 7
days. I have summarized these motion counts various ways for each
respondent, but since not all respondents have the same reporting period,
there is varying degrees of precision - some respondents have only a few
thousand counts, some have the full 10k - and most importantly, I have found
that the reporting period is associated with some of my key independent
variables (in particular, overweight children have shorter reporting periods
than normal weight children). So I want to incorporate the different
precision of the summary outcome into my regression models.

In most linear regression models I would achieve this by weighting each
observation by the inverse of the variance of the dependent variable
measurement using -aweights-. However, -svy reg - does not support weights.

My questions are all due to the fact that I am a casual user of complex
survey data. Is this because of a theoretical restriction due to the complex
designs? Or is it a mechanical problem that I can somehow circumvent? Or is
it possible with Stata, only I am overlooking the solution?

thanks,
Jeph
*
*   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/



*
*   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