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: Pweights in SUREG?


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Pweights in SUREG?
Date   Thu, 9 Jun 2011 12:36:51 -0400

Benjamin Harris <bch4@sfu.ca>:
Welcome! Don't make fweights out of pweights unless you really are
sure it's the right thing to do.  Better to ignore the weights than to
pretend you have a census (one potential outcome of making fweights
out of pweights), and better than that to use aweights and fail to
correct for heteroskedasticity or clustering. Do you have different
explanatory variables in different regressions?  If not, -suest- is
just as good for you as -sureg-. Note aweights plus clustering is 99%
of what -svy- does (aweights plus robust=pweights), and -suest- has an
option to cluster:

webuse nhanes2, clear
qui reg bpsystol age sex i.race
est sto s
qui reg bpdiast age sex i.race
est sto d
suest d s
sureg (bpdiast age sex i.race) (bpsystol age sex i.race)
qui reg bpsystol age sex i.race [aw=finalwgt]
est sto s
qui reg bpdiast age sex i.race [aw=finalwgt]
est sto d
egen c=group(strata psu)
suest d s, cluster(c)
svy:reg bpsystol age sex i.race
svy:reg bpdiast age sex i.race

On Wed, Jun 8, 2011 at 7:10 PM, Benjamin Harris <bch4@sfu.ca> wrote:
> Hello all,
>
> Long-time reader, first-time writer. I owe you many thanks for all the questions you've answered that other people asked. Unfortunately, I can't seem to find anyone who has asked my question now, so I'm writing in. I hope you can help me yet again.
>
> I am estimating a Seeminly Unrelated Regression model using survey data. The weights in the data are pweights, although the SUREG command only takes aweights and fweights, and there is no svy: option for the SUREG command either. I'm a bit at a loss about what to do. It has been suggested that I can create fweights from the pweights by:
> 1. Calculating average pweights within groups of relevant subsamples (e.g., men, women, etc.),
> 2. Scaling individual pweights by their relevant group,
> 3. Rounding the result to the nearest integer to get the "fweights"
> This seems like the wrong thing to do for many reasons. Most importantly, it creates an "fweights" that are 0.
>
> So, what I'd like to know is, (1) is there any reason SUREG doesn't allow for pweights or the svy option? That is, if I were smart enough, could I write my own code to make SUREG do what I want it to do, or is there a theoretical reason it doesn't do that? (2) Does anyone know of other software (such as SAS or R) that I could use to do what I need to do? And (3) are there any other ideas for work-arounds.

*
*   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