|Date: ||Sun, 17 Sep 2006 23:02:35 -0700|
|Reply-To: ||David L Cassell <davidlcassell@MSN.COM>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||David L Cassell <davidlcassell@MSN.COM>|
|Subject: ||Re: Weighted regression|
|Content-Type: ||text/plain; format=flowed|
b.bradbury@UNSW.EDU.AU wrote back:
>Thanks for the comments (and the SUGI paper).
>I was reading up on how Stata does this problem. As far as I can see,
>they simply apply the Huber-White covariance estimation method when
>using sampling weights. (This is for their standard regression
>procedure - maybe they have some other survey-specific procedure?). Any
>comments on the appropriateness of this?
I think that is just for generalized least-squares regression with
analytic weights, not for their survey sampling methods. SAS
has a similar feature in PROC REG, which is also not appropiate
for survey sample analysis.
>More generally, are there any papers which compare SAS, Stata and SPSS
>features for doing regression with sampling weights?
There are a few. You might look up the poster by Arlene Siller and
They also cover SUDAAN.
The bottom line is that everyone uses some manner of linearization
to get variance estimates, so they all give really, really similar
results in simple situations.
I think there was a SUGI 25 poster on this topic too. That may be
the very first paper on the topic.
You have a simple model: two strata, equal weights within strata.
It probably will not matter which software you use to do the
analysis, as long as they *have* survey sample analysis tools.
Of course, I like SAS.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Get real-time traffic reports with Windows Live Local Search