Date: Fri, 21 Sep 2001 10:10:35 -0700
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "David L. Cassell" <Cassell.David@EPAMAIL.EPA.GOV>
Subject: Re: Urgent: Proc MEANS and SURVEYMEANS problem!
Content-type: text/plain; charset=us-ascii
Hamani.Elmaache wrote [in part]:
> Hi David and thank you for all you did for my.
You're welcome. But please, direct your questions to the SAS-L list
instead of my work address. There is far more guidance available on
the mailing list.
> To calculate some mean I used both Proc MEANS and SURVEYMEANS and
> I got the same mean but NOT the same Std Error. Look at these two tables:
But I don't need to look at the tables to guess that they would have
different variances. They're SUPPOSED to have different variances.
PROC MEANS assumes [a simple random sample of] independent, identically
distributed values. It uses a variance estimator which is quite different
from PROC SURVEYMEANS, which uses a Taylor expansion methodology to get a
very tight approximation to the variance of the sampling design.
If you really do have a sampling design [since this is a follow-up to a
post on a stratified sample from a finite population], and not a sequence
independent observations from a conceptually infinite population, then you
should be using PROC SURVEYMEANS. And you should be happy that it is
giving you variances different from PROC MEANS, since you know those
David Cassell, CSC
Senior computing specialist