Date: Thu, 3 Jan 2008 04:22:41 -0800
Reply-To: RolandRB <rolandberry@HOTMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: RolandRB <rolandberry@HOTMAIL.COM>
Subject: Re: How do I code the Satterthwaite approximation?
Content-Type: text/plain; charset=ISO-8859-1
On Jan 3, 12:10 pm, peterflomconsult...@mindspring.com (Peter Flom)
> RolandRB <rolandbe...@HOTMAIL.COM> asked
> >Thanks, that is helpful and gives rise to another question. In
> >deciding whether to use the Satterthwaite approximation or not, which
> >value drives this and what should the threshhold value be. I see both
> >"significant differences in the mean value" in that article you
> >provided a link for and "equality of variances" in another article
> >that can be linked to below. Which should I use?
> Hi Roland
> The Sattherthwaite approximation is needed when the variances are different; the t-test is used to determine if the means are different
> However, this raises the question of how to tell if the variances are, in fact, different. There are different tests for that difference, and there are disagreements as to how big the difference has to be before a correction is needed. The standard methods evolved in the pre-computer era, when using the correction was a big pain in the ass. Nowadays, it's a very fast program, even for large data sets, so I would err on the side of using it.
> I'm not entirely sure why you need to code this, but if you do, one thing to keep in mind is whether it is better to use a correction to the t-test rather than some data-driven technique.
> If computers had been invented before Fisher was born, the whole field of data analysis would look very different!
You mentioned a correction to the t-test. Are you thinking maybe the
COCHRAN option in the PROC TTEST statement or something else? It seems
from what I have read that the Cochran and Cox approximation can give
a very different value to the Satterthwaite approximation and I am
wondering why that should be the case.