Date: Tue, 8 Jan 2008 07:43:27 -0800
Reply-To: RolandRB <rolandberry@HOTMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: RolandRB <rolandberry@HOTMAIL.COM>
Subject: Re: Q: another approach for Satterthwaite
Content-Type: text/plain; charset=ISO-8859-1
On Jan 8, 3:20 pm, Warren.Schlec...@TPWD.STATE.TX.US (Warren
> Since the t-test is a simplification of the F-test, and since with a
> single degree of freedom, t2=3DF, I would suggest you run only F-tests,
> and don't worry about a t-test, then make the adjustment from F to t in
> the 1 df case at the end (or report the F-test).
> This sounds like what you're suggesting, and it makes sense to me.
> Warren Schlechte
The way I implemented it was to use proc glm with hovtest=Levene (can
be changed) in the MEANS statement to get the homogeneity of variance.
I then rerun proc glm and if the HoV test was significant I use "/
hovtest=Levene welch" in the means statement or if the HoV test was
not significant then use means without them.