Date: Fri, 1 Aug 2003 09:32:24 -0500
Reply-To: Paul R Swank <Paul.R.Swank@UTH.TMC.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Paul R Swank <Paul.R.Swank@UTH.TMC.EDU>
Subject: Re: GLM output of repeated measurements
Different tests, partially different assumptions. The multivariate tests
assumes multivariate normality whereas the univariate test assumes
sphericity. If the sphericity test is violated then the univariate tends
toward the liberal side and one tends to see more significant univariate
tests than multivariate. Sphericity has to do with the structure of the
variance covariance matrix. Proc Mixed gives one more flexibility wth the
var-cov matrix than does univariate GLM. Thus, which one is more believable
depends on the assumptions.
Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
UT Health Science Center at Houston