Date: Thu, 22 Jan 2004 09:07:27 -0500
Reply-To: Bert Bishop <bert_bishop@EUDROAMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Bert Bishop <bert_bishop@EUDROAMAIL.COM>
Organization: Ohio State University
Subject: Variance Component 0 in split-plot using Proc Mixed.
Content-Type: text/plain; charset=us-ascii
I am analyzing a set of data from a split-plot experiement using SAS's
proc mixed. The resultant "covariance parameter estimate" for
the whole plot error term is 0. Someone told me that SAS suggests
removing the term from the model. With ddfm=satterth or with the term
removed I get the same results.
In both cases the denominator degrees of freedom for testing the whole
plot factor is the same as that used to test the split-plot factor.
This is inconsistent with my understanding of such things as I
understnad them from Kempthorne's 1977 article "Why Randomize". I'm
more inclinded to use
ddf=n1,n2,n2 where n1 is the number of degrees of freedom that one would
classically use i.e., (b-1) * (t -1) where b is the number of blocks and
t is the number of main plot factor levels. Here n2 is the denominator
degrees of freedom for testing the split-plot factor.
Can someone help me understand an argument for using the suggested
method rather that the one I used? Is my approach erroneous? Thanks in