Date: Tue, 11 Dec 2001 19:15:30 -0800
Reply-To: Scott Marion <smarion@VIMS.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Scott Marion <smarion@VIMS.EDU>
Organization: http://groups.google.com/
Subject: PROC MIXED code for split-plot ANOVA?
Content-Type: text/plain; charset=ISO-8859-1
SAS users,
I'm seeking help with the appropriate SAS code for an analysis. Any
assistance will be greatly appreciated. The analysis is a split-plot
ANOVA design with two between-subjects factors and one within-subjects
factor, but since the term 'split-plot' seems to mean different things
to different people, I'll specify it as clearly as possible:
The data are abundances of animals in vegetated plots at two sites.
There are three plot sizes, with three replicates of each plot size at
each site for a grand total of 18 plots. In every plot, two estimates
of animal abundance are made: one near the edge and one near the
interior of the plot. So in split-plot parlance SITE and SIZE are
between-plots factors, and POSITION is the within-plots factor. All
three factors are fixed. There are no missing cells. I have heard
that in a split-plot design PROC GLM does not estimate standard errors
for the within-subjects factor and its interaction, so I've been
pointed toward PROC MIXED.
Considerations:
(1) I'd like to include a covariate (plant density).
(2) If the SITE*SIZE interaction is significant, I plan to analyze
each site separately. How would the code change?
(3) If Proc Mixed is the answer, is it capable of creating an output
table with the analysis results (as can be accomplished in Proc GLM
with the 'outstat=' option)?
(4) When structuring the dataset, should I include a unique
identifier for each plot (as in the first case below), or will the
procedure interpret the correct relationship between the variables if
instead I designate replicates 1, 2, and 3 within each site (as in the
second set):
|
PLOT# SITE SIZE POSITION | SITE SIZE REP POSITION
1 1 small edge | 1 small 1 edge
1 1 small interior | 1 small 1 interior
2 1 small edge | 1 small 2 edge
2 1 small interior | 1 small 2 interior
3 1 small edge |vs. 1 small 3 edge
3 1 small interior | 1 small 3 interior
4 1 med edge | 1 med 1 edge
4 1 med interior | 1 med 1 interior
5 1 med edge | 1 med 2 edge
etc... |
THANKS MUCH for any help; I could wander around in the manuals for a
long time without resolving this.
-Scott Marion
smarion@vims.edu
[posted to comp.soft-sys.sas and sci.stat.consult]
| 
