Date: Tue, 7 Jun 2005 12:14:11 -0500
Reply-To: Robin High <robinh@UNLSERVE.UNL.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Robin High <robinh@UNLSERVE.UNL.EDU>
Subject: Re: ANOVA Questions
In-Reply-To: <c12d2b07acf0d877e28372975dfa4274@sbcglobal.net>
Content-Type: TEXT/PLAIN; charset=US-ASCII
On Mon, 6 Jun 2005, Elaine Pierce wrote:
> Hello,
>
> 1. My understanding was that for a fixed effect, multi-way ANOVA (as in
> ANCOVA) if you find a significant interaction (lack of parallelism) the
> analysis cannot proceed. Instead, you must stratify your data by levels
> of your interacting variable and analyze each level separately
> (although the reduction in power is a bummer).
..to the Grad Student in over her head
A quick response to your first question.. Assuming you are using ANCOVA
for the analysis, the interaction you mention is commonly stated in
various references on the subject. However, for anyone running PROC MIXED
with SAS, you should look at the explanation, formulas, and sample SAS
code provided in the first 2 chapters of "Analysis of Messy Data Vol III:
Analysis of Covariance" by Milliken and Johnson (Chapman & Hall).
Namely, the LSMEANS statement allows you to compare group means for any
value of the covariate:
LSMEANS group / diff at (cov)=( #low_value );
LSMEANS group / diff;
LSMEANS group / diff at (cov)=(#high_value);
The default computations are made at the mean of the covariate (the second
statement). The "no interaction" assumption exists because group
comparisons are assumed to be made at the covariate mean. If there is an
interaction, the default comparisons are not appropriate, since
differences may exist at the low value and not at the high value of a
covariate (or the other way). With the ODS it is easy to get data you can
plot to show this interaction visually.
Robin High
Univ. of Oregon
