Date: Tue, 16 Dec 2008 16:48:47 -0500
Reply-To: Kevin Viel <citam.sasl@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Kevin Viel <citam.sasl@GMAIL.COM>
Subject: Re: Regression: do you always need main effects with interactions?
We have covered this before. Shawn has cited his mentor, Prof. Westfall:
"Peter Westfall teaches it as the effect heredity principle."
I, like Mary, have cited Prof. Kleinbaum and called it a Hierarchically
Well-Formulated Model. In the very least, I believe including the lower
order terms makes the results robust against different coding: 0/1 verus
1/2 or -1/1, for instance. Note that I said lower order; the question can
be extended to exponentials: x**n, where x-x**(n-1) should also appear in
I can also recall a conversation I had with a committee member, Prof. Drews-
Botsch. State your model a priori and that is the one you use to test your
hypothesis. After you run the regression, you do not drop covariates or
add covariates-that is left for another study, you simply interpret your
results. Choose your model with the best knowledge and science available
In the same vein, it might be better to potentially "over-control"