Date:         Thu, 30 Jul 2009 15:55:05 -0400
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: Interaction
Content-Type: text/plain; charset=ISO-8859-1

On Wed, 29 Jul 2009 07:58:01 -0400, Billy Thompson
<bill.thompson@BROOKS.AF.MIL> wrote:

>Hopefully some of you stats ‘gurus’ can help me solve my problem.
>
>I have a project in which the PI is interested in the interaction effects
>among certain variables and not the main effects.    The problem results
>that when I run  the ‘full’ factorial model with all effects the
>interaction effects are non significant and the main effects are
>significant.  If I then remove the significant main effects the
>interaction effects then become significant.
>
>Is it appropriate to run the model with interaction effects only, thus
>ignoring the main effects?


Shawn Haskell calls it the heredity principle(1), I call in a
heirarchically well-formulated model.

(1) http://www.listserv.uga.edu/cgi-bin/wa?A2=ind0810D&L=sas-l&P=R1693

One thing to establish is *why* you are running the model.  What is the
variable of interest?  What does the science say the model should be?

Having diverted, I will say no; you should include lower order terms.

HTH,

Kevin
Kevin