Date:         Wed, 8 Dec 2004 21:24:16 -0500
Reply-To:     "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "DePuy, Venita" <depuy001@DCRI.DUKE.EDU>
Subject:      Re: no main effect but interaction effect?
Comments: To: Duck-Hye Yang <dyang@CHAPINHALL.ORG>
Content-Type: text/plain; charset="iso-8859-1"

Even though the main effects do not *appear* significant in the output, you
need to include them.
Think of it this way - the effect of var1 is shown in var1 and var1*var2 . .
so if var1*var2 has a significant p value and var1 does not, it's really
because the effect of var1 is split up into the two lines.

Hm, maybe that's not a textbook explanation, but it's how I keep it straight
in my head :)

In 99% of cases - ie anytime you're not in an obscure specialized case where
you are supposed to do it another way (I think something on SAS-L about a
year ago mentioned one case?) - you need to include ALL the main effects -
whether apparently significant or not - for the interaction effects you're
using.

Hope this helps
Venita


-----Original Message-----
From: Duck-Hye Yang
To: SAS-L@LISTSERV.UGA.EDU
Sent: 12/8/2004 9:02 PM
Subject: no main effect but interaction effect?

Hello,
I just got sas output from a proportional hazard model.
No significant main effects of two variables were found, but one of
interaction terms was significant.
I wonder how it can happen?  Is it possible to have no significant main
effect but a significant interaction effect?

Thanks,
Duckhye