home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.stat-l
- Path: sparky!uunet!utcsri!skule.ecf!torn!cunews!tgee
- From: tgee@alfred.carleton.ca (Travis Gee)
- Subject: Re: interaction effects in regression models
- Message-ID: <tgee.712543970@cunews>
- Sender: news@cunews.carleton.ca (News Administrator)
- Organization: Carleton University
- References: <9207301706.AA25813@linc.cis.upenn.edu>
- Date: Fri, 31 Jul 1992 00:52:50 GMT
- Lines: 29
-
- In <9207301706.AA25813@linc.cis.upenn.edu> matsuda@LINC.CIS.UPENN.EDU (N o s t a l g i a) writes:
-
- >Supposing that there are some grounds in putting the non-significant variable
- >which turned out to interact with some other variable(s) in the model together
- >with the interaction term, how would you deal with the multicollinearity that
- >would surely ensue between that non-significant variable and the interaction
- >term?
-
- While I agree that one should stick with what is meaningful
- theoretically, Kenjiro makes a point here that should not be
- overlooked. While I don't know the literature well enough to support
- my hunch, my gut instinct here is to treat the interaction term like
- any other collinear variable, and look at the main effects _after_ the
- interaction effects have been partialled out. This will determine the
- extent to which the unique contribution of main effects stands over
- and above that 'messy' term....just like with ANOVA, we interpret the
- interaction first, since main effects are confounded by it. So we
- orthoganalize the main effects by residualizing with respect to the
- interaction. Anybody out there with more info on this approach????
- _Please_ post, this is becoming an issue in my lab, too!
-
- Regards,
-
- ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((
- Travis Gee () tgee@ccs.carleton.ca ()
- () tgee@acadvm1.uottawa.ca () ()()()()
- () () ()
- () ()()()()()()()()()
-
-