home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!gatech!paladin.american.edu!auvm!HASARA11.BITNET!A716HOX
- Message-ID: <STAT-L%92073010402816@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Thu, 30 Jul 1992 16:30:14 CET
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: Joop Hox <A716HOX@HASARA11.BITNET>
- Subject: interaction effects in regression models
- Lines: 13
-
- I have learned that if you have a regression model, for instance
- Y=a+ b1X + b2Z + b3XZ, with an interaction, and you find that one of the
- simple efects (say b1) is not significant, you still should not omit that
- effect as long as the interaction is there. So I wrote a paper in which I
- include a nonsignificant effect in a regression model, explicitly stating
- that I do this because that variable was involved in an interation that was
- significant. I now have a reviewer who states that it is otherwise, and
- especially if the interest is in interactions and not in main effects
- (such as in multilevel analysis), all main effects should be deleted instead.
- I have difficulty believing this. Any comments?
-
- Jop Hox
- University of Amsterdam
-