home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!UWPG02.BITNET!CLARK
- X-Envelope-to: stat-l@mcgill1.bitnet
- X-VMS-To: in%"stat-l@mcgill1.bitnet"
- MIME-version: 1.0
- Content-transfer-encoding: 7BIT
- Message-ID: <01GN0T3HFBKY9TCNQY@uwpg02.uwinnipeg.ca>
- Newsgroups: bit.listserv.stat-l
- Date: Fri, 31 Jul 1992 08:34:06 -0600
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: Jim Clark <CLARK@UWPG02.BITNET>
- Subject: Re: interaction effects in regression models
- Lines: 46
-
- Discussion raises some interesting questions. A few random thoughts:
-
- a. One basic question appears to be whether you want the test of the
- interaction effect to evaluate its difference from zero or the difference
- between the slopes. With main effects included, interaction coefficient
- represents change in slopes. Without main effects, its not clear to me that
- you would be testing differences in slopes.
-
- i.e.,
- Y = b0 + b1*X1 + b2*X2 + b12*X1X2
- becomes
- Y = b0 + b1*X1 + (b2 + b12*X1)*X2
- i.e., b12 is change in slope of X2 as X1 varies (could also be
- arranged as change in X1 as X2 varies) and b1 is the change in the intercept as
- X1 varies.
-
- What would b12 represent without main effects for b1 and b2 included in
- the equation? I guess it would be the difference between b12 and 0 rather than
- differences in slope of one variable across levels of the other.
-
- b. If you want error term analogous to factorial ANOVA, then you would
- include all predictors, wouldn't you?
-
- c. Can SS predicted be unambiguously assigned to main effects and
- interactions in the overall analysis? This unambiguous allocation is helped by
- centering, in my experience. If there is variation that is shared by the main
- effects and interactions, you might have to decide theoretically which
- allocation makes more sense, or decide rationally the order in which variation
- is to be allocated, or communicate the ambiguity. In any case, seems to me
- that you would still want to (have to?) analyze and partition variation taking
- into consideration main effects.
-
- d. Here is reference to Aiken & West.
-
- Aiken, L. S., & West, S. G. (1991). _Multiple regression: Testing and
- interpreting interactions_. Newbury Park, CA: Sage.
-
-
- Best Wishes
- Jim
-
- James M. Clark CLARK@UWPG02.BITNET (note ZERO-TWO)
- Department of Psychology CLARK@UWPG02.UWINNIPEG.CA
- University of Winnipeg (204) 786-9359
- Winnipeg, Manitoba, Canada
- R3B 2E9
-