Date: Sun, 13 Sep 1998 21:49:30 GMT
Reply-To: Richard F Ulrich <wpilib+@PITT.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: Richard F Ulrich <wpilib+@PITT.EDU>
Organization: University of Pittsburgh
Subject: Re: Qualitative by Qualitative Interactions in Multiple Regression
Van Kirk, Jeff (VanKirk@PSYCHIATRY.UCHC.EDU) wrote:
: I'm running an OLS multiple regression with 6 control and predictor
: variables and several of their interactions. 2 of the predictors are
: dichotomous. Is it appropriate to run their interaction using the usual
: product term approach? It seems to me that the resulting residuals
: would by definition violate the normality assumption? Thanks
Is your question about using dichotomous variables, or
about computing and using the interaction?
Dichotomous variables don't leave extreme outlyers, which is
usual bugaboo of bad predictors. So they are ordinarily okay.
The test you get on a dichotomy is the same test as if the variable
were a factor in an ANCOVA (analysis of COvariance).
The interaction is just another dichotomy, which does not even
correlate much with the main effects if you have coded the main
effects as (-1,1) before you multiple them.
Rich Ulrich, biostatistician email@example.com
http://www.pitt.edu/~wpilib/index.html Univ. of Pittsburgh