Date: Wed, 20 Mar 1996 16:51:02 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: Interactions in regression equations.
Bisson Jocelyn (bisson@ERE.UMONTREAL.CA) wrote:
: From my understanding of analytical strategies concerning regression
: analysis, when an interaction term turns out to be significant (using
: the r2 change after simple effects have been entered), it is indicated
: to perfom separate analyses, one for each level of the interaction term.
-- Well, you CAN do it that way, and it might be handy in taking
a CLOSE look at your data....
: For instance, if gender (X1) interacts with husbands' drinking behaviors
: (X2) in explaining wives' drinking patterns (Y), one makes a separate
: regression for each gender.
-- ... especially when, say, the regression of a husband-variable
explaining a wife-variable depends on *gender*. (Is that the gender
of the husband, or of the wife?)
-- In answer to your later question: As David Nichols says, you
properly need to look at 2x3 cells, for gender and age, *if* there
is a higher interaction including THEM; in your example, a 3-way
interaction. Else, you can look at two groups, and three groups.
Rich Ulrich, biostatistician firstname.lastname@example.org
Western Psychiatric Inst. and Clinic Univ. of Pittsburgh
============remainder of original note
: My question is: What should be done when there are two or more separate
: significant interaction terms ? For example, if on top of the previous
: interaction, wives' age (3 groups) interacts with husbands' drinking
: behaviors in explaining wives' drinking patterns.
: Should one necessarily present in this case 6 (2 x 3) different regression
: equations ? One for each combinations of levels of the interacting
: variables ?
: We are using data from a population survey, and for that matter the
: analysis design is nonexperimental and nonorthogonal.