Date: Tue, 27 May 1997 11:06:57 -0400
Reply-To: "Bessemer, David" <bessemer@ARI.FED.US>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Bessemer, David" <bessemer@ARI.FED.US>
Subject: Re: covariate x treatment interaction w/ GLM
Content-Type: text/plain; charset="us-ascii"
>Peter A. Vanable[SMTP:pvanable@UIC.EDU] wrote:
>With GLM, it appears that I can enter the between subject factor in it's
>original form (i.e., as a continuous variable) as a covariate and then look
>at the covariate x treatment interaction. This works and I find a
>significant interaction, as expected.
>Is there any problem with approaching this analysis in the way I've
>described?? I've haven't found any published examples of investigators
>finding a covariate x treatment interaction. In fact, under ordinary
>circumstances, I assume that a covariate by treatment interaction would be
>problematic (i.e., a violation of the homogeneity of regression slopes
Homogeneity of regression is only required to test differences among
treatment means adjusted for differences in the covariate. Adjusted
means are no longer of interest when the regression slopes differ, since
the differences among treatments vary as a function of the covariate.
In that case, the covariate-treatment interaction is the main result.
Research on covariate-treatment interactions has been of special
interest in educational psychology using ability or aptitude test scores
as covariates. An important book reviewing aptitude-treatment
interaction research was:
Cronbach, L. J., and Snow, R.E. (1977). Aptitudes and instructional
methods: A handbook for research on interactions. New York: Irvington.
There is much research in this area since then. Many examples can also
be be found in other areas that focus on individual differences, e.g.,
developmental psychology and personality research to name just two.
David W. Bessemer
U. S. Army Research Institute
Armored Forces Research Unit
Fort Knox, KY 40121-5620