Arthur Tabachneck wrote:
> Is it legal to test for an interaction without testing for the main
> effects of the variables included in those interactions.
> (which I recall it isn't), does one still have to account for the
> degrees of freedom used by the non-tested main effects?

A proper interaction test compares the improvement to fit achieved by
the the interaction *in excess of* the main effects. If there is no
improvement, the interaction effect doesn't tell you anything you
didn't already know from main effects. So technically, the
goodness-of-fit test for interactions includes the consideration of
main effects. You have to properly adjust DOF when you're comparing
these two models: the interaction one has fewer DOF than the main
effects one.

I've assumed you're doing this in the context of multiple or logistic
regression. There are many views of what an interaction is, and many
tests.

-- 
mag. Aleks Jakulin
http://ai.fri.uni-lj.si/aleks/
Artificial Intelligence Laboratory,
Faculty of Computer and Information Science, University of Ljubljana.






.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to