Stats >> Model >> Exploring interactions in a linear regression model (August 1, 2002)

Dear Professor Mean, I have a model with two factors. When I ran the model, it showed a significant interaction between the two factors. What do I do now? --Troubled Trudy

Interactions are tricky. It depends a lot on the context of your problem. If you are looking at an intervention, then an interaction implies, perhaps, that the effectiveness of the intervention is different in different subgroups.

Ideally, you should have specified how you would handle the analysis of a significant interaction in your research protocol, but most people don't do this ahead of time.

I would look at some graphs first to try to detect the nature of the interaction. There is a PLOT button on the dialog box of the SPSS General Linear Model that is probably worth a click. I would also look at residuals to make sure that the interaction is not caused by a single outlier that pulls one cell totally out of whack.

SPSS also offers the standard contrasts, though you should read a good text on ANOVA before applying these mindlessly. Which contrast is best depends in part on whether your factor represents a nominal, ordinal, or interval variable.

You also might consider consolidating some of your levels if it makes sense in the context of your problem, and then see what happens to the interaction. Or possibly focus on just two levels that are of special interest to you.

This webpage was written by Steve Simon on 2002-08-01, edited by Steve Simon, and was last modified on 2008-07-08. Send feedback to ssimon at cmh dot edu or click on the email link at the top of the page. Category: Ask Professor Mean, Category: Linear regression