# Talk:Interaction between two continuous variables

### From PsychWiki - A Collaborative Psychology Wiki

## Question re: interpretation

I have a question. Whenever you center your two continuous variables and calculate the interaction term, would it be hard to correctly interpret the effect...? Suppose you have -2 as a centered score on x1 and -2 as a score on x2. Then the interactionterm would be 4. At the same time, another respondent with centered scores of x1=2 and x2=2 will end up with the same score on the interactionterm (4)...somehow it doesn't feel right.... So now we do not have multicollinearity, but we have an uninterpretable score...??? Can anyone ease my mind on this one :-)?

- Here is the answer from one of our statistically minded grad students:

The magnitude and value of the interaction term is not important. One examines the term for significance and then conducts the relative post hocs. Just as in ANOVA a significant interaction term doesn't tell you anything. You still need to do the post hocs to figure out the nature of the interaction. Regression is identical. After all, ANOVA is a special case of regression.

## Mean centering ineffective for reducing multicollinearity

There is some evidence (Echambadi & Hess, 2007; Kromrey & Foster-Johnson, 1998) that mean-centering does not alleviate multicollinearity issues (Echambadi and Hess wrote that it neither hurts nor help). Moreover, from my reading of Aiken & West (1991), I gleaned that their sole argument for mean-centering is that it enhances interpretability of data, not that it reduces multicollinearity. This is still a good enough argument for mean-centering in my opinion, but we should be clear about why we do it. Swisnieski 12:45, 16 January 2009 (PST)