2
$\begingroup$

I am just wondering what does "to center two IVs" mean in the context of understanding an interaction. The IVs are A and gender and the DV is B.

Is it necessary for the main effects of a new interaction variable (A*Gender) to be a product (e.g. centered main effect of A * centered main effect of Gender) before the IVs can be used in a (hierarchical) regression framework?

Improve this question
$\endgroup$
3
  • $\begingroup$ What are you quoting? $\endgroup$
    – whuber ♦
    Commented Nov 22, 2011 at 3:43
  • $\begingroup$ There is a reference here (victoria.ac.nz/psyc/paul-jose-files/helpcentre/…) but I am aware I should not be creating external links, so I did not do it in my question) $\endgroup$
    – Adhesh Josh
    Commented Nov 22, 2011 at 3:48
  • $\begingroup$ There is no problem with creating links of any sort, external or not. If for some reason you are uncomfortable with that, then at least please explain the context of the quotation. How else are we to understand what you are referring to? $\endgroup$
    – whuber ♦
    Commented Nov 22, 2011 at 3:50

1 Answer 1

Reset to default
5
$\begingroup$
  • In this context "center" denotes subtracting the mean from the variable.
  • Centering can reduce correlation between the interaction term and constituent main effect variables. This can make interpretation of regression coefficients more intuitive in some contexts.
  • Whether you center or not has no effect on the r-square change produced by adding an interaction term in a hierarchical regression after putting main effects in an initial block.

There's a bit of discussion of the pros and cons of centering here.

Improve this answer
$\endgroup$
1
  • $\begingroup$ What a fascinating article @JeromyAnglim. 13 years ago, I did my dissertation on collinearity diagnostics. There was (and maybe still is) a big debate on centering. The article you posted is one side of it, but, at the time anyway, Belsley argued that centering DID reduce problems. The book I used was Conditioning Diagnostics: Collinearity and Weak Data in Regression. The math gets a bit over my head; and I see Belsley has a new book out; perhaps this has been resolved. $\endgroup$
    – Peter Flom
    Commented Nov 22, 2011 at 12:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.