Centering level-1 interaction terms in HLM

[Linda Mueller]

Dear all,

I’m running multilevel analyses with HLM and wish to examine an interaction between two level 1 predictor variables. Following recommendations in the literature, I group-mean centered both level 1 predictor variables to ease interpretation. However, I’m left clueless as to whether the respective interaction term (computed beforehand in SPSS) should be centered as well?
If this were the case, would I let HLM center the interaction term, or should I apply some kind of centering method in the process of computing the interaction in SPSS?

Thank you for your time, I would greatly appreciate your thoughts on this.

Linda

 

[Jake]

Centering the interaction term as well would not be a bad idea, since only in that case can the intercept be interpreted as the mean response. However it is also common enough for people to not bother centering the interaction terms. (As you might already know, the product of two centered variables is not necessarily centered itself.)