My chair would like me to center my data for my hierarchical multiple regression, including several moderators and interaction effects. It seems recommended to do this, if not to help minimize multicollinearity, but additionally (and more crucially) to help with interpretability of my interactions. However, where I am uncertain is regarding what to do with some variables I had already thought to transform due to problems with skewness. Do I center these variables? I feel like that might mean transforming them, in essence, twice, but remain unsure. If I don't transform them, how can I use them in an interaction with a centered variable?
Asked
Active
Viewed 931 times
1
-
2You may find the discussion in [When should you center your data & when should you standardize?](http://stats.stackexchange.com/questions/29781/when-should-you-center-your-data-when-should-you-standardize) useful. – MånsT Nov 27 '12 at 07:20
-
2why would you transform predictors that are skewed? – Glen_b Nov 27 '12 at 07:36
-
*Centering* is just a linear shift. There is nothing wrong w/ centering a transformed variable. However, I don't think it will have any effect on the interpretability of your model. Eg, the estimated coefficients will be identical, if you only center (as opposed to scale); only the intercept will change, but people rarely care about the intercept. – gung - Reinstate Monica Aug 26 '13 at 00:46