0

Two approaches have been proposed in order to overcome the issue of multicollinearity if we have interaction variables which are mean centering and standardizing (z scores). You can check No.2 in this article for more details http://www.statisticalhorizons.com/multicollinearity

> which better do you think and why?

thank you in advance

Ben
  • 175
  • 1
  • 3
  • 16
  • 1
    It's usually customary to do *both*. And what do you mean by a logarithm variable? – Joe Jun 24 '14 at 15:11
  • I transformed one of my variable to log due to high kurtosis and i am not sure whether i can transform it now to z score. thank you – Ben Jun 24 '14 at 15:16
  • This is essentially the same as your previous question. So as not to lose the extensive comment thread there, I will close this version as a duplicate, but please feel free to edit the other one to reflect the additional information you have included here. – whuber Jun 24 '14 at 15:31
  • The answers are contained in [When should you center & when should you standardize?](http://stats.stackexchange.com/q/29781/7290), of which this is also probably a duplicate. – gung - Reinstate Monica Jun 24 '14 at 16:01
  • My question is specified to multicollinearity and the provided question is general. – Ben Jun 24 '14 at 16:21

0 Answers0