I have a variable that has 57 kurtosis, so I decided to transform it to log. However, I have multicolleanirity problem due to interacting this variable and others with another variable so I am using z scores to reduce VIF value. So ...

Is it ok to transform the log variable to z score

thank you so much

  • 2
    $\begingroup$ If you have a multicollinearity problem then using z-scores will not fix it. Also, in a regression context it is not necessary for the independent variables (of which this seems to be one) to be approximately normal. You would transform it principally to cure nonlinear relationships with the dependent variable. I realize these observations don't answer your question, but maybe they might make it go away :-). $\endgroup$ – whuber Jun 23 '14 at 14:52
  • $\begingroup$ thanks for contributing. VIF actually get down after using z scores $\endgroup$ – Ben Jun 23 '14 at 15:08
  • 1
    $\begingroup$ The fact that VIF is reduced does not mean the collinearity is substantially better. VIF is sensitive to things, such as units of measurement, that are somewhat arbitrary. The impact of collinearity ultimately is measured in terms of the covariance matrix of the estimates and particularly in the sizes of the confidence intervals; standardizing the data will not change those. $\endgroup$ – whuber Jun 23 '14 at 15:41
  • 1
    $\begingroup$ Because I know almost nothing about your data or what you are trying to accomplish with them, it would be presumptuous of me--and perhaps even misleading or confusing--to make any recommendation. Consider editing your question to provide more information of this sort to help your readers understand your situation, or look through the questions on our site that focus on transformations and collinearity. Two that might be relevant are stats.stackexchange.com/a/59811 and stats.stackexchange.com/a/16715. $\endgroup$ – whuber Jun 23 '14 at 18:02
  • 1
    $\begingroup$ it is a common problem facing multicollinearity when we have interactions. When we just check the value of VIF and correlation of the independent variables alone, the results are normal no multicollinearity. So it might be not a real problem $\endgroup$ – Ben Jun 23 '14 at 18:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.