For the selection of variables for a regression model, I did a pairwise correlation matrix between the different predictors and the response variable. From the pairwise correlation matrix, I realise there could be a linearity relationship between the log transformed predictors and the response variable. I would like to ask if it is better to check for multicollinearity by using the vif function in R after transforming the predictors or just simply check for multicollinearity on the independent variables?

Also, my response variable is symmetric after I log transformed it. Is it a sensible decision in doing so before checking for multicollinearity? Any advice or suggestions are welcomed. Thanks!


1 Answer 1


Welcome to CV.

First, for collinearity, I think condition indexes and proportion of variance explained is a better method than VIF.

Second, to your question, I think you should check for collinearity on the variables as they would be in the model.

  • $\begingroup$ Hi Peter, thanks for the advice. So you are implying that simply checking for multicollinearity between IVs would be a more sensible decision rather than transforming the IVs before checking for multicollinearity? $\endgroup$ Sep 1, 2018 at 14:29
  • $\begingroup$ I am saying that you should check for collinearity with the variables in the form they will be in in the regression - maybe transformed, maybe not. The decision of whether to transform should be made on a substantive basis prior to doing the analysis. $\endgroup$
    – Peter Flom
    Sep 1, 2018 at 15:17

Your Answer

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

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