I have a big data matrix with 6000 rows (observations) and 45 columns (44 predictive variables (categorical and continuous) and 1 response variable (0 or 1). I want to check the correlation/ multicollinearity in R. I have looked into cor() and heat map so far, but it seems like for a big data I need to use something else. Please advice.

  • $\begingroup$ you could take a look at vif from car package $\endgroup$
    – Rorschach
    Sep 17, 2015 at 18:15
  • $\begingroup$ I was about to say the same thing. vif (Variance Inflation Factor) is, in my opinion, much more useful for looking at multicollinearity as it describes the effect of the collinearity on your model. Sometimes you can tolerate the existence if it doesn't have much impact. $\endgroup$
    – Benjamin
    Sep 17, 2015 at 18:17
  • $\begingroup$ Also, Spearman's correlation (eg cor) is not a meaningful concept for multinomial (ie- categorical) variables, which you state that you have. $\endgroup$ Sep 17, 2015 at 18:25

1 Answer 1


I also like VIF's, but another way would be to estimate the mutual information between/among the various predictors as it isn't concerned solely with a linear relationship. The idea is to only use those covariates with low mutual information as they are telling you something different. Check out the infotheo or entropy pkgs.


Your Answer

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

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