I have a huge amount of features that are categorical variables and I'm trying to find a system for weeding out categorical variables that are close to being multicollinear. Is vif a reasonable measure in the context of categorical variables? Would pearson's chi-squared be a good measure of association between categorical variables? I've also heard mention of polychoric correlation.

Edit 1: For vif on categorical variables, I'd need to use the r-squared from a linear probability model right? I'm assuming you can't use pseudo-r-squared since it's a completely different animal.

  • $\begingroup$ Why do you think multicolinearity is a problem? What is your ultimate goal? Maybe regularization is a better alternative ... $\endgroup$ – kjetil b halvorsen Jul 20 at 17:15

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.