Co-linearity/Stepwise regression

Question

I was wondering if we need to check the multicollinearity between variables when we run the stepwise regression, If yes, why? Also, If we have several IV, how we should know which interaction we should include in the model?

Answer 1

Stepwise regression is not a great idea, and according to Andrew Gelman, it is "considered by statisticians to be a bit of a joke". This answer is extensive on the poorness of stepwise regression as a modeling technique.

Multicollinearity can be tested for using metrics like variance inflation factor. But ideally, you should know what is collinear as a researcher, because if you test for multicollinearity using metrics like the VIF, then we can do manipulations that get rid of such multicollinearity. Often, multicollinearity is not much of a problem. If it is of key concern, there are alternatives such as ridge regression.

If you are trying to do variable selection, look up lasso instead of stepwise regression.