I want to verify for multicollinearity between independent categorial variables. Which test I should use? First, I want to examine the relationship between the willingness to participate in medical decision making (dependent variabele - 2 categories) and education (independent variable).

Later, I would do a multiple binary logistic regression (adjusted). I have education (5 categories), paid work (2 or 3 categories) and household income (2 categories). Income was questioned using a 5-point likert scale.


1 Answer 1


There's no need to pre-test for multicollinearity in a study like this. That can be an issue in large-scale studies with many potential predictor variables, so it gets a lot of attention in machine learning courses. Even if you found multicollinearity, how would you change the modeling strategy for your studies?

All that multicollinearity will do here is inflate the variance estimates for individual coefficients.* It probably won't hurt predictions from the models at all, as the high individual variance of one coefficient will typically be offset by corresponding covariances with other coefficients.

If you are worried about multicollinearity, fit a model and then use the vif() function in the R car package to calculate the variance inflation factors (VIF) directly from the model. As this page explains, the VIF produced by that function are appropriate both for categorical predictors and for models other than ordinary least squares, situations where the usual VIF calculations are inappropriate.

*If the collinearity is perfect then there might be a problem, but that just means that some of your predictors are exact linear combinations of others. Software will often just remove such a predictor from the model before it does the fitting.


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.