I am setting up a logistic regression model with mostly categorical independent variables (answers to survey questions). Some of the variables have levels like "High-Medium-Low-NotApplicable". The problem I am facing is that two of the variables (A and B) have the same level "NotApplicable" for all the respondents in the data set. For example the two variables could be survey questions like "Did you like our service?" and "How much did you pay for our service?" This is because these questions do not get asked when the answer to "Did you use our services last month?" is "No.

This results in perfect multicollinearity when the variables are dummy coded and entered into the regression. How can I work around this and still be able to include both the variables in my model?

  • $\begingroup$ Can you include some of the empirical models you tried to estimate so we can get a good idea of what you're trying to do? What is your outcome? Is it categorical or continuous? What is your sample size? It's hard to give good advice in the absence of details about the research question and the data you're working with $\endgroup$ – Marquis de Carabas Apr 26 '16 at 21:52

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.