I'm trying to fit a model with statsmodels, Multinomial Logit Model.

This works fine, but I'm quite unsure if it is the right model for my case and how to treat my independent variables.

So my dependent variable has the following response categories: "No", "Indifferent", and "Yes". My assumption is that this conforms to the case of a Multinomial model. But my independent variables also have the same structure: "Yes", "Indifferent" and "No". Currently mapped as 1 = No, 2 = Indifferent and 3 = Yes Do I need to make dummies of these variables, or can I just take the odds and multiply with 2 to get Yes?

Also, one variable is binned age, which means the age column got the values 20-30, 30-40 etc.

tldr: Independent variables are not binary, they have 3 possible values. Can I multiply the odds by 2 to get the third value?


The primary consideration for selecting the Multinomial Model is when your Dependent Variable (DV) has a number of discreet unordered response categories. In your case, your DV response categories ("Yes", "Indifferent", and "No") represent this case. Note, when selecting the model, what is important is how your DV is measured, rather than your Independent Variables (IVs). So, IVs with different scoring types (e.g. binned age or yes/no type variables) do not affect the choice of your model.

As for the IV nominal/ordinal predictor variables, dummy-coding is a common practice.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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