topic Assessing for colinearity for categorical variables and/or binary outcome in Discussions

Assessing for colinearity for categorical variables and/or binary outcome

gstins — Fri, 21 Mar 2025 01:04:29 GMT

I am seeking to assess for colinearity among categorical variables to be included in a multi variable nominal logistic regression. I understand that VIF is useful for this, but I’m unable to find VIF for categorical variables or a binary outcome. Other methods have been less initiative like a LASSO. Is there a way to, either with or without LASSO, to evaluate for colinearity? Thanks!

Re: Assessing for colinearity for categorical variables and/or binary outcome

Dan_Obermiller — Fri, 21 Mar 2025 01:33:13 GMT

From the Fit Model dialog, have you tried switching to Generalized Regression? From there you can say the response has a binomial distribution. This should give you a fit that is very similar to the nominal logistic regression and it offers the ability to look at the Correlation of Estimates table. You can use that to understand the collinearity. Otherwise you can study the collinearity of the main effects by using Multivariate Methods > Multivariate.

Re: Assessing for colinearity for categorical variables and/or binary outcome

Victor_G — Tue, 25 Mar 2025 09:39:33 GMT

Hi @gstins @gstins,

Welcome in the Community !

In addition to the solution proposed by @Dan_Obermiller, one possible "trick" to evaluate collinearity using VIF values is to create a random "dummy" numeric column, fit a multivariate regression model, and display VIF values of the model's variables. See and the response from @cwillden.
VIF are calculated only using values from input data/factors, (see equation here : Parameter Estimates for Original Predictors), so no matter the response type, you can always calculate and analyze VIF values.

Hope this workaround may help you,