Why no vIf in binary logistic Regression

[Ava Cerrid]

Dear Statalist,

in my pursuit to asses whether there is multicollinearity in my model or not, I read a lot of articles and opinions here and in other forums. In the literature I could not find critical opinions, but some people in forums say one cannot use the variation inflation factor (vif) in binary logistic regression (blr), some say yes and some even advice not to use the Vif at all. Could some of you elaborate WHY the use of vif in blr is critical and link a source where the issue is mentioned? Because atm I do not really find quotable material on this discussion.

So far the only alternatives I see to check for multicollinearity is bivariate correlation (what is troublesome since I have many different scales of measurement in my analysis) or to look at standard errors, but I cannot really asses where the critical threshold should be.

I am very thankful for any advice!!!

[Richard Williams]

Multicollinearity is a function of the right hand side of the equation, the X variables. You can change logit to regress and get vifs, or else use the user-written Collin command from UCLA. A discussion of multicollinearity can be found at

https://www3.nd.edu/~rwilliam/stats2/l11.pdf

A lot of times you just have to live with collinearity, but the handout suggests things you might do under various circumstances.

[Ava Cerrid]

Multicollinearity is a function of the right hand side of the equation, the X variables. You can change logit to regress and get vifs, or else use the user-written Collin command from UCLA. A discussion of multicollinearity can be found at

https://www3.nd.edu/~rwilliam/stats2/l11.pdf

A lot of times you just have to live with collinearity, but the handout suggests things you might do under various circumstances.

Thank you very much Mr. Williams. I will have a look.