Hi,
I have a question about multicollinearity in multinomial logistic regression. I used VIF function and though I have intercept in my model, i received this warning:
In vif. default(fit_mlogit): No intercept: vifs may not be sensible.
I also repeated in package nnet with multinom function, the result is the same with mlogit function.
what should I do now? is there another way to check multicollinearity?
thank you
Dear Samaneh H.Bahreini
The variance inflation factor (VIF), which examines how much the variance of an estimated regression coefficient rises if your predictors are linked, is one approach to measure multicollinearity. A VIF between 5 and 10 suggests a significant degree of association, which may be troublesome.
One of the major issues that occurs with the binary logistic regression model is that the explanatory variables evaluated for the logistic regression model are highly associated with one another. Multicollinearity will result in unstable estimates and erroneous variances, which will have an impact on confidence intervals and hypothesis tests.
Have a look with this answer also: https://stats.stackexchange.com/questions/69959/test-multicollinearity-for-multinomial-logit-regression
Best Regards
Dr. Fatemeh Khozaei
Fatemeh Khozaei thank you for your answer. should i use colldiag function and multinom toghether to check multicollinearity? is using VIF incorrect?
Samaneh H.Bahreini The variance inflation factor is a generally accessible metric in software to aid in the diagnosis of multicollinearity (VIF). Variance inflation factors (VIF) quantify how much the variance of the predicted regression coefficients is inflated when the predictor variables are not linearly connected.
The variance inflation factor (VIF), which examines how much the variance of an estimated regression coefficient rises if your predictors are linked, is one approach to measure multicollinearity. The VIFs will all be 1 if no factors are connected.
Best Regards
- Badges
- Science topic