Mlogit VIF

[Peter Chan]

Hi everyone,

I am having a problem in calculating VIFs for a multinomial logit. My dependent variable contains 3 alternatives for a mlogit. Bivariate correlations of the sample show values over 0.7, therefore I want to test for multicollinearity with VIFs. However using the "vif" command does not work for mlogits. Is there another possibility to calculate VIFs for a mlogit?
I read that a linear regression is sufficient in that case, but should I include the 3 alternatives of the independet variable into the linear regression or just test the independet variables. Another source even said it is fine to treat the categorial dependet variable as continous for the linear regression.
Has anyone dealt with this before?

Thank you for your help and kind regards,
Peter

[George Ford]

VIFs are on the Xs. I suspect you could run it reg and then VIF.

You can also create VIFs by hand by regressing X1 on the other X's, one at a time, and using the R2 to get the VIF.

[Richard Williams]

Phil Ender has a nice routine called collin that computes several collinearity measures.

net describe collin, from(https://stats.oarc.ucla.edu/stat/stata/ado/analysis)

This handout discusses how to detect and deal with multicollinearity.

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

Incidentally, many feel that concerns about multicollinearity are often unwarranted.

[Carlo Lazzaro]

Peter:
probably Richard alludes to A Course in Econometrics — Harvard University Press , Chapter 23.