How to test the multicollinearity in ordered probit?

[Kazi Aiman Udoy]

Hello everyone,
Hope you are all doing well.

I am doing a ordered probit model. I have estimated the model without any error. I want to check whether there exists any multicollinearity in my model. How can I check multicollinearity in ordered probit model? What is the command? I have tried the following commands but none of these works:

1. oprobit y x1 x2 x3
estat vif, uncentered

2. oprobit y x1 x2 x3
collin independent_var1 independent_var2 independent_var3 ...

3. oprobit y x1 x2 x3
collin


Please suggest me the proper way to check multicollinearity.

[Clyde Schechter]

If you really want to do this, just recognize that multicolinearity is a property of the independent variables only, so it doesn't matter whether we are doing an -oprobit- or any other regression model. For reasons known only to the StataCorp developers, they restricted -estat vif- to run only after -regress-. So

regress y x1 x2 x3
estat vif

will do it.

That said, why do you want to do this in the first place? It is usually a waste of time. Do see Arthur Goldberger's Textbook of Econometrics, where he devotes a chapter to elaborating why multicollinearity is a pointless concept that is just a roundabout way of saying that your sample size is too small! (It's very entertainingly written, by the way--well worth the read.)

The bottom line for me is that one should never test for the presence of multicolinearity. Multicolinearity is only a problem if the standard error of your key explanatory variable's coefficient is so large that you do not have a precise enough estimate of that variable's effect to answer your research question. In that case, it may be worth verifying that it really is multicolinearity, not something else, that is causing the problem, and helping to identify which variables are involved in the multicolinearity. So that's where -estat vif- can be helpful. But if you do not have this problem of a too imprecise estimate of your key variable's coefficient(s), then there is no point in using -estat vif-. Just accept success and move on.

[Kazi Aiman Udoy]

Thanks Clyde Schechter for your detailed explanation.