Correcting the problem of Multicollinearity in categorical variables

[Steve Asare]

Dear All,

I have among my explanatory variables, a number of categorical variables and after testing for multicollinearity, I had some of the variables suffering from multicollinearity. Please, I need assistance on what to do in order to correct this problem.

Thanks.

Regards,
Stephen.

[Joro Kolev]

The econometrician Professor Arthur Goldberger, humorously mocked the "problem" of multicollinearity in his textbook A course in econometrics(1991), by writing a whole section on the "problem of micro-numerosity".

Multicollinearity is not a problem in the sense that it does not give you biases estimates or misleading standard errors. In a regression that "suffers from the "problem" of multicollinearity" everything is correct.

Now you might not like that multicollinearity results in too large standard errors, and nothing is significant in your regression. But there is nothing you can do about this, short of going out and collecting more data, which is hopefully not multicollinear.

[Jeff Wooldridge]

I agree with Joro. At a minimum, you shouldn’t even think about multicollinearity diagnostics — they are not “tests” — until you look at the estimates and their standard errors. Collinearity is properly captured in the standard errors. Do you think they’re too big to be useful? Is there a particular coefficient of interest?

You’d have to show us results, but even then I suspect I won’t suggest computing multicollinearity diagnostics.

[Steve Asare]

Sorry for the delay in response.

I am grateful, Joro and Wooldridge. I am using secondary data and i think i would go with Prof. Wooldridge's advise to conduct other tests and ignore the test for multicollinearity.

Thanks once again.

[Steve Asare]

Sorry for the delay in response.

I am very grateful for your response. Based of this advise, I would go on and conduct other tests and ignore the test for multicollinearity.

Thank you once again Profs.