Test for and remedy of multicollinearity in logistic regression with categorical IVs

#1
I was searching over net how can I test the presence of multicollinearity and what is its remedy in case the IVs are categorical in a logistic regression. I am a SPSS user and found these two links over net which can be useful. But having problem to understand them. Can anyone see it and tell me how this works, I can not understand the collinearity diagnostic that SPSS does, and what is the remedy in the end? Is this a real procedure to use dummy variables in collinearity diagnosis? Is it applicable in case of logistic regression?

The links are given below:

http://www-01.ibm.com/support/docview.wss?uid=swg21476696

http://www-01.ibm.com/support/docview.wss?uid=swg21476169
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Re: Test for and remedy of multicollinearity in logistic regression with categorical

There may be more sophisticated techniques, but yes the option in the first link is used (I have used it). Just jam your data into a ill-fitting linear model and request the collinearity diagnostics. I have also converted my categorical variables (though they typically have been ordinal in general nature, or I could order them), not sure how it functions if you jam something like gender into it. Then look at the tolerance or VIF (see other posts or internet search for definitions), which these measures have some general cut-offs for determining the presence of colinearity. If colinearity is present and seems problematic (over-fitting), and the data is collected and you have what you have, then you may opt to use good reasoning to eliminate one of the variables or possibly consolidate. This is a pretty basic reply, but throughout these steps more rigorous options may exist for you.
 
#3
Re: Test for and remedy of multicollinearity in logistic regression with categorical

Thank you. Yes, now I have understood it. Could you please note/indicate some other good approaches of checking multicollinearity with categorical IV except from this method (if a better method exists)?