06-17-2016 11:46 AM - edited 06-17-2016 11:48 AM
I'm doing a regression of a count variable (which follows a Poisson distribution), and regressors are qualitatives and quantitatives. To do the regression, I used then PROC GENMOD.
But, in order to calculate VIF, I saw here (https://communities.sas.com/t5/SAS-Procedures/VIF-in-Proc-Genmod/td-p/227155) that I have to use PROC REG to get VIFs "since collinearity does not depend on the dependent variable". The problem is that proc reg cannot work with qualitatives variables, so I used PROC GLM with TOLERANCE option. I can now compute VIF like : VIF=1/TOLERANCE.
The problem is that VIFs are calculated for variable and not for each modality of variable. For example, here are some results of tolerance option :
Type I tolerance Type II tolerance
Dummy008 0.9994865097 0.9475415437
Dummy009 7.03305E-14 0
where Dummy008 and Dummy009 are respectively "Married" "Single" (modalities of variable Family situation).
How can I interpret that ? And also and can be the reason of the Type II tolerance's value 0 (EDIT : Dummy009 correspond to a reference category) ?