Exac GEE logistic regression

[iuri_leite]

Hi All,

We are carrying out a GEE logistic analysis to see if there is difference between the two categories of a dummy variable. However, when the dummy variable assumes the value one the response variable assumes the value zero. So, we have a complete separation. So, I would like to know if there is a possibility of doing an exact test when running a GEE logistic regression and it is a solution to my problem. Thank you in advance. Regards

[StatDave · Accepted solution]

Assuming that, for each subject, you just have a binary predictor and a binary response observed at each time or occasion, then you might want to consider an conditional logistic approach or a non-model-based approach using CMH statistics in PROC FREQ. For example, for the CMH approach:

proc freq; tables subject*predictor*response / cmh noprint; run;

This note discusses and illustrates both approaches which include exact test options.

[iuri_leite]

Dear Dave,

many thanks.

Regards,

Iuri