Hi Rolando,

it depends on the cause of separation and on what you are looking at. Can you provide more detail?

As a general advice, this might help you:

http://support.sas.com/kb/22/599.html

Thanks Andre....by making the model fit less perfectly, the quasi-complete separation was removed. Thanks again.

Dr. George Heinze and Dr. David Firth (at Warwick University) has focused on the quasi-complete seperation at Logistic regression analysis when cells of zero in two way, or three way classification were found. You must apply Firth type Logistic regression (Penalized Maximum Likelihood Method) to your data.

Please see my paper related to this topic: http://journals.mui.ac.ir/jrms/article/view/2206

In fact, I recommend you to contact Dr. David Firth.

I hope my response will be useful for you.

Assoc. Prof. Ecevit EYDURAN

Thanks Ecevit. I will look into the SAS macro. I noticed one of the links provided in your article for the SAS macro is no longer valid. Thanks again.

what can i Do in Multinomial logistic regression?

Ecevit - Can you repost your article on dealing with Firth logistic regression? The link does not work. 

Thank you.