03-20-2017 08:16 AM
Hi all, I am using stratification and run a logistic regression model within each stratum, however, in some strata there are terms that have zero variance and SAS log indicates the use of generalized inverse in estimation due to singularity. If these variables are not removed from the models would that effect the coefficient of the variable of interest? The reason I am hesitant to remove these variables is that (1) they are informative in other stratum, (2) so that the model will be comparable to other estimation methods. Please help.