http://www.sinica.edu.tw/~jds/JDS-155.pdf
PROC
LOGISTIC gave the warning: “
That there is possibly a quasi-complete separation of data points. In particular, the maximum likelihood estimate may not exist.
The LOGISTIC procedure continues in spite of the above warning. Results shown
are based on the last maximum likelihood iteration. Validity of the model fit is
questionable.
” It is typical of quasi-complete separation data that the variances of the pseudo estimates are seemingly large. A slight perturbation of appropriate
data points would remove the separation, or make it complete. Such a situation
can be tested by appropriately perturbing the coefficients by a percent or so (not
any coefficients which have known exact values). If the answers vary wildly, this
suggests numerical ill conditioning, and the specific solution has limited meaning