I have same error message as in this post. However all my coefficients look normal with no inflated value or standard errors. My question is how can I make sure the error message is a sign of perfect separation or it's just some of my observations have extreme predicted probabilities?

Also, my model has Residual deviance=Inf and AIC=Inf, would they suggest that there is no perfect separation as they are measures of goodness of fit?

  • $\begingroup$ Without more information, I don't think this question is answerable. Can you say more about your data? Can you paste in your model output? $\endgroup$ Commented Sep 7, 2017 at 16:01
  • $\begingroup$ @gung It's not clear to me that the indicated duplicates will answer the question (how to detect it is distinct from what is it); yet I believe a simple answer is possible (e.g. perfect prediction). Or if they do answer it, it's not clear the answer will be sufficiently distinct from the main issue in those to be easily spotted. $\endgroup$
    – Glen_b
    Commented Nov 27, 2017 at 0:43
  • $\begingroup$ @Glen_b, I think a straightforward implication of my answer in the linked thread is that if you have predicted probabilities of 0 or 1, but the model converged, that doesn't necessarily imply separation. If you think more is needed, it's fine w/ me if you reopen this. $\endgroup$ Commented Nov 27, 2017 at 1:19


Browse other questions tagged or ask your own question.