Categorical Data Analysis Using The SAS® System, 2nd Edition

8.7. Maximum Likelihood Estimation Problems and Alternatives

If you perform enough logistic regressions, you will encounter data for which maximum likelihood estimation does not produce a unique solution for the parameters; you do not obtain convergence. In addition, for data with small cell counts, large sample theory may not be applicable and thus tests based on the asymptotic normality of the maximum likelihood estimates may be unreliable. This section discusses some of the situations in which maximum likelihood methods may not produce solutions and Section 8.8 discusses the alternative strategies based on exact methods.

8.7.1. Examples of Non-Convergence

To gain insight into the possible data configurations that result in non-convergence, ...