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

10.1. Introduction

Sometimes, the usual maximum likelihood approach to estimation in logistic regression is not appropriate. As discussed in Chapter 8, “Logistic Regression I: Dichotomous Response,” there may be insufficient sample size for logistic regression, particularly if the data are highly stratified and there are a small number of subjects in each stratum. In these situations, you have a small sample size relative to the number of parameters being estimated since you will be estimating parameters for the stratification effects. For the maximum likelihood estimates to be valid, you need a large sample size relative to the number of parameters.

Often, highly stratified data come from a design with cluster sampling, that is, designs with ...