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

13.1. Introduction

Previous chapters discussed statistical modeling of categorical data with logistic regression. Maximum likelihood estimation (ML) was used to estimate parameters for models based on logits and cumulative logits. Logistic regression is suitable for many situations, particularly for dichotomous response outcomes. However, there are situations where modeling techniques other than logistic regression are of interest. You may be interested in modeling functions besides logits, such as mean scores, proportions, or more complicated functions of the responses. In addition, the analysis framework may dictate a different modeling approach, such as in the case of repeated measurements studies.

Weighted least squares (WLS) estimation ...