Logistic Regression Using SAS®: Theory and Application

8.1. Introduction

In previous chapters, we assumed that all observations are independent—that is, the outcome for each observation is completely unrelated to the outcome for every other observation. While that assumption is quite appropriate for most data sets, there are many applications where the data can be grouped into natural or imposed clusters with observations in the same cluster tending to be more alike than observations in different clusters. Longitudinal data is, perhaps, the most common example of clustering. If we record an individual’s responses at multiple points in time, we ordinarily expect those observations to be positively correlated. But there are many other applications in which the data has a cluster structure. For example, ...