Analyzing Clustered Data
Mohamed M. Shoukri, Mohammad A. Chaudhary · 2007
CONTENTS 1.1 Introduction ................................................................................................. 1 1.1.1 The Basic Feature of Cluster Data ................................................ 2 1.1.2 Sample and Design Issues ............................................................. 7 1.2 Regression Analysis for Clustered Data .................................................. 11 1.3 Generalized Linear Models ....................................................................... 15 1.3.1 Marginal Models (Population Average Models) ........................ 16 1.3.2 Random Effects Models ................................................................. 17 1.3.3 Generalized Estimating Equation (GEE) ..................................... 17 1.4 Fitting Alternative Models for Clustered Data ....................................... 19 Clusters are aggregates of individuals or items that are the subject of investigation. A cluster may be a family, school, herd of animals, flock of birds, hospital, medical practice, or an entire community. Data obtained from clusters may be the result of an intervention in a randomized clinical or a field trial. Sometimes interventions in randomized clinical trials are allocated to groups of patients rather than to individual patients. This is called cluster randomization or cluster allocation, and is particularly common in human and animal health research. There are several reasons why investigators wish to randomize clusters rather than the individual study subjects. The first being the intervention may naturally be applicable to clusters. For example, Murray et al. (1992) evaluated the effect of school-based interventions in reducing adolescent tobacco use. A total of 24 schools (of various sizes) were randomized to an intervention condition (SFG = smoke-free generation) or to a control condition (EC = existing curriculum). The number (and proportion) of children in each school who continued to use smokeless tobacco after 2 years of follow-up is given in Table 1.1.