Premium
Analysing Categorical Responses Obtained from Large Clusters
Author(s) -
Miller Michael E.
Publication year - 1995
Publication title -
journal of the royal statistical society: series c (applied statistics)
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.205
H-Index - 72
eISSN - 1467-9876
pISSN - 0035-9254
DOI - 10.2307/2986343
Subject(s) - categorical variable , psychology , mathematics , statistics
SUMMARY Several researchers have proposed generalized estimating equation (GEE) approaches for the analysis of clustered binary or multicategory response data. When large numbers of observations are taken within clusters, these GEE methods can require inversion of extremely large covariance matrices. However, in many applications, only covariates specific to large groups of observations within the cluster are of interest for the marginal model, whereas covariates specific to individual measurements within the cluster are important in modelling the pairwise associations. In this paper, we use two medical examples to motivate a discussion illustrating that, when the marginal model contains only covariates specific to many observations within a cluster, the estimating equations used for the marginal analyses can be formulated in terms of proportions rather than binary indicator variables, while still modelling the pairwise associations with covariates specific to individual measurements within the cluster. This approach can reduce the computational burden that is inherent in the analysis of large clusters, while still allowing the potential gains in efficiency for the marginal analysis that can be obtained by modelling the pairwise associations.