Study design and analytic methods for data collected from clusters of animals

Abstract

A variety of study design and statistical methods to account for the clustering of animal health and production outcomes is outlined. We argue that the relative utility of study design vs. statistical methods in accounting for cluster effects depends primarily on the objectives of the study and the amount of prior information available. The statistical methods outlined vary from simple post-hoc adjustments of test statistics to relatively complex mixture-distribution models. Methods for normal, binomial and Poisson distributed data are presented. The various options presented are discussed with reference to their underlying assumptions and how they have been or might be used in veterinary epidemiologic studies.