355 – Speed Session #4: Topics in Epidemiology and Survey Research Methods, Part 2
Modeling Clustering Design Effects When Cluster Sizes Vary
James Chromy
RTI International
The intracluster correlation coefficient is often used to model the design effect associated with cluster sampling. Typically, the average cluster size is inserted into the design effect formula even though the cluster size will vary in practice. Cluster size variation is particularly a problem when the design effect for domain estimates is desired and the domain of interest is not evenly distributed over all sample clusters. Should an alternate way of computing the intracluster correlation be developed or is some other simple solution available to resolve this problem? This paper presents an alternate, but simple, model for accounting for both clustering and the variation in cluster sizes.