Date of Award


Document Type

Open Access Thesis


Epidemiology and Biostatistics



First Advisor

Hrishikesh Chakraborty


The present study evaluates the length of the confidence interval and the percentage of time the true parameter is captured by the confidence interval for four intracluster correlation estimators. These are the ANOVA estimator, the Pearson pairwise estimator with constant weights (PEQ), the kappa type estimator called FC, and an estimator using a Resampling method (RM) for binary data. We compared these different estimates by using a large simulation study. The data we simulated is correlated binary data, which assumes an exchangeable correlation structure. We also included different variations of the number of clusters, cluster size, cluster size variation, event rate, event rate variation and the population intracluster correlation coefficients.

The results showed that, among all the confidence limits for the 4 estimators, the confidence limits by the PEQ estimator performs best and it is the ideal one to use in most situations, but if the cluster size is very small, the confidence limits by the FC estimator performs best and is the ideal one to use. Finally, if the number of clusters is very small, the confidence limits obtained by the RM estimator performs best and this is the ideal one to use.

Included in

Public Health Commons