Within Cluster Resampling

ELAINE BORLAND HOFFMAN: Within Cluster Resampling (Under the direction of Clarice R. Weinberg) Dependence among observations from the same group is known as within-cluster correlation. Typically, distinct clusters are independent, but special methods are required to account for the existence of the within-cluster correlation. Although not accounting for the within-cluster correlation may produce consistent parameter estimates, the variances will be under-estimated. Within Cluster Resampling (WCR) is proposed as a method for analyzing any clustered data; however, this dissertation will focus on clustered binary data. The proposed idea is to sample one observation from each cluster. A set of independent observations called a resampled data set results, which simply can be analyzed by standard software. As the name Within Cluster Resampling implies, the above process is repeated by randomly sampling with replacement one observation from each cluster. A series of data sets is created by resampling a large number of times. The regression parameter is estimated by the average of the resample-based estimates. The proposed variance will account for the correlation among observations from the same cluster. The interpretation of a WCR parameter is the population-averaged difference associated with a unit change in a covariate corresponding to a randomly selected observation from a randomly selected cluster. WCR is unlike traditional marginal analysis methods. WCR is a cluster-based method that equally weights each cluster, whereas traditional marginal methods are observation-based and the weight of the cluster increases with its size.