The Intracluster and Intercluster Correlations in Cluster Randomized Trials

• Activity Number: 245
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Robert E. Johnson*+ and Tina Duong Cunningham
• Companies: Virginia Commonwealth University and University of California at Los Angeles
• Keywords: cluster randomized trials ; intracluster correlation ; ICC ; induced correlation ; finite population correction

Abstract

The intracluster correlation (ICC) plays a significant role in the planning and analysis of cluster randomized studies. Two measures sampled from the same randomly chosen cluster will be correlated. It is key that the clusters are randomly selected. If fixed-e.g., a convenience sample of clusters is used-then no correlation between intracluster measures is induced. However, if the-nonrandomly chosen-clusters are randomly allocated to treatment arms, then not only is an intracluster correlation induced, but an intercluster correlation between measures in different clusters is induced. The variance of contrast in treatment means is less than the variance when randomly chosen clusters are randomly allocated. The reduction in variance is related to the finite population correction employed in sample surveys. This correction is negligible when the number of clusters is large, but may play a role in sample size determination and analysis with few clusters. We will illustrate how the sampling method and randomization may induce correlation and discuss the impact on analysis.