The use of cluster randomized experiments to study the effects of treatments on groups of subjects has increased in recent years. Many of these experiments lack the necessary statistical power to detect practically meaningful effects of treatment. One method for improving power in cluster randomized experiments that has been advanced is to use external information about the intracluster correlation coefficient (ICC) to increase the degrees of freedom available to estimate the variance of the estimated treatment effect. This thesis contributes to the discussion about improving power in cluster randomized experiments through the use of external information about the ICC in the following fashion. First, I point out that existing proposals for incorporating external information about the ICC into an analysis of a cluster randomized experiment do not have well studied size and power properties. Secondly, I derive a method for studying the small sample size and power of existing proposals and show that none of these proposals can guarantee that the hypothesis test procedure proposed achieves the nominal type I error rate in small samples. Thirdly, I derive bounds on the amount of power improvement possible through the use of external information about the ICC. Finally, I propose a new method for incorporating external information about the ICC into an analysis of a cluster randomized experiment which results in a hypothesis test that can be shown to have the nominal level in small samples. I show how this method can be extended in order to be utilized for hypothesis testing in any linear model with normal errors and a non-diagonal covariance matrix.

Last modified

• 09/06/2018

Creator

• Christopher H Rhodes

DOI

• https://doi.org/10.21985/N2RJ0V

Subject

• Statistics

Keyword

• intracluster correlation coefficient
• external information
• cluster randomized experiments

Date created

• 2008-04-25

Resource type

• Dissertation

Rights statement

• In Copyright