Improving Small-Sample Inference in Group-Randomized Trials with Binary Outcomes

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Philip Westgate*+ and Thomas Braun
• Companies: University of Michigan and University of Michigan
• Keywords: Group Randomized Trials ; Binary Outcomes ; Quasi-Likelihood ; Wald Statistic

Abstract

Group Randomized Trials typically randomize a small number of clusters composed of a large number of subjects, resulting in over-dispersion quantified using the intra-cluster correlation coefficient (ICC). When subject-level outcomes are binary, modeling can be done using quasi-likelihood with a logistic link. In this setting, the Wald statistic used to test for a treatment effect, which asymptotically has a standard normal distribution, may have a variance less than one, resulting in a test size smaller than its nominal value. When the ICC is known, we develop a method for adjusting the estimated standard error such that the Wald statistic approximately has a standard normal distribution. We also propose a way to resolve non-nominal test sizes when the ICC is estimated. Through simulation results covering a variety of realistic settings, we examine the performance of our methods.