Inference for Changes in the Kappa Statistic in Multi-level Clustered Binary Data

• Activity Number: 333
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Rebecca Roberts Andridge*+
• Companies: The Ohio State University
• Keywords: kappa ; agreement ; group randomized trial ; clustered data

Abstract

Cohen's kappa is the most widely used measure of agreement for categorical outcomes. We consider the use of kappa statistics as outcome measures in a group randomized trial (GRT). In GRTs, identifiable groups rather than individuals are randomized to study conditions, and observations within a group tend to be correlated. In addition, we consider a longitudinal GRT design with multiple rating pairs on the same subject within cluster over time. This work is motivated by the Detroit Middle School Asthma Project, an extended nested cohort GRT assessing two in-school interventions to enhance management of asthma. In this study, parents and children provide dichotomous reports of the child's experiences with asthma in school, and it is of interest to evaluate whether the intervention increases agreement of parent and child.