Title: Statistical methods for clustered longitudinal binary data
Authors: Leilei Zeng - University of Waterloo (Canada) [presenting]
Abstract: In many settings in experimental research, clusters of subjects (e.g. families/schools/clinics) are randomly assigned to different interventions and each subject has repeated measurements over the study period. The resulting responses are then cross-sectionally correlated within clusters at a given assessment time, and longitudinally correlated within subjects over time. Methods for clustered longitudinal binary data are proposed based on transition models and generalized estimating equations. Efficiency gains for the marginal regression parameters are realized when the intra-cluster association is strong. Guidance for the design of randomized trials involving this kind of analysis is also provided. The formula for the number of clusters is derived based on the robust variance estimators from the transition models that account for the intra-cluster associations.