Two-Stage Bridge Model for Binary Responses with Informative Cluster Size

• Activity Number: 44
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xiaoyun Li*+ and Dipankar Bandyopadhyay and Stuart R. Lipsitz and Debajyoti Sinha and Debajyoti Sinha
• Companies: Florida State University and Medical University of South Carolina and Brigham and Women's Hospital and Florida State University and Florida State University
• Keywords: Bridge density ; Clustered data ; Logistic link ; Random effects

Abstract

In some biomedical studies involving clustered binary responses (say, disease status), the cluster size can vary because some components of the cluster can be absent. In this paper, we propose a novel two-stage random effects logistic regression model where both the presence of a cluster component and the binary response of disease status for a present component are treated as responses of interest. Bridge distribution is used for random effects so that both the marginal probability and the conditional probability preserve the easy to interpret approximate logistic regression forms. Simulation studies are presented to assess the robustness of the estimators in spite of wrongly specified distribution of random effects, and to compare finite sample performance of estimates with existing methods. The model can also be applied to longitudinal data analysis with informative missing mechanism.