A Bayesian Multiple Imputation Method for Correlated Measurements Subject to Limits of Detection

• Activity Number: 137
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Kai Ding*+ and Amy Herring and Suzan Carmichael and Andreas Sjodin
• Companies: University of Oklahoma Health Sciences Center and The University of North Carolina at Chapel Hill and Stanford University and National Center for Environmental Health
• Keywords: correlated data ; Gibbs sampler ; limit of detection ; multiple imputation

Abstract

In many environmental studies, exposure measurements are often subject to limits of detection. Available imputation methods that provide valid inference can only handle up to 2 exposure variables. In this paper, we propose a Bayesian multiple imputation approach using the Gibbs sampler to impute multivariate correlated measurements below limits of detection, assuming a simple mixed effects model. Simulation studies were conducted to evaluate the performance of the proposed method when the limit of detection varies only over exposures, or over both exposures and subjects, under a wide range of missing proportions. A real dataset relating polybrominated flame retardants (PBDEs) to hypospadias was used to illustrate the proposed approach.