Variational Message Passing for Semiparametric Regression with Classical and Berkson Errors

• Activity Number: 133
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Sang Il Kim*
• Companies: University of Technology, Sydney
• Keywords: Variational message passing ; semiparametric regression ; graphical models ; mean field variational Bayes ; measurement error

Abstract

Measurement error in study covariates is common in many areas of scientific research. Statistical methods to account for such errors rely upon a valid error model, particularly regarding the identification of classical and Berkson errors, without requiring the distribution of the measurement error to be known a priori. We propose a Bayesian semiparametric regression approach to accommodate error-contaminated data using Variational Message Passing(VMP), which is a fast and deterministic technique that overcomes common computational and convergence issues in standard Monte Carlo methods. We derive message passing algorithms for both classical and Berkson error models by couching semiparametric regression in a graphical models framework. The algorithms' performance and feasibility are demonstrated via simulations.