Robust Bayesian Models for Surveys with Missing Data and Summary External Information

• Activity Number: 308
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Roderick Little*+ and Sahar Zangeneh
• Companies: University of Michigan and Fred Hutchinson Cancer Research Center
• Keywords: Calibrated Bayes ; nonresponse ; poststratification ; missing at random

Abstract

The Calibrated Bayes approach to survey sampling bases inferences on Bayesian models, chosen to yield credibility intervals that have good frequentist coverage in repeated sampling. Such models need to incorporate features of the sampling design like sampling weights and clustering to be robust to model misspecification. We describe two applications of the approach to survey problems involving external information: probability proportional to size sampling where aggregate information is available for sizes of non-sampled units, and nonresponse when there is external information for poststratification. Simulation comparisons with standard "design-based" methods suggest superior frequentist properties for the Bayesian methods.