A Nonparametric Bayesian Analysis of a Proportion Under Nonignorable Nonresponse

• Activity Number: 109
• Type: Contributed
• Date/Time: Monday, August 12, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Jai Choi*+
• Affiliation(s): National Center for Health Statistics
• Keywords: Selection approach ; Exchangeability; ; Beta-Binomial model; ; griddy Gibbs sampler; ; Latent variable;

Abstract

We use a Dirichilet process prior (DPP) to restrict the pooling of nonresponse binary data from small areas which may seem to be similar. We compare a nonignorable nonresponse baseline model to a new model which has DPP. Our objective is to estimate the proportion of individuals with a particular characteristic from each of a number of areas under nonignorable nonresponse. We compare a baseline nonignorable hierarchical Bayesian model with a DPP model which is centered on the baseline model. All hyperparameters have proper prior densities. We use Markov chain Monte Carlo methods to fit the models. Our comparisons show that the use of the DPP can help to correct for the blind use of exchangeability. We show empirically that there could be difference in inference between the two nonignorable models.