ABSTRACT
We discuss the concept of Bayesian uncertainty analysis in survey sampling for categorical data when there is nonignorable nonresponse. This is important because in sample surveys, data are typically summarized in contingency tables and there are nonresponders. In a nonignorable nonresponse model there are nonidentifiable parameters, and a sensitivity analysis is necessary to study the effects of these parameters on the parameter of interest. The sensitivity analysis is typically performed by setting the nonidentifiable parameters at various plausible values. In a Bayesian uncertainty analysis, rather than performing
in Bayesian
we put a prior on the nonidentifiable parameters. We illustrate Bayesian uncertainty analysis using a three-way contingency table (i.e., a single r × c× u table) with nonignorable nonresponse.
