Binomial-Beta Hierarchical Models for Ecological Inference

Abstract: Ecological inference is the process of learning about discrete individual-level behavior by analyzing data on groups. We develop binomial-beta hierarchical models for this problem and use Markov chain Monte Carlo methods for parameter estimation. This paper focuses on 2x2 tables with or without covariates. The methodology can be generalized to RxC tables in a straightforward fashion. We illustrate our models with simulated, as well as with real data.