A central theme in the field of survey statistics is estimating population-level quantities through data coming from potentially non-representative samples of the population. Multilevel Regression and Poststratification (MRP), a model-based approach, is gaining traction against the traditional weighted approach for survey estimates. MRP uses partial pooling through random effects, thus shrinking model estimates to an overall mean and reducing potential overfitting. Despite MRP’s straightforward specification of prior distributions, the estimates coming from it are susceptible to bias if there is an underlying structure that the prior does not capture. This work aims to provide a new framework for specifying structured prior distributions that lead to more robust estimates for MRP. We use simulation studies to explore the benefit of these priors and demonstrate on US survey data.

This is joint work with Daniel Simpson (University of Toronto) and Lauren Kennedy (Columbia University).