Small Area Models for Over-Dispersed Poisson Counts

• Activity Number: 282 - New Developments in Small Area Estimation Research at the U.S. Census Bureau
• Type: Topic Contributed
• Date/Time: Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Jerry Maples* and Adam Maidman
• Companies: U.S. Census Bureau and University of Minnesota
• Keywords: small area ; count data ; poverty ; SAIPE

Abstract

Recent work in small area estimates at sub-county level has focused on estimating the rate of poverty in school-aged children (e.g. Franco 2015). To obtain counts of children in poverty require having known population counts, often assumed without error, for these small areas. In general, this is not always the case and the uncertainty due to not knowing the true population counts is often not reflected in the predictions. The sampling distributions from simulating the American Community Survey design appear more similar to over-dispersed Poisson distributions than standard Poisson. We propose a small area model for over-dispersed Poisson count data to model the number of children in poverty at the census tract level. Modeling assumptions will be tested using the simulated samples from design-based simulation.