CSDE Computational Demography Working Group (CDWG) Hosts Aja Sutton on Incorporating spatial structure into multilevel regression and poststratification for subnational demographic and small area estimation (5/22/2024)
Posted: 5/16/2024 ()
On Wednesday, May 22nd from 9:00 AM – 10:00 AM, CDWG will host Dr. Aja Sutton to introduce her research. Dr. Aja Sutton is a Postdoctoral Scholar in the Social Sciences Division at the Stanford Doerr School of Sustainability, Stanford University. She received her PhD from the Department of Geography at the University of Washington (UW). From 2020-2022 she was TADA-BSSR NIH T32 Fellow in Data Science and Demography Training at UW’s Center for Studies in Demography and Ecology, from which she also holds a Certificate in Demographic Methods. She received an MA in History from Western University, and an MSc in Palaeopathology from Durham University. Her work is focused on population health, computational social science, and epidemiology. The event will take place in 223 Raitt and on Zoom (register here).
Title: Incorporating spatial structure into multilevel regression and poststratification for subnational demographic and small area estimation
Abstract: Demography is a discipline dependent on the accurate enumeration of population-level processes, especially the count of individuals. When near-perfect census enumeration or representative survey data are unavailable for a particular outcome of interest, it is often difficult to establish the count of that outcome in a population. Instead, we may be able to make indirect estimates through small area estimation (SAE) methods; these use additional contextual information to produce statistically robust estimates of under- or unobserved subpopulations or geographic units. Multilevel regression and poststratification (MRP) is a model-based statistical method that uses national/high-level administrative survey or census data to adjust for non-representativeness in subnational surveys, and to provide small area estimation in areas where subnational survey data are sparse or nonexistent. MRP is computationally inexpensive and generally able to produce quick, consistent, and accurate estimates, and prevailing approaches borrow strength from other administrative areal data through partial (global) pooling. While this approach is effective, it does not account for any potential existing neighborhood spatial structure in the data. Including spatial specifications in MRP is a powerful way to better handle existing spatial relationships and generate more accurate area-level estimates. Join us as we learn more about MRP, Bayesian methods for exploring and handling spatial structure in data, while considering how to improve indirect estimates of 2021 COVID-19 vaccination in California using MRP with spatial structure.