A BAYESIAN HIERARCHICAL SMALL AREA POPULATION MODEL ACCOUNTING FOR DATA SOURCE SPECIFIC METHODOLOGIES FROM AMERICAN...

Area of Science:

• Epidemiology
• Biostatistics
• Demography

Background:

• Small area population counts are vital for epidemiological studies but often lack quality assessment.
• U.S. Census Bureau data sources (decennial census, PEP, ACS) have methodological differences leading to varying accuracy and errors.
• Discrepancies in population denominators can affect public health disease/mortality rate calculations.

Purpose of the Study:

• To develop a Bayesian population (BPop) model for accurate annual small area population estimation and uncertainty quantification.
• To integrate multiple U.S. Census Bureau data sources, accounting for their specific methodologies and errors.
• To produce race-stratified population estimates for small areas and predict counts for years with missing data.

Main Methods:

• Developed a Bayesian population (BPop) model integrating decennial census, Population Estimates Program (PEP), and American Community Survey (ACS) data.
• Accounted for data source-specific data generation mechanisms and errors.
• Focused on Black and White populations in 159 Georgia counties for 2006-2023, including simulations and validation.

Main Results:

• The BPop model successfully fused information from multiple U.S. Census Bureau sources.
• Comprehensive, race-stratified small area population estimates with associated uncertainties were produced.
• Model performance was validated through simulations and comparisons with existing data sources.

Conclusions:

• The BPop model offers a robust framework for improving small area population estimates and their uncertainties.
• This approach enhances the accuracy of public health research by providing reliable denominator data.
• The model is adaptable for various spatial granularities and demographic subpopulations.