INCORPORATING DESIGN WEIGHTS AND HISTORICAL DATA INTO MODEL-BASED SMALL-AREA ESTIMATION

Area of Science:

• Statistics
• Public Health
• Survey Methodology

Background:

• Bayesian hierarchical regression (BHR) is commonly used for small area estimation (SAE).
• Traditional BHR methods applied to complex survey data often ignore sampling design and weights, potentially introducing bias and increasing variance.
• The use of non-informative priors in BHR necessitates combining data over multiple years, impacting timeliness and trend analysis.

Purpose of the Study:

• To develop a design-assisted model-based approach for SAE that addresses bias and variance issues.
• To enhance the timeliness of SAE by utilizing historical data for informative priors.
• To estimate the prevalence of disability at the U.S. county level using the proposed method.

Main Methods:

• Proposed a design-assisted model-based approach integrating adjusted sample weights into BHR for SAE.
• Employed a power prior, leveraging historical data to define informative priors for improved timeliness.
• Validated the method using American Community Survey (ACS) data and applied it to Behavioral Risk Factor Surveillance System (BRFSS) data.

Main Results:

• The proposed method produced more timely estimates compared to widely-used alternatives.
• Estimates from the new method were closer to ACS direct estimates, especially for counties with limited data.
• Demonstrated the ability to estimate county-level disability prevalence effectively.

Conclusions:

• The developed design-assisted BHR approach offers a more accurate and timely solution for SAE.
• The method effectively addresses limitations of traditional BHR in complex survey settings.
• The approach is generalizable for estimating other county-level health-related measurements.