Modeling Community Health with Areal Data: Bayesian Inference with Survey Standard Errors and Spatial Structure.

Area of Science:

• Spatial epidemiology
• Health geography
• Statistical modeling

Background:

• Small-area survey estimates, such as those from the American Community Survey (ACS), are widely used in epidemiological and health geography research.
• Standard errors (SEs) accompanying ACS estimates are often not utilized in reliability assessments or modeling, despite systematic variations across demographic and geographic factors.
• Ignoring observational error in small-area estimates can lead to inferential biases and overconfidence in research findings, particularly impacting predictive models for resource allocation.

Purpose of the Study:

• To propose a conceptual framework and workflow for spatial data analysis that incorporates standard errors from areal survey data.
• To demonstrate the application of this framework using the Index of Concentration at the Extremes (ICE) to analyze spatial health inequalities.
• To encourage researchers to prioritize data quality and reliability in variable selection and model building.

Main Methods:

• Developed a workflow for spatial data analysis using areal survey data, incorporating plausible reasoning and Bayes' theorem.
• Constructed and evaluated SEs for the Index of Concentration at the Extremes (ICE).
• Employed visual diagnostics to assess an observational error model for the ICE and integrated it into a model of US county mortality rates.

Main Results:

• Demonstrated systematic variations in ACS SEs across different regions, neighborhoods, and socioeconomic characteristics.
• Successfully constructed and evaluated SEs for the ICE, providing a measure of data reliability.
• Estimated an ICE-mortality gradient by incorporating the observational error model, highlighting the impact of spatial inequalities on mortality.

Conclusions:

• Failure to account for observational error in small-area survey data can significantly impact research conclusions and resource allocation decisions.
• Integrating data reliability information, such as SEs, into spatial health models is crucial for accurate inference and robust findings.
• Researchers should routinely consider data quality as a criterion for variable selection and incorporate reliability measures into their analyses whenever feasible.