Methodology

Which communities demonstrate better health outcomes than predicted by their material circumstances?

Traditional metrics like income, education, and access to healthcare explain much of the variation in community health. But some communities do significantly better than these factors would predict. We call this "community resilience" — the capacity for a community to thrive despite challenging conditions.

This project identifies these communities by looking at the gap between predicted and actual health outcomes, potentially revealing communities with strong social ties, cultural assets, or other protective factors that aren't captured by standard demographic measures.

• CDC PLACES 2024 — Census tract-level health outcomes (BRFSS 2022) including obesity, diabetes, hypertension, heart disease, and physical inactivity. Kentucky and Pennsylvania use PLACES 2023 data (BRFSS 2021) as a fallback.
• USDA Food Access Research Atlas 2019 — Low-income, low-access (LILA) classification identifying food deserts (crosswalked to 2020 boundaries)
• Census Bureau 2020 — Population data and tract boundary relationship files for geographic crosswalking

Step 1: Measure Health Burden

We construct a composite health burden index (0-1 scale) from CDC PLACES data, combining:

• Chronic disease prevalence (diabetes, obesity, heart disease)
• Mental health indicators (depression, poor mental health days)
• Preventive care gaps (lack of health insurance, missed checkups)
• Behavioral risk factors (smoking, physical inactivity)

Step 2: Predict Expected Health

We use OLS regression with state fixed effects to predict health burden based on socioeconomic and access factors:

Health Burden ~ Poverty Rate + Education + Insurance
              + Food Access + Housing Cost Burden
              + State Fixed Effects

Step 3: Calculate Resilience Score

The resilience score is the standardized residual from this model (z-score). A positive score means the community has better health outcomes than predicted; a negative score means worse outcomes than predicted.

Note on Group Quarters

Tracts where more than 10% of the population lives in group quarters (college dorms, military barracks, correctional facilities, nursing homes) are flagged but not excluded. These populations may have health outcomes driven by institutional factors rather than community characteristics.

• Kentucky & Pennsylvania Data Vintage — These states use CDC PLACES 2023 data (BRFSS 2021) instead of 2024 data, as they couldn't collect enough BRFSS data in 2023. Their health outcomes are approximately one year older than other states.
• Temporal Gap — Health data (2024/BRFSS 2022), food access data (2019), and Census data (2020) come from different time periods.
• Model-Based Estimates — CDC PLACES uses small area estimation, not direct measurement. Some tract-level estimates have high uncertainty.
• Ecological Fallacy — Tract-level patterns do not describe individual outcomes. A "resilient" tract may still contain individuals with poor health.
• Unmeasured Confounders — The model cannot capture all factors affecting health. Positive residuals may reflect unmeasured socioeconomic advantages rather than community resilience.
• Static Snapshot — This is a cross-sectional analysis and cannot show trends or causation.

@misc{resilience-mapping,
  title = {Community Resilience Mapping},
  year = {2025},
  url = {https://resilience-mapping.fly.dev},
  note = {Census tract-level analysis of community
         health resilience using CDC PLACES 2024
         and USDA Food Access Research Atlas data,
         standardized to 2020 Census boundaries}
}