What Predicts What — County-Level Correlations

Does poverty predict poor mental health? Does education predict income? Does political lean correlate with health outcomes? The explorer below lets you test any pair of county-level measures and see the relationship across more than 3,000 U.S. counties.

A few correlations worth exploring:

Poverty → Poor mental health days (r ≈ 0.72). The strongest signal in the dataset. Counties with high poverty rates have dramatically more poor mental health days — one of the most direct illustrations of how economic insecurity translates into psychological suffering.

Income → Education (r ≈ 0.70). Median household income and bachelor’s degree rates track closely at the county level. The causal direction is debated — education raises incomes, but high-income places also attract and retain educated workers.

Income inequality → Poor mental health (r ≈ 0.45). The Gini coefficient correlates with mental health outcomes even after controlling for income level. It’s not just poverty that hurts — living in an unequal place has its own effect.

Political lean → Obesity (r ≈ −0.45). More Republican-leaning counties (state-level data) tend to have higher obesity rates. This correlation is partly explained by rurality, income, and education — all of which correlate with both political lean and health outcomes. Correlation is not causation.

Religious importance → Divorce rate (r ≈ −0.30). More religious states have lower divorce rates on average — though the relationship is weaker than expected and has notable outliers in the South, where both religiosity and divorce rates are high.

Education → Physical inactivity (r ≈ −0.65). Counties with higher college attainment rates have lower physical inactivity rates. This likely reflects access to gyms and outdoor recreation, desk-job culture, and the health behaviors that correlate with education.

A note on causation: every correlation here is observational and cross-sectional. Counties differ on dozens of dimensions simultaneously — income, race, urbanicity, climate, culture — and disentangling causes from confounders requires more than a scatter plot. Use this as a starting point for questions, not a source of final answers.