Understanding regional disparities in obesity: a multiscale geographically weighted analysis in the United States

Obesity varies markedly across U.S. counties, and global models often miss place-specific determinants. While prior studies document higher prevalence in nonmetropolitan areas, the geographic variation in its determinants-known as spatial heterogeneity-remains underexplored. We linked age-adjusted adult obesity prevalence with socioeconomic indicators, and behavioral risks for 3106 contiguous counties. A global OLS model served as a baseline, followed by estimation of Multiscale Geographically Weighted Regression (MGWR). MGWR outperformed global OLS (adjusted [Formula: see text]: 0.801 vs. 0.566; AICc: 13,580.92 vs. 15,764.21), confirming non-stationarity and revealing covariate-specific scales. Metropolitan status was generally protective, but its effect attenuated or reversed in parts of the West. Income and educational attainment are broadly inverse with minimal dispersion across counties, suggesting near-global behavior in this specification. Short sleep shows a strong positive association with little spread, while binge drinking is positive and slightly more variable. Employment is narrowly positive with almost no spatial dispersion. Bandwidth diagnostics separate near-global from local processes: metro and employment operate at large bandwidths, education and binge drinking at meso scales, and income, short sleep, and marriage at finer scales. As a benchmark, metro-only models showed a uniformly protective but locally varying metro effect that attenuated once socioeconomic and behavioral covariates were included. Findings confirm non-stationarity and argue for a two-tier translation: system-level policies for near-global factors and community-tailored interventions for localized risks, with attention to Western metropolitan vulnerabilities and Southeastern rural constraints.