Presentation (CTI)

OBJECTIVES: Obesity affects 42% of U.S. adults, contributing $170 billion annually in healthcare costs. Food environment interventions show mixed effectiveness. This study tests the Exhausted Margin Hypothesis (EMH), which predicts that interventions reducing friction on unhealthy consumption are effective only where behavioral margins remain slack, and ineffective where behavioral ceilings bind.
METHODS: We conducted a retrospective cohort analysis using county-level panel data from 2010-2022 (2,810 counties; 29,695 county-years). The primary outcome was adult obesity prevalence. Exposures included dollar store density, grocery/supermarket density, farmers market availability, and SSB taxes. We applied the Caetano, Caetano, and Nielsen (2024) bunching correction for selection bias. Heterogeneity was assessed through continuous interactions with baseline obesity. All models included county and year fixed effects.
RESULTS: Dollar stores significantly increased obesity in low-baseline counties (+1.19 pp, p=0.003) but had no effect in high-baseline counties (p=0.529). The continuous interaction with baseline obesity was large and significant (β = −0.24 pp per SD, p=0.001), consistent with EMH. Farmers markets showed no causal effect after selection correction (β = +0.03 pp, p=0.896), suggesting previous positive associations reflected selection bias. SSB taxes reduced obesity by 1.3 pp overall; effects were approximately twice as large in areas with greater grocery access (interaction β = −0.049, p<0.001). Supermarkets showed null effects, consistent with Allcott et al. (2019).
CONCLUSIONS: The findings support the EMH: dollar stores harm only where unhealthy-consumption margins remain unconstrained, while farmers markets do not causally reduce obesity. Policy implications include targeted dollar store restrictions in low-obesity counties—yielding an estimated $1.4B in annual healthcare savings—and pairing SSB taxes with grocery access for maximal impact. For policy evaluation and real-world evidence, the results highlight the importance of modeling intervention effects as functions of baseline population characteristics.