Does climate adaptation policy deliver a health dividend? Exposure, modeled health burden, and defensive capacity in China's climate-adaptive city pilots.

Abstract

Climate policy is widely expected to yield a health dividend: cleaner air can improve health without a simultaneous expansion of defensive capacity. We distinguish exposure reduction from defensive adaptation in China's 2017 climate-adaptive city pilot, using a 2011-2022 panel of 264 cities, including 21 treated units. A difference-in-differences design examines fine particulate matter (PM2.5) exposure, while the Global Exposure Mortality Model translates exposure into modeled pollution-related health burden. Physicians, hospital beds, and employee-insurance enrolment measure selected defensive inputs. The pilot is associated with a 3.87 μg/m³ lower PM2.5 (95% confidence interval -6.49 to -1.25; wild-bootstrap P = 0.0067). The estimate remains negative after pollution-convergence adjustments but is smaller and imprecise in the matched sample. The central accounting model implies an approximately 8.10% lower burden (conditional 95% confidence interval -11.32 to -4.76%), not an observed mortality reduction. Average capacity estimates are imprecise and cannot establish the absence of defensive adaptation. A physician-based coupling calibration admits positive and negative contributions and cannot bound the total defensive response. Among nine pre-policy capacity interactions, telecommunications intensity is positively associated with bed expansion (+4.40 per 10,000 residents; pointwise 95% confidence interval 2.47-6.32; adjusted P = 0.021), although pre-trend concerns limit causal interpretation. Other interactions remain uncertain. The main estimate is close to an illustrative detection threshold, warranting caution with non-significant findings. The results support an exposure-related health co-benefit under explicit accounting assumptions. Assessment incentives, coordination, and financing are possible explanations for uneven responses, but their roles are not identified by these estimates.