Instrument Validity in Air Pollution Impact Studies

Instrumental variable (IV) methods dominate empirical studies of air pollution’s impacts, yet their validity often rests on untested exclusion restrictions. Using comprehensive Chinese pollution and weather data, we study three recurring threats that can arise in common IV designs. First, simulations demonstrate co-pollutant bias, where instruments influence multiple pollutants. Second, correlations between pollution proxies and meteorological instruments can generate non-classical measurement error. Third, changes in local infrastructure can affect commonly used meteorological instruments and potentially confound exogeneity. We illustrate the empirical relevance of these threats in an application relating air pollution to daily crime in Chinese cities and provide practical guidance for diagnosing and addressing these risks in environmental IV designs.