Measurement Error Bias in Air Pollution IV Designs

Instrumental variables are often used to address measurement error in air pollution exposure. This paper shows why that logic can fail when thermal inversion instruments are paired with satellite-based PM2.5. IV recovers the effect of true pollution only if proxy error is uncorrelated with the instrument. Using monitor, satellite, reanalysis, and meteorological data, we show that satellite PM2.5 error is strongly predicted by thermal inversions. In a crime application, satellite-based proxies materially change IV estimates despite strong first stages. The key diagnostic is whether proxy error is orthogonal to the identifying variation.