Abstract:
We estimate the effect of air pollution on sentiment using social media data from a panel of Japanese cities. To address concerns about potential endogeneity from unobserved simultaneous determinants of air pollution and sentiment, as well as measurement error, we instrument for air pollution using plausibly exogenous variation in atmospheric wind patterns. We find that a one-standard-deviation increase in fine (PM2.5) and small (PM10) particle concentrations reduces overall sentiment by 0.79% and 1.64% standard deviation respectively, which is composed of a more pronounced increase in negative sentiment and a smaller decrease in positive sentiment. Our unique dataset allows us to separately estimate effects on negative sentiment categories including anger, anxiety, and sadness. Our results suggest sentiment as one candidate mechanism, besides physiological and cognitive pathways, to explain the increasingly evident non-health damages from air pollution exposure on work productivity, road safety, sleep and crime.