Tyler Stoeger

Air quality alerts can improve welfare by helping individuals avoid pollution exposure. However, lowering the alert threshold creates a tradeoff: issuing more warnings can encourage avoidance on additional days, but may also change the effectiveness of subsequent alerts. I develop a model of optimal air quality alert policy that accounts for these history-dependent behavioral responses. Determining the optimal threshold requires understanding both how alerts affect outdoor activity and how responsiveness varies with recent alert exposure. I estimate these effects using a fuzzy regression discontinuity design that exploits the jump in alert issuance at a forecasted air quality index (AQI) of 101. The analysis draws on a panel of air quality forecasts, alert issuances, and high-frequency foot traffic data from six U.S. metropolitan areas. Near the threshold, receiving an alert reduces daily visits to outdoor recreational points of interest by approximately 9 percent. However, the effect depends strongly on recent alert history. There is no detectable response when no alerts have occurred in the previous seven days; alerts reduce visits by roughly 16 percent following one or two recent alerts, but the effect attenuates after three or more alerts. I then use these estimates to calibrate the model and conduct counterfactual simulations using data from New York City. The simulations place the welfare-maximizing alert threshold at an AQI of approximately 80, below the current threshold of 101. These results show that optimal warning policy depends not only on the risks communicated by alerts, but also on how repeated warnings shape the effectiveness of subsequent alerts.