Weather, Social Distancing, and the Spread of COVID-19

Using high-frequency panel data for U.S. counties, I estimate the full dynamic response of COVID-19 cases and deaths to exogenous movements in mobility and weather. I find several important results. First, weather and mobility are highly correlated and thus omitting either factor when studying the COVID-19 effects of the other is likely to result in substantial omitted variable bias. Second, temperature is found to have a negative and significant effect on future COVID-19 cases and deaths, though the estimated effect is sensitive to which measure of mobility is included in the regression. Third, controlling for weather, overall mobility is found to have a large positive effect on subsequent growth in COVID-19 cases and deaths. The effects become significant around 2 weeks ahead and persist through around 8 weeks ahead for cases and around 9 weeks ahead for deaths. The peak impact occurs 4 to 6 weeks ahead for cases and around 8 to 9 weeks ahead for deaths. The effects are largest for mobility measured by time spent away from home and time spent at work, though significant effects also are found for time spent at retail and recreation establishments, at transit stations, at grocery stores and pharmacies, and at parks. Fourth, I find that public health non-pharmaceutical interventions affect future COVID-19 cases and deaths, but that their effects work entirely through, and not independent of, individuals' mobility behavior. Lastly, the dynamic effects of mobility on COVID 19 outcomes are found to be generally similar across counties, though there is evidence of larger effects in counties with high cases per capita and that reduced mobility relatively late.