by Simone Ferro and Chiara Serra
We combine mobile locations, weather data, and COVID-19 confirmed cases in a two-way fixed effects mediation model that accounts for time-varying case detection probability and detection lag to estimate the causal impact of weather on the COVID-19 infection rate in the United States, disentangling its direct impact from the indirect effect via the endogenous response of social activity. We show that, while temperature reduces viral infectiousness, it also increases the amount of time individuals spend out of home, which instead favours the spread of the virus. This second channel substantially attenuates the beneficial effect of temperature in curbing the spread of the virus, offsetting one third of the potential seasonal fluctuations in the reproduction rate. The mediation role of social activity is particularly pronounced when viral incidence is low, while it becomes negligible when the epidemiological situation is more serious. Wind speed and precipitation, despite being significant predictors of social activity, do not induce sufficient variation to affect infections. Estimates are robust to an alternative definition of social activity based on the number of visits to indoor venues. This suggests that the negative direct effect of temperature on infections is not driven by individuals spending less time indoors on warmer days, leaving a biological effect of temperature on the virus as the most probable mechanism. Finally, our estimates show that school closures and lockdowns are effective in reducing infections.