Estimating the Causal Effect of Mobility on COVID-19 Related Mortality

Abstract

This paper estimates the effect of mobility (social distancing) on future COVID-19 mortality in the United States. By using mobility indices that directly track cellphones, we can estimate the effect of a standard deviation increase in mobility on future mortality without using proxies for mobility. To solve omitted variable bias and measurement error issues we use rainfall as an instrumental variable; to find how far in the future mobility affects mortality we use LASSO. Finally, we decompose the bias in naive OLS into measurement error and omitted variable bias by considering two mobility indices (by Descartes Lab and Unacast). Using both datasets, we estimate two statistically similar effects. A one standard deviation spike in mobility is associated with a spike of 7.34 (15.1) deaths per million people 3 weeks in the future using the Descartes Lab (Unacast) datasets. These numbers are large in magnitude, as COVID-19 resulted in 96.7 deaths per million in the week of April 20, the last week of our data. Finally, our bias decomposition shows that measurement error is a much greater concern than omitted variable bias. This suggests that we should be careful in interpreting regression results using cellphone data that do not consider measurement error.