Projections for first-wave COVID-19 deaths across the US using social-distancing measures derived from mobile phones

Spencer Woody; Mauricio Garcia Tec; Maytal Dahan; Kelly Gaither; Spencer Fox; Lauren Ancel Meyers; James G Scott

Difference 1: real-time daily social-distancing data. We use data on Americans' actual social-distancing behavior, derived from GPS traces from tens of millions of mobile phones across the country. This data source quantifies two main types of distancing behavior: 1) changes in visitation patterns to public places like restaurants, bars, schools, parks, pharmacies, grocery stores, etc.; and 2) time spent at home versus at work. The IHME model, by contrast, uses a much coarser measure of social distancing: the timing of state-level policy implementations like school closures and stay-at-home orders. But the mobile-phone data reveals substantial differences among states in the timing and extent of peoples' actual distancing behavior, even for states with nominally similar policies. In Texas, for example, many large cities issued their own stay-at-home orders before the state did, affecting the movement patterns of many millions of people days before a statewide policy was in place-a fact that is clearly visible in the data. Our measures capture this substantial state-level and temporal variation that is obscured by regressing only on policies.