Spatial association of mobility and COVID-19 infection rate in the USA: A county-level study using mobile phone location data

Human mobility has been a central issue in the discussion from the beginning of COVID-19. While the body of literature on the relationship of COVID transmission and mobility is large, studies mostly captured a relatively short timeframe. Moreover, spatial non-stationarity has garnered less attention in these explorative models. Therefore, the major concern of this study is to see the relationship of mobility and COVID on a broader temporal scale and after mitigating this methodological gap.

In response to this concern, this study first explores the spatiotemporal pattern of mobility indicators. Secondly, it attempts to understand how mobility is related to COVID infection rate and how this relationship has been changed over time and space after controlling several sociodemographic characteristics, spatial heterogeneity, and policy-related changes during different phases of Coronavirus.

This study uses GPS-based mobility data for a wider time frame of six months (March 20-August'20) divided into four tiers and carries analysis for all the US counties (N = 3142). Space-time cube is used to generate the spatiotemporal pattern. For the second objective, Ordinary Least Square (OLS), Spatial Error Model (SEM), and Geographically Weighted Regression (GWR) were used.

The spatial-temporal pattern suggests that the trip rate, out-of-county trip rate, and miles/person traveled were mostly plummeted till the first wave reached its peak, and subsequently, all of these mobility matrices started to rise. From spatial models, infection rates were found negatively correlated with miles traveled and out-of-county trips. Highly COVID infected areas mostly had more people working from home, low percentages of aged people and educated people, and high percentages of poor people.

This study, with necessary policy implications, provides a comprehensive understanding of the shifting pattern of mobility and COVID. Spatial models outperform OLS with better fits and non-clustered residuals.