Diagnosing Prevailing Trends and Disparate Impacts of COVID-19 at the County-level

Abstract

Since the beginning of the COVID-19 pandemic, there has been a significant amount of research on the dynamics of the disease and its unique effects on different locations. There remains a lot to learn about the spatial and temporal trends of the disease, as well as the disparate impacts on different populations, especially at finer levels of resolution such as U.S. counties. This research attempts to understand the differences between how and when COVID-19 hit counties and provide a narrative for the social and political forces that helped shape those differences. To develop a spatial and temporal understanding of pandemic spread, we first apply Dynamic Mode Decomposition (DMD), a dimensionality reduction technique that allows us to find a linear approximation to a non-linear dynamic system, to county-level COVID-19 deaths. We find that DMD in it most basic form largely fails as a method when applied to COVID-19 data. To develop an understanding of the disparate impacts of COVID-19, we next applied a series of machine-learning models to predict county-level death rates across different waves of the disease using a combination of social vulnerability, demographic, political, and behavioral variables as input. We find that politics played an increasingly important role in determining disease spread as the pandemic matured. We also find that social vulnerabilities are consistently important predictors of disease impact.