Using machine learning to estimate the effect of racial segregation on COVID-19 mortality in the United States.

This study examines the role that racial residential segregation has played in shaping the spread of COVID-19 in the United States as of September 30, 2020. The analysis focuses on the effects of racial residential segregation on mortality and infection rates for the overall population and on racial and ethnic mortality gaps. To account for potential confounding, I assemble a dataset that includes 50 county-level factors that are potentially related to residential segregation and COVID-19 infection and mortality rates. These factors are grouped into eight categories: demographics, density and potential for public interaction, social capital, health risk factors, capacity of the health care system, air pollution, employment in essential businesses, and political views. I use double-lasso regression, a machine learning method for model selection and inference, to select the most important controls in a statistically principled manner. Counties that are 1 SD above the racial segregation mean have experienced mortality and infection rates that are 8% and 5% higher than the mean. These differences represent an average of four additional deaths and 105 additional infections for each 100,000 residents in the county. The analysis of mortality gaps shows that, in counties that are 1 SD above the Black-White segregation mean, the Black mortality rate is 8% higher than the White mortality rate. Sensitivity analyses show that an unmeasured confounder that would overturn these findings is outside the range of plausible covariates.