Constructing a segregation index
We leveraged a spatial and localizable segregation measure based on
census tract data for each ethnoracial group, assigned to the mean
population-weighted centroids of each tract (27, 28). Each county was
projected to the appropriate state-plane projection system, and then we
conducted a Gaussian kernel density smoothing procedure for the total
population using the spatstat package in R (29). We used the same
bandwidth to calculate the smooth for each subgroup. Each
kernel-smoothed ethnoracial subgroup value was divided by the
corresponding value for the total population to create a concentration
measure per tract. Each value was multiplied by 100 to create a percent
value. This approach relies on multiple tracts per county, which was not
true for all counties, specifically in Virginia and West Virginia. When
we encountered such errors, we merged the county with its nearest
neighbor based on centroid distances.