Notes: Does not include cases not assigned to counties.
3X3 bivariate choropleths use Fisher-Jenks breaks. For a map using quantile breaks, download the quantile_bivar_map.html file here.
| Dependent variable: | |
| Deaths per Capita | |
| Population Density | 0.18*** (0.02) |
| % 65 and Above | 0.01 (0.02) |
| % Black | 0.35*** (0.03) |
| % Hispanic | 0.14*** (0.03) |
| Median Income | 0.06* (0.03) |
| Gini Index (Economic Inequality) | 0.05** (0.02) |
| Black-White Dissimilarity Index | 0.04** (0.02) |
| Black-White Interaction Index | -0.04 (0.02) |
| % College Degree | -0.07** (0.03) |
| % Living in Correctional Facility | -0.04** (0.02) |
| % Living in Nursing Home | 0.11*** (0.02) |
| % Obese | 0.02 (0.03) |
| % Without Health Insurance | 0.07* (0.03) |
| Primary Care Doctors | 0.01 (0.02) |
| Hospital Beds | -0.02 (0.02) |
| Religious Organizations | -0.06** (0.02) |
| Civic Organizations | -0.06** (0.02) |
| Parks | 0.02 (0.02) |
| Service Orgs Focused on Elderly/Disabled | -0.02 (0.02) |
| State Fixed Effects? | Yes |
| Observations | 3,132 |
| R2 | 0.20 |
| Adjusted R2 | 0.18 |
| F Statistic | 39.85*** |
| Note: |
Variables are population weighted and scaled, with a mean of 0 and standard deviation of 1. ***: p<0.001; **: p<0.01; *: p<0.05 |
Only statistically significant variables at the .05 level are included in the coefficient plot. Other variables can be found in the table to the left.
Bands are 95% confidence intervals.
Data Sources
Data for the analysis come from various sources.
Cases and Deaths
Confirmed cases and deaths come from USAFacts, which collect the data from state and local agencies. The number of viral tests given, at the state level, come from the COVID Tracking Project, but they are not included in the analysis (see below).
Demographic and Socioeconomic Indicators
Demographic and socioeconomic indicators – including total population, racial and ethnic makeup, income measures (both median income and income inequality), and education levels – all come from five-year estimates from the Census Bureau’s 2014-2018 American Community Survey. Data for number of people living in correctional facilities and nursing homes come from the 2010 Decennial Census.
Health indicators
Health indicators – the number of people that are obese, the number without health insurance, and the number of primary care physicians – come from the County Health Rankings. The number of hospital beds comes from the Homeland Infrastructure Foundation-Level Data.
Methods
The analysis on the previous page is a regression of Covid-19 deaths per capita in the U.S. on a number of demographic, social, and economic indicators – all measured at the county level. I include state fixed effects, which tries to account for the fact that counties within a state have similar characteristics, including, in this case, state wide stay-at-home orders. Running the analysis without state fixed effects and including state-level viral testing capacity (which can’t be included in the fixed effects model) doesn’t substantively change the findings.
All of the code to put the data together and to run the analysis is available in the Github repository of this dashboard. All of the data, except the County Health Rankings (as far as I know), are publicly available.
Segregation Indices
I calculated two commonly used segregation indices, the dissimilarity index and the interaction index. The former is the most commonly used index to measure evenness, or how spatially distributed different groups of people are in an area. The latter is an exposure measure. It measures how likely someone from one group is to interact with someone from a different group. In both cases, I am interested in how spatially distributed Blacks and Whites are, and how likely they are to interact, in each county. These measures depend on how one defines the smaller and larger geographic areas. In this case, the smaller geographic area I use for the measure is the Census Tract and the larger one, to match the rest of my analysis, is the county. You can find out more about these measures here, here, and here, and how to calculate them in R here.
In this case, the formula for the Black-White dissimilarity index, for each county, is:
\[\frac{1}{2} * \sum_{i}^{n}\left|\frac{b_{i}}{B}-\frac{w_{i}}{W}\right|\] where \(b_{i}\) is the Black population in \(i\) Census Tract, \(B\) is the total Black population in the county, \(w_{i}\) is the White population in \(i\) Census Tract, and \(W\) is the total White population in the county.
The formula for the Black-White interaction index, for each county, is:
\[\sum_{i}^{n}\left[\frac{b_{i}}{B}*\frac{w_{i}}{t_{i}}\right]\] where all measures are the same as in the dissimilarity index, except that \(t_{i}\) is the total population of the Census Tract.
Although the two measures are certainly related, they are not highly correlated (\(\rho = 0.11\)), so it makes sense to include them both in the analysis.
Note: Does not include cases not assigned to counties.