Data on cumulative tests and number of positive cases and deaths are available for 217 counties with 50,000 or more population in the U.S. states of California, Florida, New York, Pennsylvania, Texas, and Washington [14]. The number of negative tests in each county was obtained by subtracting total cases from total tests. To control statistically for other factors predictive of COVID-19 mortality, data on 15 factors that were found predictive in previous research as well as the percent who voted for Donald Trump in 2016 [15] were matched to the testing and mortality data from the 217 counties.
The data included five factors that increase the probability of human interaction that would facilitate spread of a contagious virus transmitted by human breath: population density per square kilometer, average number of persons per household, average employees per business, average religious adherents per congregation, and average number of social acquaintances per person reported in a population survey. Four factors that are known to be related to the severity of the disease were included separately: percent of the population with obesity, diabetes, elderly cardiovascular hospitalizations, and persons 65 years and older. Social and economic factors that are often related to health status were also included: percent of adults with at least a high school education, median age of the population, percent unemployment, median family income, income inequality, and percent African American or Hispanic ethnicity.
Population per square mile was downloaded from the U.S. Census Bureau [16] and converted to square kilometers. Estimated 2019 population, percent unemployed, and median household income prior to the pandemic for each county were downloaded from the U.S. Department of Agriculture website based on estimates from the U.S. Census Bureau and Bureau of Labor statistics [17]. Persons per household, social acquaintances, high school graduates, economic inequality, percent 65 years or older, percent with diabetes, and percent obese were downloaded from files accumulated from various sources by the Robert Wood Johnson Foundation [18]. Medicare hospital discharges for cardiovascular diseases were obtained from CDC Wonder [19]. Numbers of religious adherents and congregations were obtained from the Association of Religious Data Archives [20]. Numbers of businesses and employees were downloaded from the Bureau of Labor Statistics [21]. Percent African American and Hispanic were obtained from Dr. Randel Olson’s website [22].
The associations of politics and testing to COVID-19 mortality, corrected statistically for the influence of the other risk factors, were estimated by Poisson regression. In addition, a dummy variable was introduced for 5 of the 6 states to adjust for the aggregate effect of differences in mitigation efforts among the states. Logarithms or square roots, as appropriate, were performed on variables that had skewed frequency distributions. Log (population) was included as an offset variable to correct for differences in population size among the counties. The form of the regression equation is:
Accumulated number of COVID-19 deaths as of April 20, 2021 = b1 (Percent of the vote for Donald Trump in 2016) +.
b2 ((Negative tests for SARS-CoV-2)/1,000,000).
b3 (log (estimated 2019 residents per square kilometer)) +.
b4 (average number of persons per household) +.
b5 (log (average employees per business enterprise))+.
B6 (log (average religious adherents per congregation))+.
B7 (log (average number of social acquaintances.
reported per person)) +.
b8 (percent of the population that is obese) +.
b9 (percent of the population with diabetes) +.
b10 (Medicare cardiovascular hospitalization discharges.
2015–2017) +.
b11 (log (percent of the population 65 years or older)) +.
b12 (percent of adults who finished high school +.
b13 (log (median family income before the pandemic)) +.
b14 (income inequality before the pandemic) +.
b15 (percent unemployed before the pandemic) +.
b16 (√Percent African American) +.
b17 (√Percent Hispanic) +.
b18 (1 if California, else 0) +.
b19 (1 if Florida, else 0) +.
b20 (1 if New York, else 0) +.
b21 (1 if Pennsylvania, else 0) +.
b22 (1 if Texas, else 0).
The other state, Washington, was not included because the model would be over specified. Coefficients on the states are adjustments relative to Washington state, corrected for the other risk factors. When several of the risk factors were found intercorrelated, the association of the risk factors with percent Trump vote and negative tests per population were analyzed using ordinary least squares regression to assess their potential effect on conclusions regarding the hypotheses.