The data sources for our index of American libertarian conservatism originate from the Rand Corporation’s TL-354 state-level estimates of firearms per household, data from the Tax Foundation giving the highest marginal tax rate for each state, and data from the National Council of State Legislatures on the stringency of state laws regulating the private use of cannabis. The data for our index of American social conservatism includes CDC/Guttmacher Institute data on the number of abortion clinics by state per million residents, Public Religion Research Institute data on pre-Obergefell restrictions on samesex marriage, and the JRANK Legal Encyclopedia’s rating of U.S. state-level support for the legality of prayer in public schools. Table 1 presents summary statistics of our variables, showing differences in characteristics and COVID-safety behaviors, cases, and deaths between counties above and below a 50% county-level Trump vote in the 2020 election. On average counties with a majority Trump vote are 3.5 years older, have $5,473 lower median household income lower income, have a smaller percentage of Black, Latino, and Asian residents, have economies that are more oriented toward manufacturing and less toward services, and have somewhat poorer baseline health, ranking about 2.7 percentile points below the mean of non-Trump counties. Raw differences in mask-wearing show a 17.1 percentage point difference in means for mask-wearing between Trump and non-Trump counties,grow rack 0.50σ lower rate of sheltering at home, 12.9 percentage point lower rate of COVID vaccination, and a 1.12σ lower COVID-safety index, all significant at p < 0.001.
Raw differences in health outcomes show an additional 2,300 COVID cases and 53.4 additional deaths per 100,000 residents across counties, also both significant at p < 0.001. Figure 2 shows spacial correlations across U.S. counties by 2020 presidential voting, COVID-safety behaviors, and COVID deaths per 100,000. Figure 3 provides a series of plots by state that show self-reported mask-wearing by county-level Trump voter support. In nearly each map, the blue dots lie to the northwest of red dots, which show higher levels of county voter support for President Trump and lower rates of mask-wearing. Table 2 provides the results of our state-level fixed-effect estimations in which we regress COVID-safety behaviors on the 2020 Trump vote, controlling for county characteristics. All estimates use U.S. state-level fixed effects so that identification is off differences in the Trump vote and COVID-safety behaviors between counties within a given state. Our results examine three COVID-safety variables: county-level mask-wearing, sheltering at home, and COVID-vaccination rates. We aggregate these three variables into a COVIDsafety index using the method of Kling et al. , which is the standardized sum of each of these three individually standardized variables. The results in Column 1 of Table 2 suggest that a 10 percentage point increase in the county Trump vote is associated with a 3.9 percentage point decrease in the number of people in a county stating that they wear a mask ”all the time” in public . Column 2 shows a tightly estimated coefficient close to zero for sheltering at home. Column 3 indicates that a 10 percentage point increase in the county Trump vote is associated with a 5.1 percentage point decrease in the vaccination rate . A visualization of the significant mask-wearing and vaccination results are shown over the scatter plots provided in Figures 4A and 4B, respectively. Our estimates in Column 4 suggests that a 10 percentage point increase in the county Trump vote is associated with a 0.23σdecline in our COVID-safety index.
Estimates on control variables across columns indicate that COVID-safety behavior generally increases with county median age, median income, percent Latino, percent Asian American, percent employed in manufacturing, education and health, and baseline level of county health. In general the controlled estimates display COVID-safety behavior associations with the Trump vote that are lower than raw differences between Trump and non-Trump counties, but differences by the county Trump vote remain and are highly significant. The exception is the sheltering-at-home variable, which we do not find to be affected by the Trump vote once age, income, and other county level covariates are controlled.Table 3 shows the relationship between the county-level 2020 Trump vote and COVID cases and deaths. Vaccines began to be widely available in the U.S. roughly toward the end of February 2021, and columns 1 and 2 show the impacts on COVID cases, pre- and post-vaccine respectively. Here results indicate that a 10 percentage point increase in the county Trump vote is associated with 789 additional cases per 100,000 residents in roughly the first year of the pandemic in the U.S., and 1,394 additional cases in the pandemic’s first twenty months . Columns 3 and 4 of Table 3 show outcomes for COVID deaths pre- and post-vaccine, where a 10 percentage point increase in the county Trump vote is associated with 11.5 additional COVID deaths per 100,000 residents in the pre-vaccine first year of the pandemic and 27.6 additional COVID deaths per 100,000 country residents in the first twenty months of the pandemic . A visualization of these results is provided in Figures 5A and 5B. A comparison of the pre- and post-vaccine results in Table 3 is important because February 2021 represents both the first month after President Trump’s departure from office as well as the month during which U.S. vaccination rates most quickly accelerated among the general public. The differences between columns 1-2 and 3-4 show that the statistical relationship between Trump support and COVID infections and deaths not only failed to narrow after February 2021, but actually grew, rising from about 65.7 added cases and 0.96 deaths per month during the first twelve months of the pandemic to 76.2 cases and 2.00 deaths per month during the subsequent eight months.
This divergence may have occurred because the easing of public restrictions that accompanied the vaccination-availability phase of the pandemic may have actually elevated risks to the unvaccinated and those not wearing masks in public spaces, a phenomenon likely exacerbated by the deepening political polarization over COVID safety behaviors. How confident can we be that the higher rates of COVID infection in Trump-supporting counties resulted from reduced COVID-safety behavior? The largest and most influential controlled study to date estimating the causal effects of mask-wearing on COVID infection is the extensive randomized trail carried out by Abaluck et al. , who report results from a mask-wearing intervention implemented among 342,183 adults across 600 villages in Bangladesh. The Abaluck et al. study, referenced worldwide by governments as both motivation and guide for COVID-safety behavior, reports that the 27.9 percentage point difference between treatment and control groups, created by those induced to wear masks through the experiment, reduced symptomatic COVID infection in the treatment population by 0.91 percentage points within a two-month time-frame,greenhouse grow tables or that every single percentage point increase in mask-wearing reduced COVID cases by an average of 0.016 percentage points per month. Our results for the United States associate a 10 percentage point increase in the county-level Trump vote with a 5.11 percentage point reduction in mask-wearing and an increase in COVID cases of 1.39 percentage points over a twenty-month time-frame, or that every percentage point increase in mask-wearing reduced COVID cases by an average of 0.014 percentage points per month. Our non-controlled U.S. results are thus remarkably similar to the Abaluck et al. controlled estimates of the causal effect of mask-wearing on symptomatic COVID cases, and they suggest that higher levels of COVID infection in the high Trump-vote counties are unlikely to be due to unobserved county characteristics but rather are substantially mediated by differences in COVID-safety behaviors.In this section we ascertain to what degree the COVID-safety behaviors are specifically tied to a political identity specifically tied to Trump voter support relative to two traditional strains of conservatism in the United States: American social conservatism and American libertarian conservatism, which for many decades have formed key parts of the conservative political coalition embodied within the Republican party. Our Social Conservatism index is made up of three standardized and equally weighted state-level variables representing political issues strongly tied to social-conservative concerns in the United States. They include 1) the availability of abortion, given in terms of clinics per million residents; 2) restrictions on same-sex marriage, existing before federal legalization under the Obergefell v. Hodges, 576 U.S. 644 2015 Supreme Court ruling; and 3) the degree of support for public prayer in schools, as given from ratings by the JRANK Legal Encyclopedia.
Our Libertarian Conservatism index also consists of three standardized and equally weighted variables, representing issues of concern to libertarian conservatives, which embody a preference for minimal taxation, regulation, and interference in personal freedoms and private life by government: 1) low state taxes at the highest income brackets; 2) estimated per-capita firearms ownership; 3) the stringency of state laws regulating the private use of Cannabis. Consistent with the procedure of Kling et al. , after summing each standardized variable within the index, each index itself is then itself subsequently standardized. Table 4 shows our estimates in which we compare associations of our Libertarian Conservatism index, Social Conservatism indices, and Trump voter support with COVID safety behaviors, cases and deaths. To facilitate comparisons with these indices, we also standardize our Trump-vote variable. Overall, results show that once the Trump 2020 vote is taken into account, the Social Conservatism index and Libertarian Conservatism indices have very little systematic explanatory power over COVID-safety behaviors, cases and deaths. In these estimations, a one-standard deviation difference in Trump support decreases mask-wearing by 5.2 percentage points . While the Social Conservatism index is also statistically significant , it explains less than half the variation of Trump support.While neither standardized Trump support nor the Social Conservatism index is statistically significant in explaining sheltering at home, the Libertarian Conservatism index is actually positive and modestly significant . For vaccination rates, a one standard deviation difference in Trump support reduces the percent of adults vaccinated by 7.9 percentage points while the Social Conservatism and Libertarian Conservatism indices have the unexpected positive sign . The aggregated COVID-safety index shows that a one-standard deviation increase in Trump support decreases COVID-safety behavior by 0.45σ . Once Trump support is controlled, Libertarian Conservatism is actually positive with about half the effect but in the direction of increased COVID safety. As is the COVID-safety index, COVID cases and deaths are strongly associated with the 2020 Trump vote but are insignificantly associated with Social Conservatism and Libertarian Conservatism indices. A one-standard deviation increase in Trump voter support increases COVID cases by 2,075 per 100,000 and COVID deaths by 52.4 per 100,000 residents. These results are consistent with the fact that there are relatively weak historical ties between, for example, an anti-vaccination stance and traditional American conservatism. Indeed, some such as Berezow and Campbell trace the anti-vaccination movement to the political left rather than the political right, although there is evidence that in the last decade that anti-vaccination stances have migrated from the libertarian left toward the libertarian right . Rather than being closely tied to conservative ideology per se, numerous both pre- and post-COVID studies find the factor most strongly tied to an anti-vaccine stance to be a belief in conspiracy theories . A number of papers have linked reduced levels of COVID-safety behaviors to the Trump presidency itself . In a quantitative study of behavioral responses to social media activity, Germani and Biller-Andorno find Twitter posts by President Trump to be the strongest driver of information contradicting established scientific evidence regarding the COVID vaccines. Thus rather than originating from a particular strain of traditional American conservatism, this body of evidence taken together with our results in Table 4 strongly suggest an identity more specifically related to Trump political support to principally drive differences in COVID-safety behaviors, along with the accompanying COVID infections and deaths. that bleed across state lines. As a robustness check, we estimate our models in Tables 2, 3, and 4 with Conley standard errors that account for large spatial correlations in COVID cases and deaths. Using longitudinal and latitudinal coordinates, we employ a Bartlett linear decay of the spatial error correlation to a distance of 500 kilometers from the center of each county. In estimations without state-fixed effects, we account for spatially correlated errors 500 km from the center of each state.