Madeleine Gates and Julie Trivitt
Short answer: Yes! It’s an effective predictor of community behavior and cooperation with public health policies.
Long answer: There is no denying that many social factors influence life expectancy and quality of life even beyond the likelihood of having health insurance or the ability to pay for medical care. Researchers, policy makers and think tanks have created a variety of different measures of these social determinants of health and have studied the relationships between the determinants and a variety of health outcomes. Recently, the differences in vaccination rates by geography that correspond with voting patterns in the 2020 election has provided a stark display of how social factors can lead to behaviors with severe health consequences. Here we look at how including voting behavior predicts vaccination rates.
One composite measure of social determinants of health is the social vulnerability index, or SVI, created by the Center for Disease Control (CDC). The latest version, released in 2018, includes measures of socioeconomic status (such as income and educational attainment), housing, transportation, ages of household members, and racial and ethnic diversity in a county to create a metric of vulnerability. This index is meant to describe how vulnerable an area might be to hazardous events, including pandemic outbreaks, so that public health officials and local or state governments can pro-actively target their responses to those who need it the most, rather than waiting to see if certain areas or communities are more harmed by an event.
Just two years after this latest release of the social vulnerability index, America faced the COVID-19 pandemic. Early in the pandemic, there were many discussions of how the pandemic was disproportionately impacting certain groups of Americans. Researchers compared the SVI to COVID-19 incidence and mortality for the first few months of the pandemic using county level data in Louisiana and Michigan, finding a strong correlation between SVI and COVID mortality rates. This led to analysis of all US counties as the pandemic progressed and recommendations for targeted policy interventions, including testing, treatment, and eventually vaccinations, to communities with higher social vulnerability.
After the vaccines were developed, states were charged with planning vaccine distribution. In time, enough vaccines were produced so that every teen or adult who wanted to be vaccinated was eligible free of charge to the individual. Despite attempts to lower barriers and opposition to the vaccine, few counties in the nation reached vaccination rates needed to achieve herd immunity. Numerous articles have pointed out the correlation between vaccination rates and voting patterns, including The New York Times in April, the Denver Post in June, and the Kaiser Family Foundation in July.
Additionally, getting vaccinated is not the only behavior with health consequences that is associated with voting patterns. Public health researchers found a significant connection between health measures and voting decisions over the previous 30 years. Voting preferences have also been shown to be related to the likelihood of evacuating ahead of a hurricane, using tobacco, and eating habits and exercise. They have also been shown to be tied to the probability of complying with COVID-19 prevention recommendations (wearing masks and social distancing) prior to vaccine availability.
To see how important voting patterns are in explaining the vaccination rate, we examine the relationships across four different models using county level data. In each model we estimate the vaccination rate as of August 6, 2021. In model 1 we use only county level SVI measures and a measure of how accurately vaccination data was recorded. Those variables explain over 60 percent of the differences in vaccination rates across counties. In Model 2 we add the ratio of Republican to Democrat voters in the 2020 presidential election. This increases the amount of differences explained by more than 11 percent.
For our third model we take out the voting patterns and add other measures that could influence willingness to receive a COVID-19 vaccine. We include an indicator that a county is in one of the heartland states, the percentage of the population without health insurance, the percentage of the population that expresses strong religious affiliation, the percent of population that got a flu vaccine in 2019, and a measure of the propensity for an arthritis diagnosis. The variables and rationale for including them are summarized in Table 1, below. This model explains about 68% of the differences in vaccination rates, which indicates that this group of variables is less useful than those in group 2 when explaining county-level vaccination rates.
Our fourth model keeps all the measures in Model 3 and adds voting patterns to it. As before, this caused the explanatory power to increase by ten percentage points, from 68 to 78 percent.
Ideally, politics should not influence public health, but we live in a world that is far from ideal. Voting patterns appear to convey information that is highly correlated with behavior and health outcomes but are not currently considered in CDC recommendations based on the Social Vulnerability Index. Voting patterns may reflect a preferred information source, a distrust of government agencies either in general or just when elected officials from opposing parties are in office, or beliefs about social versus individual responsibility. It could be the case that voting preferences and health outcomes are both influenced by a common set of factors such as political ideologies and personal preferences or values, but these traits are difficult to accurately measure or quantify. However, incorporating a measure of voting patterns in an area helps to explain considerably more of the between-county differences in vaccination rates than the current measure of social vulnerability that includes factors like education and income, as shown above.
The relationship between voting preferences, socioeconomic measures, and health behavior is undoubtedly complex and nuanced and we will leave the details of those interactions to the health economists and sociologists. But as long as local voting patterns allow us to predict health behaviors or responses to policy more accurately, the information should be incorporated to maximize the health benefits we get from using our scarce health resources.
¹ The measure of model fit in the chart above ranges from 0 to 1 and can be interpreted as the fraction of the observed difference in vaccination rates is explained by the variables we consider. In statistical terminology, we are reporting the adjusted R-squared in an ordinary least squares regression with robust standard errors. More details on the methodology and data used is available upon request.