Awareness of COVID-19 at the Local Level: Perceptions and Political Consequences

Abstract

Context: Although the COVID-19 pandemic has affected all Americans, its effects have been unequally distributed across geographic areas. These variations in the pandemic's severity—and public perceptions thereof—likely have political consequences. This study examines the factors that shape perceptions of COVID-19 at the local level and assesses the consequences of these perceptions for public opinion and political behaviors.

Methods: The authors use questions from the 2020 Cooperative Election Study linked with county-level COVID-19 rates to examine predictors of respondents' perceptions of the pandemic's severity in their county, including demographic, political, and informational characteristics. The study also examines whether these perceptions are associated with public opinion and voter behavior.

Findings: Respondents' perceptions are correlated with case rates. Liberals and Democrats estimate the pandemic to be more severe than Republicans and conservatives do, as do CNN viewers compared to Fox News viewers. The study found only limited evidence of a relationship between perceptions of the pandemic in a respondent's county and political outcomes.

Conclusions: The results add to the accumulating evidence that both news media and political predispositions shape perceptions of COVID-19, and they raise important questions about whether and how the pandemic has shaped—and will continue to shape—political outcomes.

Although the COVID-19 pandemic has had a profound effect on the lives of virtually all Americans, these effects have not been evenly distributed across the country. At any given time since it first began to spread in the United States, the pandemic's severity has varied widely by region, and also across demographic groups within regions. At times, some communities have faced the grim spectacle of total lockdowns, overwhelmed hospitals, and refrigerated morgue trucks, even as life continued relatively normally in other places. While some of the economic, social, and political impacts of the disease have been national in scope, the most disruptive and severe effects have been localized: spikes in infection rates, hospitalizations, and deaths; state and local lockdown orders and mandatory masking ordinances; and localized business closures and layoffs, to name a few. These localized outbreaks, and the policy responses to them, may have profound political consequences.

Emerging research has begun to explore how community-level public health threats shape political outcomes. This research has found, for instance, that poor public health conditions at the county level in the United States—such as reduced life expectancy resulting from suicidality and drug overdoses—are associated with political outcomes at the community level, including aggregate voting patterns (Bilal, Knapp, and Cooper 2018; Bor 2017; Goldman et al. 2019; Monnat and Brown 2017; Wasfy et al. 2020; Wasfy, Stewart, and Bhambhani 2017). In particular, research has suggested that poor community health is associated with county-level vote shifts toward Republican candidates in 2016 and 2018 (Wasfy et al. 2020; Wasfy, Stewart, and Bhambhani 2017). However, what is often missing from such studies that focus on aggregate public health conditions and political outcomes is attention to individual-level mechanisms. For localized public health conditions (including outbreaks of infectious disease) to affect political outcomes, the public must perceive and respond to those conditions. In other words, we posit that individual-level perceptions are a necessary mechanism linking public health trends and political behavior, if the two are truly related.

Why Public Perceptions Matter

Research in political science and related disciplines has demonstrated that the ways in which the public perceives a social or policy problem has political consequences. For example, research on the social construction of problems suggests that public perceptions of a problem will shape what type of policy response that problem receives, the public's support for that policy approach, and whether the public holds elected officials accountable for addressing the problem (Schneider and Ingram 1993; Stone 1989). If a problem such as an infectious disease is determined to be something natural or inevitable, then the public may not hold politicians accountable for handling it, as compared to when the story surrounding that problem emphasizes intentional causes and adverse consequences (Stone 1989).

The public's perceptions of a problem can be shaped by many sources. One source, of course, is the actual objective conditions—the real death or infection rates in a community—that can shape whether the public perceives the public health problem as severe or not. Gollust and Haselswerdt (2021) found that people living in communities with higher rates of overdose mortality were in fact more likely to perceive the problem of opioids to be more severe. While this past work demonstrates that objective indicators can shape the public's perceptions, research also demonstrates that other factors, including how these issues are portrayed by media and the organizations that monitor those conditions, also matter (Jones and Baumgartner 2005; Yanovitzky and Weber 2019). For instance, perceptions of social conditions can be shaped by news media, which tend to downplay some issues and overemphasize others (Armstrong, Carpenter, and Hojnacki 2006).

COVID-19, Partisanship, and the Media

One critical factor that shapes public perceptions of and the political response to COVID-19 is partisanship. From the very beginning of the pandemic's emergence, the public was poised to interpret the pandemic through a political lens, coming on the heels of the polarized debates over the Affordable Care Act and in the context of a polarized electorate in an election year (Gollust, Nagler, and Fowler 2020). In fact, President Trump's very first speeches addressing the pandemic downplayed the threat, and Republican elites and media commentators echoed this perspective, contributing to the almost immediate divisions among the public about how grave the threat of COVID-19 really was. By March 2020, most aspects of the pandemic demonstrated a partisan divide, with Republicans perceiving the pandemic as less concerning than did Democrats (Gadarian, Goodman, and Pepinsky 2021; Moniz 2022). These divides were observed in survey data collected throughout 2020 (Barry et al. 2021) and were apparent into 2021, with Democrats much more likely to be vaccinated against COVID-19 than Republicans (Kirzinger et al. 2021).

While examining the many reasons for the partisan divide in all aspects of COVID-19 politics is beyond the scope of this study (but see Gadarian, Goodman, and Pepinsky 2022), a few contributing factors relate directly to how Americans perceive the problem of COVID-19. First, studies confirm that news media coverage of the pandemic consistently featured politicized content (Hart, Chinn, and Soroka 2020), and the content differed based on the political lean of the news outlet (Motta, Stecula, Farhart 2020). Survey data demonstrate that people who relied on different news sources had opinions and attitudes about the pandemic that varied according to the news source in question, such as understanding of disparities (Gollust et al. 2020) and endorsement of conspiracy beliefs (Jamieson and Albarracin 2020). Second, the availability of public health information also differed across local jurisdictions; for instance, one study showed that county health departments were more likely to publicize county-level information about COVID-19 incidence in Democratic-leaning counties than in Republican-leaning counties (Hansen et al. 2021). These factors suggest that information sources will affect public perceptions of COVID-19, and that political affiliation might not only independently shape public perceptions of COVID-19 but also condition responses to local information.

The partisan or ideological lean of news and information sources is not the only relevant consideration, however. News sources also differ in terms of their geographic focus, which may have implications for how well Americans understand the severity of the pandemic in their own communities. In his analysis of the “nationalization” of American political behavior, Hopkins (2018) finds that Americans' attention to state and local politics has suffered as their focus on national politics has increased, and that this shift in attention is driven in part by changing media consumption habits. Over the last few decades, Americans have largely forsaken geographically bounded news sources (e.g., local newspapers) in favor of nationally focused media, such as cable news and the internet. It is reasonable to expect that Americans who focus on national news sources at the expense of local ones will have a poorer understanding of the COVID-19 situation in their own communities, even if they are well informed about the pandemic in general.

COVID-19 and Political Outcomes

Almost as soon as COVID-19 reached US soil in early 2020, it became evident that the pandemic would have electoral consequences. The virus had immediate and devastating consequences for Americans' health and the economy, with US workers losing more than 20 million jobs in April 2020 (BLS 2020). The pandemic continued to create economic hardship for low-income Americans throughout 2020 (Parker, Minkin, and Bennett 2020). Because COVID-19 was an extremely salient issue for Americans—one that affected self-interest, pocketbook concerns, and broader outlooks about the country's direction—it seemed obvious that Americans' perceptions of COVID-19 would also affect their perceptions of politics, and particularly of the 2020 presidential election. It is well known from political behavioral studies that voters consider incumbent performance and the broader economy in their judgements of candidates (De Benedictis-Kessner and Warshaw 2020; Fiorina 1978).

A few studies have examined the electoral consequences of COVID-19 in the United States directly.¹ An early analysis by the Associated Press suggested that counties with the highest number of incident cases went to Trump (Johnson, Fingerhut, and Deshpande 2020). In contrast, two studies contribute evidence in the opposite direction. In a study conducted and published before the 2020 general election, Warshaw, Vavreck, and Baxter-King (2020) used voter surveys to determine that people living in states with higher COVID-19 death rates were less likely to support Trump (both in terms of voting preference and overall approval), and that voters in such states were also less likely to support Republican candidates. Using actual voting data, Baccini, Brodeur and Weymouth (2021) subsequently found substantial evidence that higher COVID-19 incidence was negatively related to Trump vote share in 2020. They also found that COVID-19 incidence was somewhat related to higher voter turnout and that local unemployment rates were not related to voting. Considering the relationship from the opposite direction, another ecological-level study (Eden et al. 2021) considered whether pre-COVID election results relate to COVID-19 case and death counts in early 2021. The authors found that both case rates and mortality rates were higher in Republican-voting counties.

Baccini, Brodeur, and Weymouth (2021) suggest that their evidence indicates that voters affected by the COVID-19 pandemic may have “punished” Republicans (and especially Trump) for their mismanagement of the pandemic in 2020 and/or that their preferences became realigned with President Biden's, perhaps because of financial hardship and need for an expanded social safety net. However, such explanations require a closer look at the relationship between the pandemic and individual-level perceptions. Our research begins to fill this gap by examining the political predictors of perceptions of the severity of the COVID-19 pandemic in 2020 as well as assessing the consequences of those perceptions for public opinion and political behaviors.

Research Objectives

We have two major objectives for this study. First, we examine the predictors of public perceptions of the county-level severity of the COVID-19 pandemic during fall 2020, assessing whether survey respondents are attentive to local measures of the burden of disease. We also assess whether certain characteristics—particularly partisanship and measures of knowledge and media consumption—moderate the relationship between objective public health indicators and public perceptions. Second, we examine whether public perceptions of the severity of the COVID-19 pandemic shape selected political outcomes: public opinion about pandemic policies, turnout in the 2020 election, voting for Trump, voting for Republican candidates for the US House of Representatives, and approval ratings for national and state elected officials. By focusing on the potential political effects of perceived COVID severity rather than of objective COVID statistics, we seek to establish a microlevel mechanism linking health trends with political outcomes. Although our approach is largely exploratory, we do begin with some expectations: that actual severity of the pandemic in a county should be reflected in respondents' perceptions, especially among those with a high level of information exposure; that Democrats and liberals should perceive the pandemic as more severe than Republicans and conservatives do; and that viewers of Fox News should perceive the pandemic as less severe than viewers of other networks. Furthermore, we explore whether these same characteristics (information exposure, partisanship and ideology, and specific news outlet use) also condition the relationship between perceptions of pandemic severity and objective measures of pandemic severity.

Data and Methods

Our primary data source is a module of 1,000 respondents to the 2020 Cooperative Election Study (CES). The CES, formerly known as the Cooperative Congressional Election Study, was first fielded in 2006, and is administered by the survey research firm YouGov in collaboration with university-based researchers. The CES recruits a nationally representative sample of more than 50,000 American adults (61,000 in 2020) from YouGov's online panel through a procedure called sample matching, and sampling weights are provided to address deviations from the national population in the completed sample (see Schaffner, Ansolabehere, and Luks [2021] for details). Respondents participate in both pre- and postelection surveys. University-based teams purchase randomly selected modules of 1,000 respondents each. Roughly half of each survey consists of a battery of political and demographic questions referred to as the Common Content, which all respondents answer, while the other half is designed by the university research teams and is specific to a given module.

Our measure of local COVID-19 perceptions is a module question fielded on the 2020 preelection CES between September 29 and November 2. The question reads: “The coronavirus pandemic has affected some areas in the United States more than others. Thinking about your own county, would you say that the percentage of people who have gotten sick with the virus has been higher or lower than the national average, or do you think it's about the same as the national average?” Respondents then chose from three response options: “Higher than the national average,” “About the same as the national average,” and “Lower than the national average.” This results in a three-category ordinal variable with “Higher than the national average” as the highest value.

The use of broad, relativistic categories for this question was deliberate. It is well established that the mass public lacks precise knowledge of numerical facts (Lawrence and Sides 2014), and it is unlikely that many Americans could report precise case numbers from their counties on demand. Moreover, we argue that precise numbers are less politically relevant than an individual's perception than their county is better or worse off than the country as a whole.

Our module also includes a battery of questions on the postelection survey (fielded November 8 to December 7) measuring respondents' support for five COVID-related public health measures: mask mandates, stay-at-home orders, bans on gatherings, closing businesses, and closing K-12 schools. For each policy, respondents rated their support on a 5-point Likert scale ranging from “strongly oppose” to “strongly support.” The five items scale very well (α = .94), so we use respondents' mean response across the five items as a measure of support for restrictive COVID-19 mitigation measures.

The Common Content provides us with a number of other questions that we use in the present study as additional political outcomes: approval rating of political figures, including Trump, the US Congress, and respondent's governor; vote choice in the presidential and US House elections; and voter-file-validated turnout. The survey also includes a range of variables we use as covariates in the analysis, including party identification, self-reported ideology, education, news media use, social media use, political knowledge, race and ethnicity (which we operationalize with dummy variables for Black, Hispanic, Asian, and other minorities, leaving non-Hispanic whites as the excluded category), gender, and age. For news media use, we use a summary measure of the number of media-related activities reported for the previous 24 hours (watching TV news, reading a newspaper, listening to radio news, and using social media) in some models and more specific indicators in others, including watching local and national newscasts, Fox News, CNN, MSNBC, and broadcast network news; listening to radio news; and reading a newspaper.² Descriptive statistics for all survey variables are displayed in online appendix A, while wording for all Common Content questions is publicly available in the 2020 CES guide (Schaffner, Ansolabehere, and Luks 2021).

To determine the correspondence between perceptions and reality with regard to COVID-19, we require a county- and date-specific measure of COVID cases. We use daily cumulative county-level case counts from the Johns Hopkins University Coronavirus Resource Center.³ We then use county identifiers in the CES to match each respondent to the cumulative case number per 100 residents in the respondent's county on the day before they took the survey (since many respondents likely took the survey before case numbers were reported on that day). While this is our primary measure of county-level pandemic severity, we also collected the number of new cases and the seven-day new case average (both measured the day before the survey) as well as statistics on COVID-19 deaths (cumulative, new, and seven-day average). All measures are population-adjusted, with different denominators to make them roughly comparable in scale. Descriptive statistics are provided in online appendix table A.1.

Our analytic strategy is relatively straightforward. First, we examine the bivariate relationship between county-level case rates and perceptions of the pandemic. Then, we estimate a series of ordered logit regression models to examine whether county-level case rates are related to perceptions of the pandemic (with a robust set of covariates). Because we anticipate that the relationship between the real county-level severity of the pandemic and perceptions thereof will vary based on respondents' characteristics (partisanship and use of information, including specific news media sources reported), we also estimate a series of models with interaction terms, and we plot the resulting marginal effects.

Last, we use a set of regressions to examine how perceptions of the pandemic relate to the political outcomes described above. In these analyses, we operationalize respondents' county-level COVID-19 estimates with dummy variables indicating “below average” and “above average,” leaving “average” as the excluded category. (In a separate analysis reported in online appendix D, we replace these perceptions with the actual case rates to assess whether community-level indicators of disease severity correlate with political outcomes; we discuss these results at the end of this article's Results section.) This flexible approach does not impose a linear functional form on the relationships between estimates and political behavior, if any exist. We also include all of the covariates used in our models of perceptions. In the model of gubernatorial approval, we convert the party identification measure into a “partisan distance” measure that reverses the scale as appropriate, since the partisanship of the governor varies across respondents. For example, a strong Democratic respondent in a state with a Republican governor is coded as maximally distant in the “governor” model, while a strong Republican is coded as minimally distant.

Results

We begin with a simple cross-tabulation of respondent perceptions of COVID-19 severity in the respondent's county with the actual cumulative case rate, divided into population-weighted terciles. The results are displayed in table 1. The public's understanding of the severity of the pandemic in their communities is clearly imperfect; for example, in the hardest-hit tercile of counties there were more respondents who believed the county case rate was average than who correctly believed it was above average. Still, perceptions are clearly responsive to reality. Moving from the lowest to the highest tercile, the percentage answering “below average” declines, while that answering “above average” increases. Severe misperceptions (belief that the case rate is below average in an above-average county or vice versa) were relatively rare.

Predictors of Perceptions

Next we turn to multivariate ordered logit models, using the three-point county COVID estimate measure as our dependent variable, to test the association of the real case rate with perceptions along with other covariates. The results are displayed in table 2. Models 1–3 include covariates measuring respondents' demographic characteristics, their general information and knowledge level, and their political beliefs, respectively, while model 4 includes all covariates. In all specifications, the actual cumulative case rate in a respondent's county has a positive and statistically significant (p < .01) association with their perception of the pandemic's local severity.⁴ The magnitude of the relationship is similar across specifications, with an increase of 1 COVID case per 100 residents (slightly less than a standard deviation) increasing the probability of answering “above average” by about 5% or more.

Few other covariates have a significant relationship with estimates, but those that do bear mentioning. First, it is troubling that older respondents gave significantly lower estimates of severity than younger respondents regardless of the actual case rate, since COVID-19 is especially dangerous for the elderly. Second, we see evidence of political polarization in public health perceptions, with Republican and conservative respondents estimating the pandemic to be less severe in their counties than did other political groups (again, irrespective of the actual case rate). While the partisanship and ideology variables are each only marginally significant in model 3 and insignificant in model 4, these results greatly understate the joint impact of these highly correlated variables. A postestimation test from model 4 indicates that the coefficients for partisanship and ideology are jointly statistically significant at the p < .0001 level. This model predicts that a respondent who identifies as very liberal and a strong Democrat has only a 20% chance of answering “below average,” compared to a 46% chance for a very conservative strong Republican, holding actual case rate and all other variables at their means. Interestingly, the models show that none of the knowledge or information variables have a statistically significant relationship with respondents' perceptions of local pandemic severity.

We also note that women gave significantly more severe estimates than men. While women are not considered a high-risk group in biological terms,⁵ this may reflect the pandemic's disproportionate economic impact on women relative to men (Madgavkar et al. 2020). Lastly, while the coefficient identifying Black respondents is positive and marginally significant in model 1, which could reflect heightened concern among a high-risk subpopulation, the coefficient is considerably smaller and statistically insignificant in model 4. This suggests that the apparent difference between Black and white respondents (the excluded category) was attributable to other factors, such as partisanship and ideology. We find no other evidence of differences in perceptions across racial or ethnic groups.

We also consider the possibility that personal characteristics (specifically, a person's information level or their political beliefs) may condition the relationship between perceptions and reality. If certain types of people (e.g., people identifying with a specific party or those who are generally knowledgeable) are more attentive to public health information, the magnitude of the association of the real case rate with their perceptions should be stronger than that found among less attentive respondents. To test for this, we conduct a series of ordered logit models with interaction terms of selected covariates with the real case rate. Figures 1 and 2 illustrate the results of these models for measures of information level and political beliefs, respectively, using predicted probabilities of an answer of “above average” to the COVID county estimate question, with 90% confidence intervals.⁶ While the vertical space between the lines for different groups reflects some of the baseline differences identified in table 2, here we focus on differences in slopes indicating a conditioning effect.

In figure 1, we explore the question of whether Americans who are generally better informed and more knowledgeable are also more attentive to the COVID case rate in their counties, and we see some evidence that this is true. The slope is considerably steeper for more educated (p = .02 for the interaction term) and more politically knowledgeable (p = .08) respondents. There is also evidence of a stronger effect of the real case rate among those who consume news from higher numbers of news media sources (p = .12),⁷ although not among intense social media users (p = .48). This may reflect a difference between geographically bounded and nonbounded information sources, per Hopkins (2018), a possibility we explore in more depth below.

To our surprise, figure 2 shows only slight differences in slopes between those at opposite ends of the political spectrum, demonstrating that the relationship between case rates and perceptions of the pandemic does not vary widely based on respondents' political predispositions. The gaps between the solid and dashed lines show that overall, Democrats or liberals perceive the pandemic as more severe than their Republican and conservative counterparts, providing evidence for the polarization dynamic we discussed above. Importantly, though, polarization does not make Republicans or conservatives less attuned to the facts on the ground in their counties. In fact, if anything, Republicans may be more responsive to facts than Democrats, as suggested by the narrowing of the gap between strong Republicans and strong Democrats as the cumulative case rate increases, although this interaction term does not approach conventional levels of statistical significance (p = .30), and the narrowing is much less pronounced in the ideology interaction (p = .54).

We now consider the possibility that different types of media use may have different associations with perceptions of the pandemic. Table 3 displays the results of models incorporating more specific media use variables. Models 1 and 2 include measures of watching national and local newscasts, listening to radio news, and reading a newspaper. We find no consistent evidence that any of these sources or activities affected respondents' baseline likelihood of perceiving the pandemic as severe in their communities. However, model 2 offers some suggestive evidence that people more attuned to local news perceive the pandemic as more severe in their local communities. Models 3 and 4 replace the local and national newscast variables with indicators of watching the major cable news networks and broadcast network news. Here, the effect of the partisan or ideological lean of news outlets is striking: regardless of the real case rate, respondents who watched Fox News gave significantly less severe estimates, while those who watched CNN gave significantly more severe estimates. While the coefficients shrink once control variables (notably, partisanship and ideology) are added in model 4, they remain statistically significant and substantial in size. Holding partisanship, ideology, case rate, and all else constant, Model 4 predicts that watching Fox News reduces the likelihood of answering “above average” by about 6 percentage points, while watching CNN increases that likelihood by about 9 percentage points, resulting in a 15-percentage-point difference in predicted probability between a respondent who watched one network and not the other. Watching MSNBC (generally considered to be a liberal and Democratic-friendly outlet) and broadcast news, by contrast, have no significant association with estimates.

Next, we explore whether these different media sources make respondents more responsive to the facts on the ground, using the same interaction term approach employed above. Figure 3 displays the results for national and local TV newscasts, radio, and newspapers. Despite our expectations, the results do not tell a straightforward story about national versus local news consumption. Neither national nor local TV newscasts significantly conditioned the effect of the case rate on perceptions, nor did reading the newspaper. Listening to radio news, however, did make perceptions significantly more responsive to case rate. Perhaps this finding reflects local radio news coverage, although the question does not distinguish between different types of radio news.

Figure 4 explores possible conditional relationships involving specific cable news sources and broadcast news. Here, the pattern for Fox News viewers and nonviewers is similar to what we observed in figure 2 for partisan and ideological groups: Fox News viewers gave significantly less severe estimates than nonviewers, but they were not less responsive to the real case rate. By contrast, watching CNN appears to make respondents more attuned to the real case rate in their communities, even though it is a national news source. Watching MSNBC and broadcast news did not condition the relationship of case rates with severity perceptions.

Although we focus on how these variables condition the relationship between facts on the ground and perceptions of COVID-19’s local severity, these analyses do not directly answer an important question: do these characteristics make individuals' estimates more or less likely to be accurate? We turn to this question in additional analyses in online appendix E. We do so with caution, because our severity perception item was not designed to separate “correct” and “incorrect” answers; doing so requires us to rely on arbitrary thresholds surrounding the “average” range.

Appendix table E.1 shows the results of a simple logit model of “accuracy,” with the dependent variable coded 1 if a respondent's answer corresponded to their actual county tercile and 0 otherwise, using all demographic and political variables as well as the specific media consumption variables as covariates.⁸ We find a positive and statistically significant relationship of political knowledge with the likelihood of an accurate estimate; no other coefficients were statistically significant. Tables E.2 and E.3 distinguish between overestimates and underestimates, although doing so requires us to consider respondents from the three terciles of counties separately.⁹ While some findings are statistically significant, few clear patterns emerge across the different types of counties to suggest consistent predictors of under- or overestimation. The one exception is age: older respondents were significantly more likely to underestimate COVID severity in both above-average and average counties. This is a troubling pattern given the vulnerability of older people to COVID-19. Other significant relationships are inconsistent; for example, watching Fox News was associated with a decreased likelihood of overestimating in below-average counties and an increased likelihood of underestimating in above-average counties (table E.2), but this covariate showed no significant relationship with either form of inaccuracy in average counties (table E.3). We are reluctant to draw firm conclusions from these results given their ambiguity, the nature of the survey question, and the reduced statistical power of these subgroup analyses.

Perceptions as Predictors of Political Outcomes

We have demonstrated that objective case rates strongly predict respondents' perceptions of COVID-19 severity in their local community, as do their political orientation and the types of media they consume. To now consider whether these perceptions of COVID-19 severity have political consequences, we examine whether they are correlated with a variety of political outcomes, including support for pandemic-related public health policy, approval of federal and state-level public officials, and voting in the 2020 election.

First, we examine the relationship between county-level COVID-19 perceptions and support for pandemic-related public health restrictions. The results of our multivariate ordinary least squares regression, displayed in the first column of table 4, confirm our expectations: there is a positive, statistically significant, and roughly linear relationship between how severe a respondent considered the pandemic to be in their county and their support for restrictions. The association is moderate in size; the model predicts an estimated increase of 0.59 on the COVID policy index for a change from “below average” to “above average,” all else being equal, a difference of roughly 0.43 standard deviations. This may reflect a meaningful effect of county-level perceptions on acceptance of public health restrictions, although unobserved concern about the pandemic in general likely influences both, inflating the apparent causal effect.

We now turn to the association of these perceptions with public opinion of leaders and institutions, starting with President Trump and the US Congress. The second and third columns of table 4 display the results of ordered logit models of approval rating. We observe only a weak and insignificant negative relationship between county COVID estimates and support for Trump. For Congress, estimates of “below average” were associated with significantly lower approval ratings relative to estimates of “average,” although there was no significant difference in approval between the latter category and “above average.” It is possible that those who were less concerned about the pandemic disliked the measures Congress took to address it, although most of the more onerous measures (e.g., lockdowns, mask mandates) originated at the state and local levels. Last, we examine approval of the respondent's governor. In the governor model, we see some evidence of an accountability effect: those who answered “above average” gave significantly lower approval ratings to their governors than those who answered “average” (p = .05), although those who answered “below average” were not significantly more favorable toward their governors.

Table 5 shows the results of models of voter behavior in the 2020 election, including voter-file-validated voter turnout and reported partisan vote choice in the presidential and US House elections (for both of the latter, a Republican vote is coded as 1, while any other vote is coded as 0). In each model, the results suggest a monotonic and negative relationship between the severity of a respondent's estimate and the outcome: voter turnout, voting for President Trump, and voting for the Republican candidate in their US House race. In none of these cases are the relationships statistically significant, however.

Finally, in models examining whether ecological-level case rates are correlated with individual-level political outcomes (presented in online appendix D), as ecological studies suggest (e.g., Baccini, Brodeur, and Weymouth 2021), we find little evidence that this is the case. The only outcome that demonstrated any relationship to case rates in the respondent's county (adjusting for all other covariates) was policy opinion, wherein respondents living in counties with a higher case rate were more likely to support more restrictive COVID mitigation policies, even after adjusting for partisanship and the other individual-level factors. This finding suggests that the significant relationship between perceptions and COVID policy support shown in table 4 is not purely an artifact of individual predispositions about COVID. There was no relationship between case rate and approval ratings, voter turnout, or vote choice.¹⁰

Discussion and Conclusion

The objective of this study was to explore the political implications of perceptions of the COVID-19 pandemic in the fall of 2020. Overall, we find that people were actually quite attuned to the COVID-19 pandemic and were able to estimate the relative severity of the pandemic in their local environment. While we see an expected political filtering of severity—with conservatives, Republicans, and Fox News viewers more likely to say that the local pandemic is less severe, consistent with Republican elites' downplaying of the pandemic (Gadarian, Goodman, and Pepinsky 2022; Moniz 2022)—partisanship does not condition responses to the local objective COVID severity rates. In other words, political predispositions do not seem to shape how objective severity indicators affect perceptions. We also find no evidence that demographic characteristics condition how county-level case rates shape respondent perceptions of the pandemic's severity in their communities. We do show, however, that people who are more attuned to media (especially radio news), are more educated, and have higher political knowledge are more responsive to county-level rates.

These results describing the important role that media sources play in shaping the public's perceptions of the pandemic build on existing research. Specifically, several studies have used survey data to show that different media sources—especially those with a partisan or ideological leaning—contribute to people having divergent opinions and reported behaviors regarding the pandemic (Motta, Stecula, and Farhart 2020; Zhao et al. 2020). Fox News in particular has stood out in previous work. In one study, respondents with more trust in Fox News engaged in fewer preventive behaviors (Zhao et al. 2020); in another, those who reported Fox News as an information source were less likely to acknowledge socioeconomic or racial disparities in COVID mortality (Gollust et al. 2020). Studies have also identified relationships between Fox News exposure and pandemic-related behaviors, without relying on self-reported surveys (see, e.g., Ananyev, Poyker, and Tian 2021; Simonov et al. 2022).

These news source effects may be a result of the different content conveyed; that is, Fox News and CNN may actually be presenting different information about the pandemic's severity, which would shape the public's understanding accordingly. Indeed, content analysis work suggests that outlets differed in how they presented the pandemic, with right-leaning media more likely to present misinformation in the initial stage of the pandemic (Motta, Stecula, and Farhart 2020). Alternatively—or in addition—the relationship between the media use variables and perceptions of the pandemic could reflect the increasingly close relationship between perceived credibility of a media outlet and ideological predispositions. Because the survey items on CES ask only about use of various news sources and not about trust in them, we cannot determine whether the relationships we observe are attributable to differences in content—potentially along with selective dismissal and/or acceptance of that content—or to credibility judgements of outlets that are also correlated with pandemic perceptions (see, e.g., Druckman and McGrath 2019). The many ways in which media sources have shaped public attitudes about the pandemic remain an important research priority. Finally, we did not observe large differences between local and national news in shaping opinions about the pandemic, consistent with Hopkins's (2018) articulation of the nationalization of news and corresponding political attitudes.

Our findings demonstrate that by late 2020, the pandemic had limited relationships with political outcomes, beyond shaping individuals' attitudes about the desirability of restrictive pandemic mitigation strategies. The fact that the real case rates shaped perceptions and that these perceptions indeed correlated with attitudes (i.e., perceptions that the pandemic is more severe related to more support for mitigation) signals that the conditions on the ground did, to some extent, shape public judgements about needed solutions to the problem. However, we did not find substantial evidence that these perceptions affected public judgments of public officials or vote choice, once adjusting for partisanship and other demographic factors. While individuals who perceive the pandemic to be more severe consistently demonstrate lower levels of voter turnout, support for Trump, support for Republican House candidates, and approval of political figures, few of these relationships are statistically significant.

These individual-level findings contrast with some emerging research measured at the ecological level suggesting that the pandemic may have had political consequences; that is, that voters in counties more severely impacted by the pandemic voted for Trump at lower rates in the aggregate (Baccini, Brodeur, and Weymouth 2021). Our study aimed to explore perceptions of pandemic severity and individual-level political outcomes to illuminate a potential individual mechanism for these types of political consequences of the COVID pandemic. The consistency of the signs (where perceiving the pandemic to be more severe is associated with negative attitudes toward national Republicans) does suggest some correspondence with this aggregate finding, but more research is needed using different (larger) samples and other research designs. With a sample size of fewer than 1,000 respondents, we may not have been adequately powered to identify statistically significant effects of perceptions, especially given the robustness of partisanship and ideology in shaping political outcomes generally and in the 2020 election in particular. Our finding that perceiving the pandemic to be of above-average severity in one's county was associated with less support for one's governor suggests at least some evidence that the experience of the pandemic may have manifested in political consequences, although more work is needed to confirm these relationships.

Furthermore, the political consequences of the pandemic are likely long-ranging. As of this writing, the pandemic continues and may have contributed to other recent political outcomes, such as the limited gains by Republicans in the 2022 midterm elections. However, the introduction of vaccines in 2021 and ongoing “pandemic fatigue” in the population may have blunted the association between objective case rates (which had several peaks in 2021 and into 2022) and the public's perceptions of the pandemic's severity. That is, it could be that the relationship between incidence and respondents' concerns about the pandemic grew weaker after 2020, a speculation that future research should consider. Furthermore, with the removal of most pandemic-related policy restrictions by 2022, the pathway between public perceptions and political accountability for policy action may also have become weaker. However, more research using a policy feedback lens is needed to evaluate the many ways in which the pandemic, and especially policy-making actions to mitigate its consequences, might influence political behavior and political attitudes over a longer time frame. Such effects could manifest through changes in the allocation of resources or through interpretive pathways, such as changed interpretations on the part of government and/or beneficiary or constituent groups (Campbell 2012).

This study has several limitations worth noting. First, as described in the measures, our key measure of pandemic perceptions asked about county-level severity, framed with reference to “average.” Since this is purposefully a subjective measure, it cannot be used to establish more granular or precise estimates. A second and related limitation is that we asked respondents to consider severity in their county; however, because of inequities in racial residential segregation, occupational risks, and other social determinants of health, actual mortality rates varied at an even more local level (i.e., zip code, census blocks), so perceptions at a smaller geographic scope may be more meaningful, in terms of both public health and politics. Third, this study is cross-sectional, so we cannot state with certainty that any relationships we identify are causal; instead, our objective was to identify correlational relationships that future study designs (perhaps using experimental designs or longitudinal panel designs) should examine.

Despite these limitations, our study contributes new evidence on the individual-level political consequences of the COVID-19 pandemic, a question that will remain significant for years to come.

Acknowledgments

We presented an early version of this article at the 2021 American Political Science Association Annual Meeting, and we thank Andrea Campbell (discussant) and other members of the panel and audience for their valuable contributions. We also thank Eloise Haselswerdt for her contributions.

Notes

References