Abstract
Objectives. To assess how political polarization during COVID-19 compares with that in past disease outbreaks in the United States.
Methods. Using random-effects meta-analyses and mixed-effects meta-regressions, we searched the Roper Center for Public Opinion Research and estimated the association between polarization during COVID-19 and during non–COVID-19 disease outbreaks.
Results. The study included 170 polls spanning 13 outbreaks over nearly 70 years. Polarization during COVID-19 was 5 and 12 times greater than in past disease outbreaks in terms of concern about infection and vaccine hesitancy, respectively. After we controlled for survey-level characteristics, COVID-19 was associated with 20.23 (95% confidence interval [CI] = 12.30, 28.17) and 25.89 (95% CI = 6.63, 45.16) additional percentage points of polarization regarding concern about infection and vaccine hesitancy, respectively—far higher than historical trends would predict.
Conclusions. In terms of concern about infection and vaccine hesitancy, polarization during COVID-19 was substantially higher than in any other disease outbreak in modern American history for which we have relevant data.
Public Health Implications. High levels of polarization do not appear endemic to disease outbreaks, suggesting such division may be preventable in the future. (Am J Public Health. 2026;116(1):124–136. https://doi.org/10.2105/AJPH.2025.308226)
Political polarization can inhibit effective responses to public health crises like infectious disease outbreaks. This is because ideological polarization is associated with legislative gridlock—the growing inability of legislative bodies like Congress or state legislatures to pass laws—since members of each party have less incentive to compromise as they move apart.1,2 Legislative gridlock, in turn, is associated with lagging social and economic policies, which can hamper the government’s ability to effectively respond to constituent needs in a crisis.2,3 In situations like disease outbreaks, where conditions are changing rapidly, these policy lags are associated with waning risk protection, especially for the least powerful.4 Affective polarization—the growing tendency of partisans to distrust and dislike members of the other party5—is also associated with legislative gridlock and declining trust in government,6 which can jeopardize public health in several ways. First, preventive behaviors crucial to controlling the spread of diseases during outbreaks (e.g., masking, social distancing, vaccination) are strongly correlated with partisanship7–11 and trust in government in the United States.12 In fact, trust in government has been identified as a key determinant of COVID-19 vaccine uptake globally13–15 and influenza vaccination in the United States.16,17 Not only are vaccines integral to curbing the incidence of severe disease and death from infectious diseases, but increasing vaccination is inversely correlated with hospitalizations,18 which may prevent excess morbidity and mortality from overwhelmed health systems.19 Thus, increasing polarization and declining trust in government can threaten pandemic preparedness.
Political polarization is not new; research indicates it has been growing since the mid-1970s.20 American politics scholars largely agree that political parties have become more polarized over time and, as voters improve their ability to choose parties that align with their ideological views, society has followed.1,21–25 Despite this rise in party sorting, studies show that the views of the public remain largely moderate, even on controversial issues like abortion.23,24,26–28 However, it is unclear how levels of polarization observed during COVID-19 compare with historical trends.
This study addresses the question: How did political polarization during the coronavirus pandemic compare with polarization during past disease outbreaks in the United States? Perhaps polarization is a natural consequence of public health emergencies and, in times of crisis, we should expect that partisans will retreat to their ideological corners. If levels of polarization during COVID-19 were akin to levels during contentious outbreaks like anthrax and HIV, polarization should be reversible with existing strategies. On the other hand, if polarization during the pandemic was substantially higher, past tactics may prove ineffective. Using a meta-analysis of 170 public opinion polls across 13 disease outbreaks over nearly 70 years, we show that although some partisan separation is routine, pandemic-era levels of polarization were far higher than in any other disease outbreak in modern US history for which we have relevant data. Additionally, high levels of polarization continue to linger, indicating that research on and new approaches to reducing societal divisions are urgently needed.
METHODS
To generate a list of disease outbreaks with associated polling data, we searched the Roper Center for Public Opinion Research’s iPoll database, the largest polling archive in the world, for disease-related terms and general ones such as “flu,” “virus,” “fever,” “disease,” “illness,” “outbreak,” “epidemic,” “infection,” and “pandemic.” To compare across outbreaks, we looked for questions with wording that remained stable over time and that were used by multiple organizations to measure attitudes during a given crisis. This produced 2 outcomes of interest: concern about infection and intention to receive a vaccine. We measured the first outcome by variations of the question, “How concerned are you that you or someone you know will get infected with [a given disease]?” We measured the second outcome using variations of the question, “Do you plan to get vaccinated for [a given disease] when a vaccine becomes available to you?” Full question wording is available in Appendixes D and E (available as a supplement to the online version of this article at http://www.ajph.org).
We then combined terms related to our outcomes with disease terms to search the database. For concern about infection, our outcome-related search terms included “worried” or “concerned”; for vaccine intention, we used the terms “vaccine,” “vaccination,” “vaccinate,” “vaccinated,” “shot,” “inoculate,” and “inoculation.” These were paired with specific disease-related terms (e.g., “H5N1,” “HIV”); general terms (e.g., “immune deficiency,” “sarcoma”); and colloquial terms, including some commonly used pejorative phrases that reflected news coverage at the time (e.g., “bird flu,” “Asian flu,” “gay cancer”). For diseases associated with vaccines already in widespread use (e.g., diphtheria, chickenpox, influenza) we added search terms to focus on one’s likelihood of vaccination during a given health threat (e.g., “likely,” “plan,” “would,” and “probably”), rather than past receipt, to keep comparisons standard over time.
Inclusion Criteria
In the first screening phase, we used a database filter to return only polls with downloadable data sets. We copied these results into an Excel file. Polls were then screened for relevant question wording. Third, we reviewed full-text questions to determine whether they met inclusion criteria. A poll was considered eligible if it (1) asked about concern on a 4-point response scale or vaccine intention on a binary or 4-point scale, (2) included party affiliation, (3) did not use prejudicial terms, (4) asked about an outbreak within the first 5 years of its emergence or resurgence into the national discourse, and (5) sampled adults nationwide. Study searching occurred from April to July 2021, and an update was conducted from January to April 2024. One reviewer (C. L. M.) conducted the primary search, and both authors (C. L. M. and R. C.) conducted the update. Discrepancies were discussed and resolved. Because most major news organizations in the United States archive their data sets at the Roper Center regardless of their results, publication bias is unlikely to be a concern. Assessments of poll quality based on survey organization ratings are available in online Appendix H.
Polls that surveyed adults in certain states, of certain ages, or only registered voters were excluded for not being nationally representative. We also excluded polls that asked questions on dissimilar scales because differences in the underlying distribution of responses could be mistakenly interpreted as changes in opinion. We did not include polls that contained prejudicial terms such as “safe” or “effective” regarding vaccines and phrases such as “seriously ill” regarding concern. These qualifiers were considered meaningful departures in wording that could exaggerate differences between partisans if, for example, Democrats were more likely than Republicans to believe in the reported safety or effectiveness of a vaccine. Polls on HIV/AIDS conducted in the 1990s were also excluded to ensure that surveys asking about disease outbreaks captured attitudes while the threat remained relatively novel and thus comparable in timing to COVID-19. Additionally, we excluded polls that asked if one worried about infection yesterday and vaccine polls that asked about receipt, rather than intention.
Data Extraction
To preserve survey weights (present in k = 160/170) that ensure a poll’s sample is nationally representative at the time of its fielding, each poll was analyzed individually (by C. L. M.). The proportions of self-identified Democrats and Democratic-leaners (“Democrats”) and Republicans and Republican-leaners (“Republicans”) who said they were “very concerned” about infection or who did not intend to get a vaccine when one became available to them were estimated for each poll using the svy package in Stata version 17.0 (StataCorp LP, College Station, TX). Some vaccine polls (k = 52) used a 4-point likeliness scale for responses, ranging from “very likely” to “not at all likely.” To enable comparisons to questions on a binary yes–no scale (k = 37), this variable was dichotomized by aggregating responses into a “likely” group and an “unlikely” group.
Ten polls conducted before 1988 did not have survey weights but instead used area-based probability sampling techniques. To obtain standard errors for point estimates from these polls, we created survey weights of 1 to generate uncertainty estimates without affecting accuracy or precision. Additionally, 2 polls used a duplicate card system for survey weighting, which can artificially inflate the sample size and statistical power of a poll. To correct this, we multiplied each poll’s duplicate card weight by the inverse mean of all the weights in the sample to create a new weight centered around the total weight mean. This preserved the original sample size and the relative magnitude of the weights.
Statistical Analyses
We conducted a random effects meta-analysis to analyze the mean difference between partisans. Specifically, our principal summary measure was the percentage-point difference between Democrats and Republicans (or, for vaccine hesitancy, Republicans and Democrats). Our subgroups allowed us to compare polarization during the pandemic and during noncoronavirus outbreaks. A random-effects model was selected a priori because whereas fixed-effects models assume 1 true effect size across all studies, random-effects models allow effect sizes to vary between studies and assume that included studies “represent a random sample of effect sizes that could have been observed.”29(p71) Given the Roper Center’s archiving practices, we assumed that this condition was satisfied. We used restricted maximum likelihood estimation to estimate between-study variability.30
Mixed-effects meta-regression explored sources of heterogeneity and assessed the relationship between COVID-19 and polarization for each outcome. Moderator variables adjusted for variation in question wording and response options, where applicable; survey administration mode (e.g., telephone, web); the presence of survey weights; months until the next election; the party of the president in power; and, where sample size allowed, so-called “house effects”—differences in sampling and weighting techniques unique to individual survey organizations.31 Additionally, by fitting our meta-regression model to prepandemic data, we obtained a predicted level of polarization in December 2023 to evaluate whether COVID-19 fell along a trend of rising polarization over time. Because of sample size limitations, moderator variables in the predictive models adjusted for variation in question wording and response options, house effects (for concern about infection), the presence of survey weights (vaccine hesitancy), months until the next election, and party of the president in power. We evaluated moderator associations based on coefficients and 95% confidence intervals (95% CIs), and meta-regressions used the Knapp–Hartung adjustment to help account for uncertainty in the estimate of τ2.30 We examined residual heterogeneity by estimating τ2 using restricted maximum likelihood estimation and I2, and employing Cochran’s Q test, where the null hypothesis was that all studies share a common effect.29 We conducted all meta-analyses and meta-regressions in R version 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria), using the metafor package (Viechtbauer, 2010).
RESULTS
The search yielded 6095 surveys for preliminary screening: 2474 pertaining to concern about infection and 3621 to vaccine hesitancy. After we removed those without downloadable data sets, 2950 surveys were screened for inclusion across both outcomes and 369 received full-text review. Ultimately, 170 unique surveys met inclusion criteria, of which 90 asked questions on concern about infection and 89 asked about vaccine hesitancy. These polls included a total sample size of 158 143 partisans, of which 87 412 identified as or leaned Democrat and 70 731 identified as or leaned Republican.
Study Characteristics
Table 1 summarizes the characteristics of the included polls. The online Appendices include our PRISMA flow diagram (Appendix A), search terms and results (B and C), and full question wording with sample size for each poll (D and E). Regarding concern about infection, 42% of polls examined COVID-19, whereas the rest covered 8 other outbreaks dating back to HIV/AIDS in 1985. Seventeen survey organizations polled on this outcome, the most common of which was Ipsos (23%). Just under half (43%) asked about one’s level of concern about being infected by, becoming ill with, or getting sick from a given disease. About a quarter (28%) asked about catching, contracting, or getting a disease, and a similar proportion (22%) asked about being exposed to or affected by a given disease. Just 4 (4%) asked about being a victim of a given disease. Most (71%) were conducted via telephone, followed by web only (23%), telephone and web (4%), and face-to-face (1%). All polls used survey weights and 73% were fielded during a Republican presidential administration. On average, these polls were conducted 10 months before the next election, with about 452 self-identified Democrats (and leaners) and about 373 Republicans (and leaners).
TABLE 1—
Characteristics of Included Polls: United States, May 2, 1954–December 31, 2023
| Concern About Infection (n = 90), No. of Studies (%) or Mean (95% CI) | Vaccine Hesitancy (n = 89), No. of Studies (%) or Mean (95% CI) | |
| Disease | ||
| COVID-19 | 38 (42) | 63 (71) |
| Zika | 8 (9) | 0 |
| Ebola | 5 (6) | 1 (1) |
| H1N1 (2009) | 4 (4) | 7 (8) |
| H5N1 | 5 (6) | 0 |
| SARS | 11 (12) | 0 |
| West Nile | 1 (1) | 0 |
| Smallpox | 0 | 5 (6) |
| Anthrax | 9 (10) | 1 (1) |
| HIV/AIDS | 9 (10) | 2 (2) |
| H1N1 (1976) | 0 | 3 (3) |
| H2N2 | 0 | 2 (2) |
| Polio | 0 | 5 (6) |
| Survey organization | ||
| Associated Press | 0 | 1 (1) |
| Beacon Research/Shaw & Co. Research | 0 | 1 (1) |
| CBS News | 1 (1) | 1 (1) |
| CBS News/New York Times | 1 (1) | 1 (1) |
| Gallup | 10 (11) | 11 (12) |
| Harris Interactive | 2 (2) | 1 (1) |
| Hart-Teeter Research Companies | 0 | 1 (1) |
| International Communications Research | 7 (8) | 2 (2) |
| Ipsos | 21 (23) | 45 (51) |
| Langer Research Associates | 10 (11) | 6 (7) |
| Los Angeles Times | 3 (3) | 0 |
| Marist | 0 | 9 (10) |
| National Opinion Research Center at the University of Chicago | 4 (4) | 2 (2) |
| Opinion Research Corporation | 2 (2) | 2 (2) |
| Peter D. Hart Research Associates | 1 (1) | 0 |
| Princeton Survey Research Associates International | 11 (12) | 0 |
| Public Religion Research Institute | 1 (1) | 0 |
| SSRS (formerly Social Science Research Solutions) | 9 (10) | 1 (1) |
| Stony Brook University Center for Survey Research | 1 (1) | 1 (1) |
| The Roper Organization | 0 | 2 (2) |
| TNS Intersearch | 5 (6) | 2 (2) |
| Yankelovich Clancy Shulman | 1 (1) | 0 |
| Question wording: impact of disease | ||
| Being exposed to/affected by | 22 (24) | … |
| Catching/contracting/getting | 25 (28) | … |
| Being infected by/becoming ill/getting sick | 39 (43) | … |
| Being a victim of | 4 (4) | … |
| Survey mode | ||
| Web only | 21 (23) | 45 (51) |
| Web and telephone | 4 (4) | 2 (2) |
| Telephone only | 64 (71) | 32 (36) |
| Face-to-face | 1 (1) | 10 (11) |
| Response choice | ||
| Binary choice (yes/no) | … | 37 (42) |
| Four-point likeliness scale (very/somewhat/not too/not at all) | … | 52 (58) |
| Presence of survey weights | ||
| With weights | 90 (100) | 79 (89) |
| Without weights | 0 | 10 (11) |
| Party of the president in power | ||
| Democrat | 24 (27) | 50 (56) |
| Republican | 66 (73) | 39 (44) |
| Proximity to next election: months until Election Day | 10 (−3, 23) | 14 (1, 28) |
| Weighted sample size | ||
| Democrats and leaners | 452 (−165, 1070) | 558 (−1084, 2200) |
| Republicans and leaners | 373 (−32, 777) | 448 (−713, 1609) |
Note. CI = confidence interval. Full question wording, response options, and sample size figures are provided in Appendices D and E (available as a supplement to the online version of this article at http://www.ajph.org). Percentages may not add to 100 because of rounding.
Regarding vaccine hesitancy, 71% of polls examined COVID-19 whereas the rest covered 8 other outbreaks dating back to polio in 1954. Seventeen survey organizations conducted polls on this outcome, 51% of which were conducted by Ipsos. Half (51%) were conducted via web only, followed by telephone only (36%), face-to-face (11%), and a combination of telephone and web (2%). Most polls (58%) used a 4-point likeliness scale, whereas 42% used a binary yes–no response scale. Ten (11%) lacked survey weights, and 56% were conducted during a Democratic presidential administration. On average, vaccine hesitancy polls were conducted about 14 months before the next election, with about 558 self-identified Democrats (and leaners) and about 448 Republicans (and leaners).
Meta-Analysis
Pooled estimates of polarization regarding concern about infection during COVID-19 and non–COVID-19 disease outbreaks are displayed in Figure 1. Before the pandemic, the weighted average difference between Democrats and Republicans who said they were “very concerned” that they or someone they know would be infected with a given disease was about 5.04 percentage points (k = 38; 95% CI = 4.15, 5.93). During the pandemic, this rose to 24.66 percentage points (k = 52; 95% CI = 21.39, 27.93). Tests indicate high levels of heterogeneity (non–COVID-19: Q = 73.35, P < .05, τ2 = 3.00, I2 = 29.3%; COVID-19: Q = 242.48, P < .001, τ2 = 85.94, I2 = 86.2%), which is not unusual for public opinion over time. We can reject the null hypothesis that polarization regarding concern about infection is the same for COVID-19 and non–COVID-19 outbreaks (QM = 9.07; P < .001).
FIGURE 1—
Percentage-Point Difference Between Democrats and Republicans Expressing High Concern About Infection, by Disease Subgroup: United States, August 1, 1985–December 31, 2023
Note. CI = confidence interval; NORC = National Opinion Research Center; NSF = National Science Foundation; P-P Diff = the percentage-point difference between Democrats and Republicans who said they were “very concerned” that they or someone they know would get a given disease; RE = random effects; SSRS = Social Science Research Solutions; UMD = University of Maryland. Weighted percentages reflect the use of survey weights to match known population parameters for the adult US population at the time of the survey. This figure was abbreviated for space. For the full figure, please see Appendix F (available as a supplement to the online version of this article at https://www.ajph.org).
Figure 2 displays pooled estimates of polarization regarding vaccine hesitancy, during COVID-19 and non–COVID-19 disease outbreaks. Before the pandemic, the weighted average difference between Republicans and Democrats who said they were unlikely to or would not receive a vaccine for a given disease when one became available to them was about 1.90 percentage points (k = 26; 95% CI = −0.16, 3.97). During the pandemic, this rose to 23.85 percentage points (k = 63; 95% CI = 22.10, 25.59). Tests again indicate high levels of heterogeneity (non–COVID-19: Q = 66.32, P < .01, τ2 = 16.81, I2 = 66.5%; COVID-19: Q = 311.59, P < .01, τ2 = 37.37, I2 = 77.5%). We can reject the null hypothesis that polarization regarding vaccine hesitancy is the same for COVID-19 and non–COVID-19 outbreaks (QM = 23.63; P < .001).
FIGURE 2—
Percentage-Point Difference Between Democrats and Republicans Expressing Vaccine Hesitancy, by Disease Subgroup: United States, May 2, 1954–December 31, 2023
Note. CI = confidence interval; P-P Diff. = the percentage point difference between Republicans and Democrats who say they will not or are unlikely (somewhat unlikely or very unlikely) to receive a vaccine for a given disease when one becomes available to them; RE = random effects. Weighted percentages reflect the use of survey weights to match known population parameters for the adult US population at the time of the survey. Seven polls in this study conducted before 1960 did not employ survey weights; neither did 2 H1N1 polls conducted in late 1976, nor the HIV/AIDS poll conducted in April 1987. Three polls conducted between May 1, 1954 and May 31, 1955 asked about one’s intention to get their child vaccinated with the polio vaccine. Full question wording and information about survey weights are available in the Appendices (available as a supplement to the online version of this article at http://www.ajph.org). This figure was abbreviated for space. For the full figure, please see Appendix G.
Meta-Regression
Regarding concern about infection, COVID-19 was associated with significantly higher levels of polarization (20.23; 95% CI = 12.30, 28.17; P < .001) in our meta-regression models (Figure 3a). Regarding vaccine hesitancy, COVID-19 (25.89; 95% CI = 6.63, 45.16; P < .05), the polling firm Stony Brook University Center for Survey Research (17.03; 95% CI = 1.93, 32.12; P < .05), and months until Election Day (0.25; 95% CI = 0.08, 0.41; P < .01) were significantly associated with higher levels of polarization (Figure 3b). Meanwhile, having a Republican president (−8.03; 95% CI = −10.64, −5.43; P < .001) was significantly associated with lower levels of polarization (Figure 3b).
FIGURE 3—
Mixed-Effects Meta-Regression of the Association Between Study-Level Characteristics and Polarization for (a) Concern About Infection and (b) Vaccine Hesitancy: United States, May 2, 1954–December 31, 2023
Note. CI = confidence interval; NORC = National Opinion Research Center; PRRRI = Public Religion Research Institute; SSRS = Social Science Research Solutions.
To test whether polarization underwent a discrete jump during the pandemic, we predicted levels of polarization in December 2023 under a Democratic president using prepandemic data. Results indicate that the observed level of polarization regarding concern about infection during COVID-19 was 10.5 percentage points (1.09 times) greater than what would be predicted based on historical trends (9.64; 95% prediction interval [PI] = 3.44, 15.84). For vaccine hesitancy, results indicate that the observed level of polarization during COVID-19 was 24.3 percentage points (14.91 times) greater than what would be predicted based on historical trends (1.63; 95% PI = −11.30, 14.55).
Robustness Checks
We conducted 3 checks to test whether these findings were robust to alternate specifications. First, we replaced all polls (concern: k = 2; vaccine: k = 8) with a negative difference between partisans by using the absolute value. Second, we removed all polls that occurred before and during party realignment in the United States from 1954 to 1980 (k = 10)—a procedure that only affected vaccine hesitancy and removed all but 1 poll that did not employ survey weights (Associated Press and Media General, April 1987). Third, we tested alternative specifications of polarization. For concern about infection, we analyzed responses from the opposite end of the response scale—those who said they were “not at all concerned” about infection. One poll from the HIV/AIDS epidemic (Rolling Stone, September 1987) was dropped for this test because the lowest response category used different wording. For vaccine intention, analyses examined enthusiasm, measuring those who said they did intend to get a vaccine. Overall, these procedures had few substantive impacts on our findings; full results are available in online Appendices I through R.
DISCUSSION
Our findings indicate that political polarization during COVID-19 was about 5 times greater than in any other disease outbreak for which we have public opinion data regarding concern about infection (non–COVID-19: 5.04; COVID-19: 24.66) and more than 12 times greater regarding vaccine hesitancy (non–COVID-19: 1.90; COVID-19: 23.85). Furthermore, it appears that high levels of polarization continue to linger, suggesting reversal may be difficult. Although COVID-19 was superlative in its scale and severity, these comparisons are not straw men. They include polls conducted just before midterm elections during the Ebola crisis, when President Obama’s handling of the disease became a prominent campaign issue32; polls conducted a month after September 11, 2001, during a second round of terrorist attacks that used anthrax as a biological weapon; and during the HIV/AIDS epidemic, a disease shaped by homophobia, racism, and xenophobia. For vaccine hesitancy, the comparison also includes polls conducted a few months before the 1976 presidential election and polls before and after the release of faulty polio vaccines made at Cutter and Wyeth Laboratories, which accidentally used live virus and resulted in 204 cases of polio and 11 deaths among children.33
Even after adjustment for variation in question wording, response options, survey administration mode, sampling and weighting techniques, months until Election Day, and the party of the president in power, we find that COVID-19 was associated with 20 to 25 additional percentage points of polarization, compared with earlier disease outbreaks, and that this difference is statistically significant. Our results also indicate that outbreaks are not free from partisan differences regarding concern about infection. In fact, we should expect to see about 5 percentage points of separation during the average health crisis, with Democrats more concerned about infection. However, polarization during COVID-19 was considerably more extreme than in past outbreaks, representing a discrete jump in polarization that cannot be explained by gradual partisan separation over time. In fact, observed levels of polarization during COVID-19 were about 1.1 to 14.9 times higher than what prepandemic trends would predict, which indicates that past approaches to reducing societal divisions may not be effective. Additional research on strategies to reunify partisans is urgently needed.
Unprecedented levels of polarization may be influenced by the unique circumstances of the pandemic. The public’s response to polarization among party elites is ordinarily limited by (1) party identification, (2) awareness of party differences on policy issues, and (3) issue salience, meaning that only those who think the issue is important pay attention.34 However, COVID-19 was the most widespread pandemic in a century, raising the disease’s salience, and it coincided with a contentious presidential election that encouraged people in the United States to understand party differences and cast their allegiance on the ballot. Together, these forces may have reduced protective factors normally present during disease outbreaks, leading to a surge in polarization.
Limitations
This study has several limitations. First, it examined 2 important prevention behaviors, for which the wording and response options in the polls remained stable over time. Other outcomes with shorter trends and less stable measures may have been less divisive during COVID-19 (e.g., contact tracing or quarantine). Second, the study only analyzed data available for download from the Roper Center. Although publication bias is not a concern, not all media outlets archive their data there, and news organizations only ask about outbreaks that capture the public’s attention. Those that make national headlines may be more prone to polarization than others that do not garner polls, such as pertussis and mumps. However, this implies that our results may represent a conservative estimate of polarization between COVID-19 and non–COVID-19 disease outbreaks. Third, included polls demonstrated substantial heterogeneity. Although this can make it challenging to interpret results, it reflects the nature of public opinion during times of crisis. Lastly, it was not possible to account for variations in outbreak magnitude. This is important, given that COVID-19 was far more widespread in the United States than any other included outbreak. However, it was not possible to control for case counts leading up to each poll because of regional tracking of influenza.
Public Health Implications
From legislative gridlock and lagging risk protection to declining trust in government that can hamper public health efforts, political polarization can inhibit effective responses to outbreaks.3,4,7 Fortunately, high levels of polarization, although they continue to linger, do not appear endemic to disease outbreaks and may be preventable. This indicates that research on new approaches to reducing societal divisions are urgently needed. However, progress on reducing polarization as it pertains to disease outbreaks may be jeopardized by funding freezes for health research and the revocation of grants to study vaccine hesitancy. More broadly, the effectiveness of public health efforts to prevent or mitigate future disease outbreaks may be diminished if they are perceived to be connected to politics or motivated by political gain.35
ACKNOWLEDGMENTS
C. L. McMurtry’s work on this article was supported by the Dissertation Completion Fellowship at Harvard University.
We thank the staff—past and present—of the Roper Center for Public Opinion Research at Cornell University for the hard work they continue to invest in cataloguing, cleaning, and archiving public opinion data. Thanks also to the news and polling organizations that collect these data and share them with the Roper Center and with us, including, Mollyann Brodie and Ashley Kirzinger at KFF. Additionally, we thank Steven Worthington of the Institute for Quantitative Social Sciences (IQSS) at Harvard University, Rui Duan of Harvard T. H. Chan School of Public Health, and Laura Hatfield of the National Opinion Research Center (NORC) at the University of Chicago for their methodological mentorship. We are grateful for comments and feedback from seminar participants at Harvard University as well as conference participants at the National Research Service Award (NRSA) Trainees Research Conference, the American Public Health Association, and the American Political Science Association. Finally, we thank Robert Blendon, Jessica Cohen, and David Cutler for their help in developing and refining this dissertation paper. This study would not have been possible without their encouragement and support.
CONFLICTS OF INTEREST
The authors have no conflicts of interest to disclose.
HUMAN PARTICIPANT PROTECTION
Because this project analyzed existing anonymous data sets available through the Roper Center, the institutional review board at Washington University in St. Louis determined this study was not human subjects research (IRB ID #202506183).
See also Nuzzo, p. 26.
REFERENCES
- 1.Barber M, McCarty NM. Causes and consequences of polarization. In: Mansbridge J, Martin CJ, eds. Political Negotiation: A Handbook. Washington, DC: Brookings Institution Press; 2016:37–90. [Google Scholar]
- 2.McCarty NM. The policy effects of political polarization. In: Pierson P, Skocpol T, eds. The Transformation of American Politics: Activist Government and the Rise of Conservatism. Princeton, NJ: Princeton University Press; 2007:223–255. [Google Scholar]
- 3.McCarty NM, Poole KT, Rosenthal H. Polarization and Public Policy. Cambridge, MA: MIT Press; 2006. [Google Scholar]
- 4.Rigby E, Wright GC. The Policy Consequences of Party Polarization: Evidence From American States. In: Thurber JA, Yoshinaka A, eds. American Gridlock. New York, NY: Cambridge University Press; 2015:236–255. [Google Scholar]
- 5.Iyengar S, Lelkes Y, Levendusky M, Malhotra N, Westwood SJ. The origins and consequences of affective polarization in the United States. Annu Rev Polit Sci. 2019;22(1):129–146. 10.1146/annurev-polisci-051117-073034 [DOI] [Google Scholar]
- 6.Hetherington MJ, Rudolph TJ. Why Washington Won’t Work. Chicago, IL: University of Chicago Press; 2015. [Google Scholar]
- 7.Gadarian SK, Goodman SW, Pepinsky TB. Partisanship, health behavior, and policy attitudes in the early stages of the COVID-19 pandemic. PLoS One. 2021;16(4):e0249596. 10.1371/journal.pone.0249596 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Allcott H, Boxell L, Conway J, Gentzkow M, Thaler M, Yang D. Polarization and public health: partisan differences in social distancing during the coronavirus pandemic. J Public Econ. 2020;191:104254. 10.1016/j.jpubeco.2020.104254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bruine de Bruin W, Saw HW, Goldman DP. Political polarization in US residents’ COVID-19 risk perceptions, policy preferences, and protective behaviors. J Risk Uncertain. 2020;61(2):177–194. 10.1007/s11166-020-09336-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Fridman A, Gershon R, Gneezy A. COVID-19 and vaccine hesitancy: a longitudinal study. PLoS One. 2021;16(4):e0250123. 10.1371/journal.pone.0250123 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Painter M, Qiu T. Political beliefs affect compliance with government mandates. J Econ Behav Organ. 2021;185:688–701. 10.1016/j.jebo.2021.03.019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Suhay E, Soni A, Persico C, Marcotte DE. Americans’ trust in government and health behaviors during the COVID-19 pandemic. RSF. 2022;8(8):221–244. 10.7758/RSF.2022.8.8.1037342867 [DOI] [Google Scholar]
- 13.Pagliaro S, Sacchi S, Pacilli MG, et al. Trust predicts COVID-19 prescribed and discretionary behavioral intentions in 23 countries. PLoS One. 2021;16(3):e0248334. 10.1371/journal.pone.0248334 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.COVID-19 National Preparedness Collaborators. Pandemic preparedness and COVID-19: an exploratory analysis of infection and fatality rates, and contextual factors associated with preparedness in 177 countries, from Jan 1, 2020, to Sept 30, 2021. Lancet. 2022;399(10334):1489–1512. 10.1016/S0140-6736(22)00172-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Monfared IG. The connection between COVID-19 vaccine abundance, vaccination coverage, and public trust in government across the globe. Vaccine. 2022;40(43):6211–6217. 10.1016/j.vaccine.2022.09.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Quinn SC, Parmer J, Freimuth VS, Hilyard KM, Musa D, Kim KH. Exploring communication, trust in government, and vaccination intention later in the 2009 H1N1 pandemic: results of a national survey. Biosecur Bioterror. 2013; 11(2):96–106. 10.1089/bsp.2012.0048 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Jamison AM, Quinn SC, Freimuth VS. “You don’t trust a government vaccine”: narratives of institutional trust and influenza vaccination among African American and white adults. Soc Sci Med. 2019;221:87–94. 10.1016/j.socscimed.2018.12.020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Tenforde MW, Self WH, Adams K, et al. Association between mRNA vaccination and COVID-19 hospitalization and disease severity. JAMA. 2021;326(20):2043–2054. 10.1001/jama.2021.19499 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Janke AT, Mei H, Rothenberg C, Becher RD, Lin Z, Venkatesh AK. Analysis of hospital resource availability and COVID-19 mortality across the United States. J Hosp Med. 2021;16(4):211–214. 10.12788/jhm.3539 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Poole KT, Rosenthal H. Congress: A Political-Economic History of Roll Call Voting. New York, NY: Oxford University Press; 1997. [Google Scholar]
- 21.Abramowitz A. The Disappearing Center: Engaged Citizens, Polarization, and American Democracy. New Haven, CT: Yale University Press; 2010. [Google Scholar]
- 22.Levendusky M. The Partisan Sort: How Liberals Became Democrats and Conservatives Became Republicans. Chicago, IL: University of Chicago Press; 2009. [Google Scholar]
- 23.Fiorina MP. Unstable Majorities: Polarization, Party Sorting and Political Stalemate. Stanford, CA: Hoover Institution Press; 2017. [Google Scholar]
- 24.Fiorina MP, Abrams SJ. Political polarization in the American public. Annu Rev Polit Sci. 2008;11(1): 563–588. 10.1146/annurev.polisci.11.053106.153836 [DOI] [Google Scholar]
- 25.Jacobson GC. The electoral origins of polarized politics: evidence from the 2010 Cooperative Congressional Election Study. Am Behav Sci. 2012;56(12):1612–1630. 10.1177/0002764212463352 [DOI] [Google Scholar]
- 26.Ansolabehere S, Rodden J, Snyder JM. Purple America. J Econ Perspect. 2006;20(2):97–118. 10.1257/jep.20.2.97 [DOI] [Google Scholar]
- 27.Baldassarri D, Gelman A. Partisans without constraint: political polarization and trends in American public opinion. Am J Sociol. 2008;114(2):408–446. 10.1086/590649 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Levendusky MS, Malhotra N. (Mis)perceptions of partisan polarization in the American public. Public Opin Q. 2016;80(S1):378–391. 10.1093/poq/nfv045 [DOI] [Google Scholar]
- 29.Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. 2nd ed. Hoboken, NJ: John Wiley & Sons; 2011. [Google Scholar]
- 30.Viechtbauer W. Conducting meta-analyses in R with the metafor package. J Stat Soft. 2010;36(3): 1–48. 10.18637/jss.v036.i03 [DOI] [Google Scholar]
- 31.Flores AR. Examining variation in surveying attitudes on same-sex marriage: a meta-analysis. Public Opin Q. 2015;79(2):580–593. 10.1093/poq/nfv017 [DOI] [Google Scholar]
- 32.Campante FR, Depetris-Chauvin E, Durante R. The virus of fear: the political impact of Ebola in the US. Am Econ J Appl Econ. 2024;16(1):480–509. [Google Scholar]
- 33.Langmuir AD, Nathanson N, Hall WJ. The surveillance of poliomyelitis in the United States in 1955. CDC Report. November 9, 1955. Available at: https://stacks.cdc.gov/view/cdc/48111. Accessed August 1, 2025.
- 34.Layman GC, Carsey TM, Horowitz JM. Party polarization in American politics: characteristics, causes, and consequences. Annu Rev Polit Sci. 2006;9(1):83–110. 10.1146/annurev.polisci.9.070204.105138 [DOI] [Google Scholar]
- 35.SteelFisher GK, Findling MG, Caporello HL, et al. Trust in US federal, state, and local public health agencies during COVID-19: responses and policy implications. Health Aff (Millwood). 2023;42(3): 328–337. 10.1377/hlthaff.2022.01204 [DOI] [PMC free article] [PubMed] [Google Scholar]