Does Policy Uncertainty Boost Vaccine Hesitancy? Political Controversy, the FDA, and COVID-19 Vaccine Hesitancy in Fall 2020 Free

Abstract

Context: Vaccine hesitancy is associated with political and institutional distrust, but there is little research on how people's trust responds to political events. The authors revisit the fall of 2020, when evaluation of new COVID-19 vaccines collided with an impending US national election. Drawing on a “political Bayesian” perspective, the authors assess abrupt changes in attention to political events and test hypotheses on how subpopulation responses may differ in accordance with partisanship, educational attainment, and race/ethnicity.

Methods: The authors analyze daily changes in US news reporting and social media use in 2020, combined with detailed analysis of two-large scale surveys fielded at the time, focusing on questions of COVID-19 vaccine hesitancy and safety concerns about COVID-19 vaccines.

Findings: Vaccine hesitancy spiked in the United States from late August to early October 2020. The authors identify several plausible triggers for this spike, all pertaining to the Food and Drug Administration and electoral politics. Heightened vaccine hesitancy occurred among Democrats, Asian citizens, Black citizens, and college-educated respondents. Turbulence mainly affected those who were initially most trusting in government and vaccines. Asian American vaccine confidence recovered; that of Black Americans did not.

Conclusions: Electoral politics may destabilize citizen assumptions about vaccine authorization and boost uncertainty, thereby undermining public willingness to take approved vaccines.

Trust is a critical predictor and mediator of citizens’ willingness to use approved and prescribed health treatments. This dynamic has been observed in medication adherence (Brown et al. 2016; Hall and Heath 2021; Polinski et al. 2014) and vaccine hesitancy (Goldenberg 2021; Jennings et al. 2021; Liu, Zhao, and Wan 2023; Soares et al. 2021). However, the agents or institutions in which trust is placed, or from which trust is withheld, are critical to understanding the implications of trust for adherence and hesitancy. Well before the COVID-19 pandemic, scholars developed generalized vaccine hesitancy models in which “confidence is defined as trust in (i) the effectiveness and safety of vaccines; (ii) the system that delivers them, including the reliability and competence of the health services and health professionals and (iii) the motivations of policy-makers who decide on the needed vaccines” (MacDonald 2015: 4162). Even this simple and widely referenced summary of trust and its role in vaccine hesitancy raises questions that remain largely unaddressed in scholarship. Is vaccine hesitancy attributable to generalized distrust, to more specific factors relating to the health care system (e.g., physicians, nurses, hospitals, governmental health authorities), or to the vaccine itself? What factors shape hesitancy over time? What do we know, and what can we know, about how those factors affect subpopulations in different ways?

A scholarly focus on health politics, policy, and law will better equip researchers to examine the institutional and political context of changes in vaccine hesitancy. Yet at present, scholarship lacks analyses of how citizens respond to political events and policy uncertainty over time. The principal reason comes from the predominantly cross-sectional nature of research into vaccine hesitancy (Aw et al. 2021; Dubé et al. 2013; Rich-Edwards et al. 2021; Troiano and Nardi 2021; Willis et al. 2021). While cross-sectional studies can clarify the predictive power of an individual citizen's distrust (general or specific), they tend not to examine changes in attitudes as events occur and as citizens learn from them or react to them. Several studies including time-series and cross-sectional time series analyses of COVID-19 vaccine hesitancy have appeared in recent years, but they also have limits, as they (1) examine national contexts where political uncertainty was not as high as in the United States (for a study of Spain, see Beca-Martínez et al. 2022); (2) study racial trust differences in general and not differences in policy uncertainty or institutional trust in particular (Padamsee et al. 2022; but see also Alsan and Wanamaker 2018); or (3) they measure distrust in government more generally but not in specific institutions (King et al. 2021).

To examine how institutional trust shapes vaccine hesitancy in an uncertain environment, we revisit the months before the emergency authorization and rollout of COVID-19 vaccines in the United States. Although real-world events do not provide prospectively designed experiments, and even “natural” quasi-experiments have questionable external validity, historically situated observational analyses often provide clear settings in which high-cost actions are undertaken and in which real-world variations in political events, policies, and context take place.

In the months before the US Food and Drug Administration's (FDA) December 2020 authorization of COVID-19 vaccines based on novel mRNA technology, a historic and multidimensional collision of forces in American health politics injected substantial doubt into an already uncertain vaccine evaluation process (Gollust, Nagler, and Fowler 2020; Oberlander 2020). The devastating impact of the global COVID-19 pandemic coincided with three broader sociopolitical forces: (1) a presidential election cycle that was closely contested and that reflected a historically high level of polarization and mistrust among American citizens; (2) an incumbent president willing to break norms of deference to expertise, science, and public health agencies; and (3) the rapid development of new vaccine treatments (specifically mRNA vaccines) for the prevention and amelioration of COVID-19 and its symptoms. Particularly along lines of race, ethnicity, and partisanship (Bailey and Moon 2020; Miller, Woods, and Kalmbach 2022), the social and political fractures induced by this collision of events were unprecedented.

In the context of studying citizen trust and vaccine hesitancy, our research questions are as follows: Were there particular political or regulatory events that garnered more or less attention from citizens and from the sources they use to learn about the world? How did Americans’ vaccine hesitancy change in the months leading up to the November election, which beliefs were most affected, and which sources of policy and regulatory uncertainty were most plausibly noticed by the mass public? Specifically, but just as importantly, whose trust was shaken most as these developments unfolded, and is there evidence that trust lost was later regained? Asking questions not just about average movements in institutional and vaccine confidence but also about individuals over time is critical for understanding health politics (Gollust, Nagler, and Fowler 2020) and racial and ethnic politics (Bailey and Moon 2020). This is especially so for vaccine authorization by the federal government, not only because the vaccines in question were new but also because the mRNA technology used in them had never before been used in an approved biopharmaceutical treatment.

Existing evidence, which is scattered and largely nonscholarly, points to a crisis of confidence in the fall of 2020 that was connected to the intersection of presidential politics and the development and review of mRNA vaccines (KFF 2020). News reports based on interviews and static surveys have suggested that as the election approached, damage had been done to the public's trust in the FDA, President Trump and his administration's handling of health policy, and the promise of COVID-19 vaccines specifically (KFF 2020). Although research is scarce, several reports point in particular to deep uncertainty and hesitation among Black Americans. In a striking development showing that experts took surveys and cultural discourse as evidence of probable behavior, a group of Black physicians created a separate group to review vaccines in September 2020 (Boodman 2020). (We elaborate on this case below.) Prominent public figures and journalists criticized the Trump administration for publicly pressuring the FDA to authorize COVID-19 vaccines before they were ready and criticized the FDA for giving into this pressure (Gibson 2020; Karlin-Smith 2020). Yet to our knowledge, no systematic study to date has examined trends in public sentiment and expressions of willingness or hesitancy with regard to COVID-19 vaccinations and other treatments around these events.

We examine these issues in the United States for a number of reasons. The United States has high rates of vaccine hesitancy (Lazarus et al. 2021; King et al. 2022), and suboptimal COVID-19 vaccine uptake has contributed to public health outcomes that are measurably worse than in other countries (Aw et al. 2021). The cross-sectional correlates and causes of vaccine hesitancy (and hesitancy related to COVID-19 vaccines in particular) have been well studied and are linked to social media use (Wilson and Wiysonge 2020), political beliefs (Rich-Edwards et al. 2022), education (King et al. 2021; Willis et al. 2021), race and ethnicity (Padamsee et al. 2022; Willis et al. 2021), religion (Rich-Edwards et al. 2022), concerns about vaccine effectiveness (Haselswerdt and Gollust 2023; Lazarus et al. 2022; Padamsee et al. 2022; Willis et al. 2021), and trust in institutions and society more generally (Beca-Martínez et al. 2022; Ternullo 2022; Trent et al. 2022).

Our research examines several sources of data. First, we examine events that may have triggered greater public attention to the vaccine evaluation process or may have boosted uncertainty about it. We look for variation (especially abnormal changes) in news coverage and social media discussion of vaccine-related, regulation-related, and election-related events in the summer and fall of 2020. Second, we examine large-scale survey responses over time to understand temporal patterns of public sentiment. Previous research has examined cross-sectional differences in vaccine hesitancy, with less attention to temporal trends, especially those that might be associated with reluctance to adopt a new treatment among the population or key subgroups (Aw et al. 2021; Crane, Faden, and Romley 2021).

We develop our analysis first by describing a theoretical perspective and generating hypotheses. We then review relevant political events in the summer and fall of 2020 and test our hypotheses, examining abrupt changes in media and citizen attention and analyzing large swings in vaccine attitudes and COVID-19 vaccine hesitancy in the fall of 2020. Next, we review these tests of our hypotheses, identifying several events that coincided more than others with abrupt changes in media and public attention to vaccines, the FDA, and President Trump, and finally showing how contemporary movements in vaccine hesitancy were differentiated by partisanship, race/ethnicity, and education level.

Theory and Hypotheses

When faced with situations of high policy uncertainty and new technologies, citizens may experience uncertainty that leads them to avoid these new technologies even as they seek out and process information. Scholarship in health politics, public policy, law, and political science more generally offers several factors that may shape these judgments and reactions, including uncertainty, partisanship, race/ethnicity, and education. In formulating these hypotheses to test when examining changes in public attention and vaccine hesitancy, we draw on a “political Bayesian” perspective. Our research does not aim to develop a new theory; instead it draws on predictions and findings adduced elsewhere in the literature. Stemming from classic studies in voter behavior and public opinion (see Bartels [1987] 2020; Gerber and Green 1998), these perspectives predict that when new information emerges, it is the new information that differs the most from earlier knowledge that will most heavily revise human judgments. As Bartels (1993: 275) summarizes, “By the logic of rational (Bayesian) opinion change, only discrepant messages can produce observable change—and then only in direct proportion to the subjective uncertainty of preexisting opinions.” However, our Bayesian approach does not imply perfect rationality, as updating may be shaped by partisan, racial, and ethnic factors, or other considerations. The idea is to situate the hypotheses in a theoretically grounded process that explains reaction to short-term information but also predicts differential reaction by personal, political, and social context.¹

Uncertainty over What?

To render more precisely what we mean by policy uncertainty, we note that several forms of it might arise and shape citizens’ thinking. First, citizens might be uncertain about whether the FDA itself would be equipped to evaluate and clear a safe and effective vaccine, even in the absence of any political pressure. Second and relatedly, citizens might be concerned that FDA career officials have embedded biases in favor of new treatments and that they are too friendly to industry and to new medicines (a form of distrust of the sort that occurred after the withdrawal of Vioxx in 2004). Drawing on the literature, we focus on a third kind of policy uncertainty, namely that citizens generally trusted the FDA process but were worried that, with an impending election and a president prone to breaking norms, an independent scientific process would be compromised and the vaccine would be released so as to improve an incumbent candidate's reelection prospects. Scholars such as Moffitt (2010, 2014) and Carpenter (2002, 2010) have argued that trust in agencies like the FDA (especially trust in their career staff) is separable from trust in elected executive officials.² Other studies suggest that FDA review is premised on the reduction of uncertainty, both about drugs and about firms and disease categories (Carpenter et al. 2010), and that reducing this uncertainty may translate into reduced consumer or patient uncertainty (Carpenter, Grimmer, and Lomazoff 2010). We follow this lead and focus on trust in (or uncertainty about) the procedure of vaccine evaluation and whether it is delegated to career scientists or to elected officials and their direct appointees (such as the secretary of the Department of Health and Human Services or the commissioner of the FDA, both of whom are appointed directly by the president). This idea of procedural uncertainty also coheres well with the literature on commitment problems in delegation (see the classic treatment in Miller and Whitford 2016).

In short, policy uncertainty here is procedural uncertainty about a critical policy process—namely, the process of vaccine evaluation and whether it is (sufficiently) independent from pressure by elected officials. Based in part on qualitative evidence, particularly the announced efforts of several governors and Black physicians’ groups to establish an independent vetting process for the mRNA vaccine, we believe this form of uncertainty was the most pervasive in the autumn of 2020.

Theorized Processes and Associated Hypotheses

A. Policy or Regulatory Uncertainty

A1. Movements of Attention in Response to Particular Regulatory or Policy Events

Almost by definition, events garner journalistic and media attention (i.e., become “news”) to the extent that they are unexpected. Studies have previously shown sharp distrust in institutions in response to surprising new developments, adverse news, or scandals (Cvetkovich et al. 2002; Kleinnijenhus, Van Hoof, and Oegema 2011; Solé-Ollé and Sorribas-Navarro 2018), including in the context of COVID-19 (Jennings et al. 2021; Motta, Stecula, and Farhart 2020).

While we do not formulate it as a specific hypothesis, one can view the different triggering events we discuss as alternative potential causes of changes in distrust and associated vaccine hesitancy.³ However, in the context of uncertainty, we should expect to observe different events causing major changes in public attention (media, social media, internet search).

A2. Alternative Hypothesis: Monotonic Proximity to Election

As a source of policy uncertainty—a closely contested election—nears, institutional distrust and vaccine hesitancy will rise monotonically.

B. Subpopulation Reactions to Uncertainty

B1. Partisan and/or Ideological Reaction

American politics has become highly nationalized and polarized, particularly in recent years. Landmark studies in partisanship and political psychology demonstrate that citizens process basic news and social and economic information through a partisan lens (Green, Palmquist, and Schickler 2004), and they do so with high levels of out-partisan distrust (Iyengar et al. 2019). The COVID-19 pandemic exacerbated these tensions, giving rise to a “partisan pandemic” (Clinton et al. 2020; see also Gadarian, Goodman, and Pepinsky 2021; Jennings et al. 2021; Miller, Woods, and Kalmbach 2022; Motta, Stecla, and Farhart 2020; Ternullo 2022). Recent studies (Feldman et al. 2024) also demonstrate that trust in the FDA is highly correlated with political ideology. Given uncertainty-inducing events of this sort, a partisan-ideological reaction perspective would predict the following.

Hypothesis B1: Citizen response to policy uncertainty will be different along partisan lines; some parties will react more strongly to new information than others, and the reactions of different partisan groups may (but need not) move in opposite directions.

Corollary B1: The strongest movement to policy or regulatory uncertainty will occur among Democrats because of the presence of a Republican presidential incumbent and widespread distrust of President Trump in particular, and because Democrats have higher baseline vaccine trust than do Republicans and independents (that is, there is more trust to lose).⁴

B2. Racialized and/or Ethnically Centered Reaction

Scholars have documented significant racial and ethnic differences in COVID-19 response and vaccine uptake (Carman et al. 2021; Padamsee et al. 2022; Soares et al. 2021). Willis and colleagues (2021) found significant differentiation in trust as it related to vaccine hesitancy. The occurrence of the COVID-19 pandemic in the midst of nationally polarizing debates about race, and the fraught American history with race (Bailey and Moon 2020), imply the following.

Hypothesis B2: Citizen response to policy uncertainty will differ along racial and ethnic lines, with different reactions among Black, white, Hispanic, and Asian citizens.

Corollary B2: The strongest movement following triggered uncertainty should be witnessed among those populations most trusting in government and science (for example, among Asians).

B3. Education- and/or Class-Based Reaction

In part because more highly educated individuals have had greater exposure to science education and to others with scientific education and training, we should expect response to uncertainty to differ along lines of educational attainment. More educated individuals have traits (such as reading more science/health news and being less likely to live in rural areas) that correlate with more baseline trust in the FDA (Feldman et al. 2024).

Hypothesis B3: Citizen response to policy uncertainty will differ along lines of educational attainment.

Corollary B3: The strongest movement following triggered uncertainty should be witnessed among those populations most initially trusting of vaccines, such as those who are more highly educated.

Moving to the empirical portion of this article, we note that one difficulty in testing these hypotheses is that measurements of the factors involved in these three hypotheses (partisanship/ideology, race/ethnicity, and education) are highly correlated. Testing will require data that differentiate among these factors and that measure individuals’ responses over time.⁵

Uncertainty Triggers: Electoral Politics, White House Pressure on the FDA, Campaign Debates

The COVID-19 pandemic exploded in the winter and spring of 2020, moving from China through Europe and then to multiple locations in North America. On March 11, 2020, with tens of thousands getting sick and perishing, the World Health Organization declared COVID-19 a global pandemic. American society and its institutions appear to have been caught by surprise, and many observers have documented a lack of preparedness (Gollust, Nagler, and Fowler 2020; Oberlander 2020). Social debates about public health measures, including regulations and prohibitions on gathering; closure of schools, businesses, and other social institutions; and masking and other preventive measures have been discussed and studied elsewhere (e.g., Gadarian, Goodman, and Pepinsky 2021; Goidel et al. 2023; Gollust, Nagler, and Fowler 2020; Ternullo 2022). By the spring of 2020, however, public attention began to focus on possible treatment modalities for COVID-19, some preventive and others to treat the virus once contracted. The possibility that these therapies might be authorized more quickly than usual for use in combating COVID-19 was set when Alex M. Azar II, secretary of the Department of Health and Human Services (HHS), invoked a public health emergency declaration under the Public Readiness and Emergency Preparedness Act on March 17, 2020 (Office of the Secretary, Department of Health and Human Services 2020).

Although we cannot rule out other events as triggers of uncertainty, in our review of news and social media the following are the events that seem to have garnered the greatest amount of coverage and controversy from March 2020 to Election Day:

1. Hydroxychloroquine. The early debate over the possible therapeutic properties of hydroxychloroquine (Select Subcommittee 2022) was accompanied by a quick emergency use authorization (EUA) by the FDA on March 28, 2020. The hydroxychloroquine EUA was widely debated, although few surveys were in the field about institutional trust and vaccine sentiments in March 2020. Journalistic inquiry later demonstrated that the Trump administration worked to sidestep the FDA by getting hydroxychloroquine—for which the EUA was limited to hospitals and clinical trials—distributed to pharmacies (Rowland, Cenziper, and Rein 2020). In early April, the Trump White House secured the distribution of 23 million hydroxychloroquine tablets (enough for 1.9 million COVID-19 patients) to pharmacies around the country. By June, the FDA had reviewed sufficient evidence that it reversed the EUA for hydroxychloroquine as a treatment for COVID-19 (Kupferschmidt 2020).

2. Bleach and sunlight. On April 24, 2020, President Trump suggested in a news conference the possibility of injecting a disinfectant such as common bleach to combat COVID-19, or exposure to sunlight as another form of disinfectant. While the president later stated he was joking, it was not presented as a joke at the time of the press conference, and a number of citizens apparently took him seriously, as calls to poison centers increased and state and municipal health centers received more inquiries about the possibilities of using disinfectants to combat COVID-19 (Qamar 2020).

3. Messenger RNA vaccine emergence and Operation Warp Speed. The sequencing of the COVID-19 genome in early 2020 led to the development of an mRNA vaccine candidate by Moderna (and separately by German outfit BioNTech). On March 16, 2020, Moderna's candidate entered clinical trials. By May 2020, laboratory announcements and news reports began to emerge of promising results from these trials. Around the same time, the Trump administration announced a set of policies under the umbrella code name Operation Warp Speed (Slaoui and Hepburn 2020) focused on subsidizing and facilitating the testing and production of COVID-19 vaccines.

4. Pressure on the FDA and the convalescent plasma authorization. The Trump administration's pressure on the FDA reached new levels and broke new barriers in August 2020. In July and August 2020, Trump administration officials began discussing the possibility that a treatment known as convalescent plasma—a concentration of antibodies drawn from patients who had been infected with COVID-19—may be effective. FDA officials saw limitations to the treatment's efficacy but received constant communication from Trump White House officials (and not just higher-level HHS officials) about the possibility of an EUA for convalescent plasma and the preferred timing of such an EUA. Trump then expressed his frustration directly on Twitter on the eve of the Republican National Convention (August 22, 2020, before 5 a.m.) (Select Subcommittee 2022: 33): “The deep state, or whoever, over at the FDA is making it very difficult for drug companies to get people in order to test the vaccines and therapeutics. Obviously they are hoping to delay the answer until after November 3rd. Must focus on speed, and saving lives! @SteveFDA.”

   Trump's tweet directly called out FDA Commissioner Stephen Hahn. The next day, the FDA announced the EUA of convalescent plasma. The announcement of the convalescent plasma EUA differed from previous announcements in important ways. Commissioner Hahn made the announcement, and he did so from the White House, standing on the podium with President Trump himself, and personally thanked the president. Commissioner Hahn stated that convalescent plasma offered a 35% reduction in mortality, confusing absolute risk and relative risk estimates from treatment. (As it turns out, convalescent plasma was effective on neither an absolute nor a relative scale, but the important point is that the FDA commissioner erred on a basic principle of statistics on the podium, apparently weakening his credibility.) At the same time, FDA spokesperson Emily Miller, a political appointee, highlighted the plasma authorization and linked the decision to the ongoing Republican National Convention. Miller's tweet restated President Trump's line from the Republican National Convention—“‘Convalescent plasma will save thousands and thousands of lives.’—Pres. Trump”—and added the hashtags “#COVID-19” and #RNC2020Convention (Sollenberger 2020).

   According to our review of FDA history, August 23, 2020 marked the first time that an American head of state was personally present at the announcement of any new authorization or regulatory approval of a new treatment. Previously, long-term practices had even separated agency leadership such as the commissioner from such announcements. Newspapers and other media sources noted the president's presence and gave the event extensive media coverage (Blake 2020; Kaplan and Thomas 2020; Sachs 2020).

5. News on FDA vaccine testing requirements and Trump criticism. On September 20, the Washington Post reported that the FDA was planning to release guidance that would spell out rigorous evaluation standards for emergency vaccine authorization and might push vaccine authorization until after Election Day. On September 23, President Trump attacked the planned release of guidance as “a political move,” and journalists reported that his chief of staff, Mark Meadows, demanded detailed justification from the FDA (Goldstein and McGinley 2020). Also on September 23, New York governor Andrew Cuomo announced an independent review of any FDA-approved vaccine.

6. FDA release of testing requirements for vaccine. On October 6, 2020, the FDA released guidance for the minimal requirements for an EUA of coronavirus vaccine. The Trump administration had earlier blocked the guidance. Then President Trump tweeted his disapproval and suggested a political conspiracy against him: “New FDA Rules make it more difficult for them to speed up vaccines for approval before Election Day. Just another political hit job! @SteveFDA.”

7. Kamala Harris debate statement. On October 7, 2020, in the sole vice presidential candidate debate, when asked if she would take a coronavirus vaccine approved by the Trump administration, Senator Kamala Harris said, “If Dr. [Anthony] Fauci, if the doctors tell us that we should take it, I'll be the first in line to take it. Absolutely,” Harris said. “But if Donald Trump tells us to take it, I'm not taking it.” Trump, Pence, and other Republicans accused Harris of undermining Americans’ faith in a lifesaving vaccine.

Noted Swings in Vaccine Hesitancy, September 2020: Gallup and Pew

Before examining news events, we note that there was in fact a large and observed increase in vaccine hesitancy in the late summer and early autumn of 2020. On September 17, 2020, Pew Research Reports released a report (Tyson, Johnson, and Funk 2020) titled “US Public Now Divided over Whether to Get COVID-19 Vaccine.” The report showed substantial increases in reluctance to get a COVID-19 vaccine across a range of variables (partisanship, education, race, and gender), with a 21-percentage-point decline in willingness to get the vaccine from May to September 2020 (table 1). Separately, the Gallup Panel (Reinhart 2020) also showed a roughly 10-percentage-point increase in vaccine hesitancy from August to September 2020, after a 5-point increase from July to August. Figure 1 displays the spike in vaccine hesitancy in the Gallup Panel, adjusted for within-subject measurements (see appendix table B-01 for the corresponding regression estimates). Appendix figure B-0 presents the same analysis with survey weights, revealing a nearly identical pattern.

The series displayed in figure 1 is broad-scale trend data, and disaggregation is required to test our main hypotheses, but the aggregate time series can falsify the simplest version of hypothesis A-2, namely, that while there was an increase in hesitancy before the election, the month before November 2020 witnessed a decrease in vaccine hesitancy. Why did vaccine hesitancy peak well before the election, in September 2020, and then decline in October 2020? While our data do not permit a precise answer to this question, we find clues in examining news and social media responses to events at the time.

While figure 1 and table 1 present aggregate movements from two surveys, other snapshots point to newsworthy levels of distrust in COVID vaccines and particularly in President Trump in September of 2020. As with any such report, one cannot rule out the possibility of media or reporting bias. Yet an Axios poll (Talev 2020) revealed that “barely two in 10 Americans would take a first-generation coronavirus vaccine if President Trump told them it was safe.” A poll by Harris and STAT News (Silverman 2020) revealed that “most Americans believe the Covid-19 vaccine approval process is driven by politics, not science.” An NBC News poll (Kamisar and Holzberg 2020) revealed that twice as many Americans (52%) did not trust President Trump's remarks about a vaccine, compared to those who did trust him (26%), also reporting a 5-percentage-point drop in vaccine confidence during August.

Changes in News Reporting, Internet Searches, and Social Media Activity

The evident swings in vaccine hesitancy conceal important variation and political dynamics. The Pew study cannot reveal the timing or trajectory of change between May and September 2020. While the Gallup Panel demonstrates major movement in September 2020 (after a notable but smaller increase in August 2020), the movements can only be analyzed at the monthly level. Although other surveys reported similar levels of distrust in the vaccine and the FDA (Kamisar and Holzberg 2020; Silverman 2020; Talev 2020), few of them repeated comparable measurements of vaccine confidence over time.

To assess the events that seem to have drawn the most news attention and social media activity, we systematically examined two databases that curate news and social media use. First, to measure news coverage over time, we drew on searches of the ProQuest TDM Studio database accessed through our home institutions. (Search terms and parameters are reported in appendix A.) Second, to measure social media activity, we drew on the Twitter database of the Network Science Institute at Northeastern University (NetSI). Calendar year 2020 was a period before the acquisition (and subsequent renaming) of Twitter by Elon Musk, after which use of the platform fell precipitously. We nonetheless adjust our estimates for baseline rates of use by population attributes (especially partisanship) as measured by NetSI.

Although other databases exist, we chose these because in each case the data are curated, cleaned, and maintained consistently and historically by institutional third parties. In each case—ProQuest TPM Studio and NetSI's Twitter database—there are proprietary restrictions on exporting the full data (we can only request aggregates). Yet the aggregate data we have procured are available for replication analysis, and in the case of NetSI, their proprietary arrangement with Twitter allows them to measure user attributes.

News Coverage

We were interested in the intersection between presidential politics and vaccine authorization, so we examined the fluctuation of news stories mentioning the FDA, President Trump, presidential candidates Biden and Harris, and the vaccine. We were particularly interested in news attention to Trump and the FDA. A search for “Trump” and “FDA” in the database returned 17,825 documents. Figure 2 displays the number of these news articles by day, from July 2020 to the end of December 2020. (See appendix A for other time-series plots.) The aggregate article counts show weekly fluctuations, but three important spikes occur: on August 24, 2020, one day after the convalescent plasma announcement (noted with a vertical dashed line); on October 8, 2020, the day after the Harris-Pence vice presidential debate (noted with a vertical dashed line); and several days after Election Day (November 3, noted with a vertical dashed line), on November 8–9, when Trump and Pence praised Pfizer for vaccine progress and when candidate Biden announced his COVID task force selections.

For the period before Election Day, the spike on August 24 amounted to more than a quadrupling of the average flow of articles from July 1, 2020, to August 22, 2020. Moreover, the longer-term level of articles mentioning FDA and Trump increased substantially in the weeks following. From July 1 to Election Day, the post-plasma-announcement count of stories is 104, compared to 68 beforehand (t-test of means = 4.41; p < 0.0001). Not counting the peak of 286 stories on August 24, the seven weekly peak story counts after August 23, 2020 (averaging 157.2 stories on the weekly peak), are 50% higher than the seven weekly peaks beforehand (102.1 stories). A Bayesian changepoint analysis (see appendix A.2) suggests that August 23, 2020, corresponds to the most plausible changepoint of this time series before Election Day. Our conclusion here is not that the August 23, 2020, event caused those higher peaks, but that the period from late August through September and into early October saw much higher news coverage of the FDA and of Trump.

Analysis of Twitter Data

For the NetSI Twitter database, we retrieved Twitter aggregates by a range of keywords and keyword combinations—Trump, vaccine, COVID, FDA, Kamala OR Harris—and decomposed the aggregates by partisanship of user (as measured by the NetSI data). A fuller analysis and display of trends during 2020, with a focus on the fall of 2020, appears in appendix A.

We examine the responses of Twitter time-series keywords to the events discussed above. The first result is that many of the keyword time series do not respond strongly to these events. This is especially the case with “Trump,” (“Kamala” OR “Harris”), “vaccine,” and “COVID.”⁶ However, three events are associated with substantial immediate increases in tweets and, in some cases, longer-term increases as well.

• ▪ Hydroxychloroquine (March 28, 2020). The hydroxychloroquine authorization is associated with a 20-fold increase in the number of tweets mentioning “FDA” and “Trump” (appendix A.3.5), a level not again exceeded until August 24, 2020. The event is not associated with a large increase in tweets about either the FDA and the vaccine or Trump and vaccine, which is understandable in light of the distant prospect of COVID-19 vaccines at this time.

• ▪ Plasma (August 23). The convalescent plasma announcement is associated with another 20-fold increase in tweets mentioning the FDA (appendix A-3.4), or FDA and Trump relative to the previous week (appendix A-3.5). Among all calendar 2020 days, this series reaches its peak on August 24. The plasma announcement also occasions the largest peak in the “FDA AND vaccine” series, which was little affected by hydroxychloroquine. Bayesian changepoint analyses (appendix A.5) suggest that August 23 observed a structural change in these time series.

• ▪ FDA guidance announcement and Trump criticism of FDA (September 24). The news coverage series seems little affected by the September 20 news of the FDA's guidance and Trump's criticism of September 23, but in the Twitter data, the days from September 20 to September 23 witness peaks in “Trump AND vaccine” (appendix A.3.8) and “FDA AND vaccine” posts (appendix A.3.6). There is no observable peak for “FDA” tweets in late September, however (appendix A.3.4), which had fallen dramatically since the August 23 plasma announcement.

• ▪ Harris's debate performance and FDA guidance (October 6 and 7). These dates exhibit a roughly 12-fold increase in tweets mentioning the FDA and Trump (appendix A.3.6). These increases are about 60% of the peaks seen for the hydroxychloroquine and plasma events. The “FDA and Vaccine” series reaches a preelection peak on October 7. Tweets mentioning Harris also spike on these days, although not as much as on the day when she was chosen as Biden's running mate.

Partisan Differences in Tweets

When we examine those events for which tweeting increased, we find clear partisan differences, especially for the first two events. The increase in tweets mentioning “FDA AND Trump” or “Trump AND vaccine” or “FDA AND vaccine” is generally characterized by a much higher increase for tweets from recognizable Democrats than from recognizable Republicans (appendix A6.4, 6.5, 6.6, 6.8). Since more Twitter users were Democrats at this time, however, it is important to point out that while the compositional increase was driven more by Democrats, Republican tweeters increased their rate of tweeting on these subjects proportionately more than Democrats did.

The following figures demonstrate some of the changes we observe. Figure 3a shows that the August 23, 2020, plasma announcement is followed by a statistically significant and substantive increase in tweets containing the words “FDA” and “vaccine.” (Note that the plasma announcement was not directly concerned with the COVID-19 vaccines under development, so the Twitter users were making a separate, indirect connection to the COVID-19 vaccines.) Figure 3b shows that the increase is driven by Democrats, whose tweeting behavior before mid-August was statistically indifferentiable from that of Republican Twitter users; but from late August onward, Democratic users of Twitter more consistently mention the agency along with the vaccine. (The LOESS functions plotted use data from leading as well as lagging observations, so the functions move earlier than do the numbers themselves.)

Figure 4 demonstrates similar but subtly different patterns, with tweets mentioning both “Trump” and “vaccine” (and by extension, not the FDA). Here the total Democratic and total Republican user aggregates are similar and largely nondifferentiable before August 2020, but at that point they begin to diverge. Interestingly, while important differences occur on August 23, 2020, and in the week following, the largest partisan tweeting difference occurs on September 23, during the first controversy over FDA guidance. Note, however, that the partisan series are no longer statistically differentiable by Election Day 2020. Figures in appendix A, section 6, show similar patterns for tweets on the FDA alone.

Conclusions

The hydroxychloroquine authorization, the plasma announcement (August 23), the late September controversy over FDA guidance (September 20–23), and the combined Harris statements and FDA guidance (October 6–7) are all associated with near-term increases in Twitter activity in which the FDA, the vaccine, and President Trump are associated with one another. Importantly, these events are not associated with near-term increases in tweets about Trump himself, about the vaccine itself, about COVID, or about Kamala Harris (save for her appearance at the vice presidential debate).

The most important takeaway is that starting in late August and continuing through early October, a much higher level of Twitter activity was witnessed, and this aggregate activity was driven by Democrats.

The Fallout in Public, Professional, and Elite Opinion: Existing Accounts from Published Sources and Media Reports

We cannot say whether the convalescent plasma authorization was a cause of the subsequent media and social media attention; but the period from August 23 to the end of September 2020 was characterized by much higher news reporting and Twitter activity linking the FDA, Trump, and vaccines. It is important to place these movements in the context of a national electoral campaign that was accelerating, and the ends of the major parties’ national conventions. Still, after attention to the intersection between Trump, the FDA, and the vaccine stayed at low levels through the summer of 2020, the convalescent plasma announcement was at the beginning of approximately five to six weeks of higher news coverage. Again, overall coverage of the vaccine and COVID-19 did not exhibit discontinuous changes during this time period, and social media activity on Trump trended gradually upward until Election Day.

It is thus important to see how news organizations covered the plasma announcement and how many of them raised issues about the FDA's credibility and about the possibility that vaccine authorization might be compromised. Numerous statements of former FDA commissioners and medical experts cast doubt on the convalescent plasma authorization. Commissioner Hahn responded by correcting himself and apologizing in a tweet of his own on August 24. He then terminated FDA spokesperson Emily Miller on Friday, August 28. As two New York Times reporters interpreted the moves, “The head of the Food and Drug Administration ousted its top spokeswoman from her position on Friday in an urgent bid to restore the tarnished credibility of the agency after he made erroneous claims that overstated the benefits of plasma treatments for Covid-19 at a news conference with President Trump” (Foley 2022; Kaplan and Thomas 2020).

For many media outlets, the damage had already been done, even as Commissioner Hahn walked back his Sunday evening remarks and tried to right his ship with personnel changes. As Washington Post reporter Aaron Blake (2020) wrote, “Two months ago, the Food and Drug Administration reversed course on President Trump's much-hyped drug hydroxychloroquine, revoking its authorization for emergency use to treat the novel coronavirus. Given that Trump had pushed for the drug and then bragged about speeding the initial authorization process along, it raised very valid concerns about whether the FDA had made a hasty decision under political pressure. Those concerns have to resurface after the spectacle we saw Sunday.”

Blake had made the connection between one product authorization decision (convalescent plasma) and an earlier one (hydroxychloroquine). In both cases, he noted, the FDA had quickly admitted errors in judgment. In both cases, he concluded, there was political pressure. Yet those who were following the development of the mRNA vaccines for the novel coronavirus began to wonder anew whether the vaccine authorization process would itself be affected by electoral calculations. Prominent trade publications that are heavily read by investors—notably Pink Sheet, which has been a principal source of industry news for biopharmaceutical investors and followers for more than seven decades—wondered in print if confidence in the vaccines under development would be affected (Karlin-Smith 2020). Pink Sheet journalist Sarah Karlin-Smith's (2020) piece on this topic was titled “Plasma Authorization Raises Fears of Politically-Influenced COVID-19 Vaccine Decisions, Compromised FDA.”

In the wake of the convalescent plasma story, different actors in politics, public health, and business began to speak out about their worries regarding the credibility of any COVID-19 vaccine authorized under partisan political pressure. Microsoft cofounder and billionaire Bill Gates, who has donated billions to public health and vaccination initiatives, remarked that “we saw with the completely bungled plasma statements that when you start pressuring people to say optimistic things, they go completely off the rails. The FDA lost a lot of credibility there” (Gibson 2020). A CBS News reporter summarized the Gates interview by saying that Gates “fears that politics will erode public confidence in the eventual vaccine” (Gibson 2020).

In mid- to late September 2020, representatives of the National Medical Association (NMA), a Black physicians group founded in 1895 to counter other medical associations’ exclusion of Black American doctors, announced the creation of an independent task force to vet FDA decisions on vaccine authorization (Boodman 2020). The NMA's president said, “It's necessary to provide a trusted messenger of vetted information to the African American community. There is a concern that some of the recent decisions by the Food and Drug Administration have been unduly influenced by politicians” (Boodman 2020). Another Black physician, Uché Blackstock—not an NMA member, but the founder and CEO of consulting firm Advancing Health Equity—argued explicitly that the new effort was needed because of cratering Black trust in the FDA: “What we've seen in terms of political interference in the FDA and CDC [Centers for Disease Control and Prevention] has really undermined what little trust the Black community had” (Boodman 2020).

Finally, the governors of New York and California announced that they would establish vaccine review processes that were independent of the FDA. Governor Andrew Cuomo of New York announced on September 23, 2020, that New York would have an independent panel to review any vaccines approved by the FDA. Cuomo faulted not the FDA or Commissioner Hahn but Trump and his electoral timeline. “I don't think Dr. Hahn is running for anything,” Mr. Cuomo said, adding that “President Trump is engaged in the political process and has an Election Day” (Gold and McKinley 2020). Weeks later, California governor Gavin Newsom announced on October 19, 2020, that he would be appointing a “COVID-19 Scientific Safety Review Workgroup” charged with “independently review[ing] the safety and efficacy of any vaccine that receives FDA approval for distribution. . . . These top health experts—guided by the principles of safety, equity and transparency—will review any vaccine that receives federal approval and verify its safety, before California makes a COVID-19 vaccine available to those most at risk” (Office of the Governor 2020).

Limitations of Existing Reports

While various public sources suggest growing concern at this time, there is as yet no sustained and targeted analysis that examines changes in public opinion at this time with regard to institutional trust and vaccine hesitancy. This is important for assessing the degree of sentiment or opinion change—and, just as important, its timing. Black physicians feared that, in September 2020, the confidence of Black Americans in COVID-19 vaccines had plummeted, and that Black Americans wanted an institution other than the Trump administration FDA to validate the efficacy and safety of COVID-19 vaccines. Undoubtedly these physicians were privy to a range of community conversations; but there is little public evidence suggesting that Black vaccine hesitancy rose (as opposed to remaining at customarily high levels) in August or September 2020.

Even when we can document significant changes in expressed trust and vaccine hesitancy, we often do not know whether changes in opinion reflect the composition of different survey respondents and different times, or whether individuals’ attitudes were changing. Different polling firms (i.e., Gallup and the Kaiser Family Foundation) reported aggregates at different times (summer, fall, and winter 2020), none specifically reported fixed-effects estimates that allowed for the tracing of movements across individuals. Nor did any of these other surveys report movements at a level of temporal granularity of less than a month.

Examination of Temporal Changes in American Public Sentiment: Data and Methods

To address these shortcomings of previously published reports on public opinion regarding COVID-19 vaccine hesitancy in the fall of 2020, we gathered and analyzed data from the Gallup Panel (Gallup 2022) and Nationscape (DFVSG n.d.), two large nationally representative surveys, to examine changes in vaccine hesitancy in the second half of 2020 through the spring and summer of 2021. We examine short-term aggregate movements of surveys and trace the durability of any changes.

Gallup Panel

The Gallup Panel is a probability-based, nationally representative panel of US adults that ran from March 2020 to September 2021 (Gallup 2022). The Gallup Panel allowed us to analyze within-subject variation, that is, changes in vaccine hesitancy (as defined by expressed willingness to be vaccinated for free) for the same people at different points in time (see appendix section B1). We applied multivariate fixed-effect regressions to examine changes by month of survey wave. We estimated changes in vaccine hesitancy per month and by reported categorical variables (self-reported race, partisanship, education). For the subsample panel of respondents who, from September and October 2020 until September 2021, switched from being vaccine hesitant to being nonhesitant, we examined the cross-sectional determinants of continued vaccine hesitancy.

Nationscape

Nationscape is a large-scale representative national survey with weekly waves that ran from July 2019 to January 2021. From the summer of 2020 onward, the surveys included questions on COVID-19 vaccine confidence (on a 0–100 scale) and whether respondents had specific concerns about vaccination that would prevent them from getting vaccinated (appendix section B2). These survey data permit highly granular temporal analysis of changes across survey waves. We observed a pool of 116,853 respondents answering the COVID-19 vaccine confidence question, and 76,668 respondents answering questions about specific concerns regarding the COVID-19 vaccine. Questions about COVID-19 vaccine concerns were stopped in mid-November 2020. We display comparisons between the Gallup Panel and the Nationscape survey in appendix B.3.

An important limitation of these data is that spring 2020 views on vaccines are not easily or credibly measurable with surveys, as COVID-19 vaccine hesitancy only began to be measured in July 2020 in Gallup and in August 2020 in Nationscape. The reason for this is that COVID-19 vaccine candidates only became viable in terms of regulatory authorization over the summer. While this shapes our subgroup response hypotheses, it does not affect our examination of news and social media trends, which rely on different data, and it does not shape our analysis of responses to events after July 2020.

Comparators

To assess the hypothesis that observed changes were a function of confidence in vaccines more generally rather than COVID-19 vaccines in particular, we also used Nationscape data to examine confidence in the flu vaccine and general vaccine hesitancy (appendix section B2). We also used Gallup Panel questions about avoidance of going out in public, small gatherings, and travel (appendix section B1). Examination of these questions also helps to assess the possibility that movement in vaccine confidence or hesitancy was related to other considerations about the pandemic, especially social isolation behaviors and strategies.

Study Results with the Gallup Panel and Nationscape

As reviewed above (fig. 1), the Gallup Survey reveals a single-month 10-percentage-point increase in vaccine hesitancy from August 2020 to September 2020. This change was concentrated among Democrats, whose hesitancy rose by 25% during this month, and independents, whose hesitancy rose by 14%. While the September shift does not rule out hydroxychloroquine as a contributing factor, it suggests that intervening events either (1) primed or reinforced questions from the hydroxychloroquine authorization, or (2) supplied new reasons for uncertainty. Appendix figures C-1 and C-2 show smaller movements for Nationscape respondents.

Tests of Hypothesis B1: Partisanship

Analysis of the Gallup Poll shows statistically differentiable levels and changes in levels of vaccine hesitancy by partisanship. In particular, the September 2020 rise in vaccine hesitancy in the Gallup Panel among Democrats (25 percentage points) is far higher than for Republicans and independents. This result also provides support for corollary B1 (expected greater response among Democrats). Although we also note that initial Democratic trust in the vaccine is higher, we cannot test the mechanism that the September drop occurred because Democrats were initially more trusting. The Nationscape data give more nuanced results (see appendix figure C-2), which were smaller for the general population but large for several subpopulations. There was a 3% drop in overall vaccine confidence (August 23–September 16, 2020)—including a 10% drop among Democrats—and a 4% increase in vaccine safety concern (appendix figures C-1 and C-2).

Tests of Hypothesis B2: Race

One group for whom there was a concentrated rise in vaccine hesitancy from August to September 2020 was Black Americans, whose hesitancy rates in the panel rose 21 percentage points in a month, from 55% to 76%. As figure 5 makes clear, these hesitancy rates among Black Americans were not increasing substantially from July to August (2 percentage points, from 53% to 55%). The Gallup Panel also reports a large rise in vaccine hesitancy at the same time for Asians (roughly 30%). These results provide support for hypothesis B2.

The Nationscape survey demonstrates little movement in the vaccine confidence responses for Hispanics and Asians, a small and negligible decline for Asians, and a small but statistically differentiable decline for whites between August 26 and September 23. The Gallup results for Asians lead us to partially reject corollary B2, namely that the largest movements should occur among those racial and ethnic groups less trusting of government. Asian Americans generally show higher levels of trust in government than do Black Americans (Koch 2019), especially during COVID-19 (Li et al., 2022), but in this case the two populations reacted similarly. Still, given the levels of ethnic bias triggered against Asian Americans during the Trump presidency and the COVID-19 pandemic, racialized (dis)trust may well have played a role in Asian Americans’ rapidly changing views about the COVID-19 vaccine.

Tests of Hypothesis B3: Education

While respondents without college degrees (or having up to an associate's but not a bachelor's degree) had always displayed less confidence in the COVID-19 vaccine, there were important changes among college-degree holders from late August to mid-September in the Nationscape survey (table 2). Those three weeks saw an 8-percentage-point increase in safety concerns reported by bachelor's-holding Republicans and a 10-percentage-point increase in safety concerns reported by bachelor's-holding independents. In the following two months these levels of concern abated more for Republicans than for independents, but not entirely. Statistically nonsignificant and substantively minimal movements were seen for Republicans and independents without college degrees in the Nationscape sample. These results are consistent with hypothesis B3 and especially corollary B3.

Comparative Predictive Power

Can race, education, and gender be distinguished from partisanship, and vice versa? There are two ways of assessing this question. The first is to ask, in an analysis of variance or multivariate regression, whether there is identifiable covariation between these nonpartisan factors and hesitancy once partisanship is controlled for, and what fraction of variance is (descriptively) explained by nonpartisan variables. Table 3 displays an analysis of variance for these factors for the Gallup Panel. For July to October (before the election), party is clearly the dominant factor, with almost half the model-explained variance (47%). But the combined effect of bachelor's degree, race, and sex accounts for 40.6% of the model-explained variance. Party remains the most important factor in the Gallup Panel, but race, gender, and education matter as well. Corresponding tests for the Nationscape survey (appendix tables E-2A-1 and E-2A-2) show that party and race rival each other in total predictive contribution for the sample before Election Day, but we note that these results are necessarily cross-sectional and not, like the Gallup Panel, about different sources of movement.

The second is to examine Wald tests for differences within party, or for non-Democrats (since Democrats have the largest movement). Appendix tables E-2A-1 and E-2A-2 show such an analysis for education and party in the Nationscape sample, and they document significant movements among college-educated independents and Republicans on vaccine safety concerns from August 23 to September 16 (see the code and corresponding regression in appendix E-1, and in appendix E-2 for analyses adding race/ethnicity). We do not present an identical table for Gallup in the text, but appendix E-3 presents a table and code showing the results of this exercise for the Gallup Panel.

We retrieve a number of Wald tests showing increases in vaccine hesitancy from July 2020 to September 2020 among Democrats, differentiated by education. For example, we find such increases occur among college-educated Democratic Black women (F = 63.74; p < 0.001) but not for Black female Democrats without a bachelor's (F = 0.34; p = 0.56); among independent white women, both for those without a bachelor's (F = 5.20; p = 0.02) and for those with (F = 10.07; p < 0.01); and independent Hispanic men (F = 5.36; p = 0.02) and independent Asian men (F = 389.88, p < 0.001) without a bachelor's degree.⁷

A Heterogeneous Decline in Vaccine Hesitancy in Fall 2020 and Spring 2021

There was considerable decrease in vaccine hesitancy after early November 2020 (fig. 1) that was highly concentrated among Democrats and more educated respondents. We note that these declines in vaccine hesitancy were undoubtedly affected by the authorization of the mRNA vaccines on December 11, 2020, and the subsequent use of the vaccine itself, so they remain purely descriptive.⁸

Vaccine hesitancy did not, however, recover evenly among the populations for whom it increased heavily in August and September. The Gallup Panel permits analysis of which subjects who were hesitant in fall of 2020 switched to being less hesitant from November 2020 to September 2021. Figure 6 examines nonpartisan factors and displays relevant coefficient estimates from a linear probability model. Controlling for partisanship (coefficients not displayed), earlier growth in vaccine hesitancy among Asian Americans and Hispanics largely recovered. Yet Black respondents who became more hesitant in the fall of 2020 were almost 20% less likely than other groups to switch from hesitancy after October 2020. Similarly, respondents who lacked college degrees were 12% less likely to have switched from being hesitant to nonhesitant after the fall of 2020.

In short, while the aggregate spikes in hesitancy and vaccine safety concern in September 2020 appear to have regressed to the lower level mean, this was not the case for two important subpopulations: Black respondents and independent respondents without bachelor's degrees (fig. 6). To be sure, the continuation of hesitancy after October 2020 may have been shaped by factors that emerged only after the authorization and distribution of the vaccines.

Comparator Tests

Appendix figures D-1 through D-4 show that, in the Nationscape survey, there was no similar movement nor any statistically differentiable pattern in either the proportion of respondents expressing confidence in the influenza vaccine or in the proportion of respondents expressing general antivaccination attitudes.

Appendix figures D-5, D-6, and D-7 show that all three nonavoidance behaviors were stable during the Gallup Panel in September and October 2020. Hence, concerns about vaccine efficacy and safety were spiking even as self-reported behaviors were not changing.

Limitations and Exclusions

Our research has a number of limitations. First, we lack survey data on COVID-19 vaccine hesitancy for spring 2020 that permits comparison with fall 2020. Second, there may be reliability issues pertaining to vaccine hesitancy survey responses when there was no actual vaccine available. Third, some differences in the changes observed across surveys may reflect the survey instruments used. Gallup uses a binary indicator of vaccine hesitancy, while Nationscape uses a 1–100 scale of vaccine confidence. The binary safety concern question in Nationscape exhibited larger movements than did the vaccine confidence scale. Fourth and finally, space limitations prevent us from conducting a full analysis of the entire trajectory of COVID-19 vaccine hesitancy from 2020 to the present. There are important research questions about the evolution of vaccine hesitancy in November–December 2020 and afterward, but they are best left to other efforts.

Discussion

Scholars have long drawn connections between trust or distrust in science and public institutions and corresponding trust in health treatments, especially vaccines. Yet few of the studies available permit an analysis of changes in vaccine hesitancy over time, asking whether particular events or patterns of policy making, media reporting, and/or social discourse might shape trust and vaccine hesitancy. The summer and fall of 2020 saw the collision of several forces that heightened uncertainty: a pandemic, the possibility of a new preventive modality in mRNA vaccines, a deeply polarizing and competitive election, and a president who rejected previous patterns of deference to decisions of scientific agencies. Did institutional trust and vaccine hesitancy change along with these forces, and in ways that are differentiable from changes in trust and hesitancy before and after these events?

We approached these developments using a political Bayesian perspective that embeds a degree of learning that is affected by partisanship and related political variables. Our review of published media reports and published survey results suggests that a number of audiences reacted to broad and unprecedented political intervention in FDA processes beginning in the spring of 2020 (at the latest). Media reports and Twitter use also suggest a growing degree of attention to the intersection of electoral politics, presidential pressure, and the FDA (or the vaccine approval process) from late August to early October 2020. Three events in the autumn of 2020 appear to have garnered the most attention: (1) the August 23 authorization of convalescent plasma, (2) the controversies between September 22 and 23 with FDA guidance news, Trump criticism, and Governor Cuomo's announcement of an independent review task force for New York state, and (3) the twin events on October 7 and 8 with the final FDA guidance announcement and Senator Harris's vice presidential debate remarks.

We then showed via two different longitudinal surveys how vaccine hesitancy rose sharply in the fall of 2020 and then recovered heterogeneously across key subpopulations. The movements in these surveys were similar in direction—exhibiting an increase in vaccine hesitancy and vaccine safety concern—but differed in magnitude. We can reject a purely monotonic version of “proximity to election” as the explanation for vaccine hesitancy, as hesitancy and safety concern peaked in September 2020 and declined the next month. While our results generally support a political Bayesian perspective—those who were most initially trusting of government and vaccines lost the most trust, and along partisan lines—the fact that Black and Asian Americans both saw major increases in vaccine hesitancy from July to September 2020 is a departure from this Bayesian perspective, as one of these populations (Asians) was far more trusting in government institutions than the other (Blacks). Our results also suggest that movements in vaccine hesitancy and safety concern were not reducible to partisanship alone.

The vast scale of US society and the likelihood that hundreds of millions of Americans would be eligible for COVID-19 vaccines ensures that even small-scale changes have important implications for public health. Even some of the smaller shifts detected are statistically significant in weighted samples with linear tests of restrictions (Wald tests). If projected to the entire adult US population, these shifts would reflect many millions of adults (perhaps 10 million or more) becoming newly wary of COVID-19 vaccines in September 2020.

Finally, we note that two movements among our surveyed population appear to have persisted. Black American vaccine hesitancy spiked in September 2020, and Black Americans who reported hesitancy at this time were significantly less likely to become more confident during our study period. Second, there was no gap between Americans with and without college degrees in the summer of 2020, and the gap that opened in August and September of 2020 has never abated. Both surveys suggest that the educational gap persists even within parties.

In conclusion, we document spikes in hesitancy and safety concern about COVID-19 vaccines in September 2020, concentrated among Democrats, independents, Asian Americans, and Black Americans. After this period, vaccine hesitancy did not decline uniformly but persisted differentially among Black Americans and individuals without college degrees (especially self-reported political independents). As we have noted, public health officials and physicians seemed to notice some of these changes, although they did not report estimates of opinion change when doing so. Ours is the first analysis to statistically document these changes.

Because trust in medical products depends on regulation and trust in regulators (Carpenter, Grimmer, and Lomazoff 2010; Kaina 2008; O'Neill 2018), our findings identify potential dangers associated with political uncertainty associated with vaccine evaluation and authorization (GAO 2022). This is especially concerning given that the FDA's timing in emergency vaccine authorization was rapid, and any possible delay carried negligible plausible public health consequences (Borrell 2022). Indeed, our results suggest that uncertainty around FDA review, and public perceptions that FDA review might be caught up in electoral politics, likely undermined vaccine confidence. In situations when such controversy is unavoidable, our findings may point to a particular need for outreach among those groups most heavily affected by the uncertainty, to increase uptake of a useful novel vaccine during a future pandemic.

Acknowledgments

This work was funded by the Greenwall Foundation Making a Difference grant program. Daniel Carpenter's work is also funded by Open Philanthropy. William Feldman's work is also funded by the National Heart, Lung, and Blood Institute (K08HL163246), and his and Aaron Kesselheim's work is funded by Arnold Ventures. Outside the submitted work, William Feldman and Aaron Kesselheim receive grant funding from the FDA and serve as consultants for Alosa Health. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; review or approval of the manuscript; and decision to submit the manuscript for publication. We acknowledge Stephen Ansolabehere, Jennifer Hochschild, Taeku Lee, and Lynn Vavreck for comments on earlier versions of the manuscript.

Notes

References