On January 6, 2021, pro-Trump protesters stormed the US Capitol. The violence capped months of efforts by President Trump and allies to overturn the 2020 election results. It was also the culmination of a four-year term that threatened a range of democratic values (see, especially, Azari Reference Azari2018; Lee and McCarty Reference Lee and McCarty2019; Levitsky and Ziblatt Reference Levitsky and Ziblatt2018; Lieberman et al. Reference Lieberman2019).
The 2017–21 Trump presidency took place after decades of sorting and political polarization by citizens (Levendusky Reference Levendusky2009; Mason Reference Mason2018), and is the product of a combination of causes operating at multiple levels of the political system (Sides, Tesler, and Vavreck Reference Sides, Tesler and Vavreck2019). Here, we provide evidence on one factor that may have facilitated the threats to democracy: shifts in overall public opinion in the years preceding the 2020 election. Researchers have offered varied models of the interplay between political elites and public opinion, with some highlighting elite opinion leadership, while others emphasize the stability or thermostatic elements of public opinion (Bawn et al. Reference Bawn2012; Berinsky Reference Berinsky2009; Broockman and Skovron Reference Broockman and Skovron2018; Jacobs and Shapiro Reference Jacobs and Shapiro2000; Page and Shapiro Reference Page and Shapiro1992; Wlezien Reference Wlezien1995; Zaller Reference Zaller1992). Public opinion may also potentially serve as an early-warning system for anti-democratic drives. Here, we use a unique, population-based panel administered in fifteen waves between 2007 and 2020 to evaluate shifts in public attitudes that may have facilitated or foreshadowed Trump's presidency and its threats to democracy.Footnote 1
Specifically, we consider several types of attitudes that could plausibly contribute to democratic backsliding. The first are attitudes toward the fairness of elections and perceptions of the US system of government generally (Daniller and Mutz Reference Daniller and Mutz2019; Graham and Svolik Reference Graham and Svolik2020; Grossman et al. Reference Grossman2022). The second are attitudes toward Trump and other prominent politicians, which may have granted Trump unusual license (see Corrales Reference Corrales2018). The third are citizens' explanations for their vote choices, which allow us to identify if Trump's coalition is grounded in oppositional or anti-systemic views. In addition, we track negative sentiment by evaluating open-ended perceptions of the political parties. The Online Appendix reports other related outcomes, including vote choice, turnout, and perceptions of ideological extremity. Given the historical connections between racial hierarchy and anti-democratic movements in the US (Bedolla Reference Bedolla2021; Mickey Reference Mickey2015; Miller Reference Miller2021; Parker and Towler Reference Parker and Towler2019), the Online Appendix also considers white Americans' racial prejudices (see also Enns and Jardina Reference Enns and Jardina2021). To be sure, public opinion is inextricably connected with Trump's actions as president and his anti-democratic efforts in numerous ways: Trump became president only because American voters elected him in 2016, and support among Republicans helped him remain in office after impeachment and win renomination. Public opinion is also directly implicated in the events of January 6—in fact, in the pre-polling era, such public demonstrations often constituted public opinion (Lee Reference Lee2002). Still, our goal here is not a comprehensive assessment of the role of public opinion in producing the conditions that sustained Trump politically. Instead, it is to assess whether attitudinal trends among the public may have facilitated or anticipated the coming challenge to democracy. Such tests will also enable us to provide a limited assessment of Trump's capacity to influence public opinion in the pre-2020 election period.
The portrait that emerges is of aggregate stability—a stability that contrasts with the convulsions and anti-democratic shifts in elite-level US politics over this period. Certainly, our analyses uncover several attitudes that are worrisome in their prevalence, including dissatisfaction with the political system and strikingly negative views of the opposing party (see also Iyengar et al. Reference Iyengar2019). However, overall, our analyses do not identify much evidence of shifts in the years before October 2020 that made the ground more fertile for anti-democratic efforts. This holds true among both the population overall and Republicans. One exception is improving perceptions of Trump among his supporters between 2016 and 2020, but even this improvement is not atypical during this period.
In short, these data provide little evidence that attitudes shifted so as to facilitate the anti-democratic elements of Trump's presidency. What is more, as of October 2020, Trump's presidency had not led to demonstrable opinion shifts in his favor (see Page and Shapiro Reference Page and Shapiro1992; Wlezien Reference Wlezien1995). It is certainly vital to study opinions after the 2020 election to assess how rhetoric and events affected these same attitudes. However, understanding the precursors of Trump's anti-democratic efforts also requires increased attention to the opinions and behaviors of much smaller groups of activists and elites (Blum Reference Blum2020; Parker and Barreto Reference Parker and Barreto2014). It is those groups whose actions are likely to have immediate anti-democratic impacts, even when they are out of sync with public opinion.
Data
This article relies on what is, to our knowledge, the longest panel of a population-based sample of Americans covering Trump's 2017–21 presidency. Knowledge NetworksFootnote 2 recruited panelists offline via address-based sampling or random-digit dialing prior to the five waves administered online in 2007–08 under the auspices of the National Annenberg Election Survey (NAES); the Institute for the Study of Citizens and Politics panel has been separate from NAES since. The 2,606 respondents in Wave 6 (October 2012) were sampled from the respondents to Waves 1–5, with weights designed to produce marginal distributions that closely approximately the US population.Footnote 3 Waves 7, 8, 9, 11, and 12 sampled only respondents to the prior wave, while waves 10, 13,Footnote 4 14, and 15 resampled Wave 7 respondents. Such a sampling strategy attempts to balance sample size against item missingness. The Online Appendix provides additional information about response rates, sampling, and survey administration, including the full list of waves, dates, and sample sizes in Table A1. Wave 15 in October 2020 constituted the final wave.
As Table 1 illustrates, the panelists' demographics generally match those of the US population of English-speaking adults as of 2008 quite well. With respect to racial demographics, the October 2020 respondents are 12 per cent black, 11 per cent Hispanic, and 71 per cent white. The average respondent has fourteen years of education, while 36 per cent have at least a four-year college degree. However, this sample is more politically engaged than the target population. For example, validated voter turnout in 2016 was 73 per cent, as compared to the benchmark estimate of 60 per cent (United States Elections Project, 2021). It should also be noted that because panelists had to be eighteen in the panel's first wave, the sample includes no one younger than thirty in 2020.Footnote 5
Demographics for respondents to Waves 7 (2012–13; n = 2,471) and 15 (October 2020; n = 884)
Notes: a Variables in years. b Variable in thousands of US dollars.
Election Fairness and System Legitimacy
A foundational element of electoral democracies is that the losing side accepts the legitimacy of the election (Collier and Levitsky Reference Collier and Levitsky1997; Levitsky and Ziblatt Reference Levitsky and Ziblatt2018). As a consequence, beliefs in the fairness of elections and the system's legitimacy are crucial elements of democratic politics (Graham and Svolik Reference Graham and Svolik2020; Grossman et al. Reference Grossman2022). Our panel includes three questions that assess related beliefs (see also Daniller and Mutz Reference Daniller and Mutz2019).
One question—asked in five waves—was about perceived election fairness. Specifically, respondents were asked, “Do you believe presidential elections in the United States are generally very fair, somewhat fair, neither fair nor unfair, somewhat unfair, or very unfair?” Figure 1 illustrates the over-time trends, both overall and broken out by partisanship as of 2007–08 for the 344 respondents who answered the question in at least 2007/08, 2016, and 2020. Overall ratings lean toward the fair side, with the average response never dipping below 3.40 on a 1–5 scale, which is between “somewhat fair” and “neither fair nor unfair.” Democrats actually rate elections as less fair than Republicans in 2008 (3.53) and 2016 (3.56), perhaps because Republicans had won the most recent presidential election in both surveys (Daniller and Mutz Reference Daniller and Mutz2019). Republicans do dip from 3.91 to 3.75 between November/December 2016 and October 2020, quite possibly as a response to Trump's repeated lies about election integrity. However, even in October 2020, beliefs about electoral fairness were nearly identical for Democrats and Republicans—and there is no evidence of crisis levels for either group. Certainly, there is no evidence of long-held Republican grievances on this score.
Trends in views about election fairness and political system legitimacy over time.
The panel also included two questions about the legitimacy of the political system. The first asked respondents whether they agreed or disagreed with the statement, “I would rather live under our system of government than any other that I can think of,” with responses coded 1 (strongly disagree) to 5 (strongly agree). The middle panel of Figure 1 illustrates the results over time (n = 438), showing that 2008 Republicans are actually systematically higher than Democrats, with an October 2020 mean (of 4.50) that is slightly higher than that of 2012 (of 4.45). To be sure, this item may tap American exceptionalism or identity. However, support for the American system is high and reasonably stable. In fact, it is the Democrats who show a drop under Trump, from 4.24 in 2016 to 3.98 (or almost exactly “somewhat agree”) in 2020.
Another item illustrated in the bottom of Figure 1 instead asks respondents to agree or disagree with the statement, “At present I feel very critical of our political system,” with responses again ranging from strongly agree (1) to strongly disagree (5). Via this measure, discontent is substantially higher, with the average response (of 2.28) being nearest to “somewhat agree.” The parties are quite similar in 2012, 2014, and 2016. Yet, here too, it is Republicans who are slightly less likely to agree—and so slightly less critical of the system—as of October 2020. From the survey evidence alone, it is not clear that the system had less perceived legitimacy among Republicans than among Democrats shortly before the 2020 election.
Candidate Affect
One prospect is that Trump may engender especially intense support among the subset of people who view him positively. Research on Latin America finds that anti-democratic efforts are often the product of popular leaders seeking to entrench power (Corrales Reference Corrales2018), making an assessment of Trump's popularity vital. The panel asked respondents to assess several political figures, including Trump, in multiple years on 0–100 feeling thermometers. Figure 2 illustrates the distributions of responses for Trump in January 2016 and October 2020 (top), and for both Obama and Trump in the October they stood for re-election (bottom), for respondents who answered both questions.Footnote 6 For Trump, one key story is negativity. The median rating of Trump was 30 in 2016 and 25 in 2020. Another story is the increased positive sentiment among pro-Trump respondents, as the 90th percentile rating of Trump was 90 in 2016 but 99 in 2020. After serving as president, Trump came to be rated more positively by a subset of voters.
Distribution of respondents on 0–100 feeling thermometers assessing Trump at different times (top) or Trump versus Obama while running for re-election (bottom).
Is the extent of strong pro-Trump sentiment unusual? The bottom panel of Figure 2 and Figure A1 in the Online Appendix help answer that question. Figure 2 illustrates assessments of Obama and Trump in the October prior to their re-election bids (2012 for Obama; 2020 for Trump). Opinions of Trump are both more negative on average (mean of 40.9 versus 50.6) and more polarized. Trump does have a slightly higher fraction of people rating him 95 or above (14.6 per cent versus 11.0 per cent), but the difference is not substantively large. However, as the distributions of ratings for Clinton and Romney (see Figure A1 in the Online Appendix) illustrate, neither enjoyed the same strongly positive assessments. Trump stands out relative to Clinton and Romney, but much less so relative to Obama.
In Figure A2 in the Online Appendix, we present another evaluation of the intensity of Trump support. There, we plot vote support by lagged partisanship four years prior. If the US public were more polarized in their vote choices in 2020 than in earlier years, this is one way to measure it. In 2012, Mitt Romney won 92 per cent of those who had identified as strong Republicans four years prior. In 2020, Trump won 90 per cent of the comparable group. This is noteworthy because Trump had emerged as a political figure in 2016, meaning that identifying as a strong Republican four years before was already an identification adopted in the shadow of Trump. Thus, there is little evidence that Trump had atypical levels of support among those whose partisanship in prior years meant they were most positively predisposed toward him.
Between 2008 and 2020, US federal elections had a range of outcomes, from decisive Republican victories (2010 and 2014) to decisive Democratic victories (2008, 2012, and 2018) and more divided outcomes (2016 and 2020). How do we square the stability reported in these time-series with that electoral volatility? Shifts in vote choice in this period have received extensive consideration elsewhere (Hill, Hopkins, and Huber Reference Hill, Hopkins and Huber2021; Sides, Tesler, and Vavreck Reference Sides, Tesler and Vavreck2019). Still, Figure A3 in the Online Appendix and the associated discussion also provide evidence of over-time stability, as a striking 88 per cent of our panelists did not cross party lines in 2012, 2016, or 2020.Footnote 7 Nonetheless, in a closely divided polity with winner-take-all institutions, even small electoral shifts can tip the balance of political power.
Open-Ended Explanations of Vote Choice
Closed-ended questions written years ago may not capture shifts in public opinion in response to recent events. Accordingly, we next analyze open-ended responses to a question asked in October 2020: “Let's say a friend asked you why you were supporting [CANDIDATE] in the presidential election. In one sentence, what would you say?”
Three coders manually annotated each of the 533 responses from the 2020 survey into various non-exclusive categories to identify the reasons people gave for their vote. We then averaged the annotations; the Online Appendix provides the full codebook. Figure 3 reports the fraction of October 2020 responses coded into common categories; Table A2 in the Online Appendix provides the corresponding fractions.
Distribution of hand-coded responses to a question about why respondents supported Trump versus Biden—2020 open-ended explanations (n = 533).
Here, too, there is little that indicates the impending threat to democracy. Biden supporters are much more likely to say that they were motivated by their opposition to Trump than the reverse (35 per cent versus 12 per cent), and the fraction of responses from Trump supporters containing disaffection is just 6 per cent. Indeed, the distribution of responses from Trump supporters looks like it might for any incumbent president, with backers highlighting his personal traits (32 per cent), policies (30 per cent), experience (11 per cent), and performance in office (29 per cent).Footnote 8 In Figure A5 in the Online Appendix, we plot two of the most indicative categories—opposing the alternative candidate and expressing political disaffection—by lagged political partisanship. These figures reinforce the fact that independents who leaned Democratic or pure independents backing Biden were the most likely to explain their vote as opposition to the other candidate. Disaffection is not especially concentrated among strong Republicans.
Negative affect toward the parties is another critical metric (Iyengar et al. Reference Iyengar2019): as negative affect rises, people may dismiss or dehumanize their political opposition (Jardina and Piston Reference Jardina and Piston2016), or else see their victory as unacceptable. While we do not have a measure of affect toward the parties, we did ask randomly chosen subsets of respondents in November/December 2016 and October 2020 to describe the political parties.Footnote 9
Overall levels of negativity are high, but they did not change markedly between those two waves. Respondents rated the Democrats negatively in 60 per cent of their November/December 2016 responses, but that came down to 52 per cent in October 2020. Ratings of the Republican Party were negative 54 per cent of the time in 2016, a figure that rose to 57 per cent in 2020. A total of 84 per cent of 2016 Republicans made negative comments about the Democratic Party, which dropped slightly to 81 per cent in October 2020. To be sure, such ratings reflect worrisome levels of negativity (Iyengar et al. Reference Iyengar2019). Such high levels of negativity could well facilitate denial of Biden's victory or foster related beliefs about the stakes of the 2020 election (Kalmoe and Mason Reference Kalmoe and Mason2019). However, they do not show evidence of major shifts in recent years. Whatever changed in recent years appears to have been principally at the elite and activist levels, not among the public. (We should add that other studies using standard measurement techniques have reported discernible increases in negative partisanship over this period. For example, Lelkes et al. [Reference Lelkes2022] report that negative partisanship outpaced positive partisanship for the first time in 2020, though it also shows that such increases are more pronounced in online versus face-to-face surveys.)
Discussion
Certainly, the attitudes probed here are not an exhaustive list. Students of democratic breakdowns in Europe in the 1930s often note the intense ideological polarization of the period, which raised the perceived stakes of elections (Weber Reference Weber1996; see also Mainwaring and Pérez-Liñán Reference Mainwaring and Pérez-Liñán2013). In Figure A7 in the Online Appendix, we use questions about perceived ideology to measure whether Donald Trump or Joe Biden was perceived as unusually extreme. Perceptions of Trump did grow more conservative between 2016 and 2020—but perceptions of Trump's ideology in 2020 looked little different than perceptions of Romney's in 2012. There is no evidence that perceptions of ideological extremity were markedly different in 2020 than in 2012. Here, too, the survey data alone provide little indication of the looming anti-democratic threat.
Historically, racial hierarchy has been closely connected with anti-democratic movements (Bedolla Reference Bedolla2021; Mickey Reference Mickey2015; Miller Reference Miller2021; Parker and Towler Reference Parker and Towler2019), so Figure A8 in the Online Appendix also reports trends in white Americans' anti-black prejudices. If anything, such prejudices were declining—or at least not increasing—in the Trump era (see also Enns and Jardina Reference Enns and Jardina2021; Hopkins and Washington Reference Hopkins and Washington2020).
Overall, the story that emerges from this population-based panel is one of stability in core aspects of public opinion that may have presaged Trump's anti-democratic turn. In the period before the 2020 presidential election, we do not detect heightened concerns about election fairness or system legitimacy, either overall or among Republicans. Views of Trump while running for re-election in 2020 were more polarized than views of Obama at the comparable moment of his presidency, but Trump claimed only a slightly larger fraction of very strong supporters than had Obama. Open-ended questions display significant negative partisanship but no dramatic increases in that measure and no signs of widespread disaffection among Republicans.
Explaining the sources of this stability is beyond the scope of a single article. Still, Tables A3–A7 in the Online Appendix begin this investigation by presenting regressions of this article's primary outcomes on basic demographic measures, such as socioeconomic status, racial self-identification, and religion, as well as partisan identity measured in October 2012. For some outcomes—especially partisanship-related measures, such as the Trump feeling thermometers and vote choice—the models have consistently strong predictive power. Many of the individual-level predictors are stable in sign and magnitude too. Such results suggest the stabilizing role of partisan identity and demographic factors during a time of consistent, polarized partisan cues (see, especially, Green, Palmquist, and Schickler Reference Green, Palmquist and Schickler2002; Levendusky Reference Levendusky2009).
Yet, for measures of election fairness and system legitimacy, the models are less predictive and the coefficients less stable. Here, it is helpful to consider evidence from Griffin and Quasem (Reference Griffin and Quasem2021), who report trends in Americans' perceptions of election fairness using the Nationscape survey. In late October 2020, Trump and Biden supporters were virtually identical in their belief that the presidential election would be conducted fairly, corroborating our panel. However, immediately after the election, beliefs diverged sharply based on partisanship, with confidence growing to 93 per cent among Biden supporters and dropping to just 29 per cent among Trump supporters. After prominent partisans sent salient, conflicting cues about election integrity, their co-partisans in the public followed suit, showing within-party opinion leadership (Zaller Reference Zaller1992). However, on key issues, Trump's rhetoric also generated a backlash among Democrats (Enns and Jardina Reference Enns and Jardina2021; Sides, Tesler, and Vavreck Reference Sides, Tesler and Vavreck2019). Stability can thus result from consistent partisan cues or from their absence (Page and Shapiro Reference Page and Shapiro1992).
Conclusion
In prior centuries, public opinion was tantamount to the views of the public as they were made known to government officials (Blumer Reference Blumer1948)—and by that older definition, the January 6 storming of the Capitol clearly represented an expression of public opinion. However, since the widespread adoption of polling, public opinion is more commonly considered to be the attitudes of the public as measured through surveys. By this definition, the relationship between Trump's attacks on democratic values and public opinion is not clear ex ante.
We used a population-based panel to analyze shifts in US public opinion in the years before and during Trump's tumultuous 2017–21 presidency. Were there movements in public opinion that may have laid the groundwork for Trump's anti-democratic rhetoric or his sustained challenge to the 2020 election's legitimacy? More generally, are certain attitudes leading indicators of anti-democratic threats? Analyzing various election-related attitudes, we find little evidence that 2020 stands out from prior years. While such attitudes as negative views of the opposing party remain highly concerning, our data do not detect a rise in the run-up to October 2020. Trump did enjoy strong support among a fraction of the electorate, but that support was not atypical. Obama in 2012 had only a slightly smaller share of highly positive supporters than Trump did in 2020. The choices those two presidents made, not the intensity of their supporters, is likely the key difference between the transitions in 2016 and 2020.
To be sure, our panel stops in October 2020 and so does not cover attitudinal shifts in the post-2020 election period. Crucially, it does not consider beliefs about the 2020 election itself. As detailed earlier, the stability in views of election fairness was shattered by Trump's post-election actions, with significant majorities of Trump supporters no longer reporting confidence in US elections (Griffin and Quasem Reference Griffin and Quasem2021).
This research also suggests the limits of studying mass opinion in understanding the dynamics that can produce anti-democratic efforts. Researchers would do well to complement population-based studies with focused work on the activists and elites who act in the shadows of public opinion (see, for example, Blum Reference Blum2020; Parker and Barreto Reference Parker and Barreto2014). To borrow a metaphor from Bartels (Reference Bartels2022), public opinion reflects not a wave, but rather a reservoir that political entrepreneurs can tap into. The storming of the Capitol is a powerful reminder that small minorities can nonetheless have a powerful impact on the stability of US democracy—and that shifts among small groups can be invisible when surveying the population as a whole. Small groups of activists may also have an outsized impact on how Americans perceive polarization, heightening the belief that the public is more polarized than it actually is (Ahler Reference Ahler2014; Levendusky and Malhotra Reference Levendusky and Malhotra2016).
Supplementary Material
Online appendices are available at: https://doi.org/10.1017/S0007123422000254.
Data Availability Statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/N6MS1N.
Acknowledgments
Members of the University of Pennsylvania American Politics Working Group provided incisive feedback. Special thanks to Francois Barrilleaux, Benjamin Dorph, Nick Fernandez, Tori Gorton, and Janelle Schneider for excellent research assistance, and to Nic Dias for assistance with analysis.
Financial Support
This research was funded partly thanks to grants from the Russell Sage Foundation (Awards 94-17-01 and 94-18-07 to Hopkins and Will Hobbs, and 87-11-01 to Seth Goldman and Diana Mutz).
Competing Interests
The author declares no competing interests.
Ethical Standards
This research was approved by the University of Pennsylvania's Institutional Review Board (824036).
Open access
