Donald Trump and the Lie

The legitimacy of democratically elected governments rests in part on widespread acceptance of the outcome of elections, especially among those who lost. This “losers’ consent” allows the winners to govern, and when the incumbent is the losing party, it allows for a peaceful transition of power. What happens in a democratic system when one side not only refuses to concede but also actively perpetuates lies about the outcome? This article studies the evolution of public opinion about Donald Trump’s “big lie” using a rolling cross-sectional daily tracking survey, yielding 40 days of polls and more than 20,000 responses from US voters from October 27, 2020, through January 29, 2021. We find that the lie is pervasive and sticky: the number of Republicans and independents saying that they believe the election was fraudulent is substantial, and this proportion did not change appreciably over time or shift after important political developments. Belief in the lie may have buoyed some of Trump supporters’ self-esteem. In reaction to the lie and the threat it brought to the transition of power, there was a significant rise in support for violent political activism among Democrats, which only waned after efforts to overturn the election clearly failed. Even if these findings merely reflect partisan cheerleading, we nonetheless find significant and potentially long-term consequences of the lie. A conjoint experiment shows that Republican voters reward politicians who perpetuate the lie, giving Republican candidates an incentive to continue to do so in the next electoral cycle. These findings raise concerns about the fragility of American democracy.

B reaking with long-held tradition, former President Donald J. Trump refused to accept that he had lost the 2020 presidential election to Joe Biden. Even after the election results showed a clear victory for Joe Biden and courts rejected more than 60 lawsuits filed by his campaign, Trump continued to repeat debunked conspiracy theories claiming that the election had been stolen from him. In a shocking turn of events, a mob of Trump's supporters stormed the Capitol while Congress members met to officially certify the election for Biden. Their rampage interrupted the proceedings and ended with the deaths of a police officer and several rioters. Although a transfer of power to President Biden eventually happened as constitutionally prescribed, it did so under heavy guard from soldiers. Not long ago, this set of events happening in the United States would have been unthinkable (Almond and Verba 1963), and indeed experts in American politics deemed the events of the 2020 election to be both significant and abnormal in the context of the country's political history (Bright Line Watch 2021). 1 The legitimacy of democratically elected governments rests in part on widespread acceptance of the outcome of elections, especially among those who lost (Anderson et al. 2005). Evidence from "consolidating" democracies shows that when politicians and their supporters refuse to accept defeat, it decreases support for the political system and increases the likelihood of attempts to overthrow the government through violent means (Przeworski 1991;2005). Consequently, the willingness of incumbent politicians to accept defeat is a crucial test for democratic stability.
Of course, few like being on the losing side of an election, and some voters judge the integrity of an election based on whether their party's candidate won (Baron and Hershey 1988;Cantu and Garcıa-Ponce 2015;Sances and Stewart 2015). In countries with weak commitments to democracy, incumbents commonly make dubious charges of fraud when they do not win (Schedler 2001), and their supporters often believe them (Cantu and Garcıa-Ponce 2015). Such claims are uncommon in consolidated democracies, making Donald Trump's blatant lies all the more noteworthy and abnormal (Anderson et al. 2005). 2 The goal of this article is to study how lies shape voters' perceptions about election integrity, support for violence, and ultimately, democratic stability. Most of our knowledge about lies and politics comes from the study of authoritarian countries. In these systems, citizens are obligated to live "within the lie" and behave "as if" they believe obvious falsehoods propagated by their governments (Havel 2018;Wedeen 2015). Many come to internalize regime narratives, and it is difficult to discern who really believes what (Kuran 1991;Shen and Truex 2021). Dictators create an atmosphere in which they are continually portrayed as powerful, infallible, and the only viable option (Huang 2015;Magaloni 2006;Wedeen 2015). To do so, authoritarian regimes sometimes propagate so-called big lies: those that are so grand that it is difficult to believe that someone would have the gall to make them up.
Before the 2020 election, it was unclear whether such an approach would be effective in a long-established democratic system like the United States. On the one hand, the presence of robust political competition and a free press works to limit the ability of politicians to manipulate public opinion (Chong and Druckman 2007;Zaller 1992). A cornerstone assumption in American free speech jurisprudence is that in a "marketplace of ideas," truth wins out over falsehood (Brazeal 2011). On the other hand, as partisan polarization transmogrifies into partisan sectarianism in the United States, it provides fertile ground for lies pedaled by politicians to take root and go unchecked by their partisans (Finkel et al. 2020;Druckman, Peterson, and Slothuus 2013;Van Bavel and Pereira 2018). Social media allow "fake news" to spread rapidly through partisan networks and "echo chambers" (Del Vicario et al. 2016;Guess and Lyons 2020), and lies about politics are more likely to spread than the truth (Vosoughi, Roy, and Aral 2018).
The 2020 presidential election offers an unparalleled opportunity to study whether a big lie spread by mainstream political actors can shape public perceptions in an established democracy. To shed light on the dynamics of public opinion before and after the 2020 presidential election, we created the Election Legitimacy Tracking Survey (ELTS), a nationally descriptive online survey implemented by an established survey research firm, Qualtrics, between October 27, 2020, and January 29, 2021 (n = 20,000). Our rolling cross-sectional design involved collecting daily surveys of 500 respondents for a month around the election and then biweekly surveys until nine days after President Biden was inaugurated, allowing us to see how public opinion about the election unfolded in real time. The project is unusual in that we did not know what the primary "treatment" (the election outcome) would be in advance, only that the election outcome would likely be contested. The historical importance of the election was evident well before the election, and other research teams were conducting similar survey research before and after the election independently from us (Bright Line Watch 2020;Clayton et al. 2021;Drutman 2021;Pennycook and Rand 2021). The ELTS not only complements these surveys but also provides a richer dataset in key respects.
In this article, we further develop the concept of the big lie by integrating research on cults of personality in authoritarian contexts with research on misinformation in democratic contexts. We then describe the ELTS in greater detail and show how key public opinion indicators unfolded between Election Day and a week after the Inauguration. Five key findings emerge from these data. First, and least surprisingly, we corroborate other research teams and news reports in finding that Republicans were much more likely to say that they believed the lie that Donald Trump won the election. Roughly one in four Americans say that they do not believe the election result was legitimate or identify Joe Biden as the winner. For Republicans, these proportions hover around 50%. Voters who are older, less educated, and categorize themselves as having lower social status are less likely to perceive the Biden win as legitimate. Second, even though many respondents said they would update their opinion in the direction of the truth if particular events happened (e.g., Biden is inaugurated), belief in the lie did not shift at the aggregate level when these events occurred. Third, Republican voters largely continued to link their identity to Trump as the loss became apparent and displayed net increases in self-esteem, suggesting that believing the lie helped "cut off reflected failure" (Hirt et al. 1992; see also Boen et al. 2002). Fourth, as President Trump's attempts to overturn the election became more aggressive in early December, more Democrats expressed a willingness to engage in radical violent action aimed at the state. This willingness abated after Trump's attempts to overturn the election failed, but support for radical political action remained higher among all voters than the preelection baseline. Fifth, a conjoint experiment shows that Republican voters will reward politicians who perpetuate the lie, giving Republican candidates an incentive to do so.
Our findings have implications for other established democracies. Many democracies are experiencing an increase in polarization and the loosening of the universal commitment to democratic norms (Przeworski 2019), stimulating studies of the conditions that foster "democratic backsliding" (Bermeo 2016;Carey et al. 2019;Gandhi 2019;Graham and Svolik 2020;Levitsky and Ziblatt 2018;Svolik 2019;Waldner and Lust 2018 (Svolik 2019). A key feature of those transitions is that they were enabled by other elites and supported by partisans willing to trade off democracy for ideological aims (Carey et al. 2019;Gandhi 2019;Graham and Svolik, 2020;Svolik 2019).

Perspectives on Politics
We observe many of these dynamics in the contemporary Republican Party, both at the elite and mass levels, and show how a big lie can be used strategically by a charismatic leader to capture a party and push it in an authoritarian direction. These findings offer additional support for the position that culture alone cannot sustain democratic norms (Dahl 1989;Przeworski 2005). Even established democracies are fragile, and the strategic decisions made by elites have potentially dire consequences.

The Politics of Lies
We consider big lies to be a form of disinformation, which is commonly defined as "false information that is purposely spread to deceive people" (Guess and Lyons 2020;Lazer et al. 2018). The term has its origins in Nazi Germany; Adolph Hitler infamously describes the power of lies in Mein Kampf. 3 In comparison to run-of-the-mill disinformation, the big lie is grander in scope-the audacity of the mistruth is what is thought to make it powerful. The lie itself also originates from those in power, usually in their effort to stay in power or preserve a political advantage. Finally, a big lie usually contains elements of conspiracy, centering around the idea that a hidden group of powerful people exerts some nefarious influence on society behind the scenes (Guess and Lyons 2020;Sunstein and Vermeule 2009).
In authoritarian systems, lies, along with propaganda and political indoctrination, demonstrate the dominance of the regime (Havel 2018;Huang 2015;Magaloni 2006;Wedeen 2015). Wedeen (2015) describes how citizens in Syria were forced to publicly display loyalty to Hafiz al-Assad, repeating statements everyone knew to be patently false: Assad knew "all things about all issues" and was even the country's "premier pharmacist." As Huang (2015) shows, exposure to this sort of messaging does not necessarily increase support of the regime, but it does dissuade dissent and send a strong signal that the government can maintain order.
Political systems built on lies and domination may foster preference falsification-a disconnect between what people say they believe and what they privately believeand inflate the observed public support for the regime as a result (Kuran 1991;Shen and Truex 2021). Big lies told by elites can thus engender millions of smaller lies at the individual level, as citizens are forced to present themselves a certain way out of fear. In such settings, a central feature of political life is that, although people publicly acquiesce to a lie, no one really knows who believes what (Havel 2018;Wedeen 2015).
Research on disinformation shows us that people can readily accept mistruths when they are consistent with their preferences (Douglas, Sutton, and Cichocka 2017;Duran, Nicholson, and Dale 2017;Kim and Kim 2019;Murphy et al. 2019;Nyhan and Reifler 2010;Van Prooijen and Jostmann 2013). Whether partisans believe congenial misinformation, they are willing to spread it to disparage opposing partisans (Osmundsen et al. 2021). Consequently, lies are more likely to spread "farther, faster, deeper, and more broadly than truth" (Vosoughi, Roy, andAral 2018, 1147). Moreover, people are especially susceptible to accept misinformation that highlights threats (Blaine and Boyer 2017) and are more likely to believe and spread lies and rumors that paint outgroups as dangerous (Horowitz 2001).

Data
We administered the ELTS online through the Qualtrics survey platform; this established and respected online survey research firm recruits participants and verifies their names, addresses, and dates of birth before inviting them to join their sampling panels. It incentivizes participation in surveys by compensating respondents with money or money equivalents, such as Amazon credits, and it conducts validity checks of responses to produce a highquality sample. Importantly, the data quality and representativeness of Qualtrics samples have been independently verified by scholars to be in line with probability samples like the General Social Survey and the American National Election Studies (Boas, Christenson, and Glick, 2020). We restricted the population to US citizens who are registered to vote. 4 The project employs a rolling cross-sectional design. We received a sample of 500 new respondents every day beginning October 27, 2020, through November 20, 2020. After November 20 we collected samples on Tuesday and Friday of each week, with pauses for holidays. In total, we collected 20,000 responses over 40 daily samples through January 29, 2021. For each daily sample, we calculated poststratification weights to align the sample with known characteristics of the population. Our weighting scheme was implemented using entropy balancing and included information on gender, age, race, partisanship, education, and region (Hainmueller 2012). This process resulted in weighted daily samples that all had the same composition of these five core demographic characteristics and matched the composition of the US electorate.
The Supplementary Material contains more discussion of the sample recruitment process and how it compares to relevant population statistics. Figures SI1 and SI2 show stability in the sample composition over time. The sample matches the population of registered voters with respect to turnout, gender, and partisanship, but it skews younger and more educated than the population. Black voters are slightly overrepresented. Departures like these are common in survey research, and the poststratification weights result in daily samples that are tied to population proportion, weighting respondents from underrepresented groups slightly more heavily. Our key substantive findings are not sensitive to this weighting decision.
The core questionnaire was kept largely the same throughout the project. Respondents first answered standard demographic questions and provided information on their partisan affiliations and voting history. The next module included questions on the legitimacy of the election. The remainder of the survey included several standard question batteries to measure support for political violence (Moskalenko and McCauley 2009), the need for chaos (Petersen, Osmundsen, and Arceneaux 2018), anxiety and depression (Zigmond and Snaith 1983), selfesteem (Schmitt and Allik 2005), and support for democratic norms (Inglehart 2003). These batteries were presented in random order; the question order was also randomized within each battery. The final module was a short Word Association Test that asked respondents the first words that come to mind for several cues, including Donald Trump and Joe Biden. The wordings for key questions used in this article are included in the Supplementary Material.

Results
Perceptions of Legitimacy Figure 1 shows perceptions of the election outcome over time. The top panel shows the proportion of registered voters who identified Biden as the winner of the 2020 presidential election, and the bottom panel shows the proportion who viewed the election as legitimate. 5 These proportions stayed relatively fixed over the three months following the election. Only three out of four registered American voters believe Joe Biden legitimately won the election. For reference, we also note the major events that occurred throughout the postelection period. It is noteworthy how little perceptions moved in response to significant political events like the Electoral College vote, the Capitol insurrection, or Biden's inauguration. Americans' perceptions of the election outcome were largely fixed by mid-November.
Consistent with contemporaneous survey research (Bright Line Watch, 2020; Drutman, 2021; Pennycook and Rand, 2021), our data show a sharp partisan divergence in perceptions of the election. Figure 2 presents the same outcomes broken out by party identification. Shortly after the initial election results came in on the evening of November 3, most Democrats identified Joe Biden as the winner and perceived the outcome as legitimate. This proportion rose to close to 100% after the election results were called by most media outlets on November 7. Republican voters increasingly identified Biden as the winner as the results came in, but this proportion plateaued at around 40% and remained relatively stable thereafter. Likewise, half of Republican voters consistently said they did not view Biden as the legitimate winner. Voters who identify as independent or members of other parties increasingly accepted the legitimacy of a Biden win as the results came in from different states, but again, this proportion plateaued. Nine days after Joe Biden was sworn in as president, roughly 25% of unaffiliated voters did not view the election as legitimate.
Although there has been a "winners-losers gap" in confidence about the vote count in US elections at least since 2000, the majority of voters on the losing side in previous elections were "very confident" that the vote count was accurate (Alvarez, Hall, and Llewellyn 2008;Sances and Stewart 2015;Sinclair, Smith, and Tucker 2018). The 2000 election offers another benchmark because its outcome was reasonably in doubt. George W. Bush lost the popular vote by a half-million votes but won the Electoral College vote by a thin margin in the state of Florida. Al Gore requested a recount of the ballots in Florida, and the process dragged on in the weeks following the election. In a controversial decision, the Supreme Court stopped the Florida recount as the meeting of the Electoral College loomed, effectively awarding the presidency to George W. Bush. Despite the acrimony over the decision, Al Gore conceded the election. Even though there were objective reasons to doubt the outcome, only 18% of Americans said they believed Bush stole the election (Carroll 2001), and 61% of Americans who voted for Gore said they viewed Bush as the legitimate president (Moore 2000). In contrast, fewer voters in the aftermath of the 2020 election saw the outcome as legitimate, even though Joe Biden defeated Donald Trump by a comfortable margin in both the Electoral College and the popular vote. These findings are consistent with preelection research showing that exposure to Trump's rigged election rhetoric lowered Republicans' confidence in the election (Clayton et al. 2021) and support the notion that, by repeating the big lie about a stolen election, Donald Trump undermined the legitimacy of Biden's presidency among a substantial portion of Americans What Would Change the Minds of Believers in the Big Lie? From November 16 to December 15, we asked respondents who identified Trump as the winner a "yes" or "no" question of this form: "Would you believe that Joe Biden won the election if…", followed by a hypothetical political event. Overall, 1,245 respondents, more than threequarters of them Republican (76%), saw this question over the 11 waves of the survey in which it was included. The results are presented in table 1.
The data indicate a general reluctance to shift perceptions of the outcome, even with new political developments. Of the voters who denied the outcome, 28.7% said they would believe Biden won if Republican leaders like Mitch McConnell were to say that Biden had won more votes. About 31.0% would believe Biden won if the Electoral College were to award him a majority of votes, 866 Perspectives on Politics and 42.9% would believe Biden won if there were a Supreme Court decision to that effect. About 45.2% of people who identified Trump as the winner would believe Biden won if Trump were to concede. About 33% of respondents who did not view the Biden win as legitimate did not identify any event that would make them think he actually won.
These findings echo a similar approach used by Pennycook and Rand (2021), who also found that a minority of voters who believed the big lie said they would accept  Notes: The top panel shows the proportion of respondents who answered "Joe Biden" to the question, "Who do you think won the 2020 presidential election?" The bottom panel shows the proportion who responded "Yes" to the question, "Do you accept the election results as legitimate?" Starting on November 8, the legitimacy question was preceded by the sentence, "Major news networks have announced that Joe Biden is the winner of the 2020 presidential election." Letters mark significant political events: D =

Perspectives on Politics
Low levels of political interest could explain this lack of updating: voters might not have been paying attention to the Supreme Court decision on December 11 to reject the Texas lawsuit, or the Electoral College vote a few days later, or McConnell's subsequent endorsement of Biden's win. But even the most salient events like Biden's inauguration did not increase acceptance of the outcome. Moreover, Pennycook and Rand (2021) found that Trump voters who were the most knowledgeable about politics were more likely to believe the election was stolen. Our intuition is that, by the end of the election, the reluctance to acknowledge the Biden win among Republicans was likely driven more by motivated reasoning, selective exposure to slanted news coverage, and perhaps a degree of "expressive responding" than by low political awareness (Osmundsen et al. 2021;Schaffner and Luks 2018).
Who Is Most Likely to Believe the Big Lie? Figure 3 explores who is most likely to reject the election result. It presents the results of a simple linear probability model, regressing the binary legitimate variable on demographic covariates of interest. The left panel presents results for Republican voters, and the right panel presents results for voters who identified no partisan affiliation.
We observe that voters who are older, less educated, and categorize themselves as having lower social status are less likely to perceive the Biden win as legitimate. These relationships hold for both Republican and Independent voters, although they are more pronounced among Republican respondents. Independent voters are also much more likely to accept the election outcome. These findings correspond with recent research showing that less educated voters (Abou-Chadi and Hix 2021) and those who feel that they have lower social status (Kriesi and Schulte-Cloos 2020;Mutz, 2018) are more likely to support right-wing populist politicians. Our findings also shed some light on why contemporaneous research shows that belief in the lie is associated with right-wing attitudes (Drutman, 2021).

Support for Radical and Violent Political Action against the State
As polarization increases in the United States, so does acceptance of political violence (Kalmoe and Mason 2021). This is nothing new. Partisan political violence is tightly woven into US history from the American Revolution through the Civil War and Jim Crow eras (Kalmoe 2020), and the 2020 election may have caused many Republicans to justify using violence to prevent Joe Biden from being elected (Bright Line Watch 2020; Mason and Kalmoe 2021; Pennycook and Rand 2021). Building on this work, we explore whether the perpetuation of the lie potentially mobilized support for violent political action.
The ELTS core questionnaire included the Radicalism Intention Scale (RIS), which assesses a respondent's readiness to participate in violent or illegal political action against the legal authorities to achieve their political goals (Moskalenko and McCauley 2009). The RIS is related to, but distinct from, measures of partisan violence, such as Kalmoe and Mason's (2021) measure. The RIS specifically addresses the willingness to use violence against authorities to fight for one's political group, whereas measures of partisan violence focus on the use of violence against opposing partisans. The RIS items measure willingness to participate in a violent protest, attack police forces, encourage others to participate in illegal protests, and go to war on behalf of one's social group, among other behaviors (see the Supplementary Material for exact question wordings). Disputes over election legitimacy in consolidating democracies sometimes lead partisans of the losing side to attempt to overthrow the government through violent means (Przeworski 1991;2005), and we included the RIS on the survey because it focuses on one's willingness to target state agents. Respondents were asked their level of agreement on a scale of 1-5, and their answers were

28.7
The Supreme Court rules that Joe Biden won more votes than Donald Trump. 42.9 The Electoral College awards a majority of votes to Joe Biden.
30.9 The U.S. Congress awards a majority of votes to Joe Biden. 26.6 Donald Trump concedes the election to Joe Biden.
45.2 Joe Biden is sworn in as president by the chief justice of the Supreme Court on January 20, 2021. 37.5 n = 1,245 Note: The table displays the responses to a question that was shown to respondents who did not identify Joe Biden as the winner of the election even after the race had been called. The question was only included in the survey from November 16 to December 15, 2020. All data drawn from Election Legitimacy Tracking Survey (ELTS).
averaged over the five questions in the battery. Higher scores indicate greater agreement and willingness to participate in violence (mean = 2.67, SD = 1.21). Figure 4 shows the evolution of support for radical and violent political action over time, again disaggregating the data by partisanship. 7 We observe higher levels of support for radical action among partisans and a noticeable increase in support for radical action against the state among Democrats in the immediate aftermath of the election. This Democratic support peaked just before the Electoral College met on December 14, perhaps in response to Trump's public efforts to pressure local officials and legislators to dismiss votes from key states. Trump's perpetuation of the lie, coupled with his attempts to overturn the election result, may have pushed Democratic voters into a more radical mindset. As table 2 shows, this mindset changed substantially after the election result was secured by the Electoral College vote, the events of January 6, and Democrats regained control of the White House and the Senate; yet support for violence among Democrats remained significantly higher than preelection levels even after Biden was inaugurated. Figure 5 explores precisely which members of the US population believe radical political action is most justified. The relationships that emerge are the same for Democrats, Republicans, and unaffiliated voters. We observe that voters who are younger, male, more educated, and of higher social status are more likely to say that radical action and violence are justified to defend one's group. Given that the RIS taps a willingness to engage in state-targeted violent action, it makes sense that younger men would be more likely to contemplate such behavior.

Outcome: Election Result Legitimate
Notes: The figure shows the coefficient estimates from a linear probability model where the binary legitimate variable was regressed on demographic covariates of interest. Starting on November 8, the legitimacy question was preceded by the sentence, "Major news networks have announced that Joe Biden is the winner of the 2020 presidential election," and the estimates in this figure reflect data collected after that date. Line segments represent 95% confidence intervals. All data drawn from Electoral Legitimacy Tracking Survey (ELTS).
870 Perspectives on Politics   Psychological Motivations for Believing the Big Lie In the "Three (Football) Field Study," Cialdini et al. (1976) show that students had a greater tendency to wear their school's apparel after a football team victory than a defeat and were more likely to use the word "we" when describing a win than a loss. In a related set of experiments, Hirt et al. (1992) show that, after observing their team win, fans have higher levels of self-esteem and enhanced performance on certain cognitive tasks. This behavior has been labeled "bask(ing) in reflected glory" (BIRGing). The obverse tendency is known as CORFing-cutting off reflected failure. People dissociate themselves with perceived losers to manage their own identities and self-concepts (Boen et al. 2002;Hirt et al. 1992). Partisan ties in American politics operate through similar identity-based mechanisms (Green, Palmquist, and Schickler 2002;Iyengar, Sood, and Lelkes 2012;Lyons et al. 2021), and some researchers theorize that the "winners-losers" gap in satisfaction after elections may reflect losing voters' attempts to cope with a hit to their social identity (Sances and Stewart 2015).
To study the extent to which the 2020 election shaped social identity maintenance, we included measures of selfesteem and a novel measure of how much survey respondents' identities are connected to their support for Donald Trump. The top panel of figure 6 shows an index of identification with Trump. 8 We see some degree of deidentification with Trump among Republicans over the course of the election, but the effect size is relatively modest, only -0.17 points on a 5-point scale (about onefifth of a standard deviation). This obscures some important temporal variation, including the noticeable drop in identification (-0.28 points) with Trump just after the Capitol insurrection and then a rebound in identification with him shortly thereafter. But by the end of the election, the mean level of the index among Republican voters remained 3.62 out of 5.
The ELTS questionnaire included the 10-question battery for the Rosenberg (1965) Self-Esteem Scale. Curiously, we observe sizable increases in self-esteem among Republicans as a group over the course of the election, as shown in the bottom panel of figure 7. There is a strong upward trend in the self-esteem score among Republicans 872 Perspectives on Politics in the weeks following the Capitol insurrection and Biden's inauguration.
Our interpretation of these patterns is that the big lie opened up two alternative paths to self-concept preservation among Republicans, given the identity threat posed by the election loss (Nyhan and Reifler 2019). Voters could de-identify with Trump, "cutting off his reflected failure" by rendering him less central to their identities. Alternatively, they could accept the big lie and simply deny the election loss altogether, rejecting the failure and preserving their self-concept (Boen et al. 2002;Hirt et al. 1992;Lyons et al. 2021). From the data it appears that more than half of Republican voters are choosing that second path, which was made popular and socially  acceptable by Trump himself. Indeed, for Republican voters who deny that Biden won the election, identification with Trump actually increased (0.155 points) after the election was called. For voters who accepted that he lost, identification went down after the race was called relative to the preelection baseline (-0.566 points).

The Consequences of the Big Lie
It is possible that these results reflect some degree of "expressive responding," wherein many of Trump's supporters merely say they believe the lie to signal their support (Schaffner and Luks 2018) or because of pressure to conform (Kuran 1991). Even so, the perpetuation of the lie could still shape how Republican elites behave. As long as elected Republican politicians perceive that they will be punished for not contesting the election results, they may go along with the big lie to preserve their career prospects.
To borrow Wedeen's (2015) language, they will behave "as if" they believe.
To assess the electoral consequences of the lie, we included a paired conjoint experiment on the ELTS questionnaire beginning on January 12 (Hainmueller, Hangartner, and Yamamoto 2015;Hainmueller, Hopkins, and Yamamoto 2014;Teele, Kalla, and Rosenbluth 2018). The conjoint questions asked Republican and Independent respondents to imagine a hypothetical Republican congressional primary race, presenting them with a choice of two Republican candidates (each respondent evaluated three pairs of Republican candidates). Respondents then viewed a table of two candidate profiles with randomly assigned attributes like profession, age, ethnicity, and gender, along with their position on the big lie. Candidates who bucked the lie stated that they would have voted to certify the election for Joe Biden, whereas those who supported it said that they would not have. Figure SI3 in the Supplementary Material shows how the profiles appeared in the survey.
After we implemented the conjoint experiment, we received criticisms of the design that are worth raising Notes: Coefficient estimates from a linear model where the candidate choice outcome was regressed on the randomly assigned candidate attributes in the conjoint experiment. The estimates in this figure reflect data collected after January 12. Line segments represent 95% confidence intervals, which reflect standard errors clustered at the respondent level. All data drawn from Electoral Legitimacy Tracking Survey (ELTS).
874 Perspectives on Politics here. First, the design is a bit heavy-handed in that the text about the candidate's stance on the election is the only policy stance. It stands out from the demographic and professional attributes on the survey and perhaps primes respondents to think about that dimension more than they otherwise would. Second, our conjoint experiment could have included a third option, whereby the candidate declines to say who won the election at all. This would have more closely mirrored the actual strategic environment, given that many Republicans employed that tactic at the time of the survey ("Where Republicans in Congress" 2020). We are unable to make these changes after the fact, and readers should interpret our findings here with these caveats in mind. For what it is worth, our intention was to simulate a primary election in which the big lie becomes a central issue-perhaps stoked by the former president himself-and we believe that our conjoint design accomplishes that goal. Because these attributes were randomly assigned, we can recover the average marginal component effect (AMCE) of the attribute on selection for political office using a simple linear regression, clustering the standard errors at the respondent level (Hainmueller, Hopkins, and Yamamoto 2014). For Republican respondents in the experiment, all else equal, a candidate who said Trump won is favored by 5.7 percentage points relative to a candidate who said Trump lost. This effect is in line with and is smaller than a similar conjoint experiment independently implemented by Bright Light Watch (2021) a week after ours was completed as well as one conducted by Noble and Carlson (N.d.) a year after the election.
Further, we find an asymmetry in the data after subsetting the analysis by Republican voters who believe the election is legitimate (figure 7; panel 2) and those who do not (panel 3). For those who say the election is legitimate, a candidate who admits Trump lost is favored by about 17.9 percentage points. For those who say the election is not legitimate, a candidate who insists Trump won is favored by about 25.3 percentage points. These findings show how the big lie could be a polarizing issue among Republican voters and how potential candidates must navigate substantial uncertainty, because they might not know the distribution of the two types in their districts (Broockman and Skovron 2018). Given this asymmetry, the safest course of action for most Republican candidates may be to avoid taking on this issue, even if they do not believe the lie. This is a form of self-censorship (Shen and Truex 2021).

Discussion
Our study systematically documents a new feature of American political life: roughly one-fourth of the country and half of the Republican Party rank and file say that they buy into the idea that the election was stolen from President Trump. Across the 40 days of our study, acceptance of this big lie was pervasive, sticky, and consequential. This echoes a dynamic typically observed in authoritarian regimes, in which a charismatic leader creates blatant falsehoods to justify his hold on power and requires everyone to behave as if those lies are true. Donald Trump eschewed the long democratic tradition of stepping aside and pledging support to his opponent once the results were no longer in doubt to any reasonable observer. Likely aided by the increasingly sectarian nature of partisan polarization (Finkel et al. 2020), as well as psychological needs to maintain a positive self-image, a sizable portion of his electoral base behaved as if they believed clear falsehoods. In turn, when it was unclear whether attempts to overturn the results of the election would work, many of those who did not support Trump expressed support for radical violence aimed at attacking the state apparatus.
It would be easy to dismiss these findings as partisan cheerleading. After all, the United States is very polarized along partisan lines, and people are not always truthful with pollsters. Nonetheless, we believe that our findings have important and potentially troubling implications for American democracy. Even if every single survey respondent who said that the election was stolen from Donald Trump knew that Joe Biden was the legitimate winner of the election, it still has the power to create the impression that "this is what people believe." To borrow from the authoritarian politics literature, people are behaving "as if" they believe and choosing to "live within the lie" (Havel 2018;Wedeen 2015).
The results from the conjoint experiment illustrate that whether Republicans really believe that the election was not stolen from Trump, many will reward Republican candidates who claim that it was. As a result, it makes it difficult for Republican leaders to take a stand against the big lie and requires them to at least pretend that the foundation of American democracy-its electoral apparatus-is corrupt and broken. The big lie around the 2020 election has allowed Trump to dominate other Republican elites, in a fashion not dissimilar from how leaders in authoritarian systems force their citizens to perpetuate lies (Wedeen 2015). Many Republican lawmakers at the federal and state levels have embraced the big lie (Rutenberg, Corasaniti, and Feuer, 2021), while most stay silent about what they believe ("Where Republicans in Congress" 2020). The handful of Republican leaders who have pushed forcefully against the lie have been punished, such as Liz Cheney who was stripped of her leadership position in the Republican Party (Sotomayor and Alemany 2021).
We find this possibility troubling for five reasons. First, it makes it difficult for Republicans in Congress to work with Democratic counterparts to fashion bipartisan legislation. Second, it provides a rationale for limiting voting rights and for enacting "reforms" that would make it easier to jettison ballots, in turn accelerating democratic backsliding in states controlled by Republican leaders who either believe or feel pressure to behave as if they believe the big lie (Grumbach 2021). Third, it sets a precedent that if one does not win an election, claiming fraud will not only go unpunished by the public but might even help galvanize one's side. Fourth, if claims of fraud become a regular feature of American elections in the future, they could stoke violence and undermine support for the democratic system (Albertson and Guiler 2020;Kingzette et al. 2021). Finally, we know that executive takeovers and other forms of democratic breakdown are often enabled and abetted by opportunistic elites (Gandhi 2019;Graham and Svolik 2020;Svolik 2019), and the dynamics surrounding the big lie give Republican lawmakers pressure to cater to antidemocratic forces within the party.
Democracy often dies with the consent of the people it empowers (Moehler and Lindberg 2009). If voters do not hold their own party's politicians accountable or, worse, egg them on to undo democratic processes to achieve partisan ends (Carey et al. 2020;Svolik 2019), it threatens the stability of the American democratic system.

Supplementary Materials
To view supplementary material for this article, please visit http://doi.org/10.1017/S1537592722000901.

Acknowledgment
Our gratitude goes to the University Center for Human Values and Data Driven Social Science Initiative at Princeton University for funding support. Additional financial support was provided by Temple University. We are grateful for helpful feedback from the editor, three anonymous reviewers, David Broockman, Nathan Kalmoe, Romain Lachat, Yphtach Lelkes, Tali Mendelberg, Markus Prior, and participants at the Séminaire Général de CEVIPOF at Sciences Po.
Notes 1 Although there have been contested presidential elections in the past, those instances were all caused by genuine uncertainty over the winner: the lack of a majority winner in the Electoral College in 1800 and 1824, confusion over the appropriate slate of electors from key states in 1876, and an extremely close electoral outcome that required a recount of votes in a decisive state in 2000. Yet in each of those instances, once the winner of the election was resolved, the losing candidate(s) conceded. One could quibble over whether Andrew Jackson conceded, given that he accused John Quincy Adams of striking a "corrupt bargain" to win the contingent election in the US House of Representatives. Even though Jackson did not believe Adams won in a fair way, he did not invent and spread outright falsehoods in an attempt to overturn the outcome of the election. 2 Although there is no widespread agreement on when a democratic system becomes "consolidated" (Schedler 2001), this concept is often invoked as an indicator of democratic stability. The "two-turnover test" presumes democracy to be established after two uncontroversial alternations in power among political parties (Moehler and Lindberg 2009). Under this definition, the United States was a consolidated democracy before the 2020 election. 3 Hitler writes, "In the big lie there is always a certain force of credibility; because the broad masses of a nation are always more easily corrupted in the deeper strata of their emotional nature than consciously or voluntarily; and thus in the primitive simplicity of their minds they more readily fall victims to the big lie than the small lie, since they themselves often tell small lies in little matters but would be ashamed to resort to large-scale falsehoods. It would never come into their heads to fabricate colossal untruths, and they would not believe that others could have the impudence to distort the truth so infamously." 4 To ensure some balance in the data, we instituted quotas on gender and partisanship in the collection process. 5 The two measures have a phi coefficient of .663, indicating a positive association of moderate strength. An alternative approach would be to collapse them into a single index; we refrain from doing so in the interest of transparency and simplicity in the analysis. Figures 1  and 2 are replicated in the Supplementary Material using a simple additive index. 6 Although our approach complements that of Pennycook and Rand's (2021), our survey offered more hypothetical options and continued to collect data to observe the veracity of the hypothetical predictions. 7 Figure SI7 in the Supplementary Material shows the aggregate results. 8 This index was adapted from Huddy, Mason, and Aarøe's (2015) measure of expressive partisan identity and was constructed by taking the mean agreement score on a 5-point scale on two questions: "When people praise Donald Trump, it makes me feel good," and "When people criticize Donald Trump, it feels like a personal insult."