Leading up to the 2020 presidential election, Democratic candidate Joe Biden seemed poised to become the 46th president of the United States. An incumbent president had not lost his bid for reelection in almost 30 years, but Donald Trump was embroiled in scandal (resulting in impeachment but not removal from office), had consistently low approval ratings, and presided over a drastically shrinking economy as a result of the COVID-19 global pandemic—which many argued had been grossly mismanaged by his administration (Woodward Reference Woodward2020). Despite winning the presidency in 2016, Trump alleged “millions of illegal votes” and other election fraud.Footnote 1 He continued using this rhetoric throughout his tenure as president and ultimately suggested that if he did not win reelection in 2020, it would be because it was stolen from him.Footnote 2
Although the race was too close to call on Election Night, all signs pointed to a Biden victory, and he was declared the winner by the end of the week. Trump refused to concede defeat, declaring that the election was marked by “fraud that has never been seen like this before.”Footnote 3 Despite having no evidence and losing dozens of court cases as a result, Trump continued this rhetoric, even causing tension within the Republican Party. The party’s attitude toward Trump and his tactics was well captured by an anonymous senior Republican official who said, “What is the downside for humoring him for this little bit of time? No one seriously thinks the results will change.”Footnote 4
This study seeks to answer the same question posed by this official: What is the harm of allowing unsubstantiated claims of election fraud to propagate? Specifically, we explore how allegations of election fraud influence the public’s attitudes toward the conduct of elections, election outcomes, representation, and democracy as a whole. If attitudes about electoral integrity are shaped mostly by evidence and verifiable reports about electoral malpractice, then these allegations should not affect the public’s perceptions. However, if attitudes about this question are more malleable, then depicting the electoral system as “rigged” in some way could negatively impact the public’s assessment about the state of democracy. Our results demonstrate that exposure to claims of election fraud (even without supporting evidence) indeed reduces respondents’ faith in elections and beliefs in democratic government. As Claassen (Reference Claassen2020, 118) concluded, “[p]ublic support does indeed help democracy survive.” Therefore, our study has important implications for maintaining a strong and stable democratic system of government.
Our results demonstrate that exposure to claims of election fraud (even without supporting evidence) indeed reduces respondents’ faith in elections and beliefs in democratic government.
THEORY AND HYPOTHESES
Previous studies demonstrated that attitudes about the state of democracy are influenced by numerous individual-level and institutional factors (Norris Reference Norris1999). Most relevant for our purposes, extant analyses suggest that concerns about electoral integrity have a negative effect on voters’ evaluations of their political system. According to Norris (Reference Norris2019, 7), “[f]ree and fair elections [and] meeting international standards of electoral integrity […] strengthen public assessment of democratic performance.” This claim is supported in several quantitative studies (e.g., Fortin-Rittberger, Harfst, and Dingler Reference Fortin-Rittberger, Harfst and Dingler2017).
As Edelson et al. (Reference Edelson, Alduncin, Krewson, Sieja and Uscinski2017, 93) pointed out, “[b]elief in election fraud is a common and predictable consequence of both underlying conspiratorial thinking and motivated partisan reasoning.” Furthermore, van der Linden (Reference van der Linder2015) and Jolley and Douglas (Reference Jolley and Douglas2014) concluded that exposure to conspiracy theories (e.g., the theory espoused by the former president) indeed can have measurable, negative social consequences. Given the previous discussion, we expected popular narratives around the election—namely, allegations of a “rigged” system—to shape opinion. As such, our first hypothesis was as follows:
H1: Exposure to claims of a fraudulent election will reduce positive attitudes toward elections and democracy.
However, we did not expect this effect to be uniform across the entire public. Instead, some respondents would be particularly receptive to the argument, others would not. Existing research demonstrates that partisan identities play increasingly important roles in US political discourse (Mason Reference Mason2018). This, in turn, facilitates the formation of political in- and out-groups based on party ID, and it causes people to rely on motivated reasoning to interpret the electoral success or failure of their preferred parties (Lodge and Taber Reference Lodge and Taber2013). Indeed, Edelson et al. (Reference Edelson, Alduncin, Krewson, Sieja and Uscinski2017) showed that those identifying with the losing party in the most recent election were more likely to accept related conspiracy theories as true. Furthermore, partisan identities also have been shown to influence which types of messages people internalize in their country’s political discourse. It is well known that voters “rely upon elite cues to place events in some political perspective” (Woessner Reference Woessner2005, 94). More specifically, people’s political perceptions are shaped primarily by cues from leaders of their own party (Uscinski, Klofstad, and Atkinson Reference Uscinski, Klofstad and Atkinson2016). As Bisgaard and Slothuus (Reference Bisgaard and Slothuss2018, 459) pointed out, “When citizens receive a cue from their party, they will tend to follow it and form perceptions […] in line with the party. In contrast, if the cue comes from the out-party, they will tend to reject or just ignore it.” Given these considerations, our second hypothesis was as follows:
H2: Exposure to claims of a fraudulent election will have a more (less) pronounced effect among the co-partisans (counter-partisans) of the person making the claim.
DATA AND METHODS
To examine the effect of Trump’s election fraud claim on political attitudes, we relied on original survey data collected on November 7, 2020 (i.e., four days after the 2020 presidential election). More specifically, we recruited 991 respondents for an experiment on Amazon Mechanical Turk (MTurk) (Williamson and Justwan Reference Williamson and Justwan2022).Footnote 5 MTurk is an online platform in which “requesters” can post small work assignments (e.g., completing a survey), which then are completed by “workers” in exchange for small monetary compensation. In the past decade, MTurk has been established as a heavily used data source for experimental social science research (for an overview, see Berinsky, Huber, and Lenz Reference Berinsky, Huber and Lenz2012).Footnote 6
Our statistical analysis was based on a survey experiment. Before answering any questions, each respondent read one of two randomly assigned news stories that were similar in length, layout, and sentence structure. Both stories were presented as being published on the Fox News website. The vignette for the treatment group was a slightly shortened version of a real Fox News story from August 2020.Footnote 7 In this article, Trump accused the Democratic Party of “stealing” the 2020 presidential election with its insistence on mail-in voting. More specifically, Trump declared that the upcoming election would be “the greatest scam” and “the most fraudulent election in history” because the Democrats “are trying to steal the election.” We opted for this particular item because it contained the key claims in Trump’s rhetorical campaign against the integrity of the 2020 election: (1) criticism of mail-in voting, and (2) concerns that the Democratic Party would use this voting method to subvert the election results. Half of our respondents read this news story; the other half was placed in a control group and read a “neutral” news story about a “mysterious stone slab.” This article had been used as a placebo in previous experimental political science scholarship (Albertson and Gadarian Reference Albertson and Gadarian2015).Footnote 8
Our choice to rely on a Fox News article was motivated by the following considerations. First, we wanted to make our experimental manipulation as realistic as possible. We chose a slightly modified but “real” story from a well-known news outlet, given existing research findings that suggest that a substantial proportion of American voters rely on major TV networks to obtain information about elections and political events.Footnote 9 Second, we relied on Fox News in particular because it is notable among mainstream news organizations for providing a consistent outlet for Trump’s election-fraud claims (Pennycook and Rand Reference Pennycook and Rand2021). Thus, although this research design has limitations (most significantly, it did not allow us to study the interaction between message content and source), it boosted the external validity of our findings.
We relied on five dependent variables to capture a broad range of attitudes toward democracy and political processes. First, we measured people’s assessments of the procedural fairness of the election process. To capture this variable, we asked those respondents who had voted in the 2020 presidential election (89.5%) how much they agreed with the statement that their “vote in the 2020 presidential election was counted fairly.” Answer options ranged from (1) strongly disagree to (5) strongly agree.
Next, our survey contained an item designed to tap into views about electoral responsiveness. More specifically, we asked respondents how much they believed that “having elections makes the government pay attention to what the people think.” Subjects could choose from three different answer options: (1) not much, (2) some, and (3) a great deal.
Additionally, we assessed people’s level of (external) political efficacy. We captured this variable with two separate indicators. For both items, respondents were asked to indicate their level of agreement or disagreement on a five-point scale from (1) strongly agree to (5) strongly disagree. Item 1 was “Public officials don’t care much what people like me think.” Item 2 was “People like me don’t have any say about what the government does.” Thus, higher values on both variables indicated higher levels of external political efficacy.
Finally, our survey measured people’s general support for democracy. We asked respondents how much they agreed with the statement that “Democracy may have problems but it is better than any other form of government.” Answer options ranged from (1) strongly disagree to (4) strongly agree.
We used several control variables commonly found in the literature on political attitudes. To begin, we accounted for an individual’s self-declared Party ID. This nominal variable had three different values: (1) Democrat (46%), (2) Independent (24%), and (3) Republican (the base category; 30%). In addition to party ID, we also controlled for a person’s Voting Preference during the 2020 presidential election. More specifically, we recorded whether a person voted for or—for those who did not cast a ballot—expressed a preference for Trump (the reference category; 34%), Biden (62%), or another political candidate (4%). Next, we considered the political ideology of the survey respondents. The questionnaire asked individuals to place themselves on a seven-point scale ranging from (1) extremely liberal to (7) extremely conservative (M=3.6, SD=1.9).
Furthermore, we added several correlates to capture respondents’ general information environment. News Consumption was measured by asking them how many days per week they typically watched, read, or listened to news (M=5.1, SD=2.0). Political Interest was operationalized by asking respondents to assess their interest in politics on a five-point scale from (1) very great to (5) not interested at all (M=2.4, SD=1.0). Relatedly, we captured their level of political sophistication by asking five factual questions about politics in the United States. Thus, the final Political Knowledge variable ranged from 0 (for respondents who did not answer any question correctly) to 5 (those who gave correct answers to all five items) (M=3.4, SD=1.3).
To tap into respondents’ socioeconomic status, we asked them to indicate their highest level of formal education, ranging from (1) less than high school diploma to (6) graduate degree (M=4.4, SD=1.2). Similarly, respondents also indicated one of 12 annual income brackets from (1) less than $10,000 to (12) more than $150,000 (M=6.4, SD=3.2).
Next, we introduced a binary control variable (Swing State) that captured whether a respondent lived in Arizona, Georgia, Michigan, Nevada, Pennsylvania, or Wisconsin. These six states were criticized most directly by Trump for undermining the outcome of the presidential election. As such, political attitudes in these states (i.e., 15% of respondents) may have differed from views held by respondents in other regional contexts. Finally, we accounted for age (i.e., 41% between 18 and 34; 42% between 35 and 54; 17% 55 or older); gender (53% male; 47% non-male); race (79% white; 21% nonwhite); and ethnicity (85% non-Hispanic; 15% Hispanic).
All survey questions used to measure the variables in this study are in the online appendix. Consistent with previous work relying on MTurk data, our respondent pool was noticeably younger and more liberal, educated, and politically engaged than the US population as a whole. It is notable that extant research on political misinformation (e.g., Grinberg et al. Reference Grinberg, Joseph, Friedland, Swire-Thompson and Lazer2019) suggests that these biases should have made it more difficult to find aggregate treatment effects in our experiment. Overall, we expected younger, more left-leaning, well-informed, and politically invested individuals to be better situated to dismiss Trump’s election interference claims. As such, the sample selection set up a more difficult empirical test for our hypotheses.
DATA ANALYSIS AND RESULTS
Our data analysis was based on a series of regressions. Because all dependent variables were ordinal, we relied on ordered logistic regression.Footnote 10 Confounder-adjusted treatment-effect sizes are summarized in figure 1. We begin with an examination of the effect of Trump’s election-fraud claims on perceptions of electoral fairness. Perhaps it is surprising that our model suggests that our experimental manipulation alone did not have an effect on people’s beliefs that their vote had been counted fairly. Whereas the coefficient for our binary-treatment indicator (i.e., coded as “1” for respondents who read the story about election fraud) points in the expected (negative) direction, it was not significant at conventional levels (p=0.11). It also is notable that this aggregate finding may mask substantial variation within various subsamples.
Figure 1 Confounder-Adjusted Treatment-Effect Sizes
Note: Confounder-adjusted treatment effect sizes are summarized in this figure.
For our second dependent variable, we found that individuals who read the Trump story were less likely to believe that elections make the government pay attention to what the people think than those respondents who were assigned to our placebo news story (p<0.01). Thus, in the aggregate, public claims about election fraud had a meaningful, negative effect on people’s perceptions about the responsiveness of their country’s political system.
Similar results emerged in our analyses of the two political-efficacy items. Subjects in the treatment group were less likely to disagree with the statement that “public officials don’t care much” about the thoughts of average people than those in the control group (p=0.02). Likewise, our news story also had a small (p=0.08) negative effect on whether subjects believed that people like them have any say about what the government does. Finally, subjects in the treatment group were significantly less likely to declare that democracy is the best form of government than respondents who read the placebo news story (p<0.01). Taken together, these findings demonstrate that publicly voiced criticisms of the electoral process in the United States have broad consequences, and they affect perceptions about the responsiveness of the political system and the desirability of “democracy” as a system of government.
Taken together, these findings demonstrate that publicly voiced criticisms of the electoral process in the United States have broad consequences, and they affect perceptions about the responsiveness of the political system and the desirability of “democracy” as a system of government.
To have a deeper understanding of the effect of our experimental treatment, we calculated substantive effect sizes. Holding all other independent variables at their observed values, we estimated how much our treatment influenced the probability that a given survey respondent fell above the midpoint on all relevant dependent variables (i.e., they perceived “high” electoral responsiveness, political efficacy, and support for democracy).Footnote 11 Results are shown in figure 2. According to panel A, respondents in the control group had a 49.6% probability of believing that elections make the government pay “a great deal” of attention to what the people think. By contrast, the corresponding value for individuals in the treatment group was 40.6%.
Figure 2 Effect of Treatment on Democratic Attitudes
Panel B provides estimates for the first political-efficacy item. The graph shows that respondents who received the placebo news story had a 29.2% probability of disagreeing somewhat or disagreeing strongly that “public officials don’t care much what people like me think.” By contrast, subjects who read the news story in which Trump accused Democrats of trying to “steal the election” had only a 23.5% probability of scoring highly on this variable. Similarly, our experimental treatment affected whether respondents believed that people like them have no say about what the government does (panel C). More specifically, respondents in the control group (with a predicted probability of 37.2%) were about 4.8% more likely to disagree with this statement than subjects in the treatment group (with a predicted probability of 32.4%).
Panel D shows the substantive effect for our final dependent variable. In the control group, respondents had an 89.9% predicted probability of agreeing that “democracy is the best form of government.” This high baseline number suggests that democratic norms were well established among most of the subjects in our dataset. After reading the Trump news story, however, the estimated margin decreased to 85.7%. Although this difference is small, these calculations demonstrate that existing claims about election fraud did have a negative effect on beliefs about democratic governance.
Given previous studies suggesting that people’s receptivity to elite cues differ according to their own political affiliations (Bisgaard and Slothuus Reference Bisgaard and Slothuss2018), we assessed to what extent our treatment effects varied among Republicans, Democrats, Independents, and respondents with different candidate preferences. To investigate these relationships, we first interacted the variable that captured people’s party IDs with our binary treatment indicator. We proceeded analogously with the nominal variable that represented respondents’ vote choice in November 2020.
It is interesting that our statistical analysis did not suggest that respondents differ in how much their scores on electoral responsiveness, political efficacy, and support for democracy were affected by experimental manipulation.Footnote 12 In other words, treatment effects for these dependent variables were similar, regardless of a respondent’s self-identified party or candidate preferences. We suspect this finding may be related to voter interpretations of our election-fraud treatment. Whereas some respondents likely experienced reduced levels of faith in elections, efficacy, and support for democracy because they internalized Trump’s message about election fraud, other individuals (primarily self-identified Democrats, Independents, and Biden voters) may have interpreted Trump’s rhetorical campaign itself as a strategy to undermine the election outcome. If this interpretation were correct, then it would not be surprising to see that all voters exhibited “negative” responses to our experimental treatment.
Next, our statistical models did provide evidence that Democrats and Republicans as well as Biden and Trump voters differed in how our treatment affected beliefs about electoral fairness. There were statistically significant interactions between our binary “treatment” indicator and the variables that captured party affiliation and vote choice. Figure 3 plots the estimated treatment effects for Democrats, Independents, and Republicans in panel A and for Trump, Biden, and Other voters in panel B. Consistent with the aggregate findings reported previously, the graph shows that our election-fraud story did not influence how much Democrats, Independents, Biden supporters, and Other voters agreed that their “vote in the 2020 presidential election was counted fairly.” However, self-identified Republicans and Trump supporters who were assigned to the treatment group were noticeably less likely to believe that their ballot was counted properly than their counterparts in the control group. Substantively, Republicans who read the placebo story had an 88.4% probability of agreeing with the previous statement. The corresponding value for GOP supporters in the treatment group was slightly lower (80.0%). Similarly, our experimental manipulation decreased the probability that Trump voters believed in a fair counting of votes by 8.9% (i.e., from 88.3% to 79.4%).
Figure 3 Treatment-Effect Differences
Note: Dependent Variable: Belief That Voting Was Fair.
To gain a deeper understanding of these relationships, we compared treatment-effect sizes for four different voter groups: Trump Republicans (21.7%), Non-Trump Republicans (7.6%), Trump Non-Republicans (8.4%), and Non-Trump Non-Republicans (62.3%).Footnote 13 More specifically, we assessed how much (if at all) our treatment decreased the proportion of individuals who agreed or strongly agreed that their vote in 2020 had been counted fairly. Results are summarized in figure 4. Our experimental manipulation had no effect on Non-Trump Republicans (Δ=-6.8%; p=0.53), non-Trump Non-Republicans (Δ=-1.1%; p=0.71), and Trump Non-Republicans (Δ=5.6%; p=0.55). However, for Trump Republicans (i.e., 94% of GOP voters in 2020Footnote 14 ), our treatment substantially reduced the proportion of subjects who believed in the integrity of the 2020 democratic exercise (Δ=-12.3%; p=0.03).
Figure 4 Effect of Treatment on Belief That Voting Was Fair by Nested Voter Group
What could account for the empirical patterns in figure 4? The most likely explanation for our findings is that Trump Republicans had a higher propensity to believe the election-fraud claims contained in the experimental treatment.Footnote 15 Stated differently, this voter group seems to have had particularly high levels of trust in the political statements of the former president. This interpretation is corroborated by existing survey research that suggests that those who relied on Trump as a source for election news were more likely to believe that too little attention had been given to allegations of voter fraud.Footnote 16
Next, we examined the role of people’s political ideologies and swing-state residency. Conservatives were less likely than liberals to agree that their vote had been counted fairly. At the same time, ideology did not affect (1) whether subjects agreed that elections make the government pay attention to what the people think, (2) respondent scores on the two efficacy measures, and (3) their support for democracy. Similarly, respondents who live in Arizona, Georgia, Michigan, Nevada, Pennsylvania, or Wisconsin did not differ on any of the dependent variables from subjects in other states.Footnote 17 In a final step, we interacted these variables with the binary treatment indicator. Taken as a whole, this procedure failed to uncover statistically significant results,Footnote 18 which suggests that our experiment had similar effects across people’s political ideologies and home states.
CONCLUSION
We initially asked, “What is the harm in allowing unsubstantiated claims of election fraud to propagate?” To answer this question, we posed two hypotheses; our results provide evidence in support of both. After being exposed to claims of election fraud, Republican Trump voters were more likely to doubt that their November 2020 vote was counted fairly than other respondents. Furthermore, our results show that Trump’s claims about election fraud do not influence his core supporters’ views only about the integrity of the 2020 election but also broader political attitudes in the general electorate. More specifically, we found that his rhetoric drove down public perceptions of electoral responsiveness, political efficacy, and overall support for democracy.
Our results show that Trump’s claims about election fraud do not influence his core supporters’ views only about the integrity of the 2020 election but also broader political attitudes in the general electorate.
These results have important implications for our current understanding of politics in the United States. Even if those making the claims do not sincerely believe them, as suggested by the anonymous official cited previously, that attitude does not translate to the general public. This may help us to understand the motivation for the Capitol insurrection on January 6 and why Democratic leadership subsequently moved to impeach Trump. Furthermore, legislating often requires a level of bipartisan compromise. However, if members of the two major parties have increasingly divergent beliefs about the legitimacy of the institution in which they must operate, then that compromise will become even more difficult to reach. A democratic government is only as strong as citizens’ belief in it and in legitimate elections. Therefore, to preserve such an institution, efforts must be made to ensure that citizens have trust in the process as well as in the outcomes.
Our study leaves a few issues unaddressed, which constitute fertile terrain for future research. First, our results are consistent with previous research that shows negative societal consequences for exposure to conspiracy theories. Nevertheless, it is important to note that some of our conclusions likely are context dependent with respect to Trump and his election defeat. Thus, although the primary focus of this article is to investigate political attitudes in the immediate aftermath of the 2020 presidential election, future research should focus on other electoral-fraud claims to assess the generalizability of our findings.
Second, existing opinion polls suggest that Fox News generally is considered less credible by Democrats and Independents than by Republicans.Footnote 19 Thus, it may be the case that Democrats and Independents were more likely to discount the content of our treatment, given their underlying skepticism of the presented news source. This constitutes a limitation of our study, and it implies that we might have found larger aggregate treatment effects if our experimental manipulation had used a source that was perceived as more neutral by these voters. We hope that future research will build on our work and explore the interaction between elite cues about election fraud and varying message sources.
Third, the composition of our sample did not allow us to assess how fraud cues influence attitudes of all politically relevant demographic subgroups. Most notably, our respondent pool did not contain enough nonvoters and people at the low end of the education spectrum to provide meaningful statistical estimates. Future experimental research should extend our work by also assessing the influence of elite rhetoric on attitudes toward democracy.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the PS: Political Science & Politics Harvard Dataverse at https://doi.org/10.7910/DVN/TPGJGA.
SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit http://doi.org/10.1017/S1049096522000221.
Open access
