Increasing polarization between the Democratic and Republican political parties is a defining feature of 21st-century American politics. Many pundits and scholars argue that partisans in the American public hold strong, divided, and unmovable opinions about candidate choice and public policy. This view holds that party affiliation drives vote choice and that partisans selectively interpret political information favourable to their personal politics. Evidence in support of this view includes the large differences in presidential approval between those who identify with each party regardless of the nation’s economic performance (Sides, Tausanovitch, and Vavreck Reference Sides, Tausanovitch and Vavreck2022) and the small effects of political advertising campaigns (Coppock, Hill, and Vavreck Reference Coppock, Hill and Vavreck2020, Sides, Vavreck, and Warshaw Reference Sides, Vavreck and Warshaw2022). Polarized politics can lead to gridlock in governing (Binder Reference Binder2014; Lee Reference Lee2016; Patashnik and Schiller Reference Patashnik and Schiller2020; Drutman Reference Drutman2020; Klein Reference Klein2020). It is closely linked to ‘affective polarization’, the gap between reported feelings toward in-and out-party others, which has now reached levels that stand out in America’s recent history (Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019) relative to other nations (Boxell, Gentzkow, and Shapiro Reference Boxell, Gentzkow and Shapiro2021). Druckman et al. (Reference Druckman, Klar, Krupnikov, Levendusky and Ryan2021) find that affective polarization shapes the expressed policy positions of partisans regardless of personal preferences. Perhaps most troubling is the threat that rigid partisan polarization might pose to American democracy. Clark and Stewart (Reference Clark and Stewart III2021) document a ‘historic gap in confidence’ between Democrat and Republican trust in elections, which grew even larger after the 2020 election cycle and remained large in 2022 (Stewart Reference Stewart III2023).Footnote 1 Finkel et al. (Reference Finkel, Bail, Cikara, Ditto, Iyengar, Klar, Mason, McGrath, Nyhan, Rand, Skitka, Tucker, Van Bavel, Wang and Druckman2020) contend that political polarization in the US has elevated into political sectarianism, a ‘poisonous cocktail of othering, aversion, and moralization’ that ‘poses a threat to democracy’ (p. 533). If America’s partisan divide leads to hardened divisions over election integrity, where politicians convince citizens that electoral results are fraudulent unless their side wins – regardless of countervailing information – the threat to democracy is especially grave.
Yet, while the view that Americans have rigidly polarized views might be widely accepted, a small but growing academic literature suggests that although American public opinion is polarized, it remains responsive to new information. Democrats and Republicans in the electorate may begin with largely divergent views on an issue but as they encounter relevant facts or arguments these views evolve in the same direction (Gerber and Green Reference Gerber and Green1999; Hill Reference Hill2017; Coppock Reference Coppock2022; Tapin et al. Reference Tappin, Berinsky and Rand2023). These findings are consistent with macro-level evidence on opinion shifts in recent decades, such as the change in views on same-sex marriage and marijuana legalization. While partisan gaps endure on these issues, both Republicans and Democrats have dramatically shifted their views in the same direction (Pew Research Center 2019).
Is persuasion possible in the politicized and conflictual realm of trust in elections, or do identifiers of each party hold rigid views on election administration such that new information fails to break through? In response to firsthand experience with the decline in shared trust in American elections following the 2020 presidential election, state and local election officials across the nation began recording and airing messages to explain and make transparent electoral procedures and administration. These videos aim to fill an information vacuum, with a recent authoritative survey finding that ‘voters are not very aware of the measures election officials undertake to secure elections’ (Stewart Reference Stewart III2023, p. 30).
Messaging from these experts could be a promising avenue toward restoring voters’ trust in elections, but only if Americans of all parties are open to updating their views on election integrity after exposure to official messaging. We partnered with officials in four states to evaluate whether their efforts could be successful in breaking through partisan divides to restore confidence in election integrity. We evaluate whether official messaging can increase trust in elections through a series of survey experiments conducted after the November 2022 midterm election.Footnote 2 Our experiments expose respondents to facts about how elections work using the actual public information videos produced recently by state and local elections officials. Randomization allows us to estimate if and by how much these messages influence beliefs about the integrity of elections and trust in democratic procedures, providing strong causal identification. By testing the effects of real videos on samples drawn from their target audiences, our approach also has the external validity of rigorously evaluating a public information campaign while it is in progress.
We find that voters of all partisan leanings – Republicans, Democrats, and Independents alike – are responsive to official messaging. These treatments did not simply increase trust among those who already trusted elections; both Republicans, who currently have the lowest beliefs about electoral integrity, and respondents who reported the lowest levels of trust in a pre-survey prior to the messages reported higher trust in elections after viewing the videos.
The treatment effects we observe are both significant and substantively large given the brevity of treatment. We found effects in all five survey experiments. Respondent reactions fit logical patterns; messages by officials in their own state increase trust in their own state elections. Messages describing procedures to safeguard the vote in other states lead to more trust in the elections of other states. These patterns suggest a reasoned uptake of the new information and all effects appear among voters of all partisan affiliations.
Our clearest contribution is to the scientific literature on trust in elections, much of which seeks to identify solutions to the challenges posed by a declining faith in democracy (Clayton and Willer Reference Clayton and Willer2023; Coppock et al. Reference Coppock, Gross, Porter, Thorson and Wood2023, Voelkel et al. Reference Voelkel, Stagnaro, Chu, Pink, Mernyk, Redekopp, Ghezae, Cashman, Adjodah, Allen, Allis, Baleria, Ballantyne, Van Bavel, Blunden, Braley, Bryan, Celniker, Cikara and Willer2023). We show that the actual messages that have been produced by election officials could, if amplified, bring meaningful increases in trust across party lines. Our results also support the literature suggesting persuasion is possible even on issues that divide political parties (Hill Reference Hill2017; Coppock Reference Coppock2022; Tapin et al. Reference Tappin, Berinsky and Rand2023).
Our findings – which show that public opinion can shift with new evidence and that faith in election integrity can be restored for Republicans, Independents, and Democrats alike – are also of practical importance for those on the front lines of American democracy. We designed our surveys in partnership with election officials, using an experimental design and pre-registered hypotheses to bring rigorous evidence to bear on critical questions they identified. Since the 2020 contest, those working to restore faith in the integrity of elections have been trying innovative approaches (Voelkel et al. Reference Voelkel, Stagnaro, Chu, Pink, Mernyk, Redekopp, Ghezae, Cashman, Adjodah, Allen, Allis, Baleria, Ballantyne, Van Bavel, Blunden, Braley, Bryan, Celniker, Cikara and Willer2023) and spreading best practices (Fessler Reference Fessler2022). Our study is designed to provide an analytical basis to evaluate these best practices, conducting tests that can show whether messages work, which messages are most effective, and what types of Americans are most responsive to them. This is a crucial public policy issue not just because of the abstract principle of faith in democracy. Distrust in elections challenges the non-partisan administration of elections, and even appears to have increased threats of physical violence against public servants (Barr Reference Barr2022; Gronke and Manson Reference Gronke and Manson2022). Increasing trust is vital for the operation of our electoral system.
We first motivate our theoretical expectations about why official messaging should increase trust – even in the context of partisan polarization – by grounding our expectations in existing work on both persuasion and the effectiveness of government interventions and the literature regarding trust in elections. Next, we describe our research design, our partnership with elected officials, our survey methodology, the videos we used as treatments for our two survey experiments, and our pre-registered hypotheses. We then present our results before concluding with a discussion of the implications of our findings and ideas for future research.
Increasing Trust in Elections During an Era of Partisan Polarization
Mistrust in American elections reached new highs after the 2020 election, with trust becoming increasingly polarized in recent years (Clark and Stewart Reference Clark and Stewart III2021). Yet research on election trust, political persuasion during an era of polarization, and public information campaigns broadly provide the basis for our theoretical expectation that official messaging can bolster trust and do so across party lines. We draw on this literature to motivate our pre-registered hypotheses that exposure to official messages will increase trust in elections and decrease beliefs that fraud is prevalent. Our expectations are based on five key findings of the literature that point toward the mechanisms that underlie opinion change in this area. We explore each of them in greater depth through this section, but summarize them below:
-
1. Trust in elections is malleable rather than set in stone; past work has shown that levels of confidence are responsive to new information and messaging.
-
2. Members of both parties will be open to learning and revising their beliefs about the elections, even if they begin at very different baseline levels of trust.
-
3. Americans lack information about the basic protections put in place to safeguard the accuracy and integrity of elections but indicate that they would become more confident if they knew these safeguards were in place.
-
4. Election officials are well-positioned to be trusted providers of this information.
-
5. Public information campaigns from other realms of government have proven effective.
First, the study of trust in elections in America and democracies around the world shows that trust is a malleable rather than fixed feature of public opinion, responsive to factors such as election outcomes and the claims of candidates. This evidence suggests, then, that trust in elections is eroded by a wide range of causes but also that it can be increased. Prior work suggests that political elites have played a role in increasing mistrust by repeating unsubstantiated claims about voter fraud in messaging directed at their supporters (Beaulieu Reference Beaulieu2014, Berlinski et al. Reference Berlinski, Doyle, Guess, Levy, Lyons, Montgomery, Nyhan and Reifler2021). Another stream of research from democracies around the world finds evidence of a strong winner/loser effect, with voters changing their views on governments and elections based on whether their preferred candidate won or lost (Anderson et al. Reference Anderson, Blais, Bowler, Donovan and Listhaug2005, Sances and Stewart Reference Sances and Stewart III2015). In one recent study, Reller et al. (Reference Reller, Anderson and Kousser2022) show that trust in elections can also shift at the mass level even without elite cues, finding that the partisans of the party who lose an election express reduced trust following the loss, even when their preferred candidate does not make any claims of vote fraud. Notably, they find that election winners also see an immediate increase in trust once the results are revealed.
What are the prospects for increasing trust among the electorate, especially across party lines? If citizens purely engage in motivated reasoning, the way that they process new information will be biased in the direction of positions that they already hold (Lodge and Taber Reference Lodge and Taber2013). Because Republicans and Independents reported significantly less trust in elections after the 2020 cycle, they might be resistant to information affirming the integrity of elections. This could make it difficult to persuade these voters that elections are generally safe.
Experiments testing messages from political leaders during this period provide mixed evidence. One study found that Republicans become more trusting in elections when exposed to messages from election leaders affirming the legitimacy of the 2020 result (Clayton and Willer Reference Clayton and Willer2023). However, another found no impact of messages from Mitch McConnell and Arnold Schwarzenegger on election trust, even though we might expect such Republican leaders to be trusted sources on this issue (Wuttke et al. Reference Wuttke, Sichart and Foos2024). Some prior work testing corrections of misperceptions about elections even finds that messages can backfire among Republicans (Holman and Lay Reference Holman and Lay2018, Christenson et al. Reference Christenson, Kreps and Kriner2021) or that their corrective effect among Republicans disappeared after the 2020 election (Jenkins and Gomez 2020). This is consistent with Lockhart et al.’s (Reference Lockhart, Hill, Merolla, Romero and Kousser2020) finding that while Democrats and Independents revised their views of how elections should be conducted based on scientific projections about the COVID-19 epidemic during the 2020 election cycle, Republicans did not change their views.
By contrast, others find that while the two parties may begin with very different views, their members are open to learning and revising their beliefs (Gerber and Green 1999; Hill Reference Hill2017; Coppock Reference Coppock2022; Tapin et al. Reference Tappin, Berinsky and Rand2023). One strand of this literature draws on recent experimental evidence to show that voters of all political persuasions consider new information and update their beliefs to become closer to the new information they encounter (Hill Reference Hill2017, Coppock Reference Coppock2022). This literature suggests that members of both parties can be persuaded to trust elections, even if they begin at very different baseline levels of trust. Based on these findings, the second underpinning of our theoretical expectation is that, regardless of party, Americans will be open to persuasion on this issue.
Third, we expect that new information that fills a vacuum of knowledge about election procedures can be effective at persuasion. Voelkel et al. (Reference Voelkel, Stagnaro, Chu, Pink, Mernyk, Redekopp, Ghezae, Cashman, Adjodah, Allen, Allis, Baleria, Ballantyne, Van Bavel, Blunden, Braley, Bryan, Celniker, Cikara and Willer2023) show that new information can be effective, testing a wide range of messaging strategies aimed at reducing support for undemocratic practices ranging from information provision to empathy-taking interventions. They find that many interventions are effective at reducing this support, but, unfortunately, do not look at questions of trust in current election procedures.Footnote 3
What is the information that Americans lack about their elections that might be provided to increase their confidence? Stewart’s (Reference Stewart III2023) Survey on the Performance of American Elections highlights a large gap between how voters think election administrators secure elections and the actions election administrators take. Voters underestimate the security measures that are in place while simultaneously reporting that, were those security measures in place, their trust would increase. For instance, only 15 per cent report that they are aware of signature verification procedures and only 41 per cent know about logic and accuracy testing, but 64 per cent would be more confident in the security and integrity of their elections if they knew that signature procedures were in place, and 74 per cent if they knew about logic and accuracy testing. Informational videos, many of which detail these procedures, can thus provide new information that will build confidence among voters.
Fourth, additional survey evidence suggests that state and local election officials are well-positioned to deliver this new information because they are a highly trusted source. While existing work studies the effect of messages from politicians or academics, one of the questions that we asked in our survey indicates that state and local election officials are far better messengers because voters report that they are the most trusted source on election integrity. In our national sample of Americans, we asked, ‘Who do you trust when it comes to evaluating the fairness and integrity of elections? (Check all that apply)’; 50.4 per cent of respondents selected ‘Local and state elections officials’. This was higher than any other source: ‘Television news in my local area (28.6%)’, ‘Fox News (22.8%), ‘CNN (22.0%)’, and ‘Political leaders in my party (17.3%)’.
Fifth, previous research shows government messaging campaigns can be effective at changing opinions and behaviour. A meta-analysis of 63 public health intervention campaigns finds that government messaging can change behaviour (Anker et al. Reference Anker, Feeley, McCracken and Lagoe2016). Similarly, Snyder (Reference Snyder2006) finds a roughly 5 percentage point impact of health interventions. Research on political messaging has shown that election officials can play a role in shaping voting behaviour within their jurisdictions (Kimball et al. Reference Kimball, Kropf and Battles2006).
Election officials have begun to undertake these types of campaigns to increase trust in elections and reduce misinformation; for example, in Wisconsin, the League of Wisconsin Municipalities partnered with other local government groups to run a campaign in advance of the April 2022 local elections in the state (Associated Press 2022). One study examining the correlation between spending on voter education by election officials and voter trust suggests that states with higher spending also have higher trust in elections (Suttman-Lea and Merivaki Reference Suttmann-Lea and Thessalia Merivaki2023). Our study aims to understand if such messaging from election officials is effective and can break through polarized attitudes about election integrity to increase trust in elections across the political spectrum.
Research Design
Election Official Partnership
We partnered with election officials in the US to test messaging strategies already employed by states and counties to increase trust in elections. Our treatments were chosen in consultation with partners in Los Angeles County, Colorado, Texas, and Georgia. These partners, in addition to choosing the treatments, suggested and informed our choice of questions and outcome measures. We additionally fielded two separate official messages to a nationally representative sample.
Messaging Content
In total, we tested ten different videos with information and messaging produced by state or county election officials aimed at increasing trust in elections across our two experiments.Footnote 4 We summarize each video in Table 1 below. For example, one video was a clip of a Fox News interview with the Texas Secretary of State, where he defends and advocates Texas election procedures and integrity. One video from Georgia simply presented information about how to vote in person and what to expect. We note that these messages are much more informative, even if banal, than entertaining.
Table 1. Description of video treatments for each sample. The National Experiment corresponds to Experiment 2 while the other experiments correspond to Experiment 1
We fielded experiments on five distinct samples: one that was nationally representative and one each representative at the level of our partner (subnational) geographies. Respondents assigned to control conditions viewed a 30-second State Farm Insurance advertisement in our national experiment or a 30-second Cadillac car advertisement in our subnational experiments.
Survey Methodology
We conducted the five surveys from November 17–27, 2022, immediately following the Associated Press calling the results of the 2022 midterm elections in the US House of Representatives for the Republican Party. This timing meant that election integrity was near its highest salience due to substantial media coverage along with messaging from candidates on the subject. Each survey was administered through Cint (following the merger with Lucid) using quota sampling to produce representative results. Cint has been used frequently by researchers studying American elections and has been shown to produce treatment results that are close to those produced using other samples (Coppock and McClellan Reference Coppock and McClellan2019).
In total, we collected responses from 8,338 participants. One survey targeted a national sample (N=3,038) while the other four targeted representative samples in Colorado, Georgia, Texas, and Los Angeles County, with samples of around 1,500 respondents. We targeted quotas based on the citizen voting age population in each state or county, and respondents could elect to complete the survey in either English or Spanish.
We list the full text of the election survey questions and answers in Appendix H. We employed two attention checks drawn from Berinsky et al. (Reference Berinsky, Margolis, Sances and Warshaw2021) and terminated (in real-time) respondents who failed either check. However, we subsequently identified a set of responses that, while passing the two checks, appear to have been generated by automated computer programs (‘bots’). We first identified the problem when we noticed an unusually large number of ‘respondents’ who selected a response to one question indicating that electoral fraud happens ‘all the time’ yet selected a response to a different question indicating they trust elections ‘a lot’ or ‘some’ (the strongest options available). We document in Appendix A our procedure to identify and remove suspected bot respondents based on incoherent yet repeated open-ended text responses.
Our subnational geographies reflect significant racial, ethnic, and political diversity. The overall, combined national and state sample identified as 67 per cent White, 12 per cent Black, 16 per cent Hispanic, and 4 per cent Asian. Respondents are politically diverse – 45 per cent report voting for Democratic candidate Joe Biden in the 2020 Presidential election while 36 per cent report voting for Republican Donald Trump. Seventy-five per cent report having voted in the 2022 midterm elections.
Respondents in the survey were first asked about their overall trust in elections, their trust in specific features of the 2022 midterm elections, and their experience of voting. Respondents were then presented with up to two experiments in a row, each independently randomized. For Experiment 1, following this pre-treatment battery, respondents from the subnational samples were randomly assigned to one of three conditions: one of the two geography-specific official messages (see Table 1 above) and a control condition viewing the Cadillac commercial. We then asked subnational respondents to report a feature of the video to assess compliance with the video assignment. On average, 91 per cent accurately reported the feature they were asked about, suggesting most respondents did view the treatments.
Following viewing the video in Experiment 1, all respondents – in national or subnational samples – were then randomly assigned to one of three national experiment conditions constituting a second experiment: the State Farm Insurance advertisement (control), the Virginia message, or the Maricopa County message (Table 1). After viewing the national video, respondents were again asked questions they could not answer without having watched their assigned video; 87 per cent did so successfully. Respondents in the subnational samples were assigned to view two videos. Because the treatment assignments in Experiment 1 and 2 are orthogonal, we analyze the two experiments as independent of one another (see Figure 1 for the experimental design).
Figure 1. Survey flow for each sample.
Finally, the respondents were asked four general questions about their trust in elections, as well as whether they intended to vote in the 2024 Presidential election. The post-treatment questions were as follows:
-
1. How much do you trust the accuracy and integrity of elections in your state?
-
2. How much do you trust the accuracy and integrity of elections in other states?
-
3. It is illegal to vote more than once in an election or to vote if not a US citizen. How frequently do you think such vote fraud occurs? Please provide your best guess even if you are not sure.
-
4. Do you think that official state or county election authorities – such as your Secretary of State, registrar, or elections director – ever engage in any form of vote fraud?
These outcome variables were chosen to specifically address the concerns of over-votes (Huber et al. Reference Huber, Cho, Bokemper, Gerber, Brady, McLoughlin and Crockett2024), based on our consultations with election officials and concerns over fraud committed by election officials themselves. One of the key concerns they see is citizen beliefs about voters casting additional ballots illegally, usually either supposed instances of non-citizens voting or citizens voting multiple times. Additionally, election officials often face direct claims that they engage in vote fraud and are interested in directly reducing this belief.
Expectations of Experimental Effects
We pre-registered five hypotheses related to election integrity, each corresponding to a specific survey question. First, we expect that videos that target trust in elections serve their goal and increase overall trust in elections. However, the effect on trust in one’s own state elections versus other state elections could be distinct; a respondent might view a video from their own Secretary of State and become more trusting in elections in their own state without changing how they judge elections in different states. On the other hand, receiving any information from election officials about election integrity might increase trust in all American elections. Our two pre-registered hypotheses relating to general trust in elections were:
H1: Watching any government-produced video about the integrity of elections will increase a respondent’s trust in elections in their state.
H2: Watching any government-produced video about the integrity of elections will increase a respondent’s trust in elections in other states.
We also asked about specific forms of trust; if our treatments impact trust, is it trust that other voters act with electoral integrity or trust that election officials act with integrity? We asked about voter fraud by voters who are ineligible to vote or vote more than once and we asked about fraud by election officials. Our third and fourth hypotheses were:
H3: Watching any government-produced video about the integrity of elections will increase a respondent’s doubt that there is illegal voting occurring in elections.
H4: Watching any government-produced video about the integrity of elections will increase a respondent’s doubt that there is fraud committed by election officials.
Finally, following conversations with our election official partners, we expected our treatments to potentially increase intentions to vote in 2024. This is because officials believe that concerns about election integrity are the cause of some abstention from participation Our fifth pre-registered hypothesis was:
H5: Watching any government-produced video about the integrity of elections will increase a respondent’s intent to vote in 2024.
We tested all five hypotheses in both Experiment 1 and Experiment 2.
Pre-registered Analysis Plan
For hypotheses 1 and 2, we asked respondents how much they trust elections (a) in their own state and (b) in other states. These questions had four response categories ranging from ‘Distrust a lot’ (coded 1) to ‘Trust a lot’ (coded 4). For hypotheses 3 and 4, we asked respondents how commonly fraud by voters and fraud by officials occurs. The five response categories to these questions ranged from fraud ‘Happens all the time’ (1) to fraud ‘Almost never occurs’ (5). Finally, we measured vote intent for hypothesis 5 with a five-point scale that ranged from ‘Definitely will not vote’ (1) to ‘Definitely will vote’ (5). The full text of all questions is given in the appendix, and in every case, a higher value indicates more trust in elections or greater intention to participate.
We created within-subject change in attitudes by subtracting pre-treatment answers from post-treatment answers. This approach allows us to measure the effect of our treatments more precisely as it removes variation from the outcome variable that is due to pre-treatment differences across participants. We model our results using ordinary least squares. Because we subtract pre-treatment values of our outcome variables and our treatment is randomized, we do not include any additional control variables.
For the main results, we collapse the two treatment groups (different messaging videos) in each experiment following our pre-registration analysis plan. We did not have strong theoretical beliefs for why any one message should be more effective than any other, especially at the state and local levels where treatments were chosen in part by our election official partners. We did, however, pre-register a comparison of the two national treatment videos in Experiment 2, results are presented separately below.
Results
Patterns of Trust in Elections
Before evaluating our experimental effects, we present descriptive data showing levels of trust in elections among members of each partisan group in each of our samples. In Figure 1, we plot average levels of trust in elections reported by our respondents. The top frame presents average trust for elections in the respondent’s own state and the bottom frame for other states. Our sample reproduces the polarization in trust found by others. Those who identify as Democrats report average trust in their own state elections around 3.5 on the 4-point scale (halfway between ‘Trust some’ and ‘Trust a lot’) across geographies. Those who identify as Republican report average trust in their own state elections of around 2.75, between ‘Distrust some’ and ‘Trust some’. Independents report trust similar to that of Republicans. We repeat this figure in the Appendix B grouped by party instead of state.
Patterns of trust are similar across party identification for other state elections, with average trust modestly lower (and again consistent with prior work). Democrat averages are closer to ‘Trust some’ for other states while Independent and Republican averages are now pushing closer to ‘Distrust some’ than to ‘Trust some’.
These pre-treatment patterns establish the challenge facing American democracy. Americans express some unease about the operation of their elections, particularly elections held outside their own state. Trust is especially low among Republicans and Independents. We next present evidence that messaging from election officials can improve confidence in elections and that they bring these positive effects for members of each partisan group.
Experiment 1
Table 2 presents the results of Experiment 1 for respondents in our Texan sample. It compares changes in trust among those who watched one of the two Texas election official messages to changes among those assigned to watch the car advertisement. Texans assigned to one of the two official messages exhibited significantly greater trust in elections in their own state than did Texas assigned to the control condition (column one), being 21 percentage points more likely to ‘Trust some’ or ‘Trust a lot’. The point estimate for trust in other state elections is also positive, significant at p < 0.1, but of a much smaller magnitude (column two). The videos appear to decrease the belief that officials engage in fraud more than they decrease the belief that voters engage in fraud (column four versus column three, both variables coded so that positive values mean less fraud) but do not appear to change intention to vote (column five).
Table 2. Experimental effects in the Texas sample of the Texas election official messages, following the pre-registration plan
+p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001
In Appendix C, we present estimated treatment effects from the Los Angeles County, Colorado, and Georgia samples, mimicking the presentation of Table 2. We summarize the treatment effects presented in these tables in Figure 2. Each bar corresponds to the treatment effect in that state for that dependent variable, with whiskers extending to 80 per cent confidence intervals. For example, the bar to the farthest left shows that the treatment effect of either Colorado video on respondent trust in Colorado elections is 9.3 percentage points. For each outcome variable, we also estimate a pooled treatment effect by combining the four samples and treating the design as a block-randomized experiment (tabular results also in Appendix C).
Figure 2. Trust in own and other state elections by party identification and geography.
In each sample and in combination, we find positive and often strong effects of election official messaging on trust in respondents’ own state elections (see Figure 3). The average effect of treatment videos pooling all subnational respondents on trust in their own state elections is 0.12 points on the four-point scale. This estimate is about 20 per cent of the pre-treatment difference in trust of own state elections between Democrats and Republicans in our national sample (see Figure 1). Effects vary from 0.07 in Los Angeles County to 0.21 in Texas. The results are consistent across the four geographies and suggest that official messaging increases trust in local elections.
Figure 3. Pooled treatment effects in each geographic and pooled sample. Whiskers extend to 80 per cent confidence intervals.
We find smaller and uncertain impacts of local election official messaging on trust in elections in other states. While all coefficient estimates are positive, they are always smaller than the effect on trust in own state elections. The pooled estimate is 0.042 points with a standard error of 0.019 (significant at p<.05). Importantly, this estimate is consistent with the logical implication of the informational content. All messages tested in Experiment 1 provided information about election protections in a respondent’s own state. It stands to reason, then, that the delivery of such information increased trust in their own state’s elections more than in the elections administered by other states (something we turn to in Experiment 2).
We find similar patterns for reported beliefs about the incidence of election fraud by both voters and officials: point estimates are consistently positive (suggesting messaging counteracts beliefs about fraud), though variable. The pooled estimates are 0.066 (SE = 0.024) voter fraud and 0.083 (SE = 0.024) official fraud.
We finally consider the effect of messages on the intention to vote in 2024. The pooled estimate is 0.006 (SE = 0.015). Along with the small coefficients and variability across subnational samples, our evidence suggests that official messaging on the integrity of electoral processes does not consistently or materially increase intention to vote.
In Appendix G we present the distribution of outcome variables for treatment and control groups. These figures show that there is significant movement across the range of the scale and emphasize the effect size. Fewer respondents select ‘Distrust a lot’ and more select ‘Trust a lot’ having viewed official messaging rather than the control video.
Experiment 2
Table 3 reports the results of our national experiment on trust in elections. We compare changes in trust among those who watched one of the two election official videos – either from Virginia or from Maricopa County – to changes among those who watched the control video featuring Jake from State Farm.Footnote 5 This experiment pools the national sample with the four state and local samples.Footnote 6
Table 3. Treatment effects of the national experiment, pooling all samples, following the pre-registration plan
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
We see three important results in this table. First, the assignment to view an election official message significantly increased trust in elections: respondents were more likely to trust elections in other states, believed voter fraud was less common, and believed fraud by officials was less common after viewing either treatment video. This is a promising result for those who aim to increase trust in elections: exposure to a short video produced by election officials can move opinions on a contentious, polarized topic. While views on election fraud might have been expected to be relatively crystallized by mid-November 2022, these messages significantly shifted voters’ views even at a time of heightened polarization.
Second, the results show that the messaging reached its target but did not go beyond it: exposure to a video about election protections in other states did not significantly increase respondents’ trust in their own state’s elections. Combined with our results in Experiment 1, this pattern fits what we would expect if respondents were paying attention to the substance and geographic focus of the messages. In Experiment 1, we provided respondents with a local treatment and their trust in local elections increased while their trust in elections elsewhere in the US was largely unchanged, moving just a fraction of that amount. In Experiment 2, where we presented respondents with treatments from other states, trust in elections outside their own state increased six times more than trust in elections within their state.
While we did not pre-register an expectation of this difference, it follows a pattern that suggests respondents are not simply inferring that the study is about trust in elections and then giving the researcher what they want (demand effects). If respondents were simply updating in response to survey questions about elections or guessing the purpose of the story, we would not expect to see this pattern matching across the target of the advertisement and the outcome measures. This gives us some degree of confidence that our observed results are not the result of demand effects (consistent with the finding of limited demand effects of Mummolo and Peterson Reference Mummolo and Peterson2019).
Our third finding is that, despite the hopes of our partners, the advertisements that build trust in elections do not increase reported intention to vote in future elections (column five). It is noteworthy, however, that while perceptions of fraud have increased over the past decade, electoral participation has also increased in federal elections during that timeframe. An important area for future research will be exploring the behavioural consequences that potentially do or do not flow from shifts in trust in elections, particularly in local and state elections.
Emotional vs Informational Appeals
In addition to the main effects of being exposed to a video on trust in elections, we pre-registered three additional analyses for Experiment 2. First, we test for heterogeneous treatment effects for the two national messaging videos. We describe the video produced in Virginia as an emotion-based treatment; it appeals to American identity and emotions but provides little information about the voting process or election administration. The video produced in Maricopa County, however, provides specific information about the processes used to ensure the integrity of the vote with no appeal to emotion or patriotism. Because these videos have substantially different messaging strategies, we pre-registered testing for detectable differences between the two effects. The content of these messages broadly reflects two potential pathways to the persuasion laid out above, an informational pathway and an emotional pathway that appeals to voters’ trust in election officials.
In Table 4, we find only suggestive evidence that the more factual video had a larger effect on respondent beliefs about the incidence of voter or election official fraud. The estimated treatment effect of the fact-focused message is 0.64 and on official fraud is 0.1 compared to 0.024 and 0.064 for the emotion-focused message (both differences statistically significant at p<0.1). We see negligible differences between point estimates on the remaining outcome. Thus, it appears the fact-focused message has a greater influence on beliefs about fraud than the emotions-focused message, but does not have a greater influence on trust or vote intention.
Table 4. Categorical results of the experiment in the pooled, national sample, following pre-registration
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
Additional Analyses
In Appendix D we report results from two additional pre-registered subgroup analyses. We based these choices on groups we expected to be more persuadable: those who pay attention to the stimulus and likely receive the intended information and those who hold mixed views initially towards elections (Zaller Reference Zaller1992). First, we examined whether more attentive respondents showed larger treatment effects based on past work showing attention conditions treatment effect sizes (Read et al, Reference Read, Wolters and Berinsky2022). Following our pre-registration plan, we categorized respondents as attentive if they accurately reported the feature of the video they were assigned to watch and spent at least 20 seconds viewing the page with the treatment video embedded. The results, presented in Table D1, are both substantively and statistically the same as those presented in Table 3, implying no heterogeneity by attentiveness.
The second additional analysis looked at whether respondents who hold moderate levels of trust in elections are more persuadable than those with high or low levels of trust. Zaller (Reference Zaller1992) suggests that those with moderate views are most likely to accept new information and be persuadable. Respondents who report that they ‘Trust elections a lot’ or ‘Distrust elections a lot’ in response to all four pre-treatment election trust questions might, therefore, be less persuadable. Additionally, these respondents could only be persuaded in one direction: low-trust respondents could only gain trust and high-trust respondents could only lose it. The results in Table D2, however, are substantively and statistically the same as those presented in Table 3, suggesting against the heterogeneous opportunity for targeting persuasion at those with middling views, consistent with the findings of limited heterogeneity in persuasive messaging in Coppock, Hill, and Vavreck Reference Coppock, Hill and Vavreck2020.
Results by respondent party identification
Our main conclusion is that official messaging about election procedures can increase trust in elections and decrease beliefs about election fraud. The effects presented above, however, average across all respondents regardless of whether they identify with a political party. To confirm that these messages break through political polarization, in this section we estimate effects by the party identification of the respondents. While not pre-registered, we anticipate many readers will want to see these results to feel comfortable concluding the messages break through polarization.
In Table 5 we present the treatment effect from Experiment 2 (the national videos) interacted with respondent party identification. The results show that there is no significant difference between partisans. In fact, across the five specifications, the coefficient for Republicans is larger than the coefficient for Democrats in four cases and functionally identical in the fifth. Independents also generally exhibit higher but statistically indistinguishable effects relative to Democrats. The messages tested appear to break through political polarization; voters of all partisan attachments respond to messages from election officials.
Table 5. Heterogeneous treatment effects by party. Democratic respondents are the excluded category such that interaction effects represent deviations from the effects on Democrats
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.
Additionally, we present the conditional treatment effects of the videos by party in Appendix F. This allows us to test whether the effects are conditional on election outcomes: Californian and Coloradan Republicans and Texan and Georgian Democrats had experienced electoral losses in top-of-ticket midterm races. We find no consistent evidence of heterogeneity by party in our samples.
Conclusion
In our messaging experiments conducted with five samples representative of eligible citizens in the US, Los Angeles County, Colorado, Georgia, and Texas, we find that messages produced and delivered by election officials can increase trust in the integrity of American elections. Vitally, these treatment effects do not vary by party: Republicans, Democrats, and Independents alike report greater trust in elections when exposed to information about election protections. These effects are substantively important and statistically significant.
Our findings contribute to the set of recent works exploring innovative ways to increase trust in elections (Clayton and Willer Reference Clayton and Willer2023; Coppock et al. Reference Coppock, Gross, Porter, Thorson and Wood2023; Voelkel et al. Reference Voelkel, Stagnaro, Chu, Pink, Mernyk, Redekopp, Ghezae, Cashman, Adjodah, Allen, Allis, Baleria, Ballantyne, Van Bavel, Blunden, Braley, Bryan, Celniker, Cikara and Willer2023). They also speak to the broader literature on whether partisanship prevents opinion change during this polarized era. Work on motivated reasoning and affective polarization point to the challenges of breaking through partisan polarization with new information. Some studies find this to be true in the realm of trust in elections, finding either that Republicans are unmovable (Wuttke et al. 2022) or that correcting election-related misperceptions can backfire with GOP supporters (Holman and Lay Reference Holman and Lay2018; Christenson et al. Reference Christenson, Kreps and Kriner2021; Jenkins and Gomez Reference Jenkins and Gomez2022). By contrast, our findings are in keeping with studies demonstrating that political views can evolve together across party lines (Gerber and Green, Reference Gerber and Green1999; Hill Reference Hill2017; Coppock Reference Coppock2022; Tapin et al. Reference Tappin, Berinsky and Rand2023).
These results provide proof of concept that public information campaigns can be effective at restoring trust in American elections. Although the magnitudes that we observe are not in themselves large enough to overcome the deficits in trust evident for some groups in our survey, these effects follow after viewing a single short video. Effective public information campaigns are built on repeated and varied messaging; our findings suggest that messages from election officials work and that multiple types of messages can build trust.
We find these effects despite public positions taken by prominent politicians in recent years that might have solidified views on election integrity. Our findings suggest that a lengthy and intensive campaign consisting of factual videos addressing different types of election integrity concerns could be impactful, meaningfully restoring faith in the administration of elections. Future work should evaluate whether larger-scale advertising campaigns, outside of the controlled context of a survey experiment, can build bulwarks of trust. There are important caveats to our findings that should be addressed through future work. We do not know the durability of the effects that we observe. While we speculate that they might persist (as we were able to move trust attitudes even during a time of high saliency), future work might look at this directly. Another vital question is whether the factual information presented by election officials in these videos will still be convincing in the current noisy informational context that also includes misinformation about elections promoted by political candidates. Further work that tests the impact of these videos when respondents are also exposed to a counter-frame making charges of election fraud would be important to probe this question.
Future work could also look at the other contexts in which these messages might work. For instance, do other mediums allow officials to increase trust besides public information-style videos? Our analysis tests one such message, the Fox News video in Texas, but a broader investigation is important as officials need to not only be persuasive but they must also reach voters and those who are inclined to distrust elections. Media appearances, direct outreach, election facility tours, social media campaigns, and other strategies are worth testing to learn more about when messaging might work. Additionally, this study is limited to the context of American elections, which are run at the local level. Election officials in other countries face similar obstacles and engage in similar trust-building efforts as American officials (Ilanbey Reference Ilanbey2022). While the evidence suggests similar treatments might work elsewhere, we do not have direct evidence about the effectiveness of messaging campaigns in increasing trust in elections held outside of America. Our study suggests that investing in larger communication campaigns to reach voters is worthwhile and provides evidence of the effectiveness of such interventions.
Future research could explore the causal mechanisms that underpin the increases in trust that we observe. One area that needs further clarification is whether different types of messages – for instance, those that deliver factual content versus those that prime emotions – are differentially effective. Another question is whether the information contained in these messages affects trust through increasing factual knowledge, with only the respondents who become more knowledgeable about elections shifting their level of trust. Alternatively, do respondents simply become more likely to trust elections when they are exposed to a list of the precautions in place to protect them, increasing their general level of confidence even if they do not absorb the specific details of these protections? Determining whether and how long the persuasion effects that we identify persist would be important to inform the design of any public information campaign, including its timing around elections. Finally, to address the potentially disparate effects of messaging by race, ethnicity, or age, future work could look to identify causal mechanisms that generate heterogeneity in response to the official messaging (see Uribe et al. Reference Uribe, Aldridge, Kousser, Nichols-Smith and Tye Rush2024 for one example).
Regardless of the mechanism, our set of experiments demonstrates the significant impact that information delivered by election officials can have on Americans of all party affiliations. In the 2024 presidential election and future election cycles, robust public information campaigns carried out by state and local election officials can play a significant role in restoring faith in American elections, potentially reducing the persistent partisan gap in trust that has emerged since 2020. Indeed, officials in the jurisdictions studied here and in many others across the country are undertaking significant efforts in 2024. Of course, there are scope conditions that limit where we would expect to see effective public information campaigns emerge. Many election officials lack sufficient funding and staff resources to produce and distribute messages about protections in elections. While our findings show that even relatively simple and low-cost videos increase trust and that they are an effective use of public funds, we recognize that many officials are tightly resource-constrained. Second, officials must themselves trust the processes that protect the accuracy and integrity of elections to serve as messengers about these protections. In the 2022 elections, many candidates who denied that Joe Biden was a legitimate winner in 2020 ran for Secretary of State elections. Nearly all of these candidates lost their races, including all of those running in competitive states (States United Democracy Center 2022), but this serves as a reminder that election officials will only be strong messengers for public confidence if these officials continue to trust elections themselves.
With the rise in concern over election integrity, officials have started producing more content. We show that their content works. Of course, in the real world, citizens receive messages designed to lower their trust in elections, but the fact that these messages do work to increase trust suggests that if the balance of what citizens heard were to change in favour of messages from local election officials, trust would go up, given the effectiveness of these measures. Future research should address the effects of counter-framing as well as the persistence of the effects we find.
Our findings suggest an important place for state and local election officials in combatting misinformation and distrust in elections in the US. In our national survey, we find that these officials are far and away the most trusted source of information on election integrity. More than 50 per cent of respondents identify election officials as their most trusted source for ‘evaluating the fairness and integrity of elections’, compared with 17 per cent who turn to party political leaders. This descriptive result reinforces our experimental result, that election official messages can move the needle on trust.
Another finding from our national survey has an implication for how a public information campaign could be most effective. Overall, 72 per cent of those surveyed trust elections in their own states ‘Some’ or ‘A lot’. By contrast, 58 per cent trust the accuracy and integrity of elections in other states. This gap, with Americans trusting elections in their own states much more than they trust elections in other states, is consistent across party lines. The implication is that we cannot only rely on the existing efforts of state and county election officials with messages targeting their electorates about the elections in their own states. These local efforts need to be supplemented by cross-state or national efforts designed to explain the safeguards on elections in other states, especially the battleground states most likely to be the focus of future controversy. Our national experiment shows that such effects can help build a shared national trust in elections.
Supplementary material
Supplementary materials will be made available online https://doi.org/10.1017/S0007123424000681.
Data availability statement
Replication data for this paper can be found at: https://doi.org/10.7910/DVN/BWW9LQ. The pre-registration for this project is available online at: https://osf.io/9rjk4.
Acknowledgements
The authors thank the cooperation of election practitioners in Colorado, Georgia, Los Angeles, and Texas for their help in conceptualizing this work and identifying potential treatment videos. The election officials are not responsible for any of the final results or conclusions in this paper. Additionally, the authors thank participants at the 2023 APSA, MPSA, ESRA, and SPSA for their helpful feedback and comments.
Financial support
Funding for this project was provided by the e MIT Election Data and Science Lab ‘Learning from Elections’ programme and from the UC San Diego Yankelovich Center.
Competing interests
The authors declare none.
Ethical standard
This research was reviewed and deemed exempt by the University of California San Diego Institutional Review Board.
Open access
