Voters distrust delayed election results, but a prebunking message inoculates against distrust

Abstract

Counting and certifying election results in the United States can take days and even weeks following election day. These delays are often linked to distrust in elections but does delay cause distrust? What can election officials do to counteract distrust if counting most ballots and announcing results cannot occur on election night? Using a preregistered survey experiment of nearly 10,000 Americans, this article shows that informing voters about longer-than-expected vote counting time induces a large, significant decrease in trust in the election. However, viewing a “prebunking” video in advance of being informed of the delay in results more than makes up for the delay-induced decrease in election trust. Our findings have two important implications. First, unexpected delays in calling elections induce distrust even without misinformation from third parties. Second, providing voters with information about vote counting and the legitimate reasons for delays increases trust and mitigates the distrust induced by delays.

trust in elections, election integrity, election administration, party polarization

Significance Statement

Trust in elections is a crucial basis for democracy. As such, eroding trust is a major concern for scholars and administrators of elections. Understanding both the causes of this erosion and how to prevent it provides policy makers the tools necessary to increase trust among voters. Following recent high-profile delayed elections results in the US elections of 2020 and 2022, it is important to understand how delayed results impact voter trust directly. We emphasize the importance of timely counting to limit distrust, as well as suggest that public information campaigns can head off some of that distrust if delays in results are unavoidable.

Introduction

The Associated Press did not declare Joe Biden the winner of the 2020 presidential election until 4 days after Election Day (1). In 2022, control of the Senate was not called for 4 days and control of the House took even longer (2). Not since 2000 has it taken so long for Americans to learn the outcome of their federal elections. In both 2020 and 2022, the length of time it took to count ballots in key swing states and districts received significant media coverage from across the political spectrum, both before and after the election (e.g. (3–6)).

There are many legitimate reasons why counting ballots in the United States takes time. Some state laws prohibit election workers from counting ballots until election day. Other states allow mail-in ballots to be counted even if they arrive well past election day (7).

Unfortunately, the timeline of counting ballots has increasingly been linked to a distrust in the electoral process as politicians claim that slow counting is a sign of electoral fraud (8, 9). Election officials are aware of the potential link between delayed reporting of results and conspiracy-fueled distrust but often constrained by law in their ability to speed up the counting of ballots (10).

Do delays in election results, even absent elite-driven misinformation, cause distrust in US elections? And, if so, what steps can election officials take to counteract this distrust given existing legislative landscapes of their states? The academic literature has relatively little to say about how Americans respond to the delayed vote counts that have been major stories in recent election cycles. In this article, we examine these questions using experimental evidence. We test whether the delays in tallying election results in the United States are, on their own, enough to generate distrust in the integrity of election results.

Conducting a preregistered survey experiment using a diverse sample of nearly 10,000 Americans, we find that when respondents are informed about ballot counting delays, they express lower trust in elections. However, we find that when voters watch a “prebunking” video produced by election officials that explains why it takes time to count ballots accurately, the negative effect of the delay on trust is mostly ameliorated. We further find that prebunking videos increase trust among voters who were not told that the counting of ballots was delayed.

These results have two important implications––one for legislators and one for election administrators. For legislators, our results highlight the important impact that delaying the counting of ballots can have on trust. Policies that decrease the time needed to count ballots, such as those that allow the processing or counting of mail ballots before election day (11), may increase trust in election integrity. For election administrators, we show that when delays are expected, proactive messaging campaigns before election day can increase trust in elections and make-up for delay-induced distrust.

Background

Our experiment is meant to test the direct effect of delays in ballot processing on trust in elections. We do not study the separate effects of political misinformation about the counting process, such as former President Donald Trump's claims that delays allow for ballot dumping or other electoral fraud (12, 13). Rather, we focus on the direct effect generated by delays in the counting of votes and the reporting of election results that stretch beyond election day.

Why might delays in election results decrease trust, even without additional messages from politicians? One possibility is that the discourse since 2020 has focused heavily on fraud claims tied to vote counting and reporting. Primarily because of an increased use of mail-in ballots in the previous few years, tallying has been slower than in many past. When voters encounter unexpected delays, they might fill in the blanks using existing narratives of fraud (14, 15). Another potential reason for distrust is that unexpected delays create room for uncertainty; voters who are used to rapid vote counting see longer than normal times and naturally wonder if the delay indicates a problem. Regardless of the pathway, it is plausible that delays in vote counting by officials and in election calls made by media organizations are sufficient to induce distrust among voters without need for misinformation from elites.

The potential counter-measure we focus on is prebunking. Prebunking refers to any intervention designed to come chronologically before someone is exposed to misinformation with the goal of reducing the impact of misinformation at the time of exposure (16). For example, social media companies might proactively provide instruction to users to help them identify fake news in the lead up to an election (17).

This type of messaging strategy has become more widespread in recent years due in large part to two advantages it has over fact-checking and other post hoc interventions to reduce misinformation. First, the number of possible misinformation threats in any given area is large and the effort required to debunk is high; it is easy to generate many false reasons why, for example, someone should not trust a COVID-19 vaccination and much harder to counter each possible explanation with accurate and understandable explanations. Prebunking potentially allows misinformation practitioners to combat many possible attacks simultaneously (16).

Second, researchers and practitioners are often concerned about the possibility of belief “echoes” or lingering effects of misinformation, even when the underlying belief is corrected (18). For instance, someone exposed to false claims about the danger of COVID vaccinations might update their belief about the specific claim thanks to a fact check but remain more hesitant about receiving a vaccine than they were before they were exposed to the misinformation. Prebunking might help prevent the initial reaction to misinformation and reduce the long-term impact of misinformation. Prebunking interventions have shown to be successful in other contexts, including information provision for spotting misinformation and manipulation techniques (19, 20). Because prebunking comes before misinformation, it might be easier to protect against this information rather than change attitudes afterwards.

Additionally, election officials are well positioned to engage in prebunking. While past research often highlights prebunking by social media companies, evidence suggests that local election officials are significantly more trusted by voters than social media companies when it comes to election-related information (21). Election officials already produce messages for their constituents, actively communicating with voters before election day to increase their trust in the election process (22).

This highlights that prebunking fits within a wider portfolio of methods election officials use to build trust in elections. Election officials of both parties engage in a wide range of activities to communicate with voters and inform voters about elections, including prebunking, producing postelection audit reports, running public information campaigns, speaking to local and national news outlets, and talking to voters directly (23). While the evidence on prebunking suggests it has a place within this portfolio, this by no means suggests other communication strategies are not important.

This type of messaging differs importantly from many other treatments tested in the literature by using an intervention focused on the process of counting ballots as opposed to interventions providing information about voting (24). Prebunking that focuses on processes that voters do not usually see suggests that voter trust can be improved not just by improving their personal experiences, but by filling in informational gaps about the entire election administration process.

Methods

We tested the effectiveness of prebunking interventions related to election delays in the 2022 gubernatorial election in Arizona. This election had two useful features for our research purposes. First, Arizona is home to high-profile elections and high-profile delays in reporting results during recent cycles, making it an important setting in which to study election trust.

Second, because the counting of votes in Arizona has been delayed in the past, the Arizona Secretary of State has produced messaging designed to inform the public about why counting can be slow. These messages emphasize that the length of time spent counting ballots is, at least in part, due to prioritizing the integrity of the election through the careful work of election staff. We test the effectiveness of these Secretary of State-produced prebunking messages.

Our study involved nearly 10,000 Americans recruited online through Cint (formerly Luc.id) in the spring of 2023.^a The sample mixes one nationally representative sample with geographically targeted samples in Georgia, Colorado, Texas, and Los Angeles, California^b. This experiment was part of a larger survey on attitudes about elections and election administration in the United States and was deemed exempt by the University of California San Diego's Institutional Review Board and informed consent was obtained from all participants. The geographic samples were selected as part of research-practice partnerships with election officials in these geographies.

Past research has shown that surveys conducted through Cint produce treatment effects similar to those conducted in other settings (25). Additionally, using a sample largely outside of Arizona provides a “hard test” for prebunking. Moving opinion in other states is often seen as more challenging than moving opinion within the respondent's state.

For the experiment, respondents were first randomly assigned to either a prebunking treatment video or to no video. The 40-s video was produced by the Arizona Secretary of State for the 2022 primary elections and emphasized the message that “accurate election results take time,” explaining the reasons for carefully counting ballots and providing viewers with a website to go to for more information^c.

Respondents were next independently assigned at random to receive one of two information prompts about the outcome of the 2022 gubernatorial election in Arizona. The two prompts were identical except that the treatment delay condition included truthful information that the election took multiple days to be called. The full text of the prompt (with the different timing conditions in brackets) was:

“Consider the 2022 election for Governor of Arizona. [While the election was held on November 8, 2022, the race was not called by news agencies until November 14, 2022, 6 days later./The election was held on November 8, 2022.] The Democratic gubernatorial candidate was elected governor.”

The full 2 × 2 experimental design is summarized in Table 1. This design allows us to test effects of the delay condition, of the prebunk condition, and of the interaction between the two.

Table 1.

Experimental set-up. Independent random assignment.

	Information
Video	No delay info + no prebunk	Delay info + no prebunk
	No delay info + prebunk video	Delay info + prebunk

The outcome variable of interest was how respondents answered the following question: “How much do you trust election results in the state of Arizona?” There were four response options that ranged from “Trust a lot” to “Distrust a lot” as well as a “No opinion” option which, following our preregistration plan, was excluded from the analysis. We chose this variable to maintain consistency with prior surveys that we have conducted and to focus respondents on a theoretical construct of interest that has also been the subject of public controversy: trust in the accuracy of election results. We specified the state of Arizona so that, rather than measuring generalized trust, this measure would be targeted at elections in the state that was the subject of the treatment. Below, we present the proportion of respondents in each treatment group who report either trusting elections in Arizona “a lot” or “some” for ease of interpretation. Analysis of the full four-point scale is available in Table S2.

Results

We present results in Table 2. Respondents who were assigned neither the prebunking video nor the message about delays in counting of the 2022 results said they trusted Arizona elections “some” or “a lot” at a rate of 66.7% (upper left cell). Those who were assigned information about the delay but no prebunking video were 6.5% points less trusting (upper right cell). Learning about a delay in the reporting of election results caused a decline in trust, moving about one in ten of the respondents who would have trusted the election instead to distrust.

Table 2.

Percent trusting elections in Arizona “some” or “a lot” in each experimental condition.

No delay info + no prebunk 66.7%	Delay info + no prebunk 60.2%
No delay info + prebunk video 69.1%	Delay info + prebunk 64.1%

The second row presents trust among those who first viewed the prebunking video. Those who were not assigned to the delay information were 2.5% points more trusting of elections after the video (bottom left cell). Those who were assigned both the prebunking video and the delay message expressed trust at a rate of 64.1% (bottom right cell), almost as high as those assigned to both control conditions. This suggests that watching the prebunking video ameliorates most (4 of the 6% points) of the distrust caused by information about an unexpected delay in election results.

The increased distrust among respondents who learned about delay in reporting occurs even though the delay treatment does not provide any direct suggestion of fraud or irregularities. This supports the idea that delayed election results without outside misinformation can be enough to diminish trust in elections. Yet the bottom right cell indicates that delivering official information about the legitimate reasons for delay can counteract much of this distrust.

To evaluate the uncertainty surrounding the experimental effects, we present a regression analysis in Table S1. The results show that the effects of the delay and prebunking treatments are large enough to be statistically significant from zero, with point estimates of − 6.5 and +2.4% points, respectively.

Results by party identification

In addition to the main effects of the treatment, we also preregistered analysis of treatment effect heterogeneity by partisanship. If delays induce distrust by activating existing narratives of election fraud, we might expect these effects to be concentrated among Republicans, whose party leaders have been most prominent in rhetoric about election fraud and improper counting of ballots.

Figure 1 supports this conclusion. The plot presents estimated treatment effects from analysis of the experiment by respondents' reported party identification. The plot shows that Republican respondents exhibit a decline in trust almost three times as large as Democrats and Independents in response to the delay messaging (3.6% points for Democrats, 9.6 for Republicans).

Fig. 1.

Treatment effects by party on the percent of respondents trusting elections in Arizona either “some” or “a lot”.

The mitigating value of the prebunking message, however, is also concentrated among Republicans. Republicans report 7.5-percentage point higher in trust in elections when they view the prebunking video. Similar to the overall results, the magnitude of this effect would mean that for a Republican who both sees the prebunking video and is exposed to the delay information, their trust in Arizonan elections would be very close to if they had been exposed to neither.

Discussion

Our results provide answers to each of our research questions. First, delays in the counting and reporting of election results do increase the distrust of voters in elections, even without misinformation or other elite rhetoric amplifying concerns. This result highlights that addressing delays in ballot counting ought to be front and center for election officials who aim to foster trust in elections.

We identify an important and useful tool available to election administrators who are concerned that the length of time it takes to count ballots: prebunking and communicating with voters about anticipated delays. Our results show a basic prebunking intervention with a simple message from the Arizona Secretary of State is enough to counteract much of the decrease in trust induced by delayed election results. This result fits within the growing body of work finding that local election officials are a trusted source for voters who can effectively communicate about election integrity (24, 22). While we recognize that funding constraints prevent many election officials from disseminating these messages as broadly as they would like, our findings show that a very simple, inexpensively produced message can bring significant effects, providing both a template for future communications and evidence that investment in them can be worthwhile.

We do not find evidence of interaction between the prebunk and the delay treatments, which means that even respondents not provided information about delays in the 2022 election responded to the prebunking video with increased trust. This suggests there could be additional positive benefits to prebunking communications; instead of simply anticipating distrust caused by election delays, prebunking by election officials might induce a general increase in trust. Evidence suggests that voters prefer to hear from election officials and even simple messages can increase their trust in elections (22). Messages about the length of time it can take to count ballots and how protecting the integrity of this process takes time should be added to election officials' toolbox.

Finally, these results suggest legislators should be aware of the downsides to policies that delay the counting of ballots. Exploring policies that reduce delay without diminishing access to and integrity of voting, such as the processing and counting of mail ballots that arrive before election day (26), could produce more timely results that allay potential skepticism.

Notes

a

Preregistration available on OSF (https://osf.io/sjxhu).

b

We pool all our respondents following our preregistration plan to maximize power. The results presented here are substantively the same if we exclude the geographic samples and respondents from Arizona.

c

The video can be viewed at: https://youtu.be/E6dR1Z7zu8U.

Acknowledgements

The authors are grateful for funding provided by the MIT Election Data and Science Lab “Learning from Elections” program and from the UC San Diego Yankelovich Center.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This work was funded by the MIT Election Data and Science Lab “Learning from Elections” program and from the UC San Diego Yankelovich Center.

Author Contributions

M.L., J.G., S.J.H., T.K., M.R., and L.U. conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing—original draft, project administration, writing—review & editing.

Data Availability

All data necessary to replicate these analyses will be available upon publication.

Author notes

Competing Interest: The authors declare no competing interests.