Can Addressing Integrity Concerns about Mail Balloting Increase Turnout? Results from a Large-Scale Field Experiment in the 2020 Presidential Election

Abstract The 2020 presidential election brought expanded vote-by-mail opportunities, a rise in attacks on this process’s integrity, and the implementation of novel programs such as California’s Where’s My Ballot? system to ensure confidence in mail balloting. Can heightening awareness of this ballot-tracking system and other election protections alleviate fraud concerns and raise turnout? We assess whether messages reinforcing election integrity increased participation in the 2020 election through a large-scale voter mobilization field experiment. California registrants were mailed a letter that described either existing safeguards to prevent vote-by-mail fraud or the ability to track one’s ballot and ensure that it was counted. Analysis of state voter records reveals that neither message increased turnout over a simple election reminder or even no contact, even among subgroups where larger effects might be expected. In the context of a high-profile, high-turnout presidential election, assurances about ballot and electoral integrity did not increase turnout.

turnout. 1 To counter such concerns, election officials in California set in place a robust "Where's My Ballot?" program and other measures designed to ensure confidence and promote participation in the vote-by-mail (VBM) process.Does raising awareness of this non-partisan ballot-tracking system and other election protections alleviate fraud concerns and lead to higher levels of turnout?Or were attitudes about election integrity and intentions to vote fixed so firmly in 2020 that registered voters did not react to information about election protections?
We report on the results of a large-scale voter mobilization field experiment designed to test the impact of information about election integrity protections on turnout.The study, conducted in the context of the November 2020 presidential election, used messages to address two primary concerns raised about the expansion of VBM: that one's ballot would not be counted (or even reach the elections office) and that the process itself makes it easier for others to commit voter fraud.We examine whether providing assurances that each vote will be counted accurately, as well as providing information about and an invitation to track one's ballot, increases participation (presumably by increasing perceptions that the system is not rigged).
We find that the provision of either type of information had no impact on the decision to vote (compared to both a simple election reminder and even no contact).This is true for the sample as a whole, as well as for population subgroups that theory would suggest should be most influenced by this information.We discuss the interpretation and implications of these findings in the conclusion.

Prior research
As many states expanded access to mail ballot voting in response to the COVID-19 pandemic, then-President Trump charged that the reform created "tremendous potential for voter fraud,"2 amplifying the considerable partisan disagreement about the prevalence of voter fraud and the extent to which making voting easier opens avenues for it to occur (Levy 2021;Stewart III et al. 2016;Wilson and Brewer 2013).Those concerns contributed to a widening gulf between Republicans and Democrats in their views on expanding mail voting, a stark change as mail voting had largely avoided the partisan divisions that exist over other election reforms facilitating ballot access (Bowler and Donovan 2018;Clinton et al. 2022;Lockhart et al. 2020).Lack of experience with this vote mode and lower confidence among those who cast mail ballots (compared to in-person voters) that their ballots would be counted (Alvarez et al. 2021) may have contributed to apprehensiveness about the security and integrity of voting by mail.More broadly, exposure to voter fraud claims can reduce one's confidence in electoral integrity (Berlinski et al. 2021), and 1 https://www.washingtonpost.com/politics/republicans-race-to-promote-mail-voting-as-trumps-attacksdiscourage-his-own-supporters-from-embracing-the-practice/2020/08/03/9dd1d988-d1d9-11ea-9038-af089b63ac21_story.html.The fear that fraud claims would reduce Republican turnout carried over into the Georgia Senate runoffs (https://www.cnn.com/2021/10/14/politics/republicans-fear-trump-2022-2024/index.html)and California's 2021 recall election (https://timesofsandiego.com/politics/2021/09/03/ california-gop-now-worried-about-low-recall-turnout-due-to-ballot-fraud-claims/).lower levels of confidence in the electoral process may be associated with lower turnout (Alvarez et al. 2008).Furthermore, comparative studies explicitly link perceptions of electoral integrity and the propensity to vote (e.g., Birch 2010), though Stewart III et al. (2016) find little correlation between perceptions of the frequency of voter fraud and intended or self-reported turnout in the USA in years prior to Trump's election.
To increase confidence in the security of the voting process generally and voting by mail in particular, election officials in many states put in place information campaigns and new procedures, such as California's Where's My Ballot system, which allow voters to track their ballots' progress through the mail and counting process.Can such programs effectively counter vote fraud claims and make potential voters confident enough to participate?Despite the prevalence of beliefs about election fraud for at least the past decade, we know little about how outreach on the integrity of electoral processes influences the decision to vote.Although some survey experimental work suggests that alleviating these concerns will not be easy (e.g., Berlinski et al. 2021), two sets of literatures suggest that messaging on these issues may increase turnout.First, several field experiments encouraging people to vote using a particular method have successfully persuaded registrants to switch their mode of voting and sometimes increased turnout.These include, for example, efforts to increase absentee ballot use by providing information about its availability, sending registrants a mail ballot application, or mailing them pre-filled applications (Herrnson et al. 2019;Mann and Mayhew 2015;Monroe and Sylvester 2011).They also include efforts to get registrants to drop their VBM ballots off early before Election Day (Menger and Stein 2018).Finally, a push to get registrants to apply for and cast a mail ballot in the 2020 Pennsylvania primary by referencing the safety of this method increased rates of both activities (Hopkins et al. 2021).These efforts, by both governments and non-governmental groups, are admittedly very different from our effort to address the relationship between voting methods and the integrity of one's vote or the election as a whole.They are nonetheless encouraging for our study, however, because they demonstrate that outreach can affect how people vote.
The second relevant body of work relates to a concern some registrants have about the electoral process: the secrecy of their vote.Gerber et al. (2013a) find that a large percentage of individuals do not believe that their ballots are kept secret and that instead people can (or will) find out for whom they voted.This belief is particularly prevalent among those who have not previously voted (Gerber et al. 2013b).Field experimental work shows that a message from government officials to such individuals providing assurances about the secrecy of their vote has a sizable effect on their propensity to vote (Gerber et al. 2013b).Similar contact from non-governmental organizations has a smaller but still significant positive effect (Gerber et al. 2014(Gerber et al. , 2018)).In addition, messages from both types of senders demonstrate some evidence of persistence beyond the specific election in question (Gerber et al. 2014).This body of research suggests that voter concerns over the integrity of the election process have the potential to reduce turnout and that addressing these concerns may increase turnout.Whether this effect translates to other concerns about election integrity remains to be addressed.

Field experiment design
We conducted our field experiment in the state of California in partnership with California Common Cause.Mail voting rates are generally high in California: 57.8% of those who cast a ballot in the 2016 presidential election did so in this way, while 65.3% of those who voted in 2018 did so.In response to the COVID pandemic, for the 2020 presidential election all registered voters were mailed a ballot with a pre-paid envelope they could use to return it, unlike in previous elections where voters had to request a mail ballot and pay their own postage.Other voting options included drop-off locations and drop boxes for a VBM ballot, as well as the ability to vote early or on Election Day in-person at polling locations or vote centers (depending on the county).86.7% of all voters in this contest cast a VBM ballot, though this designation pertains to VBM ballots returned in any manner, including in-person, and even ballots cast in-person in some counties. 3e obtained a random sample of active California registrants from Political Data Inc. (PDI), a third-party vendor that maintains a statewide database of registrants.Our sample came from a population that excluded households with more than four registrants and included only one registrant per household (to avoid potential spillover effects within the household).Two separate, random samples were drawn by PDI: one for an uncontacted control group (3 million records) and one for those to be assigned to our treatment groups (275,000 records).Using the same process for both samples, we first removed incomplete, invalid, and out-of-country addresses.This left 2,763,345 registrants in the control group and 253,308 registrants in the treatment sample.We then removed at random 35,000 records from the treatment sample for an unrelated study.The remaining treatment sample cases were randomly assigned with equal probability to one of three treatment groups.After assignment, we identified 23 individuals who appeared in both the control and treatment samples (two of which were among the unrelated study's 35,000 records).We removed these registrants from the study.The final control group consists of 2,763,322 registrants; treatment sample sizes are reported below. 4hose in the treatment groups were sent a letter which was delivered roughly two weeks before Election Day and arrived after all registered voters were sent their VBM ballots. 5All letters used the letterhead of California Common Cause, a non-partisan, non-profit group, as well as its bulk mail non-profit permit, though the envelopes deliberately did not bear the group's logo or return address. 6he first treatment group was sent a reminder message (n = 72,763) that mirrored standard get-out-the-vote outreach.This mailer reminded recipients of the date of the election, noted that they would receive a vote-by-mail ballot, reported the options to drop off the ballot prior to Election Day, stated the times the polls were open on Election Day, and provided the phone number and website for the Secretary of State's office.
The second treatment group was sent the tracking message (n = 72,764), which added to the reminder mailing text information designed to emphasize that voters could verify that their ballot was received and counted.The letter focused on a description of the "Where's My Ballot?" service provided by the Secretary of State's office to track one's ballot.The tool was intended to increase confidence in VBM among voters worried their ballots would be lost in transit before being counted.The outreach noted the ability to track one's ballot regardless of how it was returned, receive confirmation (via email, text, or voice call) that their ballot was received and accepted, and that if their ballot was rejected for some reason they would receive information on why this occurred and what they could do to have it accepted.The language was mainly drawn from the Secretary of State's website to test the official purpose of the program and to increase the external validity of the treatment.Specifically: Our elected officials want to ensure that you know where your ballot is along with its status every step of the way.The California Secretary of State now offers Where's My Ballot?a new way for voters to track and receive notifications on the status of their vote-by-mail ballot.
• Sign up at WheresMyBallot.sos.ca.gov to receive automatic email, SMS (text), or voice call notifications about your ballot.It's quick and easy!• You can mail your ballot in, postage paid, or you can return it at a vote-by-mail ballot drop-off location.Either way, you can still track it using WheresMyBallot.sos.ca.gov.• You will receive confirmation that your ballot was received and accepted.All mailed-in ballots postmarked by Election Day that arrive within 17 days of the election are counted.• If your ballot is not accepted for some reason, you will receive notification of why your ballot was not accepted and instructions for steps to take in order to have it accepted.
The third treatment group was sent the integrity message (n = 72,760) that added to the reminder mailing text a description of and information about how the election process and election officials maintain the integrity of the vote.These points were taken from the Secretary of State's website.They were designed to address prominent politicians' claims that ineligible people would receive ballots, that fake ballots could be produced and cast, and that someone might steal another registrant's ballot and then cast it themselves.The letter stated: Our elected officials want you to know that the election's integrity is protected.California's Secretary of State has stated that "Voter fraud in California and across the country is exceedingly rare and that tells us that our current safeguards are working."These protections include: • Vote-by-mail ballots are sent only to active, registered voters.These are individuals who have met the requirements to demonstrate their eligibility to vote.• Each vote-by-mail ballot envelope has a unique barcode associated with a specific registered voter.This ensures someone cannot print and cast fraudulent ballots.• Election officials check the signature on a returned vote-by-mail ballot envelope against the signature they have on file.If those signatures do not match or if the ballot envelope is not signed, then that ballot is not counted.
After the election, we received turnout information from PDI (taken from post-election county voter files) that we merged to our original file to identify which individuals did and did not vote.Those no longer registered were coded as having not voted (because the treatment mailing could have affected their probability of subsequent removal from the rolls, we did not drop these individuals).

Results
The average participation rate for registrants in the control condition (the group that received no mailing) was 75.47%.In comparison, 75.47% of those in the reminder condition voted, 75.53% of those in the tracking condition voted, and 75.46% of those in the integrity condition voted.As such, all conditions are characterized by extremely high-turnout rates and substantively small, statistically insignificant differences from the other groups.Table 1 reports a series of OLS regression models where the dependent variable is whether the registrant voted (coded yes = 1, no = 0). 7The column (1) model includes only dichotomous indicators for treatment assignment (the no contact control is the out group).The estimated treatment effects are 0.01 percentage points for the reminder letter, 0.07 percentage points for the tracking letter, and 0.00 percentage points for the integrity letter.8These are reasonably precisely estimated, with the top end of the confidence intervals all well under one percentage point turnout effects.These estimates change slightly, but remain substantively similar when we include available individual-level covariates in the model in column 7 Full model results appear in Table S1.Logit specifications yield similar results (see Table S2).OLS results are presented here for ease of interpretation.Can Addressing Integrity Concerns About Mail Balloting Increase Turnout?
(2) to account for potential sources of unobserved heterogeneity or also include fixed effects for county in column (3). 9 The rest of the table looks at the possibility of heterogeneous effects in different subgroups.Previous research shows that get-out-the-vote messages often produce different effects across different groups of the population (Enos et al. 2014).We use interaction terms to test whether this is the case in our context.In the column (4) specification, we compare the treatment effects for Non-Latino Whites to those for African Americans, Latinos, and Asian Americans. 10None of the treatment effects are statistically significant overall for Whites.Linear combination of coefficients tests (reported in the replication materials, Biggers et al. 2022) reveals that none of the treatments increased turnout for any minority group except the tracking letter for Asian Americans.That effect is 1.3 points (p < 0.01, two-tailed) compared to no contact and 1.2 points (p = 0.06, two-tailed) compared to the reminder letter.
In columns ( 5) and ( 6), we assess the effects of the treatments for Republicans compared to the rest of the electorate and Independents, respectively.None of the treatment effects are statistically significant, nor are they for Democrats or Independents. 11We also test whether, like ballot secrecy messages, our treatments might be more effective for those who have never previously voted or who have never previously cast a vote-by-mail ballot.Analyses in Table S3 of the SI provide no evidence for either expectation. 12

Conclusion
The record availability of mail balloting prior to the 2020 presidential election sparked considerable backlash from some quarters about how this increased ease of participation might facilitate voter fraud.Drawing on prior work that 9 Individual-level covariates include age, year of registration, gender, race/ethnicity, partisanship, and vote history (whether the registrant voted in the 2020 primary election and the 2012, 2014, 2016, and 2018 primary and general elections).In the models reported in columns (2) and (3) of Table 1, voting in the 2012, 2014, and 2016 primaries is negatively associated with voting in the 2020 general election.These relationships appear to be explained by the model's saturation with respect to vote histories.Voters who did not vote in recent elections but did so in earlier contests are likely habitual voters who stopped voting for some reason.As such, when we control for recent vote history those older vote histories indicate former voters.As a robustness check, we re-ran the column (3) model excluding the most recent vote history variables (for the 2020 primary and 2018 primary and general elections).Consistent with this explanation, the coefficients for the primary election variables are all positive and statistically significant at conventional levels.10 Race/ethnicity is not self-reported in the state voter file.Instead, PDI estimates registrant race/ethnicity based on (1) surname (the extent to which it is unique to a particular group), (2) birthplace (if another country, which is in the state file), and (3) reported language preference (registrants can receive their ballot/printed materials in a few different languages, and if they request it in a language other than English that is recorded in the voter file).The category "White" corresponds to those not identified as African American, Latino, or Asian-American.These results are robust to separating minor party members from Independents (see the replication materials, Biggers et al. 2022).Additional analyses (not reported) show that none of the additional individual-level factors included in the model condition the effect of any of the treatments.
demonstrates the ability of outreach to both affect how people cast their ballots and assuage apprehensiveness about a different electoral integrity consideration (whether one's vote choice is secret), we conducted a large-scale voter mobilization field experiment designed to address concerns that the use of mail balloting would lead to one's vote not being counted and widespread fraud that would challenge the election's validity.Despite previous studies suggesting the potential utility of this contact, neither of the messages we tested exerted a meaningful effect on turnout.This is the case both for the entire sample and population subgroups anticipated to be more concerned about integrity issues and/or more responsive to this information.
As with any field experiment, the context in which we tested these messages may have influenced their effectiveness.For example, increasing turnout in field experiments conducted during presidential elections is generally more difficult (compared to midterm, primary, or local-level contests), due to both the smaller number of nonvoters and the fact that many in the control and reminder groups are likely "treated" by other outreach efforts (see e.g., Green and Gerber 2019).It is also possible that the stakes of the presidential election were perceived by registrants to be so high that they still voted despite any concerns about the integrity and security of the voting process and the election as a whole.Such messaging may thus be more effective in lower-salience elections in which the perceived stakes of the races are insufficient to outweigh concerns about the integrity of the process.
Alternatively, it may be the case that messages about electoral integrity need to come from official government sources to be persuasive.As noted above, while outreach regarding the secrecy of one's vote can increase turnout when sent from non-governmental groups, this information has a greater participatory impact when sent from a government office.As such, receiving this information from California Common Cause might not have been seen as authoritative enough to assuage integrity concerns for those who held them.It is also possible that stronger treatments might be needed to persuade voters of the security of the election process when faced with prominent and repeated criticisms from politicians.Third, it is possible that combining the treatment messages with the registrant's personal vote history, especially if that history includes prior use of a mail ballot, might be more impactful (Gerber et al. 2008;Panagopoulos 2011).Finally, it is possible that such messages may be effective in future elections given the considerable backlash against vote-bymail following the election by some prominent politicians (including President Trump) and their claims that it contributed to a fraudulently stolen election.Alleviating concerns about electoral integrity may be even more important moving forward, though whether outreach of this nature can do so, regardless of the source, is certainly unclear.Given the central role of President Trump in driving these doubts, it is possible that only an endorsement of electoral integrity by him would positively affect perceptions for those predisposed to abstain over such concerns.That no such endorsement is likely forthcoming in the near future, however, means that these types of messages present the only realistic manner for political actors to try to reach those with concerns about the electoral process and persuade them of the robustness of protections in place.
Additional work is necessary to clarify these matters.That said, this study provides little evidence that the messages about the electoral process tested here increased turnout in the context of a highly salient, high-turnout presidential election. 11

Table 1
Effect of mail treatment about electoral integrity on turnout