Election Day is imminent. Here is a quick analysis of what to expect once the polls close.
- The election is likely to be turbulent, and there may be prolonged uncertainty regarding the winner.
- During a potential delay in determining the winner on Election Day, there are likely to be strong claims made about the accuracy of the election—possibly including the claim that large numbers of noncitizens voted—along with vigorous litigation.
- State-run audits consistently show that cases of noncitizen voting are exceedingly rare, while related claims based on survey data overestimate the phenomenon due to measurement error.
- AI-generated content and statistical analyses may fuel narratives about election illegitimacy.
- These claims matter, because countries that fail to secure widespread buy-in to the legitimacy of their elections risk suffering a long-run loss of political legitimacy that erodes the rule of law.
- Together, these factors may create substantial uncertainty regarding the outcome, which in turn may cause significant economic and policy uncertainty for businesses and the economy.
- In this brief, we explain the sources of this uncertainty, and seek to equip readers with information regarding likely claims about the election’s accuracy so that they can be prepared to navigate a chaotic information environment surrounding a difficult-to-predict election cycle
Suspense and surprises
Elections are particularly difficult to forecast these days. It’s possible either Kamala Harris or Donald Trump will end up winning by a meaningful margin and that we’ll know the winner soon after polls close on November 5. But if current trends persist, the election could be very close—and close elections can take much longer to resolve.
If the election is close, it will come down to a small number of battleground states, and ballots can take a while to count in these states. In Pennsylvania, for example, it could take a week or longer to determine the winner. In 2020, the winner wasn’t announced until four days after the election.
And if we continue to see a strong partisan split in voting mode—with Republicans more often voting in person and Democrats more often voting by mail—then we could again see a “red mirage” or a “blue wave” in battleground states where Trump appears to lead by a wide margin based on quickly counted in-person votes, but can potentially lose ground as mail-in ballots are processed.
The combination of a likely delay in determining the winner, combined with this possible red mirage, creates a potentially fraught period of time during which claims and counterclaims about the fairness of the election and its likely outcome will circulate.
In anticipation of a likely debate about the fairness of the election, we can review evidence for some of the key claims made about American elections with an eye toward the specific arguments most likely to arise on or just after Election Day.
Do noncitizens vote?
Claims about noncitizen voting have drawn the most attention ahead of the 2024 election, with prominent figures including former president Trump and Elon Musk claiming that noncitizens will illegally influence the election’s outcome. What evidence is there for this claim?
The most reliable evidence on noncitizen voting comes from state audits, which draw on cross-referenced administrative data not accessible to external reviewers. These audits typically involve matching voter registration records with records from state departments of motor vehicles (DMV) and the US Department of Homeland Security’s Systematic Alien Verification for Entitlements (SAVE) program, a system designed to determine the eligibility of noncitizens for public benefit programs. Registrations flagged through this process are typically required to provide proof of citizenship or have their registration canceled.
In recent years, numerous states have conducted audits of this kind, the results of which tend to suggest that cases of noncitizen voter registration and voting are extraordinarily rare. For instance, a 2012 audit conducted by the Florida Department of State identified 207 noncitizens on Florida’s voter rolls, or .0018 percent of the state’s 11.75 million registered voters. Likewise, a 2022 audit by the Georgia Secretary of State found that 1,634 people attempted to register to vote despite not being US citizens. Noncitizens who become registered to vote, however, do not necessarily cast a ballot. An audit in North Carolina, for example, revealed that 41 noncitizens cast ballots in the 2016 election, or .0006 percent of the state’s 6.5 million registered voters.
State audits face two significant challenges, however, that can lead to false-positive matches and inflate the appearance of noncitizen voting. First, these audits often lack unique identifiers, such as Social Security numbers, and rely instead on names and birthdates for matches. If a US citizen happens to share a name and birthdate with a noncitizen, their registration may be incorrectly flagged as illegitimate. (Even in small populations, matches of this kind can arise with surprising frequency. In a sample of 100 people with the same name, for example, there is a roughly 20 percent chance that at least two individuals will share the exact same birthdate, including the year).
A second concern is data recency. In 2023, 878,500 individuals became naturalized American citizens, and these individuals often do not immediately update their citizenship status with the DMV, leading to outdated matches. Evidence from recent state audits suggests that these concerns about false-positive matches are important. In North Carolina’s audit referenced above, 97.6 percent of registrations initially flagged as illegitimate were later confirmed as legitimate, and a 2024 Alabama audit similarly saw nearly 25 percent of canceled registrations ultimately reinstated because they were associated with US citizens.
A less reliable form of evidence of noncitizen voting comes from surveys of the American public. These surveys are unreliable because they are based on the responses of only a handful of respondents. For example, in a peer-reviewed publication in 2014, Jesse Richman, Gulshan Chattha, and David Earnest attempted to estimate the rate at which noncitizens vote in US elections. The researchers used data from the Cooperative Congressional Election Study (CCES), which asked respondents “Are you a United States citizen?” Then, among those individuals who said no, Richman and colleagues found that, in 2008, eleven respondents who identified as noncitizens were registered to vote and matched a voter file registration—constituting approximately 3.3 percent of the noncitizen respondents in the sample. Among this sample of eleven respondents, five individuals reported voting.
These estimates were almost immediately criticized. In a paper published in the same journal, Electoral Studies, Stephen Ansolabehere, Samantha Luks, and Brian Schaffner (2015) explained that “Richman et al. (2014) . . . presents a biased estimate of the rate at which noncitizens voted in recent elections.” They went on to explain that “the results, we show, are completely accounted for by very low frequency measurement error; further, the likely percent of noncitizen voters in recent US elections is zero.”
To see why small amounts of measurement cause large problems for the measurement of noncitizens in a survey, we can consider a simple hypothetical arithmetic example, which we have modified slightly from an example in Ansolabehere, Luks, and Schaffner (2015). Suppose that we have a survey of 100,000 individuals. The sample contains 99,500 individuals who are citizens and there is an 80 percent registration rate among these individuals. We will also suppose the survey has 500 individuals who are noncitizens and exactly none of these noncitizens is registered to vote. Now, suppose that, when completing the survey, one out of every 1,000 citizen respondents erroneously reports that he or she is a noncitizen. This results in approximately 100 citizens who incorrectly reported that they are noncitizens appearing in the survey, of whom approximately 80 are registered to vote. Even if all noncitizens correctly report their noncitizen status and that they aren’t registered to vote, we will obtain a very badly biased estimate of the rate of noncitizen registration. In this simple example, the survey would find that 13.3 percent of noncitizens are registered to vote, when in reality none of those noncitizens are registered to vote.
Using the same CCES data employed by Richman et al., we find that measurement error of this magnitude is entirely plausible. Specifically, we find that 1.3 percent of respondents change their citizenship status across the three waves of the panel. This small amount of measurement error is more than enough to explain the estimates reported in public surveys. This makes surveys an unreliable tool for gauging noncitizen participation in the electoral process.
Do dead people vote?
Another common claim that is much easier to test is that large numbers of dead people are voting. A particular version of this claim that has become popular since 2020 is that mail-in ballots are sent to dead people still on the election rolls, are collected by adversarial actors, and are cast fraudulently.
Whether a person dies is a matter of public record, as is whether a person turns out to vote. States already use these records to clean their voter rolls and to prevent fraudulent voting, and we can use this public data to test how well they do. In research forthcoming in the Election Law Journal, we test this claim in the state of Washington, which runs universal vote-by-mail elections in which everyone receives a mail ballot, supposedly where this fraud is most rife.
Once we account for false positives (people who share the same name, and sometimes even the same birthday), we find no meaningful evidence of votes being cast on behalf of the deceased. Although this cannot conclusively rule out dead people’s ballots being voted in other states or in 2024, these claims must be viewed with extreme skepticism. At a minimum, any claim about votes cast for a dead person would need to explain how name and birthdate duplicates are being handled, as it is easy to find false positives without accounting for this issue.
Do nonresidents vote?
A final, popular claim since 2020 has been that large numbers of voters are illegally voting absentee in states where they no longer reside. In Georgia, a report submitted on behalf of the Trump campaign alleged that 10,651 votes were cast from individuals who had changed addresses with the national change-of-address form. Lawsuits in 2020 and in 2024 in Nevada both use the national change-of-address form as a basis for questioning the legitimacy of some voters’ registration.
We examined this particular claim in detail. Merely filing a change of address is not sufficient evidence to cancel voters’ registration in their home state. Voters can change where their mail is delivered because they are attending college, vacationing at a second home, or serving in the military and deployed at a base out of state. It turns out that it is exceedingly difficult to establish which of these voters are or are not legally counted as “residents” in the state of Georgia or Nevada. While this means that we cannot conclusively establish that no meaningful number of nonresidents voted in 2020 in Georgia, it does mean that we have no evidence that many did.
This year in Nevada, thousands of registrations have been challenged based on registered voters submitting a change of address. Yet, many of those challenges are actually voters who have been deployed in the military or are having mail redirected to a post office box, and some challenges are just based on a simple error.
AI, deepfakes, and the information environment
Thus far, AI-generated “deepfakes” have not played a significant role in the election. However, in the midst of uncertainty over the election’s winner, there may be a unique opportunity for people to promulgate salacious claims about election administration with the help of artificial intelligence. We do not view this as a problem to be solved through censorship, but rather as an opportunity to equip people with good information so that they become informed and skeptical consumers of online information about election administration.
Here are some examples of the kinds of fake videos, images, or audio we imagine may circulate:
- Election workers conspiring to alter vote outcomes
- Noncitizens voting
- Republican officials or voters intimidating would-be Democratic voters
The solution is not to try to prevent the circulation of this content—particularly because if there are issues related to the election, we want to learn about them—but rather to keep in mind that if videos, images, or audio files are unsourced and not corroborated with other pieces of information, we can no longer trust that they reflect anything that actually happened, thanks to AI.
It is also likely that election skeptics will circulate statistical analyses purporting to demonstrate fraud, and may appeal to generative AI to inflate the credibility of these analyses.
This is not idle speculation. Just recently, the Washoe County election board in Nevada temporarily refused to certify a primary election because of a ChatGPT statistical analysis that purported to show election fraud. It turned out that ChatGPT had actually concocted the election data used to conduct the analysis, and the board eventually certified the election.
While we applaud the trend of citizens engaging with our system of election administration and using open data to study it, we encourage everyone to apply skepticism to analyses, whether done with ChatGPT or not, that make wild logical leaps or purport to demonstrate strong evidence of voter fraud. While such analyses could be correct, and deserve our scrutiny given the stakes involved, our experience is that the vast majority of these analyses circulating in the wild today are grossly misinformed.
Insist on evidence
The 2024 US election will have huge consequences for society and the economy. Separately from the policies pursued by the winner, though, the election itself may induce a period of significant uncertainty for the country. If the election proves to be as close as is currently predicted, we may face a week or longer before we know who the winner is. During that delay, we will hear many claims from actors across the ideological spectrum about the fairness of the election and the true winner.
Countries that fail to secure widespread buy-in to the legitimacy of their elections risk suffering a long-run loss of political legitimacy that erodes the rule of law. Hence, we need to be prepared for a period of substantial scrutiny around our elections. We need to be prepared to identify and mitigate real issues that may arise, while also dismissing bad-faith claims about electoral legitimacy that do not provide compelling evidence.
In this brief, we have sought to arm people with knowledge about some of the key claims that are likely to arise, especially around noncitizen voting, dead people voting, nonresidents voting, and ChatGPT-generated analyses of election data.
Our position is not that there can never be voter fraud or that people should not investigate it—indeed, we have devoted substantial amounts of time to looking for evidence of fraud ourselves—but rather that we should demand that these claims be specific, falsifiable, and testable with data. We should approach the 2024 election and this period of uncertainty with an open-minded but rigorous attitude.
This article was originally published on the Hoover Institution website HERE.
Andy Hall is a senior fellow at the Hoover Institution and a Professor of Political Economy at the Stanford Graduate School of Business.
Andrew Myers is a PhD candidate at Stanford University with an interest in American politics and political methodology.
No comments:
Post a Comment