This article is made possible through Spotlight PA’s collaboration with Votebeat, a nonpartisan news organization covering local election administration and voting. Sign up for Votebeat's free newsletters here.
In April 2025, the Election Truth Alliance was a three-month-old Nevada nonprofit with a growing audience in anti-Trump online spaces and a claim that something about the 2024 presidential election did not add up.
The group had released an analysis of the results in Pennsylvania and sent letters to officials there and in Wisconsin asking them to hand-check the results of multiple elections. Now, it wanted to work with Walter Mebane Jr., a tenured University of Michigan political scientist and one of the few academics who had spent decades trying to detect the statistical fingerprints of election fraud.
Mebane was skeptical.
“The name of the group didn't really inspire confidence,” he told Votebeat. “I was thinking, ‘Oh my god, truthers.’” But Mebane thought the data would be useful for his students and a book he was working on, and by early May, he had produced a working paper analyzing Pennsylvania’s 2024 presidential results. In a note, he credited the ETA for providing the data.
His findings estimated the number of potentially fraudulent votes exceeded President Donald Trump’s margin of victory in Pennsylvania. In the working paper, Mebane noted his findings could also be attributed to factors other than intentional fraud.
The ETA has described Mebane’s work as a “collaboration” that confirmed its conclusions. Mebane described it differently: He said he had no interest in the group’s advocacy and engaged because he believed it had unique data on Pennsylvania. He told Votebeat he did not know the same data was available from the state.
“He is kind of the go-to guy for election forensics in the U.S. He’s the leading expert,” ETA co-founder Nathan Taylor said in a video posted in May 2025 announcing the findings validated their Pennsylvania concerns. Toward the end of the video, a QR code to the group’s donation page flashed on screen. The ETA urged supporters to contact Pennsylvania officials and advocate for hand-count audits, and they heeded the call, sending state and local officials emails that cited Mebane’s work to allege the results were flawed.
In a June 2025 press release, the ETA said Mebane’s findings corroborated its own. “BREAKING,” the group wrote. “Election Forensics Expert Finds Vote Manipulation Concerns in Pennsylvania.
In November, the ETA filed a lawsuit against the Secretary of the Commonwealth of Pennsylvania and three counties questioning the 2024 count and asking the district court to require the counties to “hand count physical ballots.” The group cited its own analysis but relied on Mebane’s findings to suggest abnormalities in Allegheny County.
The ETA’s leaders acknowledge that statistical analysis alone can’t prove fraud. But they say the anomalies they raise warrant independent validation of the election results.
The Pennsylvania paper Mebane gave the ETA did not fully explain the statistical model he used to identify potentially fraudulent votes, which he calls “eforensics.” Mebane later told Votebeat he had “decided not to write it in a way that would make it self contained” and instead referred the ETA to his past work for details on his methodology.
The ETA’s promotion of Mebane’s original working paper and his credentials caught the attention of a network of statisticians and political scientists accustomed to dense models, technical assumptions, and arguments over what statistical analysis of election data can really show.
Justin Grimmer, a Stanford political scientist whose 2021 peer-reviewed paper debunked statistical fraud claims about the 2020 election, said Mebane’s work has been used by the ETA in ways that give unusual weight to its claims about elections, which could otherwise be more easily dismissed as fringe. The result, he added, is a “perfect storm” — a motivated interest group armed with a professor’s academic credentials.
He said giving the original analysis to the ETA was like “handing someone a stick of dynamite and being surprised when you walk away and it detonates.”
As the ETA's profile rose and its supporters trumpeted Mebane’s findings, Votebeat consulted 10 election and statistics experts about his original Pennsylvania working paper and the statistical model Mebane terms “eforensics.” All said they doubt the model can accurately detect fraud in U.S. elections.
A different kind of election denial
Claims of election fraud and statistical anomalies are now a regular feature of American elections — often blending data, analysis by citizen researchers, and the raw emotions of voters who do not want to believe their candidate truly lost. After 2020, when Trump and his allies waged an unprecedented effort to overturn the results, such efforts became closely associated with the political right.
The ETA, which was formed as a nonpartisan nonprofit, has tried to distinguish itself from those movements. “This is not election denial. This is election integrity and auditing,” the ETA’s Taylor told David Pakman, host of the progressive online program The David Pakman Show, which has more than 3.6 million YouTube subscribers. Pakman described the 2025 segment as “one of the most requested interviews” in months.
Mebane’s work has fueled public questions about elections before: In South Korea, politicians and activists cited his analysis to allege possible election fraud in 2020. Political scientists later found such claims were flawed and damaged voter confidence.
>>READ MORE: How an American professor’s contested work was used to fuel election doubts in South Korea
Now, the ETA is using his work to raise questions about election results in the U.S. In addition to Pennsylvania, the group posted a summary of an unpublished analysis by Mebane of 2024 results in North Carolina, and it has asserted that the fact that his analysis corroborated its findings in Pennsylvania validates its analyses of other states.
“In collaboration with ETA, Dr. Walter Mebane analyzed Pennsylvania vote data for the 2024 November Election and found evidence of potential election fraud, confirming ETA’s analysis of the same Pennsylvania vote data,” an attorney for the ETA wrote in an analysis sent to California election officials, threatening litigation if the state refused to act.
“Consistent with techniques used to analyze Pennsylvania’s vote data,” the ETA said it had identified “evidence consistent with vote manipulation, including ballot-stuffing, deleting votes, and using algorithms to alter tabulator vote counts” in some California counties.
ETA leaders later told Votebeat the group’s lawyer’s use of “confirming” in the letter was “less nuanced” than their own statements, but that they stood by their use of "corroborates,” since, they said, both analyses reached similar conclusions.
The ETA has also sent letters to officials in at least six states, including Minnesota, Michigan and Kentucky, suggesting it has found abnormalities in the 2024 election results or in subsequent ones.
The letters urge election officials to take steps, such as investigating technology used in the elections and conducting hand counts, to verify the results. As of Sept. 2, Pennsylvania is so far the only state where the ETA has sued. A judge recently dismissed the case but left the door open for it to be refiled. When the group resubmitted its complaint in August, it made no mention of Mebane or his analysis. In an interview with Votebeat in April, Taylor said the ETA planned to file suit in Florida next.
The nonprofit is run by three people who came together on Reddit after the 2024 election, none of whom have an apparent background in U.S. elections. Taylor, 27, has a background in information technology; his two co-founders are a U.K.-trained financial professional now living in Arizona and a Canadian policy analyst with experience coordinating resources for British Columbia’s election authority.
The group raises money to support its work, including through donations and selling merchandise such as a “The Math Ain’t Math’n” T-shirt and a women’s sleep shirt reading “Vote Once, Count Twice, Sleep Tight.” Public records show it reported raising more than $422,000 in 2025.
While Mebane said he worked with the group, he told Votebeat that he hasn’t spoken with its founders in months. He told Votebeat that he saw “many, many things” in the ETA’s explanations that he “didn’t think were sound interpretations” of his work. But he said he didn’t want to get into a public debate with them, nor would he detail his specific concerns to Votebeat. He said fact-checking online claims is a distraction from his primary professional focus, a long-gestating book on his eforensics work.
Mebane stands by his model as valid but also said he does not see it as his responsibility to ensure the public understands his work. He stressed that his work has been done transparently and in public through working papers that can be revised.
At the same time, he dismisses the criticism from other academics, saying the experts consulted by Votebeat lack key components he has “deliberately not shown.” Grimmer and University of Pennsylvania political scientist Marc Meredith produced written reviews of Mebane’s 2024 Pennsylvania working paper at Votebeat’s request. Mebane described Grimmer’s critique as “just wrong” because “these views that you've solicited here are not based on the whole case that I'm making.” In response to questions, he frequently cited his planned book, saying the answers will ultimately be found there.
From butterfly ballots to eforensics
Mebane described his childhood as that of an inquisitive Black boy coming of age in the Civil Rights Era South, in a household where his parents, both Tuskegee graduates, emphasized the importance of social justice.
“That's why I cared about elections,” Mebane, 67, said in an April interview on the University of Michigan’s Ann Arbor campus. It seemed to him to be “the one way that wasn’t violent” that could “change who has control of some levers of how decisions get made.”
When he spoke with Votebeat, he sat in a sparsely furnished office at the university’s Institute for Social Research — a space he uses for meetings. He described his primary workspace as “grimy” and packed with books. He wore a faded T-shirt featuring a tarot-style image labeled “The Wanderer,” a symbol of a lone traveler.
Mebane said he has always been drawn to questions about human behavior. At 8, while grocery shopping with his mother, he remembers searching the cereal aisle for hidden cameras, convinced companies must be tracking what people bought to measure the effectiveness of advertisements.
Mebane, who earned his undergraduate degree at Harvard and his doctorate at Yale, was teaching government at Cornell and studying elections in 2000 when the presidential race became too close to call, plunging the country into uncertainty.
He traveled to Florida as part of a team of academic experts working pro bono on a lawsuit challenging Palm Beach County’s butterfly ballot, a confusing two-page design that led some voters to accidentally select the wrong presidential candidate. His team published their findings in the American Political Science Review.
That work was straightforward: Mebane and his colleagues started with a known irregularity — the butterfly ballots — and worked outward, estimating how many votes it may have affected. There was a specific, identifiable problem. His job was to measure it.
His approach to Pennsylvania’s 2024 election works differently.
In the eforensics framework, Mebane starts not with a known problem but with the election results themselves. He compares total votes, votes for the leading candidate, and the number of eligible voters or registrants. Using those inputs, the model looks for patterns Mebane says could be produced by fraud.
Elections are shaped by a wide range of forces, from socioeconomic conditions and demographic differences to weather, campaign spending, and the quirks of individual races. Mebane’s Pennsylvania model does not directly account for those factors, though it does adjust for broad differences in turnout and voting patterns across counties. Instead, using precinct-level vote returns, it classifies precincts as having no fraud, incremental fraud, or extreme fraud — terms Mebane says are not meant to imply intentional wrongdoing.
Critics say that is where the trouble begins. The model flags innocent variations — caused by anything from strong canvassing efforts to hurricanes — as readily as deliberate manipulation.
Mebane recognizes that the model can be affected by other factors. He says his model is unable to account for “lost votes” — votes that were never cast because voters stayed home, for example. The model also sometimes flags strategic behavior — such as a voter intentionally casting their ballot for a party they don’t typically vote for — in the same way as fraud. Critics say such phenomena are commonplace in American elections, making the model a bad fit for the U.S.
“My project is a scientific mission to see whether this estimator and this approach can measure election frauds,” Mebane told Votebeat. “I think it measures frauds in a valid way, but it's imperfect.”
Experts dispute findings and raise questions about false positives
Michael Alvarez, a political science professor at the California Institute of Technology, tested Mebane’s model with Grimmer and Meredith using simulated data built to contain no fraud. The model still found fraud at levels similar to Mebane’s Pennsylvania working paper.
Grimmer and Meredith said this kind of simulation is a basic way to pressure-test any model. They, along with Alvarez, were surprised Mebane had not done it.
In Mebane’s book manuscript, a partial copy of which he provided to Votebeat, he said he has not used simulated data tests because his preferred test is to compare the model’s results with real elections. He points to examples from France and Mexico, where courts or election authorities later annulled results or otherwise identified serious problems, and his model also found anomalies.
Had Mebane pressure-tested the model with simulated data, Meredith, Grimmer, and Alvarez said, he would have seen that the model could flag fraud even when none existed.
“Mebane’s eforensics model labels precincts as fraudulent when they show just a combination of this disproportionately high turnout and support for the winning candidate,” Meredith said at a recent academic conference. That combination was common across the country in precincts that favored Trump.
Mebane responded that such a result was “no surprise,” because an election simulated to contain no intentional fraud may still effectively simulate high mobilization or strategic behavior — forces his model may flag as eforensics frauds.
All three criticized Mebane’s use of the word “fraud,” with Meredith calling it “very irresponsible” and “frankly, not defensible.”
In his book manuscript, Mebane defines fraud as “malevolent distortions of electors’ intentions that change or potentially can change election outcomes,” even if the distortions aren’t the result of malicious or improper activity. Pressed on his use of the word “fraud,” Mebane defended it.
“I have an imperfect measure of fraud.”
If the anomalies may not be due to fraud, why use the term at all?
“Because it’s eforensics frauds.”
Why not call it an anomaly instead?
“Because it measures frauds.”
Alvarez, who co-edited a foundational academic book on election fraud, took issue with Mebane’s definition. "No one else thinks that's what the word fraud means," he said.
Mebane’s eforensics model has not been published in a peer-reviewed paper. He submitted an early version of his eforensics work to the American Political Science Review in 2020. It was rejected, and he said he does not recall the specific critique.
Nevertheless, in a press release announcing Mebane’s findings, the ETA said the model “has been validated in professional scientific publications.” The group did not specify which publications. Many political scientists, who reviewed the model at Votebeat’s request, told Votebeat that it cannot be used to support the claims the ETA is making.
Pennsylvania, a swing state that has long been the subject of voter fraud conspiracy theories, draws intense scrutiny in every election. In 2024, the state had a chief election official appointed by the losing party, bipartisan boards in nearly every county, risk-limiting audits, and smaller spot-check audits.
Meredith questions whether thousands of manufactured or stolen votes could really have slipped through all that scrutiny.
“I am concerned that the model uses fraud as the mechanism to make the model fit the data better,” Meredith wrote.
In Pennsylvania in 2024, Republican registrants turned out at higher rates than Democratic registrants in nearly every county, a pattern Meredith said is more readily explained by partisan enthusiasm than ballot stuffing.
Mebane’s working paper acknowledges this possibility, albeit seven pages in. After repeatedly asserting that his findings of fraud may exceed the 2024 margin in Pennsylvania, Mebane writes that the state is a “key battleground into which extensive and intensive campaigning and mobilization efforts were directed,” adding, “maybe most or almost all of the incremental manufactured votes are false positives prompted by electors’ strategic behaviors.”
The most conservative version of his model puts the number of estimated illegitimate votes well below Trump’s margin of victory in Pennsylvania.
In the version of the working paper provided to the ETA, Mebane described the narrower estimate as “a not negligible proportion” of Trump’s margin. Mebane has since updated that paper for his book and deleted that language, according to a revised version he provided to Votebeat in July.
Mebane also provided Votebeat with a written response to the reviews of his model from his academic peers, which highlight this limitation. In that, he says that when both malevolent distortions and strategic behaviors happen in the same election, “nothing about the model per se” can distinguish between the two.
Experts were also critical of a number of choices Mebane made in his analysis. The working paper, for example, skips steps typically standard for claims of this magnitude. It does not rigorously engage with alternative explanations, address the fact that Pennsylvania’s election results were audited, or show through simulation that the model can return a no-fraud result in a setting like Pennsylvania’s.
The Pennsylvania analysis also relies on data Mebane obtained via the ETA, even though that data was also publicly available from the state. Mebane said he did not check the data with election officials before publishing his findings, something Grimmer characterizes as a basic precaution before drawing conclusions from any model.
Grimmer and the other experts said taking that step would have helped Mebane understand and account for obvious data issues that instead distorted his findings. Votebeat separately found additional data interpretation errors in Mebane’s work. After Votebeat raised its discovery with Mebane, he acknowledged it and revised his findings of precincts labeled as “extreme frauds.”
Mebane’s updated Pennsylvania analysis also changes which counties he identifies as having significant levels of “malevolent distortions.” His original paper focused on Philadelphia and Huntingdon. The revised version provided to Votebeat instead focuses on Philadelphia and Bradford.
The new paper does not explain the change, and the earlier version — accusing Huntingdon of having high levels of fraud — remains on Mebane’s website.
‘There is an element of faith’
The ETA stressed in an email to Votebeat that its claims do not rely solely on Mebane’s working paper or model, noting that it ran multiple statistical tests “to avoid putting too much weight on any one individual method.” But some of those other tests are variations on the same basic idea as Mebane’s eforensics model.
For instance, in Pennsylvania, the group used a method developed by a Russian physicist that treats precincts with both high turnout and high support for the leading candidate as potentially fraudulent — a pattern colloquially known as a “Russian tail.” Grimmer said that kind of approach runs into the same barrier as Mebane’s model: The pattern it flags may be real, but the model cannot say why it exists.
A political methodologist and researcher at Stanford University’s Hoover Institution, Kirill Kalinin, told Votebeat that he spoke with the ETA last year and warned the group its methods could produce false positives.
The ETA’s founders also spoke with Votebeat and answered additional questions, but they did not answer a final list of questions about their methods, claims, and organization. They also did not directly respond to Grimmer and Meredith’s written critiques of Mebane’s work.
“Our team remains open to transparent, unbiased data analysis, including amending and correcting errors when they are discovered or brought to our attention,” they said in an email. “We welcome and encourage constructive criticism of our methodologies or our characterization of others’ work as we seek to ensure the accuracy and integrity of election results, and to bridge the gap between academia and the general public.”
But by both sides’ accounts, the group and the professor are no longer working together, with Mebane voicing doubts about how the group is using his work. The ETA said it repeatedly reached out to Mebane to ask if it had accurately characterized his work or should adjust its framing of it but received no response.
The ETA’s founders — Taylor, Emily Barner, and Clive Griffiths — said they are not the full extent of the ETA’s team, describing a wider network that includes members with advanced statistics training and doctorates in relevant fields, but declined to name most of them.
Barner, who goes by Lilli McGregor in her work for the ETA — which she told Votebeat is a combination of a nickname and her married name — said the group is trying to protect those members from reprisal and that supporters will ultimately have to take them at their word.
“There is an element of faith,” she said.
At least one of the founders has expressed strong views about the 2024 election. Griffiths — who goes by the nickname Jive online — said on X before the election that it was “about to be stolen.” The day after the election, he posted that “we are gonna prove” Trump cheated. After Votebeat asked about the posts, the account became private.
Mebane confirmed he also analyzed North Carolina data given to him by the group which published a summary of his findings that claims he found more than 200,000 potentially fraudulent votes. However, Mebane said he asked the ETA not to publish his full paper, leaving that summary the only public account of his findings.
Mebane said he raised his objections explicitly in “a couple of Zoom meetings.” The group disputes that, saying it only learned of his concerns secondhand, around the time of an article in The Atlantic reporting that he did not endorse their claims.
The group has also contacted other academics, including Kalinin, who was advised by Mebane as a University of Michigan doctoral student. Before graduating in 2017, Kalinin worked with Mebane on election forensics methods. Kalinin said the ETA first contacted him over the summer of 2025 and later scheduled a roughly three-hour call in October.
Over the course of the call, Kalinin said, it became clear the group was seeking advice on how to restore its relationship with Mebane following his comments to The Atlantic. He told the ETA members on the call they’d have to reach out to Mebane directly. It was in that conversation, he said, that he cautioned them about false positives.
“These are just anomalies, these are not election fraud,” he said he told them, stressing that “you need to prove it.” Without ruling out alternative explanations, he said, such patterns cannot establish wrongdoing, particularly in democratic elections where normal political behavior can produce similar signals.
Mebane told Votebeat he doesn’t have time to argue with everyone on the internet, pointing to this meme. But his reluctance to publicly address the use of his work has left a vacuum.
For example, in a July 2025 video with more than 650,000 views titled “Yes, There’s Evidence Trump Hacked the 2024 Election,” independent journalist Scott Carney cited the ETA’s interpretation of Mebane’s working paper as evidence the election may have been stolen. He acknowledged he could not fully understand the math and said Mebane didn’t answer his emailed questions.
“C’mon, Mebane!” he said. “You can’t just drop a bombshell like that and then walk away like nothing happened.”
The accountability gap
As the ETA used Mebane’s working paper for advocacy and fundraising, it reached Andrew Gelman, a professor of statistics and political science at Columbia University and a leading authority on statistical methods. It had already baffled another technically sophisticated reader, who sent it to Gelman hoping he could make sense of it.
Gelman read both the Pennsylvania working paper and the underlying eforensics methodology but found the work too opaque to fully evaluate. The only reason anyone was engaging with it, he wrote, was Mebane’s reputation. “Without his name attached to them I don’t think we’d be reading them at all.”
That combination of technical opacity and academic authority is what makes work like this hard to dislodge once it spreads, said Grimmer. Debunking work like Mebane’s requires readers to follow complex assumptions and unfamiliar methods.
Kalinin, who has spent much of his career studying elections in authoritarian countries, defended Mebane’s caveats, calling the original Pennsylvania working paper “balanced.” At the same time, he said, methods like eforensics work “well on autocracies. I’m not so sure about democracies.”
Kalinin said that he has come to believe the ETA was pursuing its own political goals and using the work of academics “as the instrument” to achieve them.
“No amount of academic caution can entirely prevent a motivated organization from stripping away those caveats to advance a specific agenda,” Kalinin said in an email. “ETA is entirely responsible for its own actions. They are clearly exploiting election forensics research for strategic or political gain – something that I cautioned them against during our private conversation.”
“It does not surprise me that ETA may have mischaracterized my work, because that's what I initially observed them doing to others' work,” Mebane said in an email. “I take responsibility for what I say, not for what they or anyone else may say.”
Other researchers Votebeat spoke with agreed that the ETA had stripped away important caveats. But they were less willing to separate the ETA’s use of the work from Mebane’s own choices — including the claims he published, the authority his name carried, and his decision to remain largely silent as those claims spread. Academia offers few clear mechanisms for addressing those concerns.
Keith Whittington, a Yale Law School professor who studies academic freedom, said universities are reluctant to police what professors research or say because ideas once dismissed can later prove correct. What responsibility any individual scholar has over how their work is used, he said, is “not at all obvious.”
Ivan Oransky, co-founder of Retraction Watch — a news organization that reports on academic corrections and retractions — said the issue is not whether preliminary or unconventional work can have value. It can, he said. But scholarship is supposed to be iterative, he said: Researchers should revise work in response to criticism and publicize those revisions. Mebane’s website does not indicate whether or how his Pennsylvania analysis has changed in response to the criticism it has received.
Without subjecting your work to that process, Oransky said, “I’m not sure you’re really doing scholarship.”
What makes this different from the dozens of other statistical analyses of the 2020 or 2024 elections he’s reviewed and found flawed, Grimmer said, is who it’s coming from. Claims like these usually come from the political fringe. It is far rarer to see them come from a tenured professor at a major university.
That distinction carries weight, especially in a political climate already primed for suspicion.
“You can see how something like this takes off,” Grimmer said. People tend to put too much stock in credentials when doing so supports their position, and “when they don’t understand, people default to the conclusion they want to reach.”
Carter Walker is a reporter for Votebeat in partnership with Spotlight PA. Contact Carter at cwalker@votebeat.org.
Jessica Huseman is Votebeat's editorial director and is based in Dallas. Contact Jessica at jhuseman@votebeat.org.