New research shows 2020 U.S. vote counts were extraordinarily accurate, contradicting fraud claims

A new study published in the Proceedings of the National Academy of Sciences provides one of the most comprehensive assessments to date of the accuracy of vote counting in the 2020 United States election. By compiling and analyzing data from postelection audits in 856 jurisdictions across 27 states, researchers found that discrepancies between the original and audited vote totals were exceptionally rare.

The average error rate was measured in thousandths of a percent, with no evidence of systematic bias favoring either major presidential candidate. The findings contradict persistent claims of widespread fraud and support the legitimacy of the election’s outcome.

Following the 2020 election, false claims that votes were miscounted or manipulated gained wide traction, particularly among supporters of President Donald Trump. Although multiple investigations found no evidence of fraud or significant error, the authors of this study noted that previous assessments often lacked the kind of comprehensive, quantitative data that could settle the matter more decisively. They aimed to fill that void by producing a nation-scale estimate of vote-count accuracy based on official audit records.

“There were many prominent allegations that Americans’ votes were not counted correctly in 2020. We conjectured that, if these discrepancies were real, many of them would have been detectable by election audits,” said study author Samuel Baltz, who conducted the research when he was the research director of the Election Data and Science Lab at the Massachusetts Institute of Technology.

To carry out this analysis, the research team assembled a massive dataset by tracking down publicly available results from postelection tabulation audits—where election officials retabulate a sample or the entirety of ballots to verify the accuracy of the initial count. These audits vary widely by state and county, not only in methodology but also in how results are reported.

“In many jurisdictions, auditors double-check the vote count by taking a sample of ballots, tallying them up a second time, comparing the result to the original vote count, and recording how many discrepancies they find,” Baltz explained. “However, these wonderful data sources are individually recorded by local and regional governments across the country in very different formats (everything from datasets on a website to tallies of votes in scanned PDFs), and are never compiled into one place. So, while there was no systematic evidence that votes were counted incorrectly in 2020, we did not have a numerically specific estimate of exactly how accurately they were counted around the country.”

The researchers painstakingly standardized these records to calculate vote-count error rates for candidates across thousands of races. In total, the dataset included information on more than 71 million candidate-level votes and over 1.2 million additional ballots across all types of elections, from local races to the presidential contest. Roughly 6 percent of all votes cast for Joe Biden and Donald Trump were included in the audited sample. While 27 states provided usable data, the researchers note that audit processes also took place in other states, but either used incompatible formats or did not release detailed results.

Across the board, the study found that vote counts were extraordinarily consistent between the original tabulations and the audits. For the presidential election, the median discrepancy in a candidate’s vote total was zero. The average shift in vote margin between Trump and Biden was just 0.007 percent at the county level—a figure that is over 30 times smaller than the margin of victory in the closest state, Georgia. Nationally, the shift in presidential vote margin as a result of audits was less than one one-thousandth of a percent.

The researchers paid particular attention to the areas and allegations that had attracted the most attention in public debate. For instance, some false claims suggested that voting machines had systematically switched votes from Trump to Biden. If that were true, audits using independent tabulation methods—such as hand counts or different software—would have revealed major discrepancies. In fact, they did not. In counties where ballots were hand-counted or retabulated using separate systems, the results were nearly identical to the originals.

Other allegations centered on absentee ballots, which were said to be vulnerable to fraud or mishandling. However, many jurisdictions require audits to include a sample of ballots from all voting methods, including absentee voting. Again, the audits uncovered no consistent problems. Even in Democratic-controlled counties, where critics alleged fraud might occur, audits conducted by bipartisan or independent teams revealed vote counts that aligned with the originals.

To see whether the high level of accuracy held across various types of elections, the researchers also examined results from congressional, gubernatorial, and state legislative races. Across all these contests, the net change in votes by party affiliation was effectively zero. Among Democratic and Republican candidates for president, U.S. Senate, U.S. House, and state legislatures, audits found no measurable shifts in overall vote totals.

“We combed through thousands of pages that document what auditors found when they double-checked more than 70 million votes cast in 2020,” Baltz told PsyPost. “In state after state, county after county, town after town, and from local races all the way up to the presidency, the number of votes that auditors counted was identical or nearly identical to the number of votes that was originally reported. In about 10 million votes cast for either Biden or Trump that auditors double-checked, the margin between the candidates shifted by a few thousandths of a percentage point. The error rate in vote counting appeared to be microscopic compared to the margins that usually decide elections.”

The study also found that larger audits tended to produce lower error rates. In counties where at least 1,000 votes were audited for a given candidate, the shift in vote total was always less than 1 percent—and often far smaller. For example, in contests where more than 10,000 votes were recounted, the error rates tended to be below one-hundredth of a percent. These findings point to a pattern of highly accurate vote counting, especially in larger jurisdictions.

Even among the handful of counties that did show small discrepancies, the effects were negligible. The largest shift toward Trump found in any single county was about 0.65 percent, based on an audit of just over 1,100 votes in Butte County, Idaho. Officials attributed this to minor issues with ballot storage. In Georgia, where a full manual audit was conducted, the shifts were well within expected margins given the volume of ballots retallied by hand.

Overall, 62 percent of the 2,317 candidates included in the audit data saw no change at all in their vote totals. Where changes occurred, they typically amounted to a difference of just a few votes. The median adjustment was about 59 votes per million counted, an amount too small to affect outcomes even in the closest races.

The ballot-level audit data painted a similar picture. Across the 15 states that reported total ballots counted and discrepancies, only about 0.04 percent of ballots showed any issue. In the vast majority of counties—about 81 percent—no discrepancies were found at all.

“I was absolutely surprised by how small the error rate was,” Baltz said. “I would not have been shocked if simple, honest human error meant that, when you double-check the vote, you sometimes find that a candidate actually did a few tenths of a percent better or worse than the original vote count. Actually, that large a shift turns out to be exceptionally rare. Almost every time auditors double-checked at least 1,000 votes, error rates were in the hundredths or thousandths of a percent.”

Despite these strong results, the study does include some important caveats. First, the analysis is limited to vote counting itself. It does not account for other types of potential problems, such as voter registration errors or voter suppression. Second, while the dataset is the most comprehensive of its kind, it only includes 27 states. In some cases, audit procedures in the remaining states were either incompatible with error rate calculations or not publicly disclosed.

Still, the study provides one of the clearest quantitative answers yet to a question that has roiled American politics since 2020: Were the votes counted accurately? According to this nationwide audit-based analysis, the answer is yes—overwhelmingly so.

“I’m not working on any follow-up right now,” Baltz noted. “I was raised, though, on Carl Sagan, to believe that there is room in science to study practical questions of great social importance, and that when you take up those questions, you are bound to follow the facts exactly where they lead you, and to share your findings honestly and plainly with the public. I’ll keep an ear to the ground for practical questions about elections that matter to people, where some component remains scientifically open.”

“There is a prior piece in PNAS called ‘No evidence for systematic voter fraud‘ (Eggers, Garro, and Grimmer 2021). I think of our article as a companion piece to that one: they show that there is no systematic evidence for many types of voter fraud, while we follow up with systematic evidence specifically against widespread errors in counting the ballots.”

The study, “Audits of the 2020 American election show an accurate vote count,” was authored by Samuel Baltz, Fernanda Gonzalez, Kevin Guo, Jacob Jaffe, and Charles Stewart III.