Last week the Trump administration put out a document that says exactly what the president has been saying for six years.

The U.S. Census Bureau released a seven-page report concluding "with high confidence" that more than 24,000 noncitizens voted illegally in the 2020 election. Trump posted it on social media and wrote that it backs up his claim that he actually won.

Then people who know how Census Bureau reports are written started reading it. And the questions started.

Start with the number the White House isn't repeating

The report says researchers looked at more than 128 million voters from 2020.

Out of those 128 million, they say 24,000 were noncitizens.

That is 0.019 percent. Nineteen thousandths of one percent. If you filled a football stadium with 2020 voters, you would expect fewer than two of them to be one of these people.

Now compare it to the margins Trump lost by. The report itself says it found 500 noncitizen ballots in Wisconsin — a state Trump lost by more than 20,000 votes. It says 1,100 in Arizona, which Trump lost by more than 10,000.

And the report cannot say who any of those people voted for. Not one of them. It has no way to know.

So even taken entirely at face value, the administration's own document does not come close to changing the 2020 result. Trump's answer to that was to say census researchers are still working through another 32 million voters, and to predict the number will "explode."

That is a promise about a report that doesn't exist yet, offered to cover a report that does.

The report has no authors

Here is the part that made career statisticians stop.

Most Census Bureau reports name the people who wrote them and spell out how reliable the underlying data is. This one does neither.

"A properly done Census report would have had that information and the fact that it is missing is glaring and makes me think that they are hiding something," said Beth Jarosz, a senior fellow at Georgetown University's Massive Data Institute who has worked as a contractor on Census Bureau reports.

Terri Ann Lowenthal spent seven years, from 1987 to 1994, as staff director of the U.S. House subcommittee that oversaw the Census Bureau. She said the report was "beneath the Census Bureau's stellar reputation." Leaving out the names of career staff, she said, suggested the work "may have been done by someone with a political agenda" and "may not reflect the sound, reputable analysis for which the Census Bureau is known."

The Census Bureau did not respond to a request for comment.

There's no margin of error — and the margin of error is the whole story

Every serious statistical analysis says how wrong it might be. This one doesn't.

Justin Levitt, an election law specialist at Loyola Law School who worked on voting rights in the Biden White House, explained why that matters here more than usual:

"In most statistical exercises, it's not unthinkable that there might be 0.019% false positives — and if that were the case here, that would turn the total conclusions to zero."

Read that again. The claimed finding is 0.019%. A perfectly ordinary error rate for this kind of record-matching is also about 0.019%.

Which means the report's entire result could be the noise. And the Census Bureau declined to print the one number that would tell you.

Where the data came from — and the woman who's surprised it was used this way

The voter records behind the analysis came from Data-Clear, a Virginia company that won an $885,000 Census Bureau contract in 2024, under the Biden administration.

That contract was for something else entirely. Carolyn Carlson, Data-Clear's president and founder, said she gave the bureau nationwide voting data as part of a project meant to prevent an undercount of minority voters in the 2030 census.

She said she was surprised to see the data she provided turn up in a hunt for noncitizen voters.

The bureau then matched those voter files against federal immigration records to guess at citizenship. The report describes how it did that only in general terms.

This is the fourth time, and the pattern never changes

The 24,000 figure isn't the administration's first swing at this. It's the latest in a run of claims that all break the same way: a big number, a thin method, and no receipts when anyone asks.

The citizenship-checking tool. The administration pushed states to run millions of voter records through a revamped federal database. Election officials found it wrongly flagged actual citizens. In June, a federal judge blocked its use, finding it threatened privacy and voting rights.

The primetime speech. On July 16, Trump said in a televised address that his administration had found more than 250,000 noncitizen voters on the rolls of just four states — New Jersey, California, Pennsylvania and Nevada. Officials in all four pushed back and said they could not figure out where the numbers came from. The administration has still not explained its method.

The prosecutions. The Justice Department has been threatening election officials with jail over a problem that audits, investigations and studies keep finding is vanishingly rare.

And now the Census Bureau. Same shape. Big headline number, no authors, no error bars, no explanation.

When the White House was asked whether this report actually supports what the president said about it, it replied with a link to a social media post by Commerce Secretary Howard Lutnick.

What it's actually for

None of this is really about 2020. It's about the paperwork you'll need next time you register to vote.

Trump said when he released the Census findings that a federal law requiring documented proof of citizenship to register is his highest legislative priority. That's the SAVE Act — a bill that would require a passport or birth certificate to register, documents that millions of American citizens don't have on hand, and that married women who changed their names often can't match.

It has stalled in the Senate. Trump's two executive orders on elections have had their major provisions blocked in federal court.

So the administration needs a number. A seven-page report with nobody's name on it, claiming a rate of 0.019 percent that experts say could just as easily be zero, is the number.

The cost of getting this wrong is not symmetrical. If 24,000 noncitizens really did vote out of 128 million, the 2020 result stands and nothing changes. If Congress passes a documents requirement based on it, millions of citizens who have voted their whole lives find out at the registration desk that they can't anymore.

That is the trade being proposed. One side of it is measured in thousandths of a percent. The other is measured in millions of people.

Source

Dion Nissenbaum, "Experts skewer new Trump administration analysis claiming 24,000 noncitizens voted in 2020", Votebeat, republished by the Ohio Capital Journal, August 24, 2026. Photo: Official White House Photo by Andrea Hanks.