A number arrived in August with the full visual authority of the United States Census Bureau behind it: 24,000 noncitizens voted illegally in the 2020 presidential election. President Trump posted it. It spread. And because it came stamped with the imprimatur of a federal statistical agency, millions of people had no obvious reason to interrogate it.
That's the mechanism. Not fabrication exactly — a document exists, a number appears in it — but something more insidious: the deliberate use of institutional credibility to launder a claim that wouldn't survive scrutiny if it arrived from anywhere else.
Let's check the denominator.
The Number Has No Author, No Methodology, and No Error Rate
Start with what a Census Bureau report is supposed to look like. According to economist Justin Wolfers writing at Platypus Economics, a standard Census Bureau research product is signed by its authors, specifies what was measured and where the data came from, provides detailed tables, reports error rates, and indicates the level of confidence you can place in the results. It's written by career civil servants using established statistical methods and goes through multiple rounds of expert review, with reviewers typically listed.
None of those elements exist in this report.
What does exist is a Disclosure Review Board number — the stamp that confirms a report was checked to ensure it wouldn't reveal personally identifiable information. That's it. The DRB number is not a quality certification. It's a privacy check. Presenting it as evidence of official rigor is like pointing to a building's fire inspection certificate as proof the architecture is sound.
NPR reported that work on the analysis was not conducted by career civil servants at the Census Bureau. Instead, the team included individuals affiliated with the America First Policy Institute, a think tank started by officials from Trump's first administration. The bureau, AFPI, and the White House did not respond to NPR's requests for comment — though AFPI did highlight the analysis on social media the same day Trump posted it, calling for "transparency" in elections.
The report itself, per NPR, draws on a 2020 commercial voter file and various federal administrative records. Data and election experts described the methodology as questionable. The seven-page document is unauthored. There are no confidence intervals. There is no sampling error disclosure. There is no explanation of how a match between a voter file and administrative records was determined to constitute confirmed illegal voting rather than a data linkage error.
That last point is where the math collapses.
Record Linkage Errors Are Not Rare — They're the Whole Problem
Here's the core statistical issue, and it's not subtle.
Matching records across databases — a voter file here, a federal administrative record there — is a notoriously error-prone process. Names are misspelled. Dates of birth are transposed. Common names generate false positives. A person named "Jose Garcia" who appears in both a commercial voter file and a federal immigration record is not automatically the same Jose Garcia. Record linkage methodology exists precisely because naive matching produces enormous numbers of false matches, and the entire discipline is devoted to quantifying and reducing that error rate.
A report that claims to have identified 24,000 illegal voters by matching records across databases, without disclosing its matching algorithm, its false positive rate, its confidence interval, or its methodology for adjudicating ambiguous matches, has not proven that 24,000 people voted illegally. It has proven that 24,000 records appeared to match under some undisclosed criteria. Those are not the same thing.
Wolfers makes this point with characteristic precision: the statistical methods are shoddy, the usual standards and procedures were ignored, and the findings bear little relation to the truth. The report doesn't tell you its false positive rate because it doesn't appear to have calculated one. In a dataset of 128 million voter records — the scale Trump cited — even a false positive rate of 0.02% would generate 25,600 spurious matches. The claimed 24,000 "illegal voters" could be entirely explained by normal database noise.
Trump's social media post claimed the final number would "explode" once the remaining 32 million records were checked. That's not how statistics works. If your methodology is generating false positives, adding more records generates more false positives. Scale is not validation.
What the GOP Convention Did With This Number
The report didn't stay in a social media post. It migrated to the Republican midterm convention, where it joined a broader pattern of statistical claims that FactCheck.org documented across the event's second night.
The convention's statistical environment is worth examining as a case study in how numbers function in political rhetoric. Several speakers credited the Trump administration for a large drop in violent crime in 2025 and a historically low murder rate — but as FactCheck noted, the recent statistics continue a trend of declining crime across recent years, and criminologists identify multiple possible contributing factors. Attributing a multi-year trend to a single administration's policies requires a causal claim the data doesn't support. Correlation is not causation, and the time window matters: a trend that began before an administration took office cannot be cleanly credited to that administration.
House Speaker Mike Johnson claimed Republicans inherited "record high" inflation from the Biden administration. FactCheck found this inaccurate: the annualized inflation rate had declined to 3% when Trump entered office — lower than the current rate. The "record high" framing requires choosing a starting line that predates the handoff, a technique this publication examined in detail in the Vance manufacturing piece from May.
Trump claimed to have built "almost 2,000 miles" of border wall. FactCheck put the actual number at roughly a third of that — and only if replacement wall in locations where barrier already existed is counted alongside genuinely new construction. The denominator problem here is definitional: what counts as "built"?
RFK Jr. claimed Democrats "lost 360,000 children" who he suggested were trafficked. FactCheck described this as distorting the findings of a 2024 Department of Homeland Security inspector general report. Kennedy also claimed the U.S. has "the sickest children in the history of the world" — a claim that fails on its face given that childhood mortality rates in the distant past, when it was common for as many as half of all children to die before adulthood, were incomparably worse than anything measurable today.
Each of these is a different flavor of the same error: a number or claim presented without its denominator, its time window, its comparison class, or its causal warrant.
The Institutional Damage Is the Real Story
The individual claims are correctable. The institutional damage is harder to repair.
The Census Bureau is not a peripheral agency. As Wolfers writes, it tells us where people live, where businesses are opening, where poverty is rising, where children need schools, where communities need hospitals, and whether government policies are actually working. It determines congressional apportionment. Its data underpins the BLS employment figures, the poverty measures, the economic indicators that businesses and policymakers use to make consequential decisions.
The Census Bureau's own poverty report for 2025 — a legitimate product, produced by career statisticians, with disclosed methodology, named authors, and explicit notes on sampling error and confidence intervals — illustrates what the institution is supposed to produce. The contrast with the noncitizen voting document is stark. One tells you its error rates. One doesn't have any.
When a government agency produces a document that looks official but isn't — no authors, no methodology, no error rates, no peer review — it doesn't just mislead people about the specific claim. It degrades the signal value of every document the agency produces. If the Census Bureau will publish partisan analysis dressed as statistical research, then every future Census Bureau release becomes slightly less trustworthy. The cost isn't paid by the people who circulated the false claim. It's paid by everyone who relies on accurate federal data to make decisions — which is most of the country.
This is the mechanism that makes statistical propaganda more dangerous than outright fabrication. A fabricated number can be traced to nothing. A number that emerges from a real agency, in a real document, with a real DRB stamp, requires the reader to understand record linkage methodology, false positive rates, and the difference between a privacy review and a quality review. Most readers don't have that background. That's not a criticism of readers — it's a description of how expertise asymmetry gets exploited.
The Comparison That Matters
The question to ask about the 24,000 figure is the same question to ask about any number: compared to what?
Compared to the total number of record linkage attempts — 128 million voter records against federal administrative databases — 24,000 is a rate of roughly 0.019%. At that scale, the question isn't whether noncitizen voting occurred at all; it's whether the matches represent actual illegal voters or database noise. Without a disclosed false positive rate, you cannot answer that question. The report doesn't try.
Multiple studies and audits, as NPR noted, show that noncitizen voting in federal races is incredibly rare. That's not a partisan claim — it's the finding of investigations conducted across administrations and methodologies. The Census Bureau report doesn't engage with that literature or explain why its methodology should be trusted over theirs.
What it does instead is produce a large-sounding number, attach it to a trusted institution's letterhead, and release it six weeks before a midterm election.
The BLS news releases page shows what legitimate federal statistical output looks like in practice: dated releases, named programs, explicit methodology links, confidence intervals flagged with asterisks when the evidence is insufficient to conclude a change is statistically significant. The August CPI release, the PPI release, the import price data — each one tells you what it measured, how, and with what uncertainty. The noncitizen voting report tells you none of those things.
That asymmetry is the story. A number without a denominator, a methodology, or an error rate isn't a finding. It's a press release wearing a government seal.
Watch for whether the Census Bureau releases the underlying matching algorithm and false positive rate — that's the test of whether this was a one-time political intervention or the beginning of a new institutional norm. If the methodology never surfaces, that's your answer.
