At first glance, the report might have seemed like a bombshell: a government-backed claim that thousands of noncitizens had cast ballots in a recent election. In a country already torn apart by arguments over election security, a number like that lands with force. But for people who spent their careers inside the machinery of American statistics, the assessment raised alarm bells not just about what it concluded, but about how it was assembled. The dataset is enormous, the sources are unclear, and the methods are missing. It is a combination that makes it difficult, if not impossible, to tell whether the headline-grabbing number is a real measure of voter fraud or the product of mismatched records and misunderstood categories. The word “red flags” kept coming up. Amy O’Hara, a former Census official who once ran the agency’s administrative data linkage unit, told WIRED that the breadth of data involved is a problem in itself. Different agencies and commercial data vendors collect information about citizenship in different ways, at different times, and for different purposes. One source might count a person as a citizen because they applied for a passport. Another might miss a naturalization that happened last year. Another might rely on old records that never get updated. Unless you know exactly where those records come from, how often they are refreshed, and whether people’s legal statuses have changed, any attempt to combine them is built on shaky ground. O’Hara put it plainly: “We don’t know whether or how the sources are updating their citizenship info.” She also stressed how hard it is to join and deduplicate files, particularly across years of changing addresses, name spellings, and life events like marriage and divorce. This is not a simple spreadsheet merge. It is a massive, messy operation, and the report does not explain how the people behind it handled that complexity. If the underlying files are outdated, incomplete, or simply wrong, then every conclusion built on top of them deserves to be questioned. That question deserves an answer, not a headline.
At the center of this mess is a useful but imperfect tool called the Protected Identification Key, or PIK. The Census Bureau has used it for years to let researchers and government employees study sensitive records without exposing anyone’s identity. Instead of names and birth dates, the system assigns a unique code to each person. When a researcher needs to combine data, they can match those codes, not the identifying details. It is a clever design, and it has made countless studies possible while protecting privacy. But PIKs are not infallible. They depend on a series of decisions—which variables to use, how much to tolerate mismatches, and what to do when a record does not fit cleanly into the system. The sources who spoke with WIRED said they do not know which data points were used to match PIK data across the many datasets in this particular report. An address could have been used for cross-referencing, but without documentation it is impossible to be sure. That absence of transparency matters because matching people across files is a notoriously difficult task. A person may appear in one database under a maiden name and in another under a married name. They may have moved, changed jobs, or gotten a new driver’s license. They may share a birthday with someone who has a similar name. All of these small discrepancies can cause records to be split, duplicated, or attached to the wrong person. The Census Bureau itself has acknowledged this. A study published earlier this year, dated January 2026, noted that PIK assignment is generally accurate, but that some person records are not successfully assigned a PIK. That might sound like a technical quirk, but it has real consequences. If certain people are systematically left out or misclassified, analyses built on linked data can end up with sample selection bias—meaning the numbers reflect the flaws in the linking process, not the reality of the population. In other words, the same tool that makes this kind of research possible can also lead researchers astray if it is misused or rushed.
Now consider what the report actually claims. It says there is a specific number of noncitizens who voted, and it describes the finding with the phrase “high confidence.” For people who understand statistics, that phrase is a promise. It should mean the numbers have been tested, the uncertainty has been measured, and the conclusion still holds up. But the sources who looked at the tables say that promise is empty. One Census Bureau source, who agreed to speak only anonymously out of fear of retribution from the administration, told WIRED that no confidence intervals are shown. The report says “high confidence” without offering any way to check how confident we should be. That is not just a technical omission. It is a sign that the report may have started with a conclusion and worked backward, rather than letting the data speak honestly. The source also pointed out that the alleged number of noncitizen voters is small compared with the total population of the state in question. That matters because small numbers are more vulnerable to error. A few thousand mismatched records can easily produce an estimate that looks meaningful but is actually just noise. The source believed that the flagged records probably came from PIK errors, not from real people casting illegal ballots. Imagine a citizen whose records are tangled up with someone else’s citizenship status. Imagine a naturalized voter who appears in an old database as a noncitizen because the database was never updated. Imagine an address shared by two people whose files got merged by accident. In each case, the data would say one thing while the truth is something else. None of those scenarios involve fraud. They involve the ordinary, messy limitations of administrative records. But the report presents them as if they were proof of a coordinated threat to American elections. In other words, the report may be manufacturing a crisis where none exists.
John Abowd, a former Census Bureau chief scientist, has seen these dangers up close. He warned that false match rates—when a voting record is assigned the wrong PIK, and then “matched” with the wrong citizenship information—are likely to be higher when analysts use commercial data. That is a major concern because the report appears to rely on exactly that kind of data. Commercial datasets are often built for marketing or credit purposes, not for statistical analysis. They may contain errors, duplicate entries, or outdated information. They may include people who are not eligible to vote, but they may also include citizens who have recently naturalized or who have changed their names. When you join those records to a voting file, you are combining two imperfect systems. If something goes wrong, the result is a false positive: a record that looks like a noncitizen voter but is actually a citizen whose data was mismatched. Abowd put the problem in stark terms. The false match rate, he said, is much greater than the estimated noncitizen voting rate. That means the alleged noncitizen voters are more likely to be data errors than real people who broke the law. This is a basic matter of arithmetic, not ideology. If you are looking for something that is very rare, you need to be extremely careful about false positives. A test that flags one percent of records by mistake will seem to find thousands of “cases” even when there are only a handful of real ones. The only way to correct for that is to know the error rate and adjust the numbers accordingly. The report gives no such details. Abowd said it is impossible to estimate the false positive rate without more technical information—and the report does not provide it. That is not a small oversight. It is the difference between a credible finding and a speculative one. As he told WIRED, the alleged noncitizen voters are more likely to be the result of data errors than of illegal voting. Without that information, the public is being asked to take a very large leap of faith.
Boyd, another researcher who spoke with WIRED, is worried about the long-term damage. She has spent years studying the Census Bureau’s data and methodology, and she knows how much the country depends on its products. Census numbers help determine how federal funds are distributed, how congressional districts are drawn, and how scholars understand the changing population. That kind of legitimacy is not easy to build. It comes from rigorous peer review, transparent methods, and a consistent willingness to acknowledge limitations. This report, Boyd said, has none of those things. There are no authors, no methods, no discussion of the limitations of the data or the Bureau’s approach. Yet it is being put out with the imprimatur of the Census Bureau, so it looks legitimate. That combination is dangerous. It gives the public a false sense that the numbers have been vetted, when in fact they have not been vetted at all. Boyd worries that this could undermine more than just trust in voting and elections. If academics and researchers begin to doubt the quality and reliability of the products that the Census Bureau produces, the damage will reach far beyond this one report. The entire statistical system could be compromised in the public’s mind. That is especially troubling because the Census Bureau has already been dragged into political battles. Republicans have attacked the accuracy of the 2020 census, claiming that an algorithmic process used to protect data privacy had skewed district lines in favor of Democrats. Those claims did not survive close scrutiny, but they helped fuel suspicion about the Bureau’s work. In August, WIRED reported that an internal memo circulated by the Department of Commerce, which oversees the Bureau, proposed excluding immigrants from the count entirely. That idea, if put into effect, would not just be an administrative change. It would radically reshape how congressional districts are apportioned. Now this report adds another layer to that pattern. It uses the Bureau’s good name to advance a political argument that the data cannot support.
The bigger picture makes the report even harder to ignore. This is not a one-off mistake. It is part of a broad attack on the integrity of American elections that has been underway for years. The Trump administration has weaponized the Department of Justice, using federal law enforcement to reward political allies and investigate enemies. It has dismantled voter protections that made it easier for eligible people to register and cast their ballots. It has attempted to wrest control of election administration away from the states, which have always run their own elections under the Constitution. And it has pushed an endless barrage of baseless conspiracy theories about voting in America. Some of those theories are about machines being rigged. Some are about ballots being dumped. Some are about phantom voters who do not exist. This report adds another one to the list: noncitizens casting votes in numbers large enough to change elections. The design of the report makes it clear that it was built to support that conclusion, not to test it. There is no method, no transparency, no honest accounting of uncertainty. There is only a dramatic number, presented with the authority of the Census Bureau and the weight of “high confidence.” That is not how science works. It is how propaganda works. The former officials and researchers who spoke to WIRED were not trying to defend illegal voting. They were trying to defend the difference between fact and speculation. In a democracy, that difference matters. We need to be able to trust the numbers that tell us who lives in this country, how our communities are changing, and who is participating in our elections. If those numbers are turned into marketing tools for political campaigns, we lose something essential. The real danger of this report is not the tiny number of people it claims voted illegally. The real danger is the willingness to abandon rigor in order to advance an agenda. That is a threat to every American, no matter which party they support, because it makes it harder for all of us to know the truth. And once we lose the ability to know the truth, the entire idea of self-government begins to unravel.
