Michael Strain explains why a shoddy, politically motivated US Census Bureau report should concern us all

When US President Donald Trump fired the head of the US Bureau of Labor Statistics—claiming falsely that the agency’s economic data “were RIGGED” to make him “look bad”—we could at least take solace in the fact that the agency’s critically important work remained unadulterated.

Unfortunately, Trump’s attacks on the integrity of official statistics have continued. Last month, the US Census Bureau published an official brief that claims to have “readily determined with high confidence” that, of the 160 million voters in the 2020 presidential election, more than 128 million were citizens and more than 24,000 were noncitizens. “Over 32 million voter records remain to be analyzed.”

This report clearly falls short of the Census Bureau’s own standards of quality, integrity, and accountability. For example, it does not name its authors or provide relevant contact information for members of the public interested in asking questions about the report. Examining these unusual omissions, NPR reports that “the analysis was not conducted by career civil servants at the Census Bureau,” but by individuals affiliated with the Trump-aligned America First Policy Institute.

Then there is the fact that the brief’s subject matter is nakedly political. Illegal voting is a topic that the bureau previously would have bent over backward to avoid, lest it be accused of engaging in partisan politics. Not any longer. On the day the report appeared, Trump himself posted about it on social media, highlighting the 24,000 illegal vote figure and claiming that “The Census Bureau is going to analyze the next 32,000,000 Voters, and this number will explode. I WON THE ELECTION!”

The report’s analysis links a variety of databases together in an effort to determine if all the people who voted in the 2020 election were citizens. One is a private-sector database of voter registration and turnout records, which was used here to figure out who voted in the election. The authors then linked this database to internal government databases with the goal of determining the citizenship status of the individuals in the private-sector data on voter records.

This type of analysis is challenging for several reasons. Large private-sector databases of the type used by the brief’s authors often contain missing values for key variables (including home address, birth date, and name) that one would use to ensure that the same individual is being observed across multiple data sources. Normally, a study relying on this sort of data linking would provide extensive, detailed, and assessable information about the challenges the researchers faced and how they overcame them. That is how one convinces professional statisticians, economists, and demographers that one’s conclusions are credible. (Indeed, the bureau has reporting requirements to this effect, which the authors of this brief seem to have violated.)

In this case, special attention should have been paid to any limitations and missing information in the private-sector data on voters, and to the researchers’ methods of linking the different data sources together (especially since it appears that the private-sector data do not include voters’ Social Security numbers). On that standard, the brief does not measure up, by dint of the fact that it is only seven pages.

Nor does it address false positives, when the researchers misidentify a case of a noncitizen casting a ballot. John Abowd, the former Census Bureau chief scientist, argues that for commercial data of the type used in this analysis, the false match rate is at least seven times as large as the share of voters the study’s authors claim are noncitizens. He concludes that, “The alleged ‘noncitizen voters’ are therefore much more likely to be data errors than noncitizen voters.”

There is also an amateurish quality to the report. The claim to have “readily determined with high confidence” that 24,000 noncitizens illegally voted in the election flies in the face of the authors’ seeming obliviousness to the underlying statistical issues they failed to address.

Or consider the language in the conclusion: “The Census Bureau’s library of administrative records makes it uniquely capable of compiling and publishing statistics on election crime and other forms of fraud.” No, it does not. Fraud is a legal determination that the bureau’s data would be hard-pressed to support or refute on its own.

Moreover, publishing the report while leaving 32 million voter records unclassified is itself evidence of work below the gold standard exemplified by previous Census Bureau analyses. (Nor is it even clear to me that 32 million is the right number. From the brief’s Table A, it looks like the authors have yet to analyze 18.7 million records, not 32 million.) As Abowd points out: “Allowing such statistical nonsense to be published on census.gov without the usual internal peer reviews and without the scientific detail required to assess the quality of the results is at best irresponsible but more likely pure partisan manipulation of the Census Bureau’s reputation for reliable statistical products.”

That reputation matters. Trust in official government data is one of the pillars of American prosperity. Millions of businesses rely on government data for myriad decisions (such as whether to operate in specific parts of the country, and what products to carry), as do financial markets. Moreover, it is official data that help determine the flow of billions of dollars of federal funds across the states.

In a recent paper coauthored with Nicholas Bloom, Erica L. Groshen, and Duncan Hobbs, I estimate that the marginal reduction in trust in government statistics caused by Trump’s firing of the head of the Bureau of Labor Statistics increased policy uncertainty and thereby reduced economic output by roughly $20 billion, or by $25 for every $1 spent on the agency’s budget.

Sadly, more shenanigans are coming. The Census brief described itself as “a beginning analysis,” and Trump has already foreshadowed a follow-up report. Before that happens, the authors of the first one should step out of the shadows and subject their analysis to interrogation by professional statisticians, economists, and demographers. That sort of scrutiny is how US statistical agencies built the sterling reputations Trump is wantonly trashing—reputations that are a foundation of American prosperity.

Michael R. Strain, Director of Economic Policy Studies at the American Enterprise Institute, is the author, most recently, of The American Dream Is Not Dead: (But Populism Could Kill It) (Templeton Press, 2020). This content is © Project Syndicate, 2026, and is here with permission.