The Trump administration has ordered the Census Bureau and the Bureau of Economic Analysis to stop using “noise infusion,” a statistical technique that protects individual privacy by making certain data fuzzy, according to a Commerce Department order issued this month. The policy, which data experts say could render large swaths of federal statistics unpublishable, leaves the two statistical agencies with only two options going forward: releasing “coarsened” statistics with fewer details or not releasing some statistics at all.
Federal law requires the Census Bureau to keep people anonymous in the data it produces from surveys and government records. Noise infusion has been one of the bureau’s main privacy protection techniques for decades, ensuring that individual people — including members of minority communities — cannot be identified in published statistics.
Beth Jarosz, a senior fellow at Georgetown University’s Massive Data Institute and vice president of the Association of Public Data Users, said the order puts neighborhood-level and rural data at risk. “There are some counties that are only a couple hundred people, and you might not be able to publish data for those counties anymore,” Jarosz said.
John Abowd, a former chief scientist at the Census Bureau who served during the first Trump and Biden administrations, said the order upends privacy protection systems for multiple ongoing surveys and other datasets. For the once-a-decade head count, the bureau did not add any statistical noise to the state-level 2020 census results used to redistribute congressional seats and Electoral College votes. But the bureau applied privacy protection methods involving noise to the detailed demographic data used for redrawing maps of individual voting districts.
Abowd said under the ban on statistical noise, plans for 2030 census redistricting data “will have to be completely redesigned, and not just the confidentiality protections.” He added that “the only confidentiality protection available is coarsening. It is guaranteed to reduce the level of detail drastically.” Asked if some political mapmakers may find that kind of redistricting data unusable, Abowd replied: “I’m pretty sure most would.”
The bureau’s public information office did not respond to requests for comment. In a statement, Commerce Department spokesperson Kristen Eichamer said the order prioritized coarsening as the preferred privacy protection technique to “maintain public confidence in our data while upholding our duty to safeguard the privacy of those who provide information.” Eichamer also stated that “indiscriminate use of noise infusion — even when not mandated by law — ultimately undermined confidence in the department’s products and cast doubt on their integrity.” Asked for specific examples of such use, Eichamer did not respond.
The department’s order could be revoked before the 2030 census under a new presidential administration. But census watchers expressed concern about lasting repercussions from a policy change coming in the middle of key preparations for the upcoming national tally and shortly after a thinning of the bureau’s expertise amid the Trump administration’s slashing of the federal workforce.
Some current staffers expressed alarm about the proposal. “It would not be a major overreaction to say that this is cataclysmic,” said a bureau employee who asked not to be named because they are not authorized to speak to the press. “From our perspective right now, if this policy stays in effect, it’s the end of a lot of our data production.”
The use of statistical noise in certain 2020 census data sparked controversy within the statistical and redistricting worlds in the lead-up to its release in 2021. As the bureau’s chief scientist, Abowd led the adoption of a new privacy protection system based on a mathematical concept known as differential privacy. Bureau officials said the shift was needed to keep up with advances in computing and broader access to voter registration lists and commercial data sets that have made it easier to reidentify individuals within purportedly anonymized statistics.
Early tests of the system’s effect on redistricting data raised alarm among many data users, who feared the statistics would ultimately be unusable. Republican state officials in Alabama sued the bureau to try to block the new privacy protections. The case was ultimately dropped, and maps for voting districts across the country were drawn using the noise-infused 2020 census redistricting data.
Last year, America First Legal, a law group co-founded by Stephen Miller, President Trump’s deputy chief of staff for policy, filed a lawsuit challenging the bureau’s differential privacy system in an attempt to force the release of new 2020 census results, even though the released state population numbers were not adjusted with any statistical noise. After a three-judge court ruled it was past the time limit to sue, the group refiled their case, which continues to challenge another statistical technique the bureau used.
Jarosz said she is concerned that the ban on statistical noise was released with little explanation. “The way that it should work is that scientists, experts do their best job to come up with a plan. They present that to the public. There are opportunities for feedback. There are opportunities for criticism from other experts that aren’t in the federal government. And we end up with a system where it’s had really good transparency, lots of checks and balances,” Jarosz said. “This new order upends all of that. It takes the public out of the process. It takes the experts out of the process. This feels very much like a political choice.”
Until the bureau starts publicly explaining how the order is affecting its data releases, Jarosz added, it remains an open question whether participants in the bureau’s surveys should be concerned about the confidentiality of the personal information they have shared with the government. “The Census Bureau and all of the statistical agencies are bound by laws to protect the privacy of our information. And so they had put together the best tools that they could find to do that,” Jarosz said. “If they keep publishing the same amount of data and they don’t have those tools, then yes, privacy could potentially be at risk.”