A New Commerce Order Bans the Privacy Method Behind Detailed Census Data

On June 4, 2026, the Commerce Department issued an order called Disclosure Avoidance for Statistical Products, or DAO 216-26, which bans a family of privacy methods known as noise infusion across every data product from the U.S. Census Bureau and the Bureau of Economic Analysis. It took effect immediately, with no period for public comment.

Noise infusion is the main tool the Census Bureau uses to publish detailed local statistics without exposing individual households, so restricting it forces a choice between publishing less detail and protecting people less.

What noise infusion actually is

When the Bureau publishes a figure for a small area, say median income for a census tract of 3,000 people, the raw numbers behind it come from real households. Publish enough cross-tabulated detail about a small enough place and someone can work backward and identify a specific family. That risk grows every year as computing power and outside datasets improve.

Noise infusion solves it by adding a small, carefully measured amount of randomness to the data before publication. Differential privacy, the version used for the 2020 census, is the strictest form. It adds noise according to a mathematical formula that lets the agency state precisely how much privacy protection it's providing and how much accuracy it gave up to get there. That accounting is the point. It's what lets the Bureau publish granular numbers and defend them at the same time.

The 2010, 2000, and 1990 censuses used a different technique called swapping, where certain households are quietly exchanged with similar households nearby. It isn't clear whether swapping survives the new order either.

What the order allows instead

DAO 216-26 directs the agencies toward coarsening, which means rounding numbers off, lumping categories together, or publishing ranges instead of figures. Where coarsening isn't enough, the fallback is suppression, meaning the number simply doesn't get published.

Both approaches cost detail. A rounded, bucketed, or missing number is less useful than a precise one, and critics argue coarsening also leaves confidential data less protected than noise infusion does, so the tradeoff runs the wrong way in both directions.

The scale of the effect isn't hypothetical. The Census Bureau ran a simulation on the Quarterly Workforce Indicators, an economic dataset, and found that only about 10 percent of what it currently publishes could still be released using the remaining methods, according to reporting by Talking Points Memo. That figure covers one dataset, and the Bureau hasn't published equivalent estimates for others.

How this could affect data on our site

Nearly everything on CensusEasy below the state level is based on the Bureau's ability to publish detailed small-area data. Our ZIP code pages, our roughly 85,000 census tract pages, and the neighborhood rollups we build from tracts all exist because disclosure avoidance currently permits that level of granularity. If future releases are coarsened or suppressed, the data behind those pages gets thinner, and so does every site and study built on the same foundation. However, note that we don't know yet how far that goes, because the agencies haven't said how they're implementing the order.

What researchers are saying

The Population Association of America, along with four other statistical and population research organizations, published a joint statement on June 17 arguing the order "subverts processes developed over decades to foster transparency and public trust" and that the result will be "either less privacy for our personal information, or less usable data, or both."

Health policy researchers have been among the loudest. Aubrianna Osorio, a research fellow at Georgetown University's Center for Children and Families, wrote that the center's work on child uninsured rates by age, race, income, county, and school district all runs on detailed American Community Survey data, and that this work would be severely limited without it. She also noted that Census population totals and BEA per capita income feed the formula that sets each state's federal Medicaid matching rate, though the order explicitly does not cover data shared between Census, BEA, and the Bureau of Labor Statistics.

Census Bureau figures guide the distribution of roughly $3 trillion in federal funds a year, more than half of it through the Department of Health and Human Services.

What nobody knows yet

The Commerce Department hasn't published a detailed rationale, the agencies haven't described their implementation, and there's no public list of which datasets will change or when. The Bureau took down several webpages and working papers about privacy methods after the order and has since restored some of them. The Chief Scientist post, which would normally lead the technical response, is vacant.

Two dates are worth watching. The 2025 American Community Survey data usually arrives in mid-September, with the release date normally announced by mid-July, and as of late July no date had been announced. Preparations to apply differential privacy to the 2030 census were already underway and now have to be reconsidered, which puts the 2030 redistricting files in the same uncertainty.

For related coverage of how the 2030 count is being shaped, see our reporting on the reversal on race and ethnicity standards, the proposed residence and question changes, and proposals to make the 2030 count more accurate. For how the survey behind most of this data works, see what the American Community Survey is, and for our own data methods, the methodology page.

Sources

The order is DAO 216-26, Disclosure Avoidance for Statistical Products, issued by the U.S. Department of Commerce on June 4, 2026. The joint researcher statement is from the Population Association of America, June 17, 2026. Health policy analysis and the federal funding figures are from Aubrianna Osorio at the Georgetown University Center for Children and Families. The Quarterly Workforce Indicators simulation figure was reported by Talking Points Memo.