On June 4, 2026, the Commerce Department issued Department Administrative Order (DAO) 216-26, Disclosure Avoidance for Statistical Products, which bans a class of privacy techniques called “noise infusion” in all data products coming out of the Census Bureau and the Bureau of Economic Analysis (BEA). This order, which was made effective immediately without opportunity for public comment, is highly unusual because it overrides statistical agencies’ technical expertise and limits the methods available to protect confidential information in public data releases. As far as we know, this is unprecedented. And while we are still waiting to see how the agencies implement it, this could have disastrous effects both for health policy research and the larger federal data infrastructure.
Background
Federal laws, regulations, and guidance require the Census Bureau to balance protecting private information with publishing useful statistics, and to determine its own best methods for doing so. Often, the agency relies on noise infusion methods that protect confidentiality by adding randomness to the underlying data. For example, differential privacy, widely considered the gold standard, involves adding a carefully calculated amount of statistical “noise” to the data to achieve a guaranteed level of privacy, which then allows researchers to publish detailed estimates without worry that someone could reidentify specific households or individuals. Importantly, this method also allows the agency to quantify and document how they balance privacy and usefulness. The Commerce Department has been discussing noise infusion since at least the late 1970s and as privacy methods have improved, public data has become more granular, available even for small geographies and disaggregated by demographic, social, and economic characteristics.
Without noise infusion, many of these estimates may be considered too vulnerable to reidentification. For example, based on the Census Bureau’s own simulation, just 10% of the data currently published in an economic dataset called the Quarterly Workforce Indicators would be released only using other privacy methods. (The order directs the agencies to instead use “coarsening,” which includes rounding, aggregating, and reporting in ranges, or to suppress data that is still too risky. Both of these methods, however, would effectively reduce the level of detail the agency can publish, making statistics less useful while still leaving confidential data unprotected.)
The 2020 decennial census relied on differential privacy, while the 2010, 2000, and 1990 censuses used a technique called “swapping” (switching certain households with others nearby or with similar characteristics), but it is not clear that either of these methods will remain permissible under the noise infusion ban. That means the Census Bureau would only be able to address the privacy risks associated with the computing advances and data availability of today’s world with privacy methods not preferred since the 1980s. In short, as five statistical and population research organizations wrote after the order was issued, “there will either be less privacy for our personal information, or less usable data, or both.”
The Key Role of Census Data in Health Policy Research
CCF relies heavily on Census data to provide the evidence in our evidence-based policy analysis. Oftentimes, that means turning to the American Community Survey (ACS), including its survey microdata. The ACS provides the detailed data that powers our State Data Hub breakdowns of child uninsured rates by age, race and ethnicity, and income; our Annual Report on Child Coverage Trends; our rural health reports and reports on uninsured rates for young children and for Latino children; our estimates of Medicaid coverage by counties, school districts, and congressional districts; and other analyses of state program eligibility changes or outreach targeting, for example. Our ability to do this work will be severely limited without the same level of detailed ACS data that the current use of noise infusion allows the Bureau to publish.
Census data also play a role in Medicaid and the Children’s Health Insurance Program (CHIP) financing, and the order banning noise infusion could potentially have a negative impact. For example, the Federal Medical Assistance Percentage (FMAP) formula, also called the Medicaid matching rate, determines the share of Medicaid costs the federal government covers for each state, paying a higher regular FMAP rate in states with lower per capita incomes compared to the national average. The formula relies on state population totals based on the decennial Census and per capita income data calculated by the Bureau of Economic Analysis (the BEA adjusts some income figures using ACS data, but the order “does not include any data shared among the Census Bureau, the Bureau of Economic Analysis, or the Bureau of Labor Statistics”). The FMAP is also the basis for the enhanced CHIP match rate, and child population totals used to adjust the spending levels that determine each state’s CHIP allotment annually. The state population totals produced from the 2020 census did not include any added noise, and new data have continued to be released as scheduled since the noise ban was issued. However, the most recent population estimates may have been processed and cleared for release before the order was issued, and the detail and availability of future estimates remain unclear without more information.
The Census Bureau’s own research has determined that linking decennial and ACS data to Medicaid enrollment records can help fill in race and ethnicity information missing from Medicaid data and allow researchers to better evaluate health disparities. Data quality assessments of the Centers for Medicare & Medicaid Transformed Medicaid Statistical Information System (T-MSIS) are also benchmarked to ACS data for demographic measures like enrollee race, ethnicity, and language. The Census Bureau’s Survey of Income and Program Participation (SIPP), which tracks respondents over several years to collect information on eligibility and participation in social programs, including Medicaid/CHIP, is also likely impacted by the noise ban. Finally, while some large states do have their own surveys, such as the California Health Interview Survey, the ACS is the only national survey that produces the level of detail states need to conduct outreach, monitor health and coverage inequities, and perform budget analyses.
Census Bureau data play a critical role in the distribution of as much as $3 trillion in federal funds, more than half of which goes through the Department of Health and Human Services for programs including Medicare; community health centers; grants to monitor and improve rural health, maternal, infant, and early childhood health, and mental and behavioral health; federal data projects like the Behavioral Risk Factor Surveillance System; and more. More than one-third goes to programs serving children, primarily through Medicaid and CHIP.
What Happens Next?
Ultimately, there is still a lot we don’t know about what prompted this order, the intended policy goals, how it is being implemented, or which datasets it will impact, or how, or when—or whether the agencies’ answers to these questions will be made public. What we do know is that the announcement of the noise infusion ban, the jarring lack of transparency, and the chaotic aftermath that has followed fit patterns we have seen before, with the Census Bureau taking down several webpages and working papers related to privacy and noise infusion (they have since reposted some of these), Census staff members reporting that they are alarmed about what this order could mean for their ability to publish data, and a key leadership position, Chief Scientist and Associate Director for Research and Methodology, sitting vacant.
Preparations were already underway to apply differential privacy to the upcoming 2030 decennial census data as well, meaning those plans—and particularly the fate of redistricting data—will need to be reconsidered. And heightened concerns about whether American’s personal information is protected could damage already-declining survey response rates, leading to undercounts of certain groups like young children or individuals with Medicaid coverage, or biasing the datasets.
For now, CCF will keep monitoring Census information and waiting for a release date to be set for the 2025 ACS data, which is typically released in mid-September. (In the past, Census has announced the release date by mid-July. We do not know whether an announcement has not come because a date has not yet been set or because the ACS data release will be delayed). And no matter what, Georgetown University CCF plans to publish a new report on child coverage trends this fall.