A new Department of Commerce Policy will delay the release of 1-Year American Community Survey data from September until sometime later in 2026, and is sending ripples through many other census products.
New 1-Year American Community Survey (ACS) data are typically published about mid-September each year. That event kicks off a five month cycle of product releases that provide important information on the changing demographic, economic and social characteristics of the nation’s populace, from the state level down to the neighborhood.
But this year’s release is on hold for now.
A new Department of Commerce administrative rule issued June 4, 2026, bars several of the confidentiality protection techniques the ACS has long relied on. That led the U.S. Census Bureau, the federal statistical agency responsible for producing the statistics, to postpone the 2025 release until later in the year while it works out how to produce the survey under the new rules.
The policy, Departmental Administrative Order 216-26, applies to two federal agencies under the Commerce Department’s purview – the U.S. Census Bureau and the Bureau of Economic Analysis. It takes several methods off the table that federal statisticians use to protect the confidential government records and survey responses that are the basis for published statistical products. Specifically, the order takes aim at statistical “noise infusion” – a disclosure avoidance method that modifies a dataset “by adding random values, or noise, to certain entries.” The process nudges figures away from their true values so that no single person, household, or business can be reidentified or the database reconstructed from the results – an ironclad requirement under sections 8 and 9 of Title 13 in the U.S. Code.
- Additive noise – random values added to entries
- Multiplicative noise – entries multiplied by random factors centered near one
- Swapping – records exchanged between similar units in nearby geographies
- Formal privacy / differential privacy – noise infusion calibrated to a mathematical privacy budget
- Rounding (Probabilistic) – values rounded up or down at random
- Interval binning – values grouped and reported as ranges
- Geographic aggregation – combining small area data into larger areas
- Permitted only when coarsening is prohibited by law
- Or when coarsening would substantially defeat accuracy or usability
In its place, the order prioritizes rounding of values, aggregating smaller areas into larger geographic units and reporting data in ranges or “bins”. Data suppression – where responses are redacted to prevent respondent identification – is permitted only as a last resort, when coarsening is prohibited by law or would substantially defeat the accuracy or usability of the product.
The Census Bureau’s July 2026 determination that several of the techniques used to protect respondent confidentiality in the ACS constitute noise infusion means that statisticians will be back under the hood to reevaluate the tools the order now prohibits:
- Swapping – exchanging the characteristics of households with those of similar households in the same or nearby areas
- Partial synthesis of sensitive attributes, such as income
- Input noise infusion applied to a survey response quality measures
In practice, “noise infusion” encompasses a family of disclosure avoidance techniques used across a variety of Census products, likely making the ACS the first of many that will require an exception or reengineering to comply with DAO 216-26.
| Description | Examples of Implementation |
| Random values are added to (additive) or multiplied against (multiplicative) individual cell values or microdata entries before release, distorting the true value while preserving approximate aggregate accuracy. Partial synthesis of sensitive attributes is treated as a prohibited form of noise infusion when used as the final disclosure protection for a disseminated product |
|
| Description | Examples of Implementation |
| Records or specific values for households flagged as high disclosure risk are exchanged with statistically similar records, typically across geographies. The value itself is unchanged; the record is relocated. The DAO’s noise infusion ban is interpreted to include record or value swapping. |
|
| Description | Examples of Implementation |
| The formally private disclosure avoidance system was built for the 2020 Census. It injects calibrated random noise into counts under a differential privacy framework, then “post-processes” top-down from the national level to blocks, enforcing hierarchical consistency and invariant totals (e.g. housing unit counts). Privacy loss is governed by a tunable parameter (epsilon/privacy loss budget). |
|
Some Other Census Products Get a Temporary Exception from the Rule – including new Tennessee Higher Education Data
Post-Secondary Employment Outcomes (PSEO), along with other data products under the Longitudinal Employer-Household Dynamics (LEHD) umbrella produced by the Census Bureau’s Center for Economic Studies, looks to be in line for a temporary reprieve from the new order.
Built from data the states supply to the federal government, these products are created by linking those records across different sources. PSEO, for example, links state unemployment insurance employment records with degree data from colleges and universities, enabling earnings comparisons one, five, and ten years after graduation. That analysis is key to understanding the payoff of completing a postsecondary degree. Veteran Employment Outcomes (VEO) is a companion product showing post service employment outcomes by branch of service and state of residence. Other linkages, such as the LEHD Origin-Destination Employment Statistics (LODES), are based on home and workplace locations and are used to measure things like the dispersion of workers across industry – laborsheds.
Post-Secondary Employment Outcomes Coverage Area as of Q4, 2025 (Source: U.S. Census Bureau)
From inception, the LEHD suite has relied on some form of noise infusion to protect the confidential data. The Quarterly Workforce Indicators (QWI) and Job-to-Job Flows (J2J) lean on input noise infusion, while the newer earnings outcome products – PSEO and VEO – use differential privacy. LODES links workers’ home and workplace locations and uses both approaches. It applies input noise infusion on the workplace (employer) side and differential privacy on the residential side.
Because of the confidential nature of these records, data use agreements between state and federal agencies stipulate that noise infusion will be used. Those agreements mean that current disclosure avoidance practices can continue until new agreements and changes to the products can be put in place that are compatible with the new order.
For Tennessee, that comes as good news, since data from the Tennessee Higher Education Commission (THEC) is slated to be added to the delayed, second quarter PSEO release. That release is now tentatively slated for late August, according to a state consortium.
The State of Tennessee has contributed to the Local Employment Dynamics (LED) program at the Census Bureau since 2004, and the state’s first data from the Quarterly Workforce Indicators was posted in 2006.
Releases from one other noise infused product have also continued. Business Formation Statistics (BFS), which are a component of the Bureau’s Index of Economic Activity and a key indicator of entrepreneurship and new business dynamism, have also remained available. That product tracks new business applications at the state- and county- level and indicates how many of those are likely to translate into future employer firms. BFS relies on noise infusion for disclosure avoidance and appears to be among the products granted an exception to remain compliant with other provisions.
What’s Behind the Order and What Does it Mean for the Bureau’s Other Data Products
DAO 216-26 was announced without advance notice from the Department of Commerce. Much of what has been learned since comes from comments federal officials made at a July statistical conference in Boston, where they emphasized that transparency was the key objective behind the policy shift, noting that rules for coarsening and suppression provide a more transparent framework for determining data availability than noise-based methods.
How that principle translates into practice remains to be seen. More clarity should come as the Bureau works through implementation product by product – and, officials suggest, with renewed opportunities for data users to weigh in. But the direction is already clear: coarsening and suppression, rather than noise, will almost certainly mean less granular data or fewer published tables to maintain the same levels of confidentiality protection.
ACS delays aside, the new order halts further work to revamp the Bureau’s much debated 2020 Census disclosure avoidance tools for the 2030 decennial census data release next decade – which, by its practical effect, may well be the rule’s most consequential outcome. Internal working groups had preliminarily identified “formally private noise injection” as the primary tool for protecting detailed, small area population counts, and had committed to finalizing that system by the conclusion of fieldwork in the 2028 Census Dress Rehearsal.
That sets up another late decade switch to a new disclosure avoidance framework, similar to 2018, when the Bureau dropped swapping in favor of a pivot to a system based on differential privacy. The change came in response to growing evidence that faster computing and widely available commercial data could let outsiders piece published tables back together to identify individual responses.
With both swapping (used from 1990 through 2010) and the 2020 noise infusion-based TopDown Algorithm out of the picture, 2030 decennial products will require an entirely new disclosure avoidance regime. That makes it likely that a combination of rounding and geographic aggregation will form the backbone of the agency’s approach – which in turn could mean revisiting how many tables are published and the level of detail, in both the redistricting summary files and the Demographic and Housing Characteristics data. Similar to the 2020 process, it would likely include a round of public comment on prototype products to ensure the new methodology maintains confidentiality and serves the needs of redistricting officials, researchers, and other stakeholders before the products are finalized.
But for the small communities, researchers, and policymakers who depend on the Bureau’s most detailed statistics, the effects of DAO 216-26 may arrive well before the 2030 Census does. As noise infusion is stripped out and the available detail in existing products is reevaluated to maintain confidentiality, the trade-offs at the heart of the new rule – simpler, more transparent methods in exchange for less granularity – will start to show up in the products they have long relied on. For them, the question is not just how soon the change is coming, but if the same detail can remain when it does, in an environment where disclosure risk is rising.
This license requires that reusers give credit to the creator. It allows reusers to distribute, remix, adapt and build upon the material in any medium or format, even for commercial purposes.