Attention: You are using an outdated browser, device or you do not have the latest version of JavaScript downloaded and so this website may not work as expected. Please download the latest software or switch device to avoid further issues.
| 3 Aug 2026 | |
| Blogs |
By Nick Hart, President and Chief Executive Officer of the Data Foundation
On August 2, 2026, at the Joint Statistical Meetings in Boston, Census Bureau leadership devoted a session to the Commerce Disclosure Avoidance Order (DAO-216-26), and then took questions from the audience for more than an hour. Michael Hawes chaired the session, and George Cook, Acting Director of the Census Bureau, and Michael Lachanski, Deputy Director for Data, Policy, and Science, participated in a facilitated dialogue. The session at the annual statistical community meeting was the fullest public account so far of how the order will be implemented, and the sustained back-and-forth signaled a genuine willingness to engage a community that has had a lot of questions about implementation and impacts. The Data Foundation was in the room alongside statisticians, data scientists, and fellow privacy and data policy experts.
The single most important and explicit takeaway can be a reassuring one for the community: data products are continuing. Census Bureau leadership articulated that no statistical products are being discontinued because of the order. However, affected products will transition over time, in some cases across a year or two, and the change is intended, in Commerce’s account, to improve the accuracy of published data going forward, rather than unwinding what has already been released. Much of the speculation among data users over the past month has run ahead of that clarification from Census leadership.
A second theme ran through nearly every answer from Census Bureau leadership. The panel returned again and again to the tradeoff at the heart of disclosure avoidance, among the accuracy of published statistics, their availability and level of detail, and the confidentiality protection applied to them. As described during the session, when accuracy, availability, or confidentiality are affected, the others are also affected; the order does not remove that tension, but shifts the prioritization of the tradeoff.
The Data Foundation continues to provide resources to the data, evidence, and privacy communities to understand what the Commerce Order does and does not mean, and to provide a neutral place where the community can be heard and provide feedback.
Several key points from the session may allay fears circulating online and among data user professional societies.
George Cook, the Acting Director, described the order's core intent in plain terms: it "was really meant to make a fairly blunt statement that the Census Bureau will collect data and disseminate more or less what it's collected" while highlighting a priority toward accuracy. The order's scope is deliberately broad, covering information products from Census Bureau staff, contractors, and grantees. Michael Lachanski offered a simple rule of thumb: "if you see a number in a report that would be protected by Title 13, it would be affected by the DAO." (Title 13 is the part of U.S. Code that “authorizes the Census Bureau to conduct surveys and censuses, and mandates that any information obtained from private individuals and establishments remains confidential.”) Work presented at conferences or through the Federal Statistical Research Data Center (FSRDC) system that carries the standard "views are my own" disclaimer was described as outside the order's reach, since it is not designated a “dissemination event” by the Census Bureau (with an ongoing review about joint projects that directly involve Census Bureau personnel).
The order is intended to be forward-looking. Already-published products are not being rewritten and there is no wholesale discontinuation of data products or series. For example, Population Estimates derived from the 2020 Census baseline continue, since the noise predates the policy. Instead, affected products are planned to transition over time to implement the new order.
Cook and Lachanski offered concrete illustrations in the discussion. For example, the American Community Survey (ACS) will phase toward full compliance, with the first fully compliant five-year file expected in 2029. The Longitudinal Employer-Household Dynamics (LEHD) suite relies on voluntary state unemployment-insurance data, and roughly a dozen of those state agreements contain provisions requiring noise infusion. Because LEHD is treated as a single product, it will keep using noise infusion until state provisions are revised in coming years, which would enable time for ongoing compliance. Internal benchmark and frame files already produced with noise can keep being used, provided no additional noise is added.
On methods, the Census Bureau intends to treat swapping and final-layer synthetic data as forms of noise infusion, so both will not be available for disseminated products. DS025 draws the line by use: synthetic data remains available for imputation, small-area modeling, simulation, and validation-server code development, while synthesis used as the final protection for a disseminated product, such as partial synthesis of sensitive attributes in the ACS, is treated as prohibited noise infusion. Michael Lachanski explained the Census Bureau's determination: "the conventional academic literature, at least in the last maybe 20 years, and the Census Bureau itself had defined swapping as a type of noise infusion, and therefore our determination is that it changes the values of the statistics of the records that would be protected under Title 13, and therefore would not be allowed for disseminating statistical products." DS025 makes this explicit, stating that the Census Bureau interprets noise infusion to include record or value swapping. Coarsening, suppression, and non-deterministic coarsening such as probabilistic rounding remain available, though the latter excludes 2030 Census products and requires DSEP review and approval product-by-product. Planning for the 2030 Census is being recalibrated, with demonstration products expected in mid-to-late 2027 and public feedback feeding decisions in 2028.
If one idea recurred more than any other across the hour of questions, it was the three-way tradeoff among accuracy, availability, and confidentiality. Moving along one dimension comes at a cost to one or both of the others, and the panel raised this repeatedly rather than in passing. In the field, these ideas are often described as dials, indicating that they are not binary concepts.
Two things stood out, and together they are the honest bottom line. First, the interpretation of confidentiality protections has not changed for Title 13 or the Confidential Information Protection and Statistical Efficiency Act (CIPSEA) data to the Census Bureau. Reconstruction and reidentification of individual responses remain unacceptable, and detecting a reidentification in a product still means pulling that record. Second, holding protection constant while prioritizing accuracy has real costs elsewhere. As Hawes indicated during the session: "There will be consequences. That's unavoidable," a reading consistent with the tradeoff the Census leadership presented throughout.
The mechanics of implementation are straightforward. Suppression withholds some statistics from publication, and coarsening reduces the granularity of what is released. Shifting from noise infusion toward these methods tends to reduce granularity or availability for small-area and small-population statistics. That is precisely the data many rural areas, small communities, and small demographic groups rely on. This issue was acknowledged during the two-hour discussion as a difficult problem the Census Bureau has wrestled with for decades. Stating the tradeoff is not a verdict on the order, but it offers context the data-user community and the Census Bureau needs in order to plan and implement the next steps.
Behind the tradeoff sits a distinction the session helped sharpen, between two kinds of decisions that often get blurred: policy decisions and statistical ones. A decision about whether to prioritize accuracy and how much confidentiality risk is acceptable, are policy judgments that belong to accountable policy officials. How to carry the choice out in the data belongs to the agency's statistical and technical staff. Under the Foundations for Evidence-Based Policymaking Act, or Evidence Act, invoked only once during the session by Michael Hawes, this distinction is key because of a regulation called the Public Trust Rule for short, that delineates the interactions between statistical agencies and their “parent agencies.”
Michael Hawes was direct that the balance itself is a policy question: "Finding that optimal balance is inherently a policy decision. It is not a technical one. It is a policy decision, and the DAO is about informing those policy decisions."
Much of the confusion in the data and privacy community comes from treating policy and technical decisions as the same thing. In fact, the first commenter in the audience treated the issue as settled, while numerous questions from the audience blurred the distinction.
A useful example surfaced during the session: the choice of a differential-privacy parameter such as “epsilon” was long presented by Census Bureau staff as a technical matter when it was really a policy call about how much disclosure risk is acceptable. Raising this point from the floor, the Data Foundation's President & Chief Executive Officer, Nick Hart, noted that "the order itself is clearly a policy decision, but there are other policy decisions that have happened over time that were blurred as statistical," and pointed to epsilon as "one of those masked decisions that hides as a technical [matter] but is actually a policy distinction." Naming that distinction plainly, and applying it consistently, is a step that serves the policy and statistical communities alike.
The Census Bureau leadership framed transparency as a central goal, pointing to the Open Census effort and a broader move toward releasing code and modernizing production pipelines. One argument for the shift is that coarsening and suppression are more visible to users than noise infusion, so their effects can be seen and accounted for.
On implementation guidance, the panel noted that some material is already public and that more is forthcoming through a Census Bureau blog. (This post will link to that guidance once it is published.) Existing guidance includes a July 31, 2026, update to the Census Bureau’s Disclosure Review Board documentation (DS025).
That leaves the question raised from the floor still unanswered. Pressing the point, Hart asked "what does engagement look like going forward?" and, more pointedly, how the community can give "meaningful feedback" and what "mechanism" exists to stay looped in, whether on the regulatory process ahead, on products that are changing, or on the granular questions that arise in day-to-day use. Panelists welcomed feedback and committed to showing up at venues like the statistical meeting. Building a standing channel is an area that will continue to be fruitful moving forward.
While the Census Bureau leadership answered dozens of questions, there are many remaining questions to address in implementation for policymakers, scientists, privacy experts, and the user community. For example:
The Data Foundation is committed to helping our stakeholders and the policy community navigate what the order means in practice, and we will update this post as the Census blog and implementation guidance become public. The more clearly the policy questions and the statistical questions are separated, and the more openly both are worked through, the better served everyone will be, data users and the policymakers alike.
Resources:
Census Disclosure Review Board DS025 (July 31, 2026)
Commerce DAO 216-26 (June 4, 2026)
