FOX 24 US
Census Bureau officials last Sunday conceded that a new rule banning certain privacy protection methods will make it impossible for the agency to publish certain data that it has historically released, a development that could especially harm minority groups. Representatives of the bureau, however, denied that the policy change had political objectives, though experts and advocacy groups believe it does.
Officials from the Census Bureau spoke publicly at a major statistical conference, answering questions about a policy shift first outlined in a June memo issued by the Commerce Department. The memo banned an umbrella of statistical privacy protection tools called disclosure avoidance. At the time, experts decried the sudden switch-up and told TPM the department failed to seek expert input as was customary for this kind of policy update, “making it an administration political process instead of a public process with expertise in oversight.” One expert who was at the conference said that the way in which the OMB went about implementing the policy change was highly unusual.
The rule change, they said, would likely adversely harm small populations including ethnic and language minorities, and rural communities. They saw a political objective: Disclosure avoidance had long been targeted by right-wing groups, including those associated with Russell Vought, director of the Office of Management and Budget, and Stephen Miller, White House Deputy Chief of Staff.
Fielding questions from a group of scientists and researchers, bureau officials — including Acting Director George M. Cook and Michael Lachanski, deputy director for data, science and policy — confirmed that entire data sets would be pulled from publication as a result of the Commerce Department’s disclosure avoidance ban, according to live notes taken by conference attendees and shared with TPM.
“Generally speaking, small populations, small subgroups have historically been the hardest to protect,” an official told the group, according to the notes.
The individual is unnamed in the notes.
“They are also historically the ones who get hurt the most by disclosure avoidance,” the person continued, adding: “There is important work that needs to be done to meet those communities’ concerns.”
The Commerce Department and the Census Bureau did not immediately respond to TPM requests for comment.
Disclosure avoidance can include noise infusion, in which “noise” is used to fuzzy data in order to obscure individual respondents’ identities; one form of noise infusion is called differential privacy. The practice allows the federal government to publish granular data down to the micro-geographic level while also minimizing the risk that bad actors could use the data to identify specific people or groups. The bureau is required by federal law to protect respondent privacy and, per the June memo, will instead of differential privacy use two much more limiting privacy practices that will force staff to either “suppress” or not publish certain data sets, or to “coarsen” or aggregate up, from, for example, the block level to the county level, if individual privacy can’t be guaranteed otherwise.
“I saw the order as taking us back several decades in the tool kit for disclosure avoidance and privacy protection,” Salil Vadhan, a computer scientist who advised the Census Bureau on its use of differential privacy and who attended Sunday’s panel, told TPM.
Vadhan said suppression and coarsening have checkered histories regarding their efficacy.
“They can often be reverse engineered and undone,” Vadhan said, “and that was a lot of the motivation for the start of research on differential privacy.”
Conference attendees’ concerns about the lack of expert input bubbled over into questions about whether the motivation behind the rule change was political in nature.
“This appears to be a decision imposed on the Census Bureau,” Steve MacFeely, chief statistician of the Organisation for Economic Co-operation and Development, said. “Usually disclosure control would be considered a technical decision that should be made by professionals. How worried are you that this may be perceived as political interference?” he asked Sunday.
“I found it very curious,” MacFeely later told TPM, “that they described this as a policy issue rather than a technical one.”
Bureau officials said OMB was involved in coordinating this policy change, and that “all kinds of technical standards” come from the White House. John Abowd, former chief scientist at the Census Bureau, told TPM this is not necessarily unusual; OMB has historically been involved in policy management.
But, typically, policy setting of this magnitude comes with rigorous stakeholder engagement and feedback that often takes years.
“This is really unprecedented,” Beth Jarosz, a senior fellow at the Georgetown Massive Data Institute, told TPM in July. “This is taking the decision out of that accountability framework and making it an administration political process instead of a public process with expertise in oversight.”
America First Legal, a group co-founded by Stephen Miller, and the Center for Renewing America, which was founded by Russell Vought, have each targeted the use of differential privacy in the 2020 Census, claiming the policy caused an over-count in states with larger undocumented immigrant populations. That alleged overcount could give greater political power to states that tend to vote for Democrats. The practice, America First Legal has argued, interfered with the population-based division of congressional representation for each state, called apportionment, and took electoral power away from red states.
Those claims are untrue, Abowd and other experts have stressed to TPM.
Differential privacy does not impact the number of people counted and used for apportionment. But as right-wing groups push to dilute political representation for people of color, eliminating the amount of data the Census Bureau can publish about protected groups could have an adverse effect on those communities. If demographics about racial and ethnic minorities cannot be published, there’s a risk that relevant information that could be used to draw fair political maps will be obscured.
“There will be an impact on how granular the data is about what demographics of people [there are], and what characteristics of people there are, at what counts, in small geographic areas,” Vadhan said. “Exactly what those effects will be, I think, remains to be seen.”
