The US Department of Commerce, through its Office of Privacy and Open Government, has posted an order for federal statistical agencies to stop using randomization as a (worthwhile!) mean of protecting confidential data. Another instance of going against scientific arguments and validated procedures, and of dismantling decade long expertises, and of stopping a long series of observations, and of destasbilising federal agencies, from the Trump administration. To wit:
b. Statistical Policy Directive No. 1, codified at 44 U.S.C. § 3563, requires each statistical agency, as enabled by the head of the parent agency, to adopt policies, best practices, and appropriate procedures to ensure the confidentiality, accuracy, and objectivity of its statistical activities and to disseminate timely and relevant statistical products.
SECTION 5. POLICY.
.01 Ensuring the Accuracy, Confidentiality, Objectivity, and Relevance of Statistical Products
a. The Department shall, as a primary objective, aim to fulfill its statistical obligations by providing the public with accurate and objective information.
b. The Department is firmly committed to striking a balance of accuracy, confidentiality, objectivity, and relevance for each statistical product that is consistent with its statistical obligations and the applicable legal requirements.
c. Any use of noise infusion is inconsistent with the Department’s policies.
.02 The Census Bureau and the Bureau of Economic Analysis shall adhere to the following order of priority when considering and applying Disclosure Avoidance:
a. Coarsening shall be the preferred category of Disclosure Avoidance methods for all statistical products.
b. Suppression shall be permitted as a last resort, only to be used when coarsening is prohibited by law or would substantially defeat the accuracy or usability of a statistical product.
c. Noise infusion shall not be used for any statistical product.
.03 This policy shall not be interpreted to conflict with any constitutional, statutory, regulatory, or other legal provision.
So here go (to the drain) differential privacy, enforced by the Census Bureau under the direction of John Abowd, as well as synthetic data, subsampling, et cetera! This will directly impact the coming 2030 census. With the asinine arguments by Commerce Department Kristen Eichamer that this is done to “maintain public confidence in our data while upholding our duty to safeguard the privacy of those who provide information” and that “indiscriminate use of noise infusion—even when not mandated by law—ultimately undermined confidence in the department’s products and cast doubt on their integrity.” NPR relates this order to the introduction of differentially private measures around the 2020 census and irrational Republican attacks on the procedure. Which reminds me of an earlier opposition to random sampling instead of “exhaustive enumeration” to correct for undercounts, which brought statisticians from both sides almost to blows! (With a defence of the correction procedure by my later friend Steve Fienberg.)
