Commerce Department Order Bans Noise Infusion from Census and BEA Statistical Releases

The order directs agencies to prefer coarsening and suppression over any randomized disclosure avoidance method. It explicitly reverses the 2020 Census disclosure avoidance system that injected calibrated noise after reconstruction attacks demonstrated swapping failures on 2010 data. Primary technical documentation from the Census Bureau showed differential privacy parameters were tuned to the minimum epsilon that still blocked record reconstruction while maximizing utility under the new threat model.

Reconstruction experiments published in 2018-2019 by Census researchers established that prior swapping left 2010 block-level data vulnerable to linkage with commercial datasets at rates exceeding 70 percent for many attributes. The 2020 implementation reduced successful re-identification below statistical thresholds but introduced measurable error in small-area counts used for redistricting and funding formulas. No subsequent public benchmark has compared reconstruction risk under the mandated non-random methods against the 2020 parameters.

Demographic research pipelines and state redistricting commissions will absorb the largest operational impact. Small-area population estimates that previously carried documented noise margins will now embed undisclosed coarsening artifacts, complicating longitudinal studies and Voting Rights Act compliance calculations. Agencies have 180 days to publish revised suppression thresholds; absence of randomized noise removes any mechanism for quantifying residual disclosure risk.

Census Bureau production schedules for 2025 county-level estimates already list coarsening rules as the default, with suppression applied only when cell counts fall below new unpublished floors.