Lifetime Citizen Portal Access — AI Briefings, Alerts & Unlimited Follows
Witnesses urge oversight and alternatives to Census Bureau's differential privacy method
Loading...
Summary
Witnesses at a House Judiciary subcommittee hearing described the Census Bureau's differential privacy approach as introducing intentional 'noise' into sub‑state counts; members from both parties called for review, testing, and alternative privacy methods to preserve confidentiality while protecting redistricting and funding data integrity.
WASHINGTON — Technical witnesses at a House Judiciary subcommittee hearing on Wednesday urged Congress to review the Census Bureau's use of differential privacy after the 2020 Census, saying the Bureau's method for protecting respondent confidentiality may distort block‑level data used for redistricting and local planning.
Wade Miller, a senior adviser at the Center for Renewing America, told the subcommittee that differential privacy "deliberately injects noise" into census results at and below the census tract level and that the practice "adds fake people to where they do not live and subtracts real people from where they do live," arguing the approach can affect redistricting and federal funding allocations.
Trey Mayfield, who served as counsel to a Census Bureau director, said differential privacy was adopted by a small group of bureau civil servants and was not subjected to peer review in ways the witnesses considered sufficient. "They refused to allow their method to be peer reviewed or to disclose the data upon which they concluded that differential privacy was needed," Mayfield said.
Johnson Yang of Asian Americans Advancing Justice acknowledged the confidentiality rationale for differential privacy, but stressed it was not applied at the state level and thus does not change apportionment totals. Still, Yang and other witnesses told members the method needs transparent evaluation, independent testing and better communication with state and local data users.
Members across the aisle discussed practical alternatives that witnesses proposed, including returning to data swapping for confidentiality, limiting swapping where appropriate, turning off characteristic (detailed) data release until geographic levels reach statistical reliability, and authorizing cleared statisticians to audit the underlying files. Several members sought unanimous consent to add a Government Accountability Office report on 2020 coverage errors to the record.
No policy change was enacted at the hearing. Members asked staff and witnesses for follow‑up materials and emphasized bipartisan work to reconcile privacy protections with the need for accurate sub‑state data before the 2030 Census.
