US bans differential privacy in Census data
5 hours ago
- #policy impact
- #data privacy
- #statistical methods
- The U.S. Department of Commerce banned 'noise infusion' from Census Bureau and Bureau of Economic Analysis statistical products.
- Statistical agencies use techniques like suppression, coarsening, swapping, and noise addition to protect data privacy, with differential privacy considered the gold standard.
- The 2020 Census adopted differential privacy to balance utility and privacy after swapping was found unsafe, but this reduced accuracy and angered some users.
- The new order targets differential privacy and prefers coarsening and suppression, which may severely harm data utility or privacy, making releases useless or unsafe.
- Removing noise-based techniques simplifies privacy attacks, as accurate statistics are easier to reverse-engineer, threatening confidentiality.
- Possible motives for the ban include enabling gerrymandering, suppressing demographic data, or ignoring the inherent privacy-utility trade-off in statistical releases.