# Differential Privacy A formal mathematical framework for limiting what can be learned about any individual from published statistics, by injecting calibrated random noise into released tables and accounting for the total privacy budget spent across all releases. The Census Bureau adopted it as the disclosure-avoidance system for the [[2020 Census]] after demonstrating that its earlier methods allowed substantial reconstruction of individual records from published tables. The trade-off is explicit and contested: more noise means stronger confidentiality guarantees and less usable data at small geographies, which matters most for race and ethnicity data in small places, exactly the data that voting-rights litigation depends on. State-level apportionment totals are published as enumerated and are not noised, so the framework affects the redistricting and funding layers rather than the seat math. See [[How the Census Apportions Congress]] and [[Census Bureau]]. **Update, 2026:** Commerce Department order [[Disclosure Avoidance for Statistical Products]] (DAO 216-26), effective June 4, 2026, states that "any use of noise infusion is inconsistent with the Department's policies" and directs the Census Bureau toward coarsening first and suppression as a last resort. Already-published 2020 Census data is unaffected; the method for 2030 is not yet on the public record. ## See also - [[Census Data Products Tree]] — the products the TopDown Algorithm protected, and what the 2026 order does to them. - [[Census]] · [[Census Bureau]] · [[2020 Census]] For the routes by which a person or a researcher can reach census data at all — and the disclosure-protection gate standing on each one — see [[Census Data Access Crosswalk]].