Via Kottke – The U.S. Census Is Wrong on Purpose: “…Full census data is only made available 72 years after the census takes place, in accordance with the creatively-named “72 year rule.” Until then, it is only available as aggregated data with individual identifiers removed. Still, if the population of a town is small enough, and census data for that town indicates, for example, that there is just one 90 year old woman and she lives alone, someone could conceivably figure out who that individual is. So the census bureau sometimes moves people around to create noise in the data that makes that sort of identification a little bit harder…”

See also the U.S. Census: “Modern computers and today’s data-rich world have rendered the Census Bureau’s traditional confidentiality protection methods obsolete. Those legacy methods are no match for hackers aiming to piece together the identities of the people and businesses behind published data.  ​ A powerful new disclosure avoidance system (DAS) designed to withstand modern re-identification threats will protect 2020 Census data products (other than the apportionment data; those state-level totals remain unaltered by statistical noise). ​ The 2020 DAS is based on a framework for assessing privacy risk known as differential privacy. It is the only solution that can respond to this threat while maximizing the availability and utility of published census data…”