The 1970 U.S. Census asked something it had never asked before. On the long form that went to a fraction of households, respondents found: "Is this person's origin or descent—" followed by circles for Mexican, Puerto Rican, Cuban, Central or South American, Other Spanish, and "No, none of these."
The word "Hispanic" did not appear. That came in 1980. What appeared was a list of six options, replacing the indirect measures the Bureau had leaned on for decades: Spanish surname in five Southwestern states, mother tongue, birthplace. Let people identify themselves, the Bureau reasoned, and you will capture the ones those proxies missed.
It captured different ones. The self-identification question counted 9.1 million people of Spanish origin. A language-and-surname measure applied to the same census counted 10.1 million. One population, one year, two statistical nations, depending on which instrument you trusted. The "Central or South American" checkbox produced a smaller confusion of its own: respondents living in the central or southern United States selected it, reading the words as geography where the form intended ethnicity. The Bureau dropped that category for 1980. None of this was error in any useful sense. It was what the printed categories made it possible to say.
The larger design choice was to keep the Spanish-origin question separate from the race question. The Office of Management and Budget's 1977 Directive No. 15 formalized that separation across the federal government, and the 1997 revision made it mandatory. The architecture assumed that ethnicity and race were independent dimensions: answer one, then the other, and the pair locates you.
For millions of people the assumption failed, quietly and for fifty years. In the 2020 Census, 43.5 percent of self-identified Hispanic or Latino respondents either skipped the race question or chose "Some Other Race" alone. More than 23 million people, whose racial identity the available categories could not describe.
OMB's March 2024 revision reversed the design. Federal forms will present a single combined question with seven co-equal categories, including Middle Eastern or North African as distinct from White. Hispanic or Latino alone is now a complete answer.
The infrastructure built on the old design will take much longer to go. The American Community Survey plans to begin collecting under the new standard in 2027, and its first five-year estimates produced entirely under the revised categories arrive in December 2032. Everything in between requires bridging methods — statistical crosswalks that let analysts read data gathered under one category system alongside data gathered under another. Those crosswalks carry the discarded assumption forward inside the new numbers, which is a different thing from abandoning it.
Consider what the old machinery did when someone refused the boxes. Before 2000, a write-in of "Black-White" was assigned to Black. "White-Black" was assigned to White. Word order determined official race. The rule was consistent, documented, defensible, and entirely invisible to the person whose identity it settled.
I wrote previously that every field on a form encodes a decision about how authority moves through an organization. A census category decides something adjacent: which populations a country is equipped to perceive, and which show up in the data only as evidence of the instrument's limits.