US Census Set for Changes

Of the topics carefully reviewed by Constance Citro in “Protecting the Accuracy of the 2020 Census” (Issues, Summer 2019), I select one for elaboration: she writes that every census is a lesson in how to improve the next one. This will be so for 2020, but with an unprecedented outcome: the 2030 census will look less like the 2020 census than the 2020 resembles the 1790 census.

This rash statement is not predicated on the serious threat posed by current treatment of the census as a tool to gain partisan advantage. That is being equally seriously resisted at state and local levels and by commercial and advocacy groups, thus far with success, as evidenced by events surrounding the proposed addition of a citizenship question. The American people do not want to lose what the Constitution gave them—a reliable census that, among many other benefits, allows them to hold politicians accountable for their often-exaggerated election promises.

The turning point that will follow 2020 is not in politics but in a new data science designed for twenty-first century conditions. The Census Bureau will still anchor the nation’s information platform, as it did across the nineteenth and twentieth centuries, but it will provide greater demographic and geographic granularity; it will have the capacity to continuously update key variables in areas such as health, economics, transport, and agriculture; and it will offer stronger privacy protection than currently available.

The American Community Survey will remain, and now joined by two new data flows: one created by linkage across federal and state administrative records, and the other by arrangements to draw from third-party data—especially commercial transactions and social media. We will learn from 2020 how soon, effectively, and accurately we can begin to replace (expensive) survey data with (already paid for) administrative record data. (The nasty, wasteful legal battle in 2019 over adding a citizenship question to the 2020 form did, after all, end with an agreement to use administrative data, as was, from the outset, strongly recommended by the Bureau.)

Third-party data is less ready, first needing extensive scientific attention to barriers and challenges regarding such areas as privacy, proprietary constraints, standardization, data security, the protection of trend lines, and, of course, public trust. In some areas, progress will be quick and impressive, as is true for the census address file and today’s economic statistics; in other areas, slow and frustrating. But the overall picture points to a future in which third-party data will provide much of what census and survey data now make available.

This transforms the Census Bureau from its exclusive focus on data collection to an additional task of curating data—assessing the accuracy, coverage, privacy, and costs of commercially provided data products, and making decisions about which can be safely incorporated into the nation’s statistical system. And this is why the 2030 census will look less like the 2020 census than the 2020 resembles the 1790 census.

Kenneth Prewitt