By Adriel Cepeda Derieux*
Late last month, the Supreme Court heard argument in Department of Commerce v. New York. The case will determine whether the Trump administration can change the census by adding a question that asks the citizenship status of all persons in the United States. Despite petitioners’ misleading arguments, that question has no precedent or tradition. When a citizenship-related question has been asked at all—as it is currently, in the American Community Survey (ACS) sent annually to 3.5 million households—it has been directed only to a fraction of the population. It will also hurt our democracy. By the government’s own admission, the proposed question will cause a likely drop in self-response of at least 6.5 million people. The resulting undercount will cost states like Arizona, California, Florida, and New York a seat in Congress, votes in the Electoral College, and billions in federal funding.
At argument, several justices focused on Commerce Secretary Wilbur Ross’s single, publicly stated goal for the citizenship question: to collect better data for the Justice Department to enforce the Voting Rights Act (VRA). Chief Justice John Roberts, for one, described the Citizen Voting Age Population (CVAP) data that the question will purportedly generate as “the critical element in voting rights enforcement.”
But the VRA’s unbroken application without such data gainsays this characterization. As District Judge Jesse Furman stressed in his judgment setting aside Secretary Ross’s decision, no VRA case has ever been thwarted by a lack of data that the citizenship question could generate. In the VRA’s 54 year history, “DOJ ha[s] never before cited a . . . need for citizenship data from the decennial census; never before asserted that it  failed to bring or win a VRA case because of the absence of such data; and never before claimed that it had been hampered in any way” without it.
At argument, plaintiffs explained why a citizenship question cannot help VRA enforcement. Under Thornburg v. Gingles, the VRA requires that a “majority-minority” district be drawn when a minority group represents a majority of the voting age population. But since, as Secretary Ross acknowledged, noncitizen self-responses to the citizenship question on sample surveys are consistently inaccurate—“about 30 percent of the time” on the ACS—the data a census question generates will be near worthless to determine whether minority voters are a majority in areas with relatively large noncitizen populations.
A brief exercise illustrates the problem: assume a school has 10 classrooms, each with 10 students. Now assume the school sets out to determine how many of those 100 students did their homework by asking “did you do your homework—yes or no?” In truth, all students in 9 of the 10 classrooms did; but none of the students in the remaining classroom completed the assignment. However, when asked, 99 students (all 90 who actually did their homework, and 9 of the 10 who didn’t) say “yes.” The school’s question now has an overall accuracy rate of 91 percent. That may sound impressive, but it is wrong as to 90 percent of students in the tenth classroom, and in no way helps the school identify students who did not complete their assignment—or learn that they sit together.
If, as on the ACS, at least a third of noncitizen responses to the citizenship question are consistently inaccurate, the data a census question generates will be similarly unhelpful to identify majority-minority districts in areas that have a large noncitizen population belonging to the minority group.
The government offered that most respondents accurately report their citizenship status on existing Census Bureau sample surveys. But that is only because the vast majority of people in the United States are citizens who answer the question correctly almost all of the time. The same will not be true precisely in the areas where citizenship data is used to enforce the VRA—i.e., places with large noncitizen populations, where the data helps assess whether voters of color constitute a majority of a district’s citizen voting-age population.
Census Bureau technical advisors understood all this. They advised Secretary Ross that estimates based on administrative sources already found in legal documents—such as Social Security Administration records—“better me[t] DOJ’s [VRA-related] uses” could be used without “harm[ing] the quality of the census count.” For decades, the Justice Department and private plaintiffs (including plaintiff ACLU and various amici) have successfully relied on such estimates. Indeed, the defendants could never identify a single VRA case that failed because of a data shortfall. The one thing that has changed is that the Trump administration needed a reason to add a citizenship question to the census. Perversely and incorrectly, the VRA supplied one.
* Adriel Cepeda Derieux is a staff attorney in the ACLU Voting Rights Project.