Good detailed and balanced analysis:

After a long career as a banker and investor, Commerce Secretary Wilbur Ross is no doubt familiar with cost-benefit analyses. That seems to have carried over to his political work. In a memo declaring that the 2020 census would ask U.S. inhabitants whether they are U.S. citizens,¹ he wrote, “I find that the need for accurate citizenship data and the limited burden that the reinstatement of the citizenship question would impose outweigh fears about a potentially lower response rate.” The inclusion of the question was a request of the Justice Department, which says that it needs the information to enforce the Voting Rights Act.

But Ross isn’t the only one weighing costs against benefits when it comes to the census — respondents do it as well. Demographers and civil rights groups are concerned that under a president who has called for a ban on Muslims and immigrants from certain countries, dramatically reduced the number of refugees allowed into the country and cracked down on undocumented immigrants without criminal records, a citizenship question will push more people to decide that the risks of responding accurately to the questionnaire, or responding at all, outweigh the benefits. And the groups that seem most likely to be put off from responding — immigrants, members of households with immigrants, people living in poverty, among others — are the same ones that are already at highest risk of being uncounted.

There’s a lot at stake: The census has been used for hundreds of years to determine how many U.S. House members each state will have,² and it currently helps determine how hundreds of billions of dollars in federal spending is divvied up. “The risk that really troubles me is that there’s a big undercount and then there’s a big lack of representation,” said John Thompson, who was director of the U.S. Census Bureau until he resigned last year (the bureau is still without a director).

Many groups were already less likely than others to respond to the census. Some of the non-response trends are geographical. The rural South and the Texas-Mexico border, for example, had many areas with low response rates during the last census, in 2010, according to data from the Center for Urban Research at the City University of New York’s Graduate Center.

But there are pockets with low response rates almost everywhere, said Steven Romalewski, director of the center’s mapping service. “Every state has them,” he said. “Most congressional districts have them. It’s urban, rural and suburban, and they are scattered throughout the country.”

That’s at least partly because there are differences in mail-in response rates among demographic groups. African-Americans and Hispanics respond at lower rates than non-Hispanic white people. Immigrants (particularly the undocumented), people who rent their homes and those living in poverty have been less likely to mail back the form.

Those are also some of the groups that have historically been undercounted. For example, young children — the group most likely to be undercounted – disproportionately live in households with parents who are young, who earn poverty wages, and who are Hispanic or African-American.

The citizenship question could exacerbate the problems of non-response and undercounting. In pre-census focus groups, respondents have expressed concerns that other government agencies will be able to access data related to immigration and that it could harm their residency status (even if they are authorized). Community groups across the country have been educating undocumented immigrants and their families about their rights, encouraging them not to let law enforcement officials into their homes. This could make it more difficult for census workers to access households. It’s not just the undocumented who are at risk of not responding or not showing up on the census. The 23 million non-citizensliving in the U.S. often live with U.S. citizens as well — if the door doesn’t open, citizens are at risk of not showing up in the census, too.

Researchers believe that a resistance to sharing any personal information and the fear that one’s information will not be secure are among the reasons that people don’t respond to the census. Lawmakers themselves, most recently Republicans, have expressed concerns about the broad nature of census questions, calling as recently as this decade to end the American Community Survey — an annual survey also conducted by the Census Bureau that does ask about citizenship status. We don’t know how much public fears and political rhetoric have affected people’s willingness to participate in the census over time, but we do know that when the bureau began spending millions of dollars on advertising campaigns to assuage those concerns (“Your answers are protected by law”), response rates went up.

There’s a tradeoff between privacy and accuracy, said Kenneth Prewitt, who was census director from 1998 to 2001. The more infringement there is on information that people view as private, the less accurate the results will be. And this close to the 2020 survey, it’s likely not only the citizenship question that puts the census at risk, Prewitt said. That the census is now mired in a national political conversation about immigration, as well as various court cases pushing to keep the question off the survey, polarizes it in a way that could hurt response rates.

We don’t know whether the addition of the citizenship question will make the data that the census collects less accurate as a whole, though census workers have heard an alarming increase in concerns around immigration and privacy in focus groups conducted in advance of 2020. We can’t know what the question may do because it hasn’t been tested in a way that follows standard scientific practice, said Terri Ann Lowenthal, who is a former staff director for the U.S. House census oversight subcommittee and now works as a consultant. In surveying, many things can change response rates and the truthfulness of responses, including the order of questions, the wording on instructions and the way it’s laid out visually. The only way to know how well a question will work is by testing it repeatedly, over a number of years, she said.

“It is somewhat puzzling, in my opinion, that Secretary Ross — who is a well-respected businessman — would agree to move forward with something that I’m sure he knows in any other setting, whether scientific or business, wouldn’t pass muster in terms of readiness,” Lowenthal said.

But even though a citizenship question hasn’t been tested for the current census (or in the current political environment), there’s good reason to believe the answers will be inaccurate for those who do fill out the form, at least among non-citizens. According to Ross’s memo, some 30 percent of non-citizen respondents on the American Community Survey are believed to give incorrect responses.

There’s no good way to fix the census if there is a problematic count — we’re stuck with it for a decade. In the late 1990s, the bureau floated plans to use statistical methods to make up for chronic undercounts of groups like kids, renters and certain minority groups. The House sued, and the Supreme Court ultimately ruled that because of the way the Census Act is written, statistical sampling can’t be used for apportionment. The census is a one-shot deal.

More than a dozen states are suing to block the citizenship question from appearing on the 2020 census. And civil rights groups say they are holding out hope that Congress, which has jurisdiction over the survey, will intervene.

In the end, as Ross seemed to hint at, the citizenship question is about tradeoffs. It may provide additional information about the number of citizens and non-citizens in the U.S., but only if people respond. Because the question hasn’t been tested, understanding how it will affect the outcome is difficult. But a chorus of experts, including people who have worked at the Census Bureau, say that there’s real cause for concern and that our representation at the federal level is at stake.