President Donald Trump has ordered the U.S. Census Bureau, to do something it would prefer to avoid: Determine the citizenship status of every U.S. resident, and where they live, without asking them. And he wants the answer by early 2021.
The presidential directive, issued in July, is an outgrowth of a failed attempt by the Trump administration to put a citizenship question on the 2020 census to be conducted next April. And meeting his short deadline will be a stern test for the nation’s largest statistical agency.
The technical challenges are enormous. The Census Bureau has never before collected and disseminated citizenship data at the fine level—as small as one or two city blocks—that the president wants, and the bureau is still trying to figure out how to produce what it calls the Citizen Voting Age Population (CVAP) file.
“We need to understand what the expectations are about that product, and at the moment we do not,” John Abowd, chief scientist at the Census Bureau in Suitland, Maryland, told reporters last week after making a presentation on the topic to the bureau’s top-level scientific advisory committee. But he also assured the committee that it could be done.
Another complication is the bureau’s decision to adopt a more rigorous approach to protecting the identity of everyone whose information is contained in any data it releases. Bureau officials are still tinkering with that tool, called differential privacy, which will be used for the first time in reporting the results of the 2020 headcount that begins on 1 April.
Some researchers fear that, by complying with the president’s executive order, the Census Bureau could tarnish its stellar reputation for nonpartisanship. The CVAP data are already sensitive because they relate to the politically explosive question of how many people are living in the country without proper documentation. Trump’s order could further inflame that debate because he wants states to be able to use the citizenship data to draw new boundaries for state legislative and congressional districts in 2021. (As its name implies, a CVAP file contains information on people 18 years and older who are citizens and, thus, eligible to vote.)
Redistricting is a quintessential political exercise, and some Republicans believe removing noncitizens from the tally of eligible voters would be a boon to their candidates. But many legal experts say it may be unconstitutional for states to ignore noncitizens when setting boundaries for the nation’s 435 congressional districts. And last week, the bureau’s parent agency, the Department of Commerce, revealed that no state has yet requested citizenship information for the purposes of redistricting.
Given that absence of demand, some census experts are unhappy that the Census Bureau is being asked to make such a herculean effort to generate the data. “If nobody has requested [a block-level CVAP] for that purpose, then why does the president think we need it at that level of granularity?” asks Terri Ann Lowenthal, a Connecticut-based consultant on census issues. Nor, she says, are such detailed data on citizenship needed to help federal agencies allocate some $750 billion in federal funds, another major use of the data that the census collects.
Are you on file?
How did we get here? On 27 June, the Supreme Court shot down a controversial decision that Commerce Secretary Wilbur Ross had made 15 months earlier to add a citizenship question to the 2020 decennial census. A court majority found that the administration’s rationale—to ensure enforcement of the Voting Rights Act—was “contrived.” Ross then dropped the idea. But on 11 July, Trump instructed the Department of Commerce to collect the information “in connection with the census [and] by other means.”
Instead of asking residents directly about their citizenship status, the department was told to use so-called administrative records, or information already on file within various government agencies. The 10-page executive order gave Ross the authority to request administrative records relating to citizenship status from a long list of agencies, including the Department of Homeland Security (DHS), which keeps records on immigration and naturalization processes, and the Department of State, which issues passports.
An interagency working group is now reviewing exactly which records would help the Census Bureau do the president’s bidding. Census officials must then figure out whether the records could be incorporated into the bureau’s existing data systems in time to produce CVAP data along with the rest of the results from the 2020 census in February and March of 2021.
The discussions began after Ross decided in March 2018 that such records would be used to supplement information obtained from the question that he planned to add to the 2020 census. And Abowd said last week that the negotiations among the agencies “have been proceeding quite smoothly … we are pretty far along, and we anticipate successfully completing them.”
What’s your address?
The goal of the decennial census, which is required by the U.S. Constitution, is to place each person at a particular—and correct—address on Census Day and collect certain demographic data—notably age, race, and gender—about them. The executive order, in effect, adds citizenship status to that list.
The bureau already uses administrative records and informed scientific guesswork to fill in the gaps of those who don’t return the census form. But compiling citizenship data that way from scratch will be a formidable challenge, says economist Amy O’Hara, a research professor and co-director of the Census Bureau’s Federal Statistical Research Data Center at Georgetown University in Washington, D.C.
She led the Census Bureau’s effort to use administrative records until leaving in mid-2017, so she knows what the bureau is up against in trying to link a person to a particular address. And she cites her recent peregrinations as an example of the challenge it faces.
“If you tried to identify me by looking at the last 5 years of tax records, you’d discover that I’ve been at two Maryland addresses, one California address, and one Virginia address,” O’Hara says. “Which one would you pick?”
That’s where administrative records can come in handy. The Census Bureau already has an agreement with the Social Security Administration (SSA), which covers the approximately 90% of U.S. residents who have been given a Social Security number because they are citizens.
But what about someone who may have entered the country without proper documentation, moves frequently, and is not a citizen? In addition, SSA files don’t contain a current address, so knowing that someone is a citizen still isn’t enough to locate them if they don’t participate in the decennial census. (O’Hara says she expects to fill out the 2020 census promptly.)
DHS records could help, but they are far from a panacea. DHS’s Citizenship and Immigration Services division gives anyone approved as an official immigrant an alien registration number (A-number). It stays with them as they apply for benefits and later, in some cases, to become a naturalized citizen. But only permanent residents receive an A-number. Someone who overstays a temporary visa and remains in the country won’t have one, for example, nor would anyone who is undocumented.
In addition, someone receiving an A-number is likely to have no fixed address. And even after they are settled, any address they eventually provide could be that of a relative, a place of employment, or even someone helping them manage their immigration status.
Can the American Community Survey do it?
The Census Bureau already releases some data on citizenship. They come from the American Community Survey (ACS), a 72-item questionnaire that the bureau sends out every month. In 2005, the ACS replaced an extended version of the decennial census. But it annually reaches only about 2.5% of all households, compared with the roughly 16% that received the long form. And the CVAP file it generates is based on responses averaged over 5 years, with weighted statistics used to stretch its small sample across the entire country.
The executive order says the ACS’s small distribution has resulted in “substantial margins of errors” that limit its usefulness as a source of citizenship data. For all those reasons, the ACS product has never been good enough for state officials to use in drawing election maps.
“Everyone in the statistical community agrees that the ACS should be at least doubled in size, although most are justifiably nervous about the rationale in the EO [executive order],” says Steven Ruggles, a demographer at the University of Minnesota in Minneapolis.
Indeed, many demographers find it ironic that the executive order suggests a larger ACS. A bigger survey would cost more money, they note, and Trump has consistently requested less money for the ACS and other census activities in his annual budget submissions to Congress.
Scientists have another reason to question Trump’s commitment to better demographic data, Lowenthal says. “It is the president’s most ardent supporters in Congress who believe the ACS is too long, too nosy, and shouldn’t be mandatory,” she notes. Some legislators have even tried to make the ACS voluntary, arguing that the government doesn’t need all that information.
Where’s the bar?
The Census Bureau has another 6 months, to 31 March 2020, to determine “the viability of each potential administrative data source on citizenship,” Abowd says. He says that date is also when researchers will learn about the “methodology and format of the block-level CVAP data” to be released in 2021.
But O’Hara says the bureau has some wriggle room if officials decide the presidential directive represents a mission impossible.
“Every U.S. statistical agency has quality standards that must be met before any data can be released,” she says. “And that applies to the executive order as well. The question for Census is how to assess if the [citizenship] data are above the bar, and what the bar should be?”
“This is not something that they want to do,” she adds about her former employer. “But they say they can do it.”