The US Census Bureau announced on Monday night that the 2020 census will ask every American household to record which members of their family are US citizens.
Only a few hours later, it was facing a lawsuit over the decision. A group of 14 states, led by California, are trying to force the government to back down and leave the citizenship question off the census when it goes out to American households two years from now.
The government’s justification for the question sounds simple enough: Asking about citizenship will provide more information about who is in the United States, and more information is always good. It claims it’s simply reinstating a question that’s been part of every census except 2010’s.
But the critics are skeptical that the Trump administration intends to use citizenship data for good reasons. The not-so-subtle implication, critics say, is that that it’s part of a broader project by Attorney General Jeff Sessions and company to take America back to the pre-civil rights era.
It’s not just a symbolic issue. Critics are seriously concerned that adding a single citizenship question to the 2020 census could scare away millions of immigrants from filling out their mandatory surveys — throwing off the count of who’s present in America that’s used to determine congressional apportionment for the next decade, allocate federal funding for infrastructure, and serve as the basis for huge amounts of American research.
A skewed census would hurt the places in America where Latinos are most likely to live — cities and blue states — fueling both the lawsuit and the suspicion that the Trump administration is engaging in deliberate subterfuge.
The concern was real even before the citizenship question was added. Last year, a bureau researcher flagged to a census advisory committee that focus groups and field tests were having serious problems getting immigrants to complete the survey.
During one field test, a respondent fled her home when she started getting worried about the questions. Another family moved abruptly after an interview with a census employee, and others halted the questions or deliberately lied.
Three years ago, researchers said, they hadn’t had problems like this. But now, one respondent told an interviewer, “the possibility that the census could give my information to internal security and immigration could come and arrest me for not having documents terrifies me.”
The government can’t actually do that. Federal law strictly prohibits the Census Bureau from sharing information. But under Trump, it’s really hard for any government official to persuade immigrants — or US-born Latinos — that she can be trusted to protect them.
Adding an explicit citizenship question makes the challenge that much harder. And if the 2020 census fails, it could throw Congress’s representation for the next decade — and our understanding of who exactly lives in America — entirely off-kilter.
What the Constitution says about the census
The census is one of those rare parts of American life that is actually mandated by the Constitution — right at the top. In the 230 or so years since the Constitution was written, social science has developed sophisticated ways to estimate populations without literally counting every single person. But the Constitution still requires an “actual enumeration” to determine how many members of Congress each state will get, so that’s what the census does:
Representatives and direct Taxes shall be apportioned among the several States which may be included within this Union, according to their respective Numbers [...] The actual Enumeration shall be made within three Years after the first Meeting of the Congress of the United States, and within every subsequent Term of ten Years, in such Manner as they shall by Law direct. The Number of Representatives shall not exceed one for every thirty Thousand, but each State shall have at Least one Representative [...]
As the Supreme Court has ruled, that means the census can’t use any statistical modeling or estimation to fill in the gaps if it’s worried that not enough people have turned in their forms. But the Census Bureau collects a lot of data that goes beyond the basic mandatory every-10-year count — information that then gets used for congressional funding allotments, as the basis for other federal and academic research, in carving up congressional districts within a state, and in deciding the total number of Congress members the state gets. And for all those purposes, Congress, in 1976, amended the Census Act to encourage the government to use demographic sampling to collect data.
For a while, the government collected part of that data at the same time as the census: Five out of every six households on the government’s massive census list got the basic “short-form” version of the census, and the sixth got a “long-form” version that asked them extra questions about the people living in their household. After 1950, the question about citizenship became part of the “long-form” census — so only a minority of census respondents got asked about it.
After the 2000 census, the government decided to get rid of the long-form census and replace it with the smaller but more frequent American Community Survey. The ACS only goes out to 2.5 percent of Americans — way fewer than the long-form census — but it’s conducted every year. At present, that’s our primary source of official information about how many citizens live in the US.
This explains why the Trump administration is defending its decision to add a citizenship question to the 2020 census by saying there has been a citizenship question as part of every recent census except the 2010 one. It also explains why critics are saying the Trump administration is reinstating something that hasn’t existed since 1950.
They’re both right.
The Trump administration is correct in the technical sense: 2010 was the only year that no survey conducted as part of the US census asked about citizenship. But the critics are also correct: Citizenship hasn’t been a question on the mandatory census survey since 1950.
The real point of contention, therefore, is whether the American Community Survey — the sampling-based, more frequent survey that can’t be used for allotting seats in Congress — gives the government all the information it needs about citizenship, or whether it needs to have mandatory, enumerated official census data.
Jeff Sessions says he wants to use census citizenship data to “better enforce” the Voting Rights Act. Democrats are ... skeptical.
The Commerce Department, which is in charge of conducting the census, is acting on a request sent last fall by the Department of Justice, headed by Attorney General Jeff Sessions, that asked for a citizenship question on the 2020 census. The DOJ’s reasoning, adopted by the Department of Commerce in its new memo, was that to appropriately enforce the Voting Rights Act, the DOJ needs to know where eligible voters, and specifically eligible voters of color, live — and so they have to be able to distinguish citizens from noncitizens.
The DOJ’s letter floats a hypothetical situation in which not having citizenship data would dilute the power of voters of color: a district that was drawn as a “majority-minority” district based on total population but where minority voters didn’t make up a majority of eligible voters, and therefore couldn’t elect their preferred representatives even in a district designed to allow them to do just that.
Many federal circuit courts have decided that in Voting Rights Act suits, allegations that a state has diluted the minority vote have to be backed up with stats on the citizen voting-age population — not just the total population. Right now, the federal government and voting rights advocates have to rely on ACS data instead of census data to make that case.
There are good arguments that the ACS isn’t a great basis for Voting Rights Act suits. For one thing, it offers multiple estimates at once (one-year, two-year, and five-year averages); for another thing, because it keeps generating new data throughout the decade, it doesn’t mesh with the decennial redistricting process as well as the decennial census does.
But critics of the Trump administration are skeptical that those good reasons are Sessions’s reasons. This administration hasn’t shown a tremendous amount of interest in defending voting rights in general, and Obama-era DOJ officials were able to do a lot of voting rights enforcement using data from the ACS.
And after all, Republicans haven’t demonstrated that responsibly conducting the 2020 census is a huge priority for them. Previous field tests had to be canceled because in 2016 and 2017, the Republican-controlled Congress wasn’t funding the Census Bureau at the levels it needed to conduct a full test.
The omnibus passed in March gave the Census Bureau more than advocates expected, temporarily assuaging the concern that the census simply won’t be well-funded enough to be accurate. But the next week, the Commerce Department released its decision about the citizenship question — despite the concerns its employees had already voiced about immigrant response rates.
To critics, it all reeks of bad faith.
Democrats worry the real purpose of having the census count citizens is to change how seats in Congress are allocated
There is one respect in which the Trump administration cares a lot about voting — it cares a lot about the complete and utter fiction that large numbers of noncitizens are able to vote. The DOJ letter on the citizenship question plays into that belief, quoting a court decision that says “the dignity and very concept of citizenship are diluted if noncitizens are allowed to vote.”
The census doesn’t determine who gets to vote. But it does determine how votes count. And voting rights advocates fear that generating citizenship data from the “actual enumeration” of the census would give the federal government the information it needed to apportion congressional seats based on how many citizens lived in each state, rather than how many people — something that would likely hurt Texas and California.
It could also encourage state efforts to draw congressional districts based on citizen population. The Supreme Court has routinely ruled that states are allowed to use total population when drawing districts — including in a 2016 decision where the Court sided 8-0 with Texas’s use of total population — but it hasn’t explicitly said that they have to.
A conservative state government that wanted to allocate its representatives based only on people who could vote would already be able to do that using ACS data (because redistricting, unlike reapportionment, is allowed to use sampled data). But it would be that much easier if that data were part of the essential census package.
There is, however, one big problem. Demographers studying the ACS data keep noticing that people often mark themselves as citizens when there’s pretty much no way they could be citizens — they’ve only lived in the US for a year, for example.
According to Jennifer Van Hook, a professor of sociology and demography at Penn State University, demographic groups that have large shares of unauthorized immigrants are the ones most likely to inaccurately mark themselves as citizens: Mexican men of working age, for example.
Maybe they don’t understand the question — or maybe it’s because they’re worried about what will happen if they tell the truth.
Immigrants (wrongly) worry the government will use census data to track them down
In an environment where ICE is routinely referred to as the “Gestapo,” the knee-jerk first assumption for many people is that the Trump administration would use citizenship census data for immigration enforcement — going door to door, or something, to all the people who said they had noncitizens in their household.
That is very much not the concern that elected Democrats, demographers, or voting rights and immigrant advocates are voicing.
The assumption is wrong: Federal law is pretty darn explicit that the Census Bureau doesn’t get to share information about individuals with other federal agencies. The laws have been circumvented before: During World War II, the Census Bureau helped the government round up Americans of Japanese ancestry into internment camps by telling the government how many Japanese Americans were in a given area (thus avoiding telling them about individuals per se). But experts generally believe that the legal restrictions, which have been strengthened since then, wouldn’t allow that sort of thing again.
Additionally, the proposed question doesn’t ask about immigration status, just whether or not someone is a citizen. (If the Trump administration really wanted to round up all noncitizens, for that matter, and engaged in wholesale violation of federal law to do so, it could start with the tens of thousands of people who’ve filled out surveys as part of the ACS — or the many, many immigrants whose addresses are in various immigration forms and databases.)
The much more real and imminent concern, among Democrats and advocates, is that many Latinos will worry so much that the Census Bureau will send their information to ICE — or will freak out at the sight of any government mail, much less a government official on their doorstep — that the Latino response rate to the census will decrease substantially.
And that’s, in a way, a much bigger problem. Because (as Rose Eveleth documented beautifully in an episode of her podcast Flash Forward about the census) when already marginalized communities go uncounted in the census, they get further marginalized: underresourced and ignored.
The US census doesn’t have a way to “fix” undercounts
In most demography — say, polling — when one group isn’t as likely to respond as another group, researchers can use statistical modeling to make sure they’re not underrepresented in the results.
The census can’t do that. It’s bound by the requirement for enumeration: counting. What it gets is what it gets.
Over the past couple of censuses, the final census count has been pretty close to accurate: In 2010, 95 percent of households were counted accurately — in fact, overall, the 2010 census slightly overcounted the US population.
But that takes a lot of effort.
Only 67 percent of Americans responded to the initial census survey by mail in 2000, and 72 percent in 2010. To get the rest, the Census Bureau dispatched half a million temporary census takers to visit houses up to six times in the hopes of speaking with someone in person — often hiring census takers from within the community, to reduce the anxiety of speaking to a government official.
If that didn’t work, they’d try to get information from neighbors about who was living in the silent house — and if that didn’t work, they’d have to do “nearest-neighbor imputation,” or basically assuming that the household looks identical to that of the most similar household nearby. The Supreme Court has ruled that imputation is okay because it’s just filling in known gaps in the data, rather than speculating about where the gaps would be.
Those methods don’t get everybody. Van Hook points out that people who return their census forms might not be telling the truth about how many people are in their household because they’re trying to protect, say, relatives who are unauthorized immigrants. Or a census respondent might not count a family friend who’s sleeping on the couch for a few weeks while he finds a job because he doesn’t really live at the house — but he doesn’t live anywhere else, either.
And then there are the places where people live that aren’t on the master list of addresses the Census Bureau pulls together at all. “They might live in the back of a restaurant, or a trailer in the back of the farmhouse,” places the Census Bureau doesn’t think are homes and therefore doesn’t send a survey to begin with.
Unsurprisingly, these sorts of problems are more likely to show up in certain groups than others. African Americans, particularly African-American men, have been undercounted in the past two censuses.
So have Latinos, who were undercounted by 0.8 percent in 2000 and 1.5 percent in 2010. The Census Bureau says the change between the two censuses wasn’t statistically significant. But if the Latino response rate plummets in 2020, the undercount rate will soar.
Even before the citizenship question, census takers were having serious trouble getting immigrants to talk to them
Even though the 2000 and 2010 censuses didn’t ask about citizenship, experts and advocates were still worried that immigrants (particularly unauthorized immigrants) would be intimidated out of filling out a government form or speaking to a government interviewer.
In 2000, the Clinton administration agreed not to conduct any immigration raids during the time the census was in the field, to reduce the amount of anxiety being stirred up in immigrant communities. In some neighborhoods, census takers put up signs saying “NO INS. NO FBI. NO CIA. NO IRS.”
In 2010, when immigration enforcement under President Barack Obama was at its most aggressive, ICE made no such reassurances (and, indeed, continued to conduct raids in some areas while census takers were trying to conduct follow-up surveys). But the Census Bureau itself made a big push to reach out to Latinos, especially Spanish speakers, to reassure them that the census was for them and that it was important for them to fill out.
The effort appears to have worked. A survey conducted by the Pew Research Center at the time found that foreign-born Latinos were enthusiastic about the census and extremely confident that their data wouldn’t be used against them; US-born Latinos (who were by definition US citizens), on the other hand, were distrustful.
Mark Hugo Lopez, director of the Pew Hispanic Trends Project, hypothesizes that this was a result of the Census Bureau’s huge promotional push on Spanish-language media — something much more likely to target foreign-born Latinos than American-born ones.
Advocates are worried that the Trump administration has created an atmosphere way too fearful for a few radio ads to fix. The Census Bureau’s November presentation quoted one researcher: “The politics have changed everything recently.” Immigrants — not just unauthorized immigrants but some types of legal immigrants as well — are highly anxious about their security under Trump and leery of interacting with any government officials as a result. And if US-born Latinos were skeptical of the government in 2010, they are likely to be even more skeptical now.
The citizenship question might not damage response rates. But it’s a huge risk.
The Trump administration has rejected the idea that asking about citizenship is likely to create a substantial undercount that didn’t exist before.
“While there is widespread belief among many parties that adding a citizenship question could reduce response rates, the Census Bureau’s analysis did not provide definitive, empirical support for that belief,” the memo from Commerce Secretary Wilbur Ross says.
That’s true. There simply haven’t been any studies or experiments either way. And it’s totally possible that a citizenship question wouldn’t have that big an impact — whether because it wouldn’t matter that much or because immigrants will already be afraid of the census no matter what it’s asking.
The problem, say demographers and advocates, is that the government is risking a lot for the sake of getting this question back on the form.
Usually, census questions go through a pretesting process that can take years. But the 2020 census is already well past the development stage. As the Commerce Department announced the addition of a citizenship question, the Census Bureau was conducting its only full “dress rehearsal” field test for the 2020 census, in Providence, Rhode Island.
Advocates and researchers don’t have a ton of data to help them predict the impact of adding a question that’s never been tested. But for a social scientist, that in itself is reason for concern.