WASHINGTON ― The Trump administration’s justifications for adding a question about citizenship to the 2020 census aren’t particularly strong.
Commerce Secretary Wilbur Ross said his decision to add the question was based on three things. First, he wrote in a memo, while the census’ long-form American Community Survey (ACS) already asks about citizenship, the data it collects aren’t specific enough for the Department of Justice to best enforce the Voting Rights Act. Second, he said, he saw “limited empirical evidence” that a citizenship question would decrease response rates. And third, a citizenship question was already well-tested because the decennial census had asked about it in the 19th and 20th centuries (it stopped after the 1950 census), and the ACS questionnaire asks about it.
The Trump administration’s justifications, which have already shifted, are going to come under more scrutiny as civil rights groups and Democrats file lawsuits challenging the decision. The challengers say the decision to add a citizenship question is politically motivated and will lead to an undercount of immigrants and minority groups, who will be afraid to respond to the survey.
The ACS has asked about citizenship since 2005, but John Thompson, the director of the Census Bureau from 2013 to 2017, noted that the ACS is a completely different form than the decennial census. The ACS consists of a few dozen personal questions about things such as income, rent, disabilities and transportation to work, in addition to citizenship. It goes out to 2.6 percent of the population each year. The 2010 decennial census was just 10 questions long and asked very basic questions, including how many people were living in a given household, telephone number, race or ethnicity, and housing status. Thompson noted that there is much more attention paid to the short-form census that’s meant to be answered by everyone every 10 years.
Justin Levitt, who served as deputy assistant attorney general for civil rights at the Department of Justice from 2015 to 2017, noted that the ACS has lower stakes than the decennial census, which he said is the first responsibility the Constitution gives the federal government. If there’s a low response rate on the ACS, he said, there are ways for officials to correct their data. But the Constitution mandates that the federal government count all people. An inaccurate count could disproportionately affect certain states. For instance, if immigrants are not fully counted in an immigrant-rich state because many fear answering a citizenship question, that state could lose congressional seats as well as a fair allocation of hundreds of billions of dollars in federal money.
“If we get the percentage of population wrong that happens to have indoor toilets or outdoor toilets, it’s not of constitutional magnitude,” Levitt said. “If you’re asking people for a lot of information, and they’ve already signed up to deliver that other information, then also asking them for information about whether they’re a citizen, feels ― many people have said ― less aggressive and less worrisome.”
Two former top officials at the Justice Department charged with enforcing the 1965 Voting Rights Act, which is intended to protect the voting power of minorities, say the Trump administration’s idea that existing ACS citizenship data aren’t good enough to enforce the law is untrue.
“I oversaw the enforcement of the Voting Rights Act for just under 2½ years in the prior administration and can tell you in clear terms that rigorous enforcement of the Voting Rights Act has never required the addition of a citizenship question in the census,” said Vanita Gupta, the president and CEO of the Leadership Conference on Civil and Human Rights, who served as head of the Department of Justice’s Civil Rights Division from 2014 until 2017. The Leadership Conference has lobbied against adding the citizenship question.
Levitt said that most jurisdictions were large enough that any error in the ACS estimate was eliminated. The only places where the ACS did not produce reliable citizenship estimates were tiny towns and municipalities. Any increase in accuracy from asking about citizenship, Levitt says, would be outweighed by the potential for error caused by adding an untested question.
“By attempting to increase that precision, from 9.99999 to 10, they are introducing such profound chaos and such intentional and predictable undercount that the amount of error they will introduce into the process swamps, by exponential levels, any incremental accuracy that they are purporting to bring,” he said in an interview.
“If the reason for this is: ‘There are some cases we can’t bring because the data on the survey isn’t accurate enough,’ to believe that adding this question in this environment, without testing it ― and that is unthinkable ― it’s essentially setting off with the Titantic saying, ‘We’re pretty sure that this won’t create a problem, but just in case, yeah, let’s throw a massive iceberg in the way.’”
“If we get the percentage of population wrong that happens to have indoor toilets or outdoor toilets, it’s not of constitutional magnitude.”
Even though the Census Bureau can spend years testing the wording of a question, it hasn’t done any testing at all on the effect of adding a citizenship question. It will go into the 2020 test essentially blindfolded, taking the risk that a citizenship question won’t suppress response rates. Civil rights groups say that immigrants and other minority groups are less likely to respond to a citizenship question because of fear that the Trump administration could use their answer to find them and deport them or deport someone they know.
Six former directors of the Census Bureau warned Ross against adding the question, saying it was too risky. But Ross and the White House have downplayed the risk, saying its presence on the ACS is a test and there’s “limited empirical evidence” that fewer people will respond.
The Census Bureau took the citizenship question off in 1960 and started asking about it only on the longer questionnaires in 1970 as the practice of sampling the population became more popular in the 20th century, said Margo Anderson, a historian of the census at the University of Wisconsin-Milwaukee. But asking a question on those long-form surveys isn’t a substitute for testing the effect of the question on the actual decennial form.
“1950 is a long time ago,” Anderson said. “The normal method would be to say, ’OK, if you want to do this, let’s test it.”
Kathleen Joyce Weldon, director of data operations and communications at the Roper Center for Public Opinion Research, said pollsters are careful about including questions about citizenship in polls out of fear that people won’t respond. The seriousness of the census, a government-conducted national survey, could only further discourage people from responding.
Beyond the differences in content between the long-form ACS and the decennial census, it’s risky to assume that a question that works well on one survey will work on a different one. Social scientists are careful about the way any question on a survey is asked and even the way questions are ordered because the wording of a single question can influence how a person responds to all other questions or if they choose to continue responding at all.
In a January letter to Ross, the American Statistical Association warned that adding a citizenship question could invalidate all of the careful testing and calibration the Census Bureau had done so far to prepare the questions for the decennial survey.
“Adding a question at this late stage of the census process does not allow time for adequate testing to incorporate new questions, particularly if the testing reveals substantial problems. Further, a new question undermines the validity of the extensive testing of the current questions carried out to date.... We believe the potential for a citizenship question to create problems argues strongly against its inclusion at this advanced stage. The most likely results would be a lower response rate, more non-participation, and ultimately a sharp increase in cost for non-response follow-up.”