Skip to main content

Why a Census Undercount Spells Trouble for Social Scientists

Social science groups say a citizenship question could lead to population data flaws that compromise research for a decade.
New York City, New York.

New York City, New York.

For the first time in decades, the census will include a question asking respondents if they are United States citizens.

The Department of Commerce, which oversees the Census Bureau, announced the decision just days before the March 31st deadline to submit all 2020 census questions to Congress. The change was prompted by a Department of Justice (DOJ) request late last year. DOJ officials made the case that, in order to enforce Section 2 of the Voting Rights Act, which bans racial discrimination in voting, the department needed a more accurate count of the voting-age population.

But critics say that it could discourage immigrants, even those who are here legally, from responding, resulting in flawed population data. "This chilling effect could lead to broad inaccuracies across the board, from how congressional districts are drawn to how government funds are distributed," a group of Democratic senators wrote in response to the DOJ's request.

Several states, including New York and California, have sued or said they plan to sue the Trump administration over the addition of the question. The research community is similarly up in arms. The census provides a uniquely rich data set to social scientists, whose research could suffer for the next decade if U.S. population data is flawed.

It's hard to say for sure how a citizenship question might effect census data at this point, because the question did not go through the typical vetting process.

Every 10 years, the Census Bureau makes its constitutionally mandated count of U.S. residents. Since the first census in 1790, when U.S. marshals fanned out on horseback across the original 13 states and territories to make the first headcount, the number and content of the census questions have often changed—but never so close to the start date for data collection. "Normally, content decisions are fixed about five years prior to the census," says Gary Gates, a demographer who formerly served on the Census Bureau’s Scientific Advisory Committee.

And typically, the bureau rigorously evaluates any changes to census questions before implementing them—an often years-long process that involves surveys and evaluations. "This is very, very late in the game to be adding an untested question," says John Thompson, executive director of the Council of Professional Association on Federal Statistics. Research on how to improve race and ethnicity questions for the upcoming 2020 census began as far back as 2010, Thompson, a former Census Bureau director, notes. Despite findings that revisions to race and ethnicity questions could improve the accuracy of the survey results, the Census Bureau announced in January that it would keep the same categories used in the 2010 census.

Thompson's chief concern is that all the work the Census Bureau has done to improve census response rates in minority communities will be undone by the citizenship question. Beginning in 2000, the bureau upped their advertising and on-the-ground efforts in local communities to get the message out to the public about why it's important to respond to the census, but also that the data would remain confidential.

"We saw drops in the undercounts of some of the populations at risk. For example, the undercount of the Hispanic population in 1990 was 5 percent, and it had dropped to a little under 2 percent in 2010," Thompson says. "This question has the potential to un-do that, because people will say: 'What are they going to do with that data? Why do they need it?'"

In the best-case scenario, that would just mean more work for the Census Bureau, which sends out staff to visit households that don't respond to the initial survey. "Lower response rates mean higher costs for the Census Bureau," Gates says, "as it would require more non-response follow-up that constitutes one of the bigger expenses of any decennial census." In 2010, census takers visited some 50 million households, an endeavor which cost more than $2 billion.

But an undercount of any demographic group could undercut the validity of public health and social programs based on census data for the next decade. Census data is "used to make sure that [other government surveys] are accurate and representative," Thompson says, including the American Community Survey, a household survey that gathers information in such fields as ancestry, educational attainment, income, and disability. "So if there's a lack of representation in the census, that carries over for 10 years into these other surveys."

Public-health officials use census data to identify trends in health conditions and outbreaks, and to target interventions. In social science, census data has been used to show that substantial disparities exist in American women's access to abortion providers depending on their location; that cities with larger black and Latino populations relied more heavily on court fines and fees for revenue; that there is a strong political bias in who gets Internet coverage.

The greatest risk, as the Consortium of Social Science Associations, a social science advocacy non-profit, wrote in a statement, is that "We have no way of knowing what future insights will be lost if this data is compromised."