The Washington PostDemocracy Dies in Darkness

2020 Census may have undercounted Black Americans, new analyses say

A woman fills out a 2020 Census form with an outreach worker on Sept. 26, 2020. (Sarah L. Voisin/The Washington Post)

Two new analyses suggest the 2020 Census may have undercounted Black people at a significantly higher rate than usual, raising concerns about whether minority communities could lose out on fair representation and funding over the next 10 years.

The Census Bureau has not yet released data that would allow comparisons of 2020 Census results with earlier estimates to assess the survey’s accuracy. But a simulation comparing the bureau’s estimates for 2020 with results from 2010 indicates that the country’s Black population may have been undercounted at a rate up to three times as high as in 2010. And a second report suggests the undercount of Black children could be up to 10 times as high as a decade ago.

The findings align with concerns that some jurisdictions and civil rights advocates have had about lower-than-expected totals in the 2020 Census.

If the analysis holds up, that means the 2020 Census count of people who identified as Black alone could be approximately 2 million lower than it should be. The undercount could have profound implications for a wide array of services that are based on population counts, including Medicaid and Medicare, the Supplemental Nutrition Assistance Program (SNAP), highway planning and construction, Section 8 housing vouchers and Head Start.

“This might be our greatest undercount since 1960, or 1950,” said Marc Morial, president and chief executive of the National Urban League, which sued the bureau last year to stop the count from ending early.

Even in the best of times, the census tends to overcount some populations and undercount others, with the highest undercounts among minorities, renters, low-income people and children. But the 2020 Census was fraught with challenges, including Trump administration efforts to add a citizenship question, the coronavirus pandemic, natural disasters, and legal battles over the count’s end date. All of these raised concerns among experts about whether the undercounts would be more significant this time.

“It was a perfect storm for an undercount on multiple levels,” said Rep. Brenda Lawrence (D-Mich.). Many people in poor and minority communities are already reluctant to respond to questions about their household members, a problem that was exacerbated by the additional challenges, she said. “I’m hopeful that the official numbers are not as low as the ones that the analysts are putting out, but the numbers that we’ve seen from these analysts are disturbing.”

The simulation, an independent analysis conducted by Connie Citro, a statistician who is also a senior scholar at the Committee on National Statistics at the National Academies of Sciences, Engineering, and Medicine, presents three possibilities for a net undercount of people who check Black and no other race, based on the bureau’s low, middle and high independent population estimates released in December. Citro calculated a net undercount of between 3.24 and 7.25 percent, compared with a 2.3 percent net undercount for that group in 2010.

For people who check Black in combination with other races, Citro’s analysis found a range between a 0.28 percent overcount and a 4.36 percent undercount, compared with a 1.1 percent undercount for that group in 2010.

The full extent of the survey’s undercounts and overcounts will become clearer next year when the bureau releases what is known as its modified race file, a tally that reassigns people who marked “some other race” alone into Black and non-Black categories. A post-enumeration survey, conducted by the bureau after each decennial census, will further assess the accuracy of the 2020 count.

In the meantime, Citro looked at how the bureau reallocated people who filled out “some other race” alone in 2010, then applied those ratios to the 2020 Census’s race and ethnicity data, which came out in August — adjusting for the fact that the number of people who marked that category increased in 2020.

While her analysis is only a simulation and is not peer-reviewed, Citro said, “it gives a clue that is backed up with other clues.”

Given the challenges in 2020, she added, “it would be surprising if this census did not have more errors than 2010 and 2000. They did an outstanding job with the hand they were dealt, but it was not a good hand to be dealt.”

The bureau said it is too early to draw conclusions about the survey’s accuracy. “The data to do that are just not available,” said Eric Jensen, the bureau’s senior technical expert for demographic analysis. “Any attempt to do that at this time would just be an approximation.”

Noting that more respondents marked “some other race” in the 2020 Census than in 2010, Jensen said, “That’s why we want to be really careful and make sure that we are using 2020 data for doing that process.”

An independent report released last month by the American Statistical Association said its experts did not have enough information to determine the quality of the 2020 Census

A new system of fuzzing out some data to protect the privacy of respondents could further complicate attempts to assess the survey’s accuracy at smaller geographic levels.

The bureau’s demographic analysis estimates contain only Black and non-Black categories because the estimates rely on some information, such as birth records, that did not identify other races until more recently.

But an analysis published last week that includes Citro’s findings and focused on children suggested a high undercount for both Black and Hispanic children.

That report, by William O’Hare, author of the book “The Undercount of Young Children in the U.S. Decennial Census,” calculated that the net undercount may have increased from 0.6 percent in 2010 to 5.8 percent for Black-alone children and from 2.1 to 4 percent for Hispanic children. The calculation showed an increase from 1.5 to 4.2 percent over the past decade for children who were Black-alone or in combination.

The report was posted on the website of the Count All Kids Committee, a coalition of organizations originally formed to see that all children were counted in the 2020 Census.

“I’m very concerned,” O’Hare said. “The biggest implication has to do with funding — federal funding and state funding. Places that have large numbers and percentages of Blacks and Hispanics the census data [misses], they won’t get their fair share of funding and resources.”

The 2020 Census was the first in which respondents were encouraged to fill out the form online, which made responding easier for some groups and possibly helped boost the self-response rate. “The self-response rate in 2020 was almost the same as 2010, but I have to wonder whether that might have worked well for middle-class and upper-class White families but not so well for Blacks and Hispanics,” O’Hare said.

Citro concurred, saying, “The self-response rate dropped dramatically in census tracts that were overwhelmingly minority, and when the self-response rates drop, the data gets flakier and flakier. So that’s why this seems to me somewhat plausible, what I am simulating.” She said that she had shared the analysis with the Census Bureau and that “they said the method that I’m using makes sense, given the data that are available.”

Some jurisdictions began voicing concerns about the 2020 Census after its results showed smaller populations than expected.

When race and ethnicity data was released in August, Detroit Mayor Mike Duggan said the census had undercounted his city by at least 10 percent, based on the number of residential households with current electricity accounts, and threatened legal action. A Duggan spokesperson said this week that the mayor is awaiting the results of a University of Michigan study before deciding whether to sue.

D.C. also expected a higher population count than the census found, and immigrant-heavy states such as Texas and Florida did not grow as much as anticipated, leading some census experts to suspect undercounts in these places.

One of those experts is Arturo Vargas, chief executive of NALEO Educational Fund, who professed skepticism in April after state population totals reflected lower-than-expected counts in states with large Latino populations. “As I said before, I smell smoke, and there is definitely fire when it comes to an undercount in 2020,” he said last week.

“This is not a blame game about the bureau’s competency,” Vargas said. “There were just such extraordinary circumstances, both internal, with the Trump administration mischief, and external, meaning pandemic, that despite the bureau’s yeoman’s effort, we’re just not going to have a better census in 2020 than in 2010.”

An undercount of poor and minority groups exacerbates inequality, Rep. Lawrence said, adding that when more-affluent communities are accurately counted, “then there’s resentment in the [poorer] community that ‘Why are they getting more money to repair their roads when ours are the ones in horrible condition? Why are they getting more senior and HUD programming?’ That’s why the census is so important.”

The Congressional Black Caucus, of which she is a leader, has formed a committee to look into preventing problems such as a fight over the survey’s end date, which was the subject of heated legal battles in October 2020 as the Trump administration pushed to end the count earlier than the bureau had planned. “We’re going to have to legislate so that cannot be an option, to stop the count early,” Lawrence said.

Although congressional apportionment cannot be adjusted to compensate for an undercount, and it would be hard to change redistricting after the fact, it might be possible to revise the way funds are allocated. Lawmakers could adjust formulas to take into account disparities that come to light when the numbers are released, said Terri Ann Lowenthal, a former staff director of the House census oversight subcommittee and a consultant on census issues. “Congress hasn’t really considered that in the past, but should it? Maybe.”

Vargas agreed, saying that if the post-enumeration survey shows a higher undercount than expected, “there’s no reason why the Census Bureau shouldn’t proceed with trying to come up with corrected numbers.”