Census undercounts spark concern among experts over data reliability

New data from the U.S. Census Bureau showing surprising population losses in American cities has some demographers questioning the reliability of the figures after the bureau acknowledged missing substantial numbers of minority populations during its decennial count.

The figures released last month showed 62 of America’s 100 largest counties losing population over a one-year period between July 2020 and July 2021.

At least 10 of the country’s largest metropolitan areas lost more than 1 percent of their population over that time, amounting to what would be a historic exodus. The data show San Francisco losing 6 percent of its population, while more than 1 in 50 residents moved out of New York City, Boston and Washington, D.C.

But to some who study population data, the figures hint at the possibility that Trump-era decisions to restrict the count harmed those big cities, all of which have substantial communities of color.

“It is in big cities, like San Francisco, Los Angeles, where the folks who are hard to count live. So if you conduct the census in such a way as to severely, and I would argue deliberately, undercount hard-to-count communities, then you are going to get lower population counties in those cities,” said Melissa Michelson, a demographer at Menlo College in the Bay Area. “I think that is part of what the Trump administration was doing, was to undercount people of color, to undercount immigrants, to undercount mixed status households.”

The Census Bureau said last month its decennial count had missed only a small number of Americans when it enumerated 331 million people living in the United States. But it was far more likely to undercount Black Americans, Latinos and American Indians and Alaska Natives.

Data from a survey the bureau conducts after every count showed the Black population was undercounted by 3.3 percent, while the bureau missed 5 percent of the Hispanic or Latino population. Both figures were far higher than the undercount that occurred a decade ago.

“It was a larger undercount than we’ve seen in the past,” said Mark Hugo Lopez, director of race and ethnicity research at the Pew Research Center. “We’re still waiting to get some more data on that to figure out how extensive that is.”

On the other hand, the bureau overcounted non-Hispanic whites by 1.6 percent, nearly double the overcount rate of the last census.

“The results of the post-enumeration survey have demonstrated that the Bureau has replicated this decade problems that it’s seen in several decades leading up to 2020, which is persistent undercounts of communities of color,” said Tom Wolf, deputy director of the Brennan Center for Justice’s Democracy Program and a census expert. “We’re now several decades deep into the same story.”

The undercount carries significant ramifications for federal programs that use census data to dole out billions of dollars in aid. Missing a few people — or in the case of some cities, tens of thousands of residents — can cost those governments dearly for a decade to come.

Cities, towns and counties that receive that money can challenge the results of the census if they believe their populations have been undercounted. Twenty-two local governments have challenged census counts so far this year, from Wichita, Kan., to Bennett, Iowa, Tumwater, Wash., to Jonesboro, Ga.

Last week, Detroit became the largest city in the country to challenge the census count. In a letter, Mayor Mike Duggan (D) said an undercount would have “disastrous financial consequences for the City.”

“With every census you will find there are some cities or counties or even states that will challenge the count,” Lopez said. “How much those places receive from the federal government for various services are impacted.”

In an interview with The New York Times published Thursday, U.S. Census Bureau Director Robert Santos said his agency was exploring a number of ways to correct the count, including using administrative records generated by other government agencies and even data collected by private firms.

“We are exploring ways of identifying data from the commercial sector, acquiring it and seeing to what extent there is utility,” Santos said. “The one danger, of course, is that if you become too reliant on commercial data, there’s a danger that you’re one IPO or one change in a CEO away from not having access.”

Wolf, of the Brennan Center, said the data the bureau has produced remains unique in its breadth and scope, even though the bureau missed counting more than 18 million people.

“When you have significantly high omissions, you start to run into additional data quality problems,” he said. “These questions of fitness for use have to be answered in comparison to the alternatives. The alternatives don’t exist, so it’s hard to pursue that question any further.”

Some demographers have been looking for evidence of an urban exodus, either before the pandemic or as a result of the ensuing lockdowns and the rise of remote work. Most caution that the figures released so far will not yet capture the full picture of how Americans reacted to the pandemic, though the drops in population are an enticing data point.

“Maybe we should all be taking a step back from these dramatic pronouncement about the changes, and maybe question whether we’re analyzing false data,” Menlo College’s Michelson said. “Without an accurate count, it’s hard to know what’s actually happening.”