The census scales back a critical step: Checking its own work

With scores of people displaced because of the coronavirus pandemic and other disasters, the U.S. Census Bureau is facing an especially daunting challenge of meeting its once-a-decade goal of tallying every person living in the country “once, only once and in the right place.”

After counting is set to end on Sept. 30, the bureau has about three months to process all of the information it’s gathered this year for the once-a-decade, constitutionally mandated head count. “If you want an accurate census, the quality checks are as important as the initial enumeration itself,” says former Census Bureau Director Ken Prewitt, who oversaw the 2000 count.

But the last-minute schedule changes the Trump administration directed the bureau to make have left the agency’s staff scrambling to decide what quality checks to trim or toss out.

In recent weeks, Commerce Secretary Wilbur Ross, who oversees the bureau, has been pushing bullish claims that the new approach is not “sacrificing quality.”

A growing number of former Census Bureau officials and other census advocates, however, are raising the alarm that the truncated time for processing responses is likely to further undermine the accuracy of the data and exacerbate undercounts of people of color, immigrants and other historically undercounted groups.

“It is literally impossible to say that with less time than we’ve ever had before, with more problems than we’ve ever had before, we’re going to have better results than we ever had before,” says Arnold Jackson, the chief operating officer for the 2010 census who also served as a consultant on the 2020 count.

‘Still working’ on steps to ‘streamline’

A report released this week by the Government Accountability Office, the investigative arm of Congress, warns that the bureau is behind in finishing tests of the dozen IT systems it uses for processing census responses.

“If the Bureau does not complete all required testing, it may face an increased number of system defects or other issues,” wrote Chris Mihm, managing director of the GAO’s strategic issues team, and Nick Marinos, a director on its IT and cybersecurity team, “which could affect the quality and accuracy of the census count.”

Any technical problems could also slow down the bureau’s staff, many of whom are already spread thin trying to carry out President Trump’s request from last year to use government records to produce data on the U.S. citizenship status of every adult living in the country, as well as his recent call for information on unauthorized immigrants.

Even without any IT issues, the bureau is hard-pressed to swiftly sort through what could turn out to be an unusually complicated mess of census responses.

Because of COVID-19, many residents have scattered from where they were living on Census Day, April 1, including renters and homeowners who can no longer afford their housing and college students who fled campuses shortly before the bureau had planned to start counting them in person at their school addresses.

Some census watchers are also worried about whether the bureau can verify all of the online responses it has been collecting without a 12-digit “Census ID” that’s assigned to each known home address.

The bureau has been preparing fixes for these kinds of situations for years. But cutting short the time left to correct errors heightens the risk that the 2020 census results will count some residents more than once and at the wrong location, which could result in unfair distributions of federal funding and political representation for the next decade.

In a charged political climate churning with anti-immigrant sentiment and government distrust, the curtailed schedule could also leave the bureau with more gaping holes in information about unresponsive households than in past counts.

What’s more is that the bureau — known for releasing detailed plans years before carrying out its operations — has yet to put out any specifics about how it will, according to a statement by the bureau’s current director, Steven Dillingham, “streamline” its processing work. Meanwhile, the bureau emphasizes that it is committed to meeting the legal deadline of Dec. 31 for delivering the latest state population counts to the president. Those numbers are used to reapportion the 435 seats in the House of Representatives among the states before they start receiving more detailed data for redistricting next year.

“We are still working to identify which processing steps may need to be adjusted,” the bureau tells NPR in a statement that offered no timeline for providing updates to the public. “Our focus is to identify processes that can be started earlier, run in parallel, or be deferred until the redistricting phase of processing.”

Frank Vitrano, a former associate director for the 2020 census who helped manage the 2010 count and retired in 2018, says the bureau generally builds into its plans a cushion of time to give itself some wriggle room when carrying out the count.

“Part of streamlining could potentially be removing some of that contingency time,” Vitrano says. “But then what happens when something unexpected happens?”

A quiet cut to a quality check

This month, the bureau quietly cut at least one of its quality-checking efforts short, rattling many census advocates.

While the bureau sent an email to representatives of the states and territories involved with the program, it has not made a public announcement about the change.

Relying on the local expertise of demographers, the count review operation has helped the Census Bureau identify more than 240,000 housing units and 6,500 group living quarters, such as nursing homes and prisons, that were missing from the bureau’s records, the agency confirmed to NPR in a statement.

Qian Cai, a demographer who directs the Demographics Research Group at the University of Virginia’s Weldon Cooper Center for Public Service, had been preparing to do another review of the bureau’s data files in September to help make sure college student housing in her state was correctly counted.

“We lined up the resources and the personnel ready to provide input,” Cai says. “And all of sudden, that’s canceled.”

Huda Alkitkat, the manager of the Population Estimates Program at Portland State University’s Population Research Center, said she was relieved that her team had already completed most of its work on data about Oregon.

“We are lucky that we were able to finish this phase before the pandemic,” Alkitkat says.

But at the Arkansas State Data Center, demographer Diego Caraballo says he feels “a bit nervous” now that his team has lost an opportunity to review the Census Bureau’s files again.

“The thing that concerns me and my coworkers the most are nursing homes, since they tend to pop up and close down pretty often,” Caraballo says. “Some of them we know for a fact are located near county boundaries or city boundaries, so it could cause some funding problems down the line if that’s not correct.”

Thomas Louis, a former chief scientist at the Census Bureau who oversaw its research and methodology, has cautioned against curtailing the bureau’s count review operation.

“Eliminating it will help the Bureau achieve the December 31, 2020 deadline for delivery of apportionment data, but will do so at a considerable cost in the quality and credibility of that data,” Louis wrote in a court filing this week for a federal lawsuit led by the National Urban League that’s trying to get the administration to go back to the extended schedule the bureau developed in response to COVID-19.

Instead, the 2020 census is hurtling into a fog of uncertainty in its final months as bureau officials rush to deliver numbers from which Trump says he wants to exclude unauthorized immigrants despite the 14th Amendment‘s requirement to include the “whole number of persons in each state.”

“A lot of those changes, without a very clear explanation or justification, make data users wonder, ‘How much can I trust the census?’ There’s a big trust issue,” says Cai, the University of Virginia demographer.

‘Endangering all of government statistics’

Katherine Wallman, who retired in 2017 after close to 25 years as the chief statistician within the White House Office of Management and Budget, overseeing the federal government’s statistical policies, worries that skepticism about the quality of 2020 census data could taint public perception about the reliability of other critical information the federal government releases that is based in part on census data.

“We’re actually endangering all of government statistics because it just feeds the notion,” Wallman says, “that those guys, they just make up the numbers, right?”

Wallman, who was part of the approval process for the 2000 and 2010 census forms, says she finds the rush to produce the 2020 results amid the pandemic perplexing.

“I am not finding any rationale that I can accept as an objective reason for rushing through, ending up with crummy data and destroying the trust in a system that depends almost entirely on the voluntary cooperation of the American public,” Wallman says.

The recent additions of three political appointees at the bureau — including two new deputy directors at an agency where day-to-day operations have been overseen for decades by one deputy director who is a career official — have heightened worries that the Trump administration is trying to manipulate the 2020 count to benefit the Republican Party.

To help assuage concerns, many census advocates are calling for the bureau to release more detailed indicators of the level of public cooperation door knockers are experiencing among households that have not yet participated in the count, many of whom are among historically undercounted groups with high levels of distrust of the government.

Tim Olson, the bureau’s associate director for field operation, tells NPR that in the coming days, the bureau is planning to release data “a little more detailed below the state level” on its website.

“There’s nothing to hide,” Olson says. “When people do see that greater level of detail, they will see areas that are way ahead and areas that are behind. And that’s the normal churn.”

“At this point, it will take an unprecedented level of transparency for this count to have the credibility it needs,” says Denice Ross, a senior fellow with the National Conference on Citizenship who once co-directed the Data Center in New Orleans. “When they’re out there in the field and they’re counting a household, did they actually reach people in that household or did they ask a neighbor, for example?”

These technical details could help put into focus exactly how precise this once-a-decade portrait of the U.S. population is.

The bureau says it has “enumerated” more than 80% of housing units in the country as of Thursday. But that rate is not a clear indicator how much of the U.S. population has been counted so far. Some of those housing units could have been verified by the bureau’s workers as vacant or not actual home addresses. Others may have been added to the head count through dubious information door knockers have collected from neighbors. That’s one of the bureau’s back-up methods for filling in information gaps, in addition to looking to government records and, as a last-ditch effort, generating educated guesses through a statistical technique known as imputation.

But the more the bureau relies on these alternative processes to finish the count, the higher the risk of producing inaccurate information about historically undercounted groups, including people of color, resulting in national statistics that show a country much whiter than it actually is.

In 2010 and 2000, close to 1.2 million people were added to each count through imputation. In a court filing in support of the National Urban League’s lawsuit, former Census Bureau Director John Thompson, who has worked on four past U.S. head counts, warns there may be a greater need for imputation for this year’s count given the challenge in retaining enough door knockers and the time crunch they’re working under.

Thompson also cautions against census workers relying more on “proxy” interviews with neighbors to get households counted — a method that the bureau found was more than twice as likely to produce an error compared to the overall error rate for the 2010 census.

All of this uncertainty is leading many census advocates to wonder what kind of data the bureau is preparing to give to the president by the end of this year — and to the states next year.

Louis, the bureau’s former chief scientist, spelled out in the court filing concerns about how census data are used in “key pillars of our democratic society.”

“The Bureau will most likely release numbers at the end of the census process,” Louis wrote. “But if the quality of those numbers is low, fair apportionment and redistricting will be compromised.”

Copyright 2020 NPR. To see more, visit https://www.npr.org.