Originally Published in The New York Times
Michael Wines - August 24, 2020
WASHINGTON — With the 2020 census into its final stage, more than one in three people hired as census takers have quit or failed to show up.
Many still on the job are going door to door in areas that largely track places where there are elevated rates of coronavirus infections, according to calculations by the National Conference on Citizenship, Civis Analytics and The New York Times.
And with 38 million households still uncounted, state and local officials are raising growing concerns that many poor and minority households will be left out of the count.
Covid-19 and rising mistrust of the government on the part of hard-to-reach groups like immigrants and Latinos already had made this census challenging. But another issue has upended it: an order last month to finish the count a month early, guaranteeing that population figures will be delivered to the White House while President Trump is still in office.
Unlike the Postal Service, another fundamental American institution suddenly under siege and where problems have unleashed a furious public backlash, the census is racing toward a finale largely out of sight. But many experts are increasingly convinced that a public reckoning over a deeply flawed count may be unavoidable.
“If the current situation holds, I do not expect a census of the quality that the Census Bureau will even want to release the data,” Kenneth Prewitt, the Columbia University professor who oversaw the 2000 census, said at a University of Virginia forum this month.
Mr. Prewitt’s view is shared by many state and local census officials and private experts. “This is truly, truly, hair-on-fire awful,” said one government research contractor long involved in census issues, who declined to be identified because of an employment prohibition against being quoted.
The Census Bureau roundly disagrees. “We are sufficiently staffed, with high productivity, and we continue adding people to do the work,” Timothy P. Olson, who manages the census on a day-to-day basis, said in an interview. “I believe we’re in a really good place to complete the data collection by Sept. 30.”
Mr. Olson said the last part of the tally — when census-takers count the 61 million households that have not submitted a census form — was proceeding a third faster than predicted. The bureau projected it would be 28 percent done by now, he said, instead, it is 37.3 percent done.
He said shifting paper census surveys online, giving census takers iPhones and access to a mobile app, and offering performance bonuses had made the count far nimbler than it was a decade ago. Census-takers, he said, are two-thirds more productive than forecast.
And although the bureau has struggled to find door-knockers, he said, over 160,000 new hires being trained now or in the coming days will fill its need.
That said, the bureau has perhaps more uncounted households than at this stage in any previous census — and under the worst circumstances in memory. Just as crucial, the speedup in the deadline gives experts less time to check the data than ever before. Inside Census Bureau headquarters, officials are assessing which quality checks must be jettisoned and data-processing software rewritten to finish on time.
And on the ground, the early door-knocking has been riddled with kinks like sloppy training, a clunky mobile app and unsettling encounters with people not wearing masks and who were unconcerned about spreading the coronavirus to the stranger on their porch.
Shortfalls could mean a severe undercount of the poor and people of color, and an overcount of whites — skewing both political representation and federal largess further away from already undercounted populations.
But none has been rejected as fatally flawed. Indeed, no metric for a flawed census exists. Congress, which has legal authority over the census, could make that judgment, and a lawsuit could seek to. Either would put the census in uncharted waters.
The count faces two crushing deadlines — to compile an accurate tally by Sept. 30, and to process and double-check the numbers in time to deliver population totals to the president by Dec. 31. The Census Bureau earlier had told Congress it needed to push the delivery of population totals to April 2021 because of the pandemic.
The Trump administration ordered the speedup, critics say, because it wants to subtract undocumented immigrants from the totals before sending them in January to Congress to reapportion the House. That plan — which faces multiple court challenges — would reshuffle House seats to give a modest advantage to the Republican Party.
It comes as the administration has installed political appointees in the Census Bureau’s top ranks — two in June and a third named last week to a new post: deputy director for data. Critics say the administration wants to change crucial statistical methodologies to give Republicans an even greater edge.
While exact comparisons are not possible, six states appear to have a better rate of response to the census than at a roughly comparable time in 2010. The remaining 44 states, the District of Columbia and Puerto Rico are lagging, sometimes badly.
The bureau said Saturday that it had reached enough nonresponders to raise the total share of households counted to 74.8 percent, a healthy rise from the 64.1 percent that had voluntarily sent in census forms when door-knocking began.
But by definition, the remaining households are ever harder to reach, and the obstacles to reaching them are formidable. They include predictions of an unusually active hurricane season and fears that Mr. Trump’s anti-immigrant policies will deter census responders.
“Sane people are questioning the government,” said Esteban L. Bovo, a Miami-Dade County commissioner and liaison to the Census Bureau. “As much as I tell people federal law prohibits the government from giving their information out, I don’t know that they’re buying it.”
There are successes. Getting residents to shift from paper surveys to an online census has worked almost flawlessly. But the effort to find those who did not participate has been hamstrung by software and bureaucratic snafus like scrambled assignments of addresses for door-knocking — flaws that experts say rigorous testing that was canceled last year for lack of money might have uncovered. One door-knocker in central Seattle said he reproduces his list of nonresponders in a spreadsheet each morning before setting out to work.
A persistent problem involves center-city apartment buildings, where nonresponders are identified to census-takers only by addresses and alphanumeric codes, but building directories often identify residents only by name.
Such problems are typical of any census, Mr. Olson said, adding that the bureau has established procedures to ensure that residents are counted.
Then there is the pandemic, which forced a three-month delay in door-knocking to August — and whose nationwide average of new cases is about 50 percent higher today than when that delay was imposed last spring.
Nearly four in 10 U.S. counties recorded at least 100 virus cases per 100,000 residents in the past week. More than half of the 38.2 million residents who remain to be counted — 20.1 million — live in states with Covid-19 rates above that level, according to an analysis by the National Conference on Citizenship.
That poses a daunting problem for door-knockers, said William F. Pewen, an epidemiologist working with the National Conference on Citizenship to assess the coronavirus crisis’s impact on the census. Above the 100-cases-a-week level, he said, the chance of a sudden surge in infections rises sharply, as does the individual risk of infection. So does public wariness about interacting with strangers.
“Doors are not going to open,” Dr. Pewen said, “and we could miss thousands or millions of people.”
But Mr. Olson said that although many potential door-knockers apparently declined jobs for fear of getting sick, the virus did not significantly seem to affect residents’ willingness to talk to census workers.
That said, census takers say there is resistance. One said that doormen of high-rise buildings had denied entry in roughly three out of every four locations in an upscale Chicago neighborhood, usually citing Covid-19 concerns. The problem is bad enough that New York City last week sent a notice ordering building managers to let census workers in.
In Miami-Dade County, where four in 10 of the county’s 870,000 households have not filled out census forms, the weekly rate of virus cases is 281 per 100,000 residents, dwarfing the national average of about 13.5.
But rural areas also are at risk. Largely untouched by the coronavirus outbreak a month ago, the 13,500 residents of Montana’s Big Horn County have recorded 81 cases in the past week — 608 per 100,000. Fourteen people have died.
In 2010, about half the county’s households filled out census forms. This year, only about one in five has done so, and a Crow Indian reservation is in a lockdown to prevent spread of the virus, deterring census takers from going there.
“The initial plan was to go into communities when Covid rates were manageable,” said Denice Ross, a senior fellow at the National Conference on Citizenship. “That’s why they needed extra time. By compressing it, they’ve lost that flexibility.”
Even if the count goes well, some of the biggest challenges lie ahead.
In the past, census-takers’ submissions have undergone exhaustive accuracy checks, from reviewing buildings deemed vacant to resampling door-knockers’ work to detect “curbstoning” — making up survey responses. The latest operations update cites plans for “streamlining backend operations,” raising fears that some of that scrutiny will be scrapped.
Censuses have long used such gambits to plug holes in data. Courts have endorsed them because guesses were better than no data at all, and because their use was so sparing that the overall accuracy of the count was never in doubt.
The question now is whether this census will change that.
The Census Bureau is mission-driven, and by Dec. 31, “they’ll get something” to the White House, said John H. Thompson, who headed the bureau from 2013 to 2017.
“The quality of it,” he added, “remains to be seen.”
Sheelagh McNeill contributed research