What do you do when the gold standard turns to bronze?
The Census Bureau’s decision to cut its collection period short by one month in the midst of an already challenging pandemic has made pollsters and other statisticians nervous that this year’s census could deliver faulty data. That would leave pollsters without the baseline population portrait they use when crafting surveys and analyzing results.
“Every demographic survey I’m aware of, they use the census,” John Thompson, a former director of the Census Bureau, said in an interview. “If there’s undercounts in the 2020 census, and they’re large, that means that these surveys and these polls won’t be as accurate, because they’ll be under-representative.”
Observers have been quick to point out the immediate implications of a census undercount, both statistically and politically. It would most likely affect the representation of non-English speakers and low-income people, who are typically among the hardest for demographers to reach, and who tend to tilt Democratic.
But it also appears likely that an undercount would disproportionately affect rural communities, a group that is part of President Trump’s political base.
What happens when census workers can’t reach someone
When the Census Bureau can’t reach a household, it must often send field collectors to gather data by physically knocking on the door. The coronavirus inherently makes work harder for these collectors — known as enumerators — because many Americans may be wary of answering a house call in the middle of a pandemic, particularly in hard-hit areas.
Other groups may also have their own reasons for wariness and could therefore be harder to count. Experts have repeatedly expressed concern that Mr. Trump’s attempts last year to include a citizenship question on the census may have frightened some people in immigrant communities away from participating, although the Supreme Court ultimately rejected the move.
“Everybody is not counted, no matter what,” said Peter Miller, a retired public opinion expert at Northwestern University who spent seven years as the senior researcher for survey measurement at the Census Bureau. “You miss some people. Well, you’re going to miss many more now.”
If households can’t be reached, even by enumerators, then census takers rely on a process known as imputation — that is, they use data from demographically similar respondents to take a best guess at what the missing data ought to say.
“This year I can imagine imputation being much higher, and that will itself be a source of controversy — because imputation involves assumptions,” Dr. Miller said. “No matter what you do at that point, you’re going to have a bunch of places around the country that are unhappy with the numbers, and are going to sue. So there’s going to be a lot of controversy around this.”
Where more imputation is needed, Dr. Miller said, the door opens a bit wider for statistical wrangling — and, potentially, more political influence.
In June, in an unprecedented move, the president created two high-level positions at the Census Bureau and filled them with political appointees.
How pollsters use census data
The census is used to redraw congressional and local voting districts, and to determine how about $1.5 trillion of federal funds should be allocated. And it’s just as crucial to the work done by public and private pollsters, as well as academic statisticians.
“All public opinion polls are somewhat flawed in their raw form, without any statistical adjustment,” Courtney Kennedy, the director of survey research at the Pew Research Center, said in an interview. “We calibrate all of our surveys to that statistical portrait that we get from the Census Bureau.”
Because the census occurs only every 10 years, the bureau keeps its data current by regularly updating two major demographic resources using rolling estimates: the Current Population Survey, a source of statistics on the labor force, and the American Community Survey, which tallies information on households and the general population.
It is against those data that pollsters usually base their demographic presumptions. If a poll is conducted and the raw results have a considerably lower share of, for instance, Hispanic respondents than indicated in the American Community Survey, the pollsters will add more “weight” to the survey’s Hispanic respondents.
The Census Bureau also prepares what is called a Hard to Count index, assessing which areas and populations within the country were most elusive to reach. This resource also becomes useful to pollsters, particularly when they are surveying especially difficult-to-reach areas of the country, as they take added precautions to ensure the randomness of their sample.
How the administration has (and hasn’t) adjusted to the pandemic
Originally, the Census Bureau’s period for contacting hard-to-reach people was slated to begin in mid-May and run through July 31. But with the coronavirus scrambling things, the bureau moved that period back, from Aug. 11 to Oct. 31, roughly mirroring the original two-and-a-half-month timeline.
But this week the agency announced that it would end field data collection by Sept. 30, meaning a shorter period in the middle of a pandemic.
Mr. Thompson and three other former directors of the Census Bureau signed a letter urging the administration to reverse the move, echoing another plea written days earlier by Robert Coats, the chair of the national steering committee of the State Data Center, which acts as a liaison between the Census Bureau and state and local governments.
“We thought it was going to be a challenge doing it by Oct. 31,” Mr. Coats said in an interview. “Shortening it to Sept. 30 really ups the bar. It’s a really difficult challenge to accomplish in a shorter amount of time.”
The plan to move the data collection deadline back to Oct. 31 also included a proposal to shift the delivery of reapportionment data from December 2020 to April 2021. Aside from concerns over data collection and accuracy, the most obvious political consideration is that taking extra time would lead to delays in the delivery of reapportionment data — the numbers that control how House districts are drawn — until next year, after the president’s current term ends.
Mr. Trump announced last month that he would seek to exclude undocumented residents from reapportionment data, something that a Democratic administration could most likely undo before the data is released.
Democrats in the House are pushing for the next round of coronavirus stimulus legislation to include a stipulation moving the reapportionment deadline back to next year. This would presumably free up the Census Bureau to continue collecting field data for a longer period. But the Senate’s Republican leadership has so far expressed no interest in the proposal.
“I’m hoping that the Senate will include a provision in the Covid bill that gives the census the time that it had been counting on to deal with the ramifications of this disease, which tied it up in knots for months,” Dr. Miller said. “They’re in a very, very hard position now, and unless we get more time, this is just going to be a train wreck.”
These kinds of debates are not entirely new — and this year’s fandango has a particular historical resonance with the census of exactly 100 years ago.
In 1920, the country was also reeling from a pandemic. It also had just been through World War I, and many displaced people had flooded into the country from across Europe, prompting questions about how — and whether — to count them.
In the end, the census was conducted with such great difficulty that its data was never fully adopted, and no congressional reapportionment happened until after the 1930 census.
“Do we want to go down the same road?” Mr. Coats said. “We still have a little bit of time to make this work.”