WASHINGTON — Beyond the reports of undercounts and overcounts in population totals, there is another takeaway from the post-mortem of 2020 census data issued on Thursday: This could be the last census of its kind.
The next census will be taken in a nation where Amazon may have a better handle on where many people live than the Census Bureau itself. For some advocates of a more accurate count, the era in which census-takers knock on millions of doors to persuade people to fill out forms should give way in 2030 to a sleeker approach: data mining, surveys, sophisticated statistical projections and, if politics allows, even help from the nation’s tech giants and their endless petabytes of personal information.
The Census Bureau itself has yet to leap very far into that new era. But it has hinted recently at a “blended” approach in which official census figures could be supplemented with reliable data from government records and other sources.
That would depart from the longstanding model in which the population tally, taken once every 10 years, was the marker for everything that followed, even if it was long outdated by year six or seven.
“The model we have been using since 1790,” when the first census was conducted, “has run its course,” Kenneth Prewitt, the Columbia University scholar who oversaw the 2000 census, said in an interview. “There’s an amazing amount of work going on about how to improve census forms, to use alternate data sets and administrative records, about working with other places that have a lot of data.”
Businesses and researchers have been using those techniques for years, if not decades. But while some changes in the census seem inevitable, how much remains an open issue.
The Census Bureau has stuck with more traditional survey work in part because of legal constraints — a 1999 Supreme Court ruling barred the bureau from using statistical estimates in population counts to reapportion seats in the House of Representatives. But politicians, experts, local governments and industries that consume Census Bureau data worry that seizing on data from other sources without vetting it could make the bureau’s snapshots of the nation less accurate, not more.
It is an article of faith among data experts and the Census Bureau itself that data obtained directly from people are more reliable than secondhand or thirdhand data from other sources. And experts are wary that other data can raise privacy issues or allegations that it was cherry picked to fit an agenda.
The bureau itself considered tapping secondhand sources like state records to fine-tune its 2020 portraits of the population, but it often shied away unless it could find corroborating information elsewhere, according to Amy O’Hara, a former Census Bureau official who is now the executive director of the Federal Statistical Research Data Center at Georgetown University.
Professor O’Hara said the gusher of public and available data opens new avenues to a far more accurate census, but only if the numbers can be proven accurate and the Census Bureau can navigate the tricky boundary between tapping private research and issuing public statistics.
“There is no significant buy-in yet” to major changes in the census, Terri Ann Lowenthal, a longtime census expert and consultant to governments, businesses and other census “customers,” said in an email. “Too early without research, testing and transparency on those sorts of questions. And there probably will be even greater caution about using third-party commercial data.”
That said, she added, many users of census data agree that better use of outside records, conducted in a way that preserves privacy and credibility, could increase the accuracy of the head count and reduce its staggering cost — $14.2 billion, or about $117 per household counted in the 2020 census.
What seems clear is that the existing way of tallying the nation’s population is pressing its limits. The first census by mail was conducted in 1960. Ever since, the nation has counted itself by tallying census forms filled out on millions of kitchen tables, then dispatching an army of census takers to collect data from the millions of others who didn’t fill them out.
The 2020 census streamlined that process by moving most of the form-filling from cumbersome paper surveys to the internet, and equipping census workers with iPhones and census-taking apps instead of clipboards and paper forms. Online census forms proved a resounding success, census officials say, because they were easier, cheaper and quicker to process, and because the Census Bureau’s computer operations handled them virtually without problems.
Yet despite those improvements, the share of residents who opted to complete census forms remained stuck at two-thirds of all households, where it has sat stubbornly for four decades. The so-called nonresponse follow-up, known as NRFU, of the remaining third, conducted by census workers, was hamstrung by hurricanes, forest fires, political interference and rising suspicion of the government among partisans on the political right and among racial and ethnic groups.
Steve Jost, a former senior census official who is a consultant to the Census Project, a group advocating a more accurate count, lamented that. Tracking down nonresponders eats up roughly half the cost of each census, he said, yet the census still fails to reach 2 to 3 percent of households.
“That disproportionately impacts communities of color,” he said. “How long are we going to beat our heads against it?” The bureau needs to do something new, he added.
Mr. Prewitt and other experts say some solutions are obvious. For decades, the Census Bureau has undercounted some groups, including poorer residents and children, in part because they can be harder to find — they move more frequently, for example — and because census forms can be more confusing to people with less education or poorer language skills.
But state governments maintain accurate birth and death records and manage a range of federal programs aimed at the poor and children, such as Medicaid; the Supplemental Nutrition Program for Women, Infants and Children, or WIC; and the SNAP program once known as food stamps. None shares data with the bureau, but an agreement to do so “could probably put a bigger dent in the problem than putting more enumerators on the street,” Mr. Jost said.
There are countless other ways to improve census results. Public and private utility records, for example, assiduously track which residences are occupied or vacant, potentially making it easier for the Census Bureau to compile a more complete and accurate list of households to survey.
The adjustments disclosed on Thursday about flawed counts of people of color stemmed from a two-year spot check of census results that covered 10,000 census blocks, the smallest geographic unit used for counting population. The results of that spot check were extrapolated to the count at large to produce what bureau officials believe were accurate estimates of undercounts and overcounts.
That and other surveys — the Census Bureau conducts hundreds, from its rolling American Community Survey of the general population to reports on labor trends — could be used to update the base census totals. But for decades, Congress has resisted mixing the pen-to-paper, face-to-face census data with numbers politicians might regard as conjured up from smaller samples of people.
Mr. Prewitt’s view is that statistical samples are everyday features of things as common as medical tests, where a small vial of blood can accurately reflect the body as a whole, and that using them to make the census more accurate is overdue. He recalled urging the former House Speaker J. Dennis Hastert to let the bureau rely more on such samples, only to be rebuffed.
“I told him, ‘Next time you get a blood test, have them take all of it out,’” he said.