The decennial census is supposed to be a tally of everybody living in the United States. But it actually starts out as a master list of addresses. That list, updated once a decade by the U.S. Census Bureau, is then used to send out an army of workers to collect basic demographic information from whoever answers the door at those addresses.
The approach allows the agency to meet its constitutional requirement to provide Congress with the data needed to apportion seats in the U.S. House of Representatives. But the 2010 census also cost $12.3 billion, and future censuses that use that approach may be prohibitively expensive (see our story yesterday on issues facing the 2020 census). It also assumes that households, not people, should be the key unit of measurement.
That assumption may no longer be valid in today’s more mobile and less family-centric society, say an independent group of prominent scientists that advise the U.S. government. The group, known as JASON, has proposed an alternative to starting with addresses that would use existing digital records to compile a master list of individuals. This “in-house enumeration” would be based on federal tax and employment records and then mined to answer the 10 questions on the census form. (The 2020 census will ask the name, sex, age, relationship, and race/origin of each individual at an address, as well as whether the housing unit is owned or rented, the telephone number, and an email address.)
That method might be able to identify as many as 90% of Americans, the researchers estimate. Then the Suitland, Maryland–based Census Bureau could carry out a much smaller—and presumably much cheaper—field operation in 2030 to find and count anybody it’s missed, along with any missing information about those already in the database.
“In no way are we criticizing what the Census Bureau does now,” says Sallie Keller, a professor of statistics at the Virginia Polytechnic Institute and State University’s Arlington campus. She and economist Jonathan Levin of Stanford University in Palo Alto, California, led the JASON group that wrote a report, entitled Alternative Futures for the 2030 Census, that Census officials requested and released last fall. “But we are saying that an in-house enumeration offers a lot of possible advantages.”
Pros and cons
The JASON report posits that its approach is more in keeping with the constitutional mandate for apportionment, which includes providing enough demographic information on each state so that state legislatures can draw congressional districts that satisfy federal laws to ensure equal representation. “There’s nothing in the Constitution about knowing a person lives at 102 Main Street,” Keller notes. “Maybe block data would be sufficient.”
The key to the JASON approach is relying on so-called administrative records. The Census Bureau already has information-sharing agreements with the Internal Revenue Service and the Social Security Administration, the report notes, and those records may tell the full story for most Americans.
But “most” isn’t good enough for the decennial census. The prospect of an undercount fuels the Census Bureau’s relentless pursuit of information from those who fall outside that meritorious group called self-responders. For residents who don’t respond to the agency’s initial invitation to fill out the form, census takers might have to visit a single address half-a-dozen times in the hopes of enumerating its residents. Making things even more difficult is the possibility that the unit may be vacant.
Relying on administrative records rather than a master address file would eliminate the problem of returning repeatedly to an unoccupied dwelling. But it would still leave the Census Bureau with the challenge of finding those who don’t pop up in an electronic database.
“The hard-to-count people [in the traditional Census approach] are also hard to find in administrative data,” says economist Amy O’Hara, who this month joined Stanford University’s Institute for Economic Policy Research after heading the Bureau’s Center for Administrative Records Research and Applications. Even those who receive some type of federal assistance might be hard to track down if their records are incomplete, she adds, because they may not leave other electronic fingerprints such as credit cards, utility bills in their names, and other types of financial transactions.
For those reasons, the JASON report recommends that the Census Bureau conduct a dry run to determine the thoroughness of federal records and what geographic areas and what subpopulations are most likely to be missing. It should also test “alternative ways of reaching missing populations and filling in imperfect records,” the report says. That could include the use of sensor technology that can assess population densities and citizen volunteers who can provide information about their communities.
The goal is to identify outliers that warrant further analysis. “It is important to be clear that JASON is not advocating that this information replace the actual census enumeration data,” the report warns. Still, the group says the Bureau should get started now so that it can decide by 2022 what direction to take for the 2030 census.
Keller says the group “didn’t have enough information” to assess the potential cost savings from a people-first approach, saying the experiments need to be done before the agency makes any reasonable estimate. But the report points to several other possible benefits.
A people-first approach could be the first step toward conducting a rolling census, rather than a once-in-a-decade push. Regular updates would likely generate smaller fluctuations in the agency’s budget than under the current approach, in which costs can vary 100-fold over the 10-year cycle of the decennial census. It would also allow the Census Bureau to convert its current master address file, which must remain confidential and be used only for census purposes, into a national housing file that could be shared with other public agencies. Uses could include disaster relief, long-range planning, and various economic development activities.
A revised ACS
The shift could also be a boon to the American Community Survey (ACS), the successor to the long form of the decennial census. Begun in 2005 as an annual survey of some 3.5 million households, the ACS suffers from declining response rates and the need for costly nonresponse follow-up, the report notes. “Many of the ACS questions have known answers in administrative records,” the report asserts, citing questions about the type of dwelling, age, and cost of housing units as well as questions about family income and expenses.
Some members of Congress say the ACS is also a burden on the public and an invasion of their privacy. The JASON report acknowledges that criticism and notes that using administrative data might make ACS seem less intrusive because it wouldn’t need to ask residents some questions deemed sensitive. In place of those questions, the report suggests, the ACS could explore topics “not available in other data sources, such as belief and measures of subjective well-being.”
Keller says that the Census Bureau is well-positioned to act on the JASON report, which complements some research already underway. And John Thompson, who stepped down last month as Census director, says the group “had some good suggestions” that are worth exploring.
But whether that happens will be up to the agency’s new leadership. President Donald Trump has yet to nominate a permanent director, and right now Census officials are focused on whether Congress and the Trump administration will give them enough resources to carry out a complete and accurate 2020 census.
That battle is rightly the agency’s highest priority at the moment, Keller says. “Whatever they decided to do in 2030,” she says, “this is not the time to cut [the] Census Bureau’s budget.”