Counting on Administrative Records

With each decade, conducting the U.S. census becomes a more onerous task. One particularly enduring—and costly—challenge is accurately counting people who are very hard to count, including those who are transient, have complex living arrangements, and distrust the government.

The U.S. Census Bureau routinely introduces innovations meant to address this challenge. For the first time ever, the 2020 Census allowed households to respond online. While this option helped with data collection and processing, it did little to change the universe of respondents; as it turns out, the people most likely to respond online are also more likely to respond no matter what.

Administrative records—existing forms like tax records and beneficiary information used to fill in data gaps when households didn’t respond—proved more effective in 2020. While not perfect, this practice has improved the quality of data for the hardest-to-count populations, and it’s expected to see even more widespread use in 2030.

With Census Day fast approaching, let’s break down what we know about how administrative records were used in 2020, what 2030 could look like, and why transparency around their planned usage is so important.

How Were Administrative Records Used in 2020?

 In 2020, administrative records were primarily used to help with the most expensive and challenging part of the operation: nonresponse follow-ups. That year, the Bureau spent $5.6 billion, or 41% of the total census budget, sending thousands of enumerators into communities across the country to knock on the doors of people who didn’t respond on their own.

Administrative records helped with these follow-ups in several ways, including determining whether addresses were valid, helping to avoid sending field staff off course.  They also helped enumerate people and households who still could not be counted after in-person follow-up. To do so, the Bureau enlisted “high quality” administrative records—including data from the IRS, Medicaid, and government surveys, corroborated across multiple sources; toward the end of the count, it later switched to one “high quality” data source to help estimate the number of people in a household.

In the end, 4.6% of addresses in the 2020 Census were enumerated through administrative records. Although these data are considered lower quality than data collected directly from a household, they are higher quality than other alternative sources, lower some costs, and have documented potential for use in future efforts, including the 2030 Census.

What Are the Plans for 2030?

Early plans for the 2030 Census suggest that administrative records will play an even greater role, and perhaps earlier in the process. During the planning phase, they’ll be used to better identify residential addresses. Early in enumeration, they’ll help match addresses to responses submitted online. Later, high-quality records will be used to represent unresponsive households.

2030 will also likely include using the “demographic frame”—a supplemental file organized around individuals rather than addresses or households. Currently under development, this file is being constructed from multiple administrative record sources. Early tests show very high consistency between the demographic frame and existing records. While more details are to come in forthcoming operational plans, these efforts point toward a more accurate and complete count of groups that are hard to reach.

What Are the Benefits and Drawbacks?

Administrative records matter because they provide usable data at a lower cost in the absence of any data at all. Self-enumeration—when a household completes the census on their own—is the gold standard for data collection but has been trending downward in government surveys in recent decades, which pushed the Census Bureau to seek alternative methods. Meanwhile, the unpredictability of federal funding has caused uncertainty in the planning and execution of census activities. Administrative records are proving to be a suitable solution to both problems.

These records have also proved critical for plugging unexpected data gaps. The 2020 Census was affected by both COVID-19 lockdowns and by an active hurricane season. Administrative records were used to better count college students and to fill in data gaps in Louisiana parishes affected by two different hurricanes. Unexpected events frequently happen to disrupt census data collection, and administrative records can provide an efficient solution in such cases.

But these records are not perfect. Young children under age 5 have been undercounted for decades, and while administrative records show promise for improving the overall count, they may also perpetuate longstanding inequalities in the quality of data about children in low-income and transient households. And when groups of people limit their interaction with government agencies—as is happening with recent immigrants and foreign born individuals—participation in census activities and in the government programs that generate administrative records will be low.

Further, fears about data being weaponized in a polarized political environment have put a damper on sharing administrative records. States are increasingly passing legislation to prevent data-sharing with the federal government to protect residents’ personal information. Following allegations of the U.S. Department of Government Efficiency’s breaches of confidential government data systems, the public may also have less appetite for data sharing between federal agencies.

What Do We Still Need to Know?

As the Bureau finalizes its approach, it is mandated by law to present its plans in a timely and transparent way. However, there are indications that these updates may be delayed, including reports that small-scale testing may happen much later into the census lifecycle than usual. As a result, detailed plans for administrative record usage may not emerge until 2029—after the dress rehearsal is finished and just one year before fieldwork begins.

In the meantime, the public would benefit from more information about the quality of administrative data sources. Researchers are working to understand how to improve the count of young children with these records, but it’s unclear how successful this approach will be, or whether other demographic groups might also be better counted this way instead of through traditional enumeration. Identifying which groups are underserved by each method would help states, who have a stake in high-quality census data, target and mobilize get-out-the-count activities sooner in those communities.

Ultimately, the 2030 census should produce the most accurate count of the nation’s population possible—and if administrative records can help achieve that at scale and at lower cost, then the investment is worthwhile. But this work doesn’t happen in a vacuum. With the Census Bureau’s advisory committees now disbanded, stakeholders have fewer channels to provide feedback. As partners in the heavy work of motivating public interest in the census, they deserve transparency into the Bureau’s plans as soon as possible.