Skip Header

Executive Summary

The U.S. decennial census, codified in the U.S. Constitution, has taken place every decade since 1790, with 2020 marking the 24th census. Since 1880, the Census Bureau has relied on maps, and later actual addresses, to find locations and then enumerate the population at those locations. Since 1970 master address lists have been created prior to the enumeration and confirmed by canvassing the country. At each “address,” including homeless shelters and institutions, individuals who “live and sleep here most of the time” are counted and geo-located at that address.

The Census Bureau faces pressure to efficiently and effectively conduct an accurate and precise decennial census. The public response rate in recent censuses has declined from nearly 80% to the mid-60% range requiring huge increases in door-to-door efforts and corresponding costs for the non-response follow-up (NRFU). In 2010, enumerators made over 100 million visits to housing units. The total Census cost per housing unit was approximately $94, a 34% increase from 2000 (in constant 2010 dollars).

The Census Bureau typically starts planning the new decennial census at least 8-10 years ahead of time. This, too, presents challenges as the social, economic, demographic, political, environmental, and technological factors will change in the intervening period. The Census Bureau asked JASON to consider alternative futures for 2030 and to propose a starting point from which the Census Bureau can begin to develop a 2030 strategy.

JASON’s main recommendations takes advantage of the increased availability of high-quality government administrative data; e.g. data collected by the Internal Revenue Service (IRS) and Social Security Administration (SSA). Research at the Census Bureau and by academics suggests that 90% or more of the U.S. population could be located in a combination of IRS and SSA records; these data contain most of the variables collected in the census short form. JASON recommends that the Census Bureau consider starting the 2030 Census with an “in-office” enumeration of the population using existing government administrative records. That would be followed by a second step using additional data and more traditional methods to find people not present in government records and to “fill in” variables that might be missing in these records.

A transition to an “in-office” enumeration using administrative records also suggests a paradigm shift in the way the Census Bureau conceptualizes the enumeration. Historically, the Census Bureau has used the Master Address File (MAF), a list of housing units, as a frame, enumerating the people in each unit. Government administrative records are organized at the level of individuals, indexed by Social Security Numbers or Tax Identification Numbers. A census that starts with administrative records involves identifying individuals and assigning them to their appropriate residences as opposed to the historical process of identifying residences and then populating them. This shift of focus aligns with the declining size of households and the current mobility estimates of less than 1/3 of the population maintaining the same residence during the decade, both of which reduce the efficiency of organizing the census around a list of housing units.

JASON’s recommendations build on innovations in the Census 2020 Operational Plan, specifically the Census Bureau’s proposal to make new use of administrative records. The 2020 plan calls for using records from the 2010 Census and U.S. Postal Service (USPS) to construct approximately 75% of the MAF “in-office,” with the remainder of the work done “in-field.” Later in the enumeration process, administrative records from the IRS and SSA will be used to optimize NRFU visits and in certain cases to impute a housing unit, response if an initial visit fails.

JASON recommends that the Census Bureau use the data collection efforts already underway for the 2020 Census to undertake research on: (i) how much of the population is covered in IRS and SSA records, (ii) how many of the census short-form variables can be collected using those records, (iii) which populations and variables are missing from the records, and (iv) alternative data sources and “in-field” methods to be used to complete an accurate enumeration. Considerable research, testing and experimentation will be needed to move to a census that starts with administrative records. JASON recommends that this process begin as soon as possible in order that the Census Bureau be in a position to make a decision in five years about whether tomove forward with the approach for 2030.

Although outside of the direct tasking for this report, JASON considered several other topics. The first is the American Community Survey (ACS). The ACS replaced the census long form in 2000 and it suffers from declining response rates and costly NRFU.Many of the ACS questions have known answers in administrative records, e.g., type of housing unit, property value, age of structure, and household income. The “in-office” approach for the decadal census could establish much of this underlying data and so provide an opportunity to re-think the ACS, perhaps shifting its focus toward information not available in other data sources such as beliefs and measures of subjective well-being that require survey elicitation.

A second additional topic relates to the MAF. The Census Bureau invests considerable resources to construct the MAF but cannot share it with outside parties despite its potential value as a public good. If the Census Bureau adopts JASON’s suggestion of relying mainly on administrative records for enumeration, it may not be necessary to construct a MAF at all in 2030. However, JASON believes the Census Bureau could create value by working with the USPS, Department of Transportation (DOT), local governments, and private sector firms to construct a continuously maintained National Address File that would be public information and not subject to Title 13 restrictions on data sharing. JASON also notes that a continuously maintained address file potentially could be integrated with a continuously maintained population register, forming the basis for a rolling census that could be verified every decade to satisfy constitutional requirements.

The report concludes by considering some issues relating to historical census records. The ability to link historical census records is important for understanding trends and intergenerational issues within the United States. Currently, the Census Bureau has not been able to efficiently digitize the hand-written names in the 1950-1990 census records, in part because they are protected under Title 13. The report discusses the potential for using Optical Character Recognition to parse names into individual “words” that would be exempt of Title 13, allowing the use of cheap and fast transcription.

JASON offers sixteen specific recommended actions to be taken by the Census Bureau. These recommendations are described in greater detail in the main text.

Regarding moving to an “in-office” enumeration

  • Re-conceptualize the census by organizing it around people rather than housing units.
    (a) Identify people, then place them.
    (b) Start the count from “ninety percent” by using “in-office” enumeration.
    (c) Use field activities to fill in the gaps and validate what is known.
  • Start enumeration with administrative records from IRS, SSA and past census data to construct an “in-office” census that is as complete as possible.
  • Develop a multifaceted strategy to find people who do not appear in the “in-office” enumeration.
  • Use research and near-term experimentation to explore who will not be enumerated with this approach, what data fields will be lacking, and strategies for gap-filling.
  • Continue and expand efforts to acquire data from other agencies, which will be critical to the success of “in-office” enumeration.

Regarding trade-offs

  • Create a set of metrics and criteria by which an “in-office” approach can be evaluated against traditional “self-response plus NRFU” approaches.
  • Examine the utility and cost of expanding the use of administrative records to be a rolling census that would provide an up-to-date population to satisfy enumeration requirements between decadal censuses.
  • Develop a list of options detailing the estimated cost of the 2030 Census as a function of the “accuracy and coverage” desired, which could be used by the Census Bureau and Congress to decide “how good is good enough.”

Regarding research and testing

  • Develop and start a set of experiments now to test the “in-office” enumeration concept.
    (a) Utilize massive administrative data linkage tests to confirm percentage of population enumerable by this strategy.
    (b) Confirm ability to identify subpopulations that will be consistently missed.
  • Explore and test alternative approaches to reach the remaining hard-
    to-count populations (gaps).
    (a) These could include remote and street-level sensing, crowd-sourced Citizen Enumeration, novel data such as state Medicaid records for low-income population, and partnership with HUD to count the homeless.
  • Continue to plan for and test enumeration strategies in the face of natural disasters, terrorist attacks, and temporary dislocation of large numbers of people.
    (a) Create partnerships with FEMA to locate (place) temporarily displaced individuals and trusted civil servants (e.g., firefighters).

Regarding the American Community Survey (ACS)

  • Reconsider ACS and how much data the Census Bureau elicits from different people at different times, keeping in mind replication of administrative data and high cost of asking the first question.
    (a) Use ACS to follow trends relevant to future Census Bureau data collection, such as how long people maintain their email addresses and cell phones versus their residences.
    (b) Use ACS to monitor public trust (a subjective measure).
    (c) Commission a study on what are the high-value data per dollar.

Regarding issues beyond Census 2030

  • Create a public National Address File outside of Title 13.
  • Be alert for new public and private sector data sources that may become available.
  • Develop a pilot “rolling census” project relying mainly on administrative data.
  • Explore new methods for name recognition and OCR to digitize 1950- 1990 censuses.

Back to Header