III. The Defining Operational Change: From Direct Collection to Data-Driven Enumeration
The most consequential change in the 2030 Census design is the Census Bureau’s plan to make much greater use of information that already exists about people, households, and addresses. Rather than relying primarily on questionnaires and interviews, Census increasingly plans to combine direct responses with administrative records, commercial and other supplemental data, linked person-and-address files, statistical models, and automated decision-making.
This shift extends across much of the census. Outside data may help Census update the address list, identify people associated with a housing unit, decide how to contact a nonresponding household, resolve conflicting responses, identify possible duplicates, and evaluate whether collected information is complete. Many of these uses could improve overall data quality. Administrative records may help Census find an address that would otherwise be missed, recognize that a housing unit is occupied, or identify a case that requires additional fieldwork.
The most consequential use, however, is “in-office enumeration:” counting some households that have not responded using administrative and supplemental data rather than obtaining their information through a completed questionnaire or interview. Census describes this as a way to improve accuracy, make better use of field resources, and reduce costly in-person follow-up. But it also changes the point at which Census may decide that it has enough information to stop trying to reach a household directly.
A second new operational area, Person Characteristic Frame Management, will provide much of the infrastructure for that decision. The Person Characteristic Frame will link people to addresses and assemble information about their characteristics and household composition. It will also assign quality measures intended to help other census operations determine when outside data are reliable enough to use. The frame is expected to be continuously updated as Census receives new datasets and more recent versions of existing records.
Together, these operations represent more than a technical improvement to the 2020 design. They shift part of the census from asking households who lives there to determining who probably lives there from information collected for other purposes. That distinction matters because administrative records are not created to conduct the census. Tax, benefits, education, health, correctional, and commercial records have different definitions, update schedules, coverage gaps, and legal constraints. A record may be accurate enough for its original purpose while still being outdated, incomplete, or misleading when used to determine where someone should be counted on Census Day.
The central question is therefore not whether Census should use administrative and supplemental data. These sources can clearly improve some parts of the count. The question is when outside data should supplement direct collection and when Census will allow them to replace it. Answering that question requires more than assessing whether a dataset usually contains a name and address. Census must determine whether the records accurately represent people who move frequently, live in complex households, face language barriers, distrust government, or are otherwise more likely to be missed.
This part of the guide examines that shift from several perspectives. It first explains how in-office enumeration and the Person Characteristic Frame are expected to work and identifies their potential benefits. It then considers the ways administrative data may omit people, place them at the wrong address, misunderstand household composition, or provide incomplete characteristics. Particular attention is given to the still-unreleased rules that will determine when fieldwork is considered complete, when records are judged sufficient, and how Census will resolve conflicts between direct responses and outside data.
The sections that follow also examine whether large-scale in-office enumeration would satisfy the Constitution’s requirement for an “actual Enumeration,” as well as the privacy, cybersecurity, and misuse risks created by assembling detailed linked records about people and households. The final subsections identify the research, transparency, and operational safeguards needed to ensure that new data sources help Census reach historically undercounted communities rather than provide a justification for ending direct collection too soon.
The stakes are substantial. A carefully limited and well-tested system could help Census resolve difficult cases and use field staff more effectively. A poorly designed system could direct less fieldwork toward the people who most need it, accept incomplete records as finished enumerations, and deepen existing disparities in who is counted accurately. Because the final decision rules have not yet been released, advocates still have an important opportunity to shape where Census draws the line between using data to support enumeration and allowing data to substitute for it.
Where to look: 2030 Census Operational Plan, Figure 14 and the overview of “Count the Population,” pp. 27–29; section 3.2.3, “In-Office Enumeration,” pp. 32–34; and section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48. See also the 2030 Census Strategy and 2030 Census Operations Strategy and Roadmap for the broader emphasis on administrative records, person-location information, and reductions in field workload.
A. What “In-Office Enumeration” Means
Traditionally, Census has tried to count a household by obtaining information directly from someone who lives there. Households first have an opportunity to respond online, by mail, or by telephone. If a household does not respond, the address generally moves into Nonresponse Follow-Up, or NRFU, where an enumerator attempts to contact someone at the home. If no household member can be reached, Census may seek information from a neighbor, landlord, building manager, or another knowledgeable source.
For 2030, Census plans to add another pathway: in-office enumeration. Under this approach, Census may complete the count for some nonresponding households using administrative records and other outside data rather than obtaining a questionnaire or completed interview from the household.
The current Operational Plan describes a general sequence. Census will first solicit a response from household by sending an invitation to respond by mail. If the household does not respond, it will enter field follow-up and receive at least one in-person visit. Models and available administrative data may help Census decide how to handle the case, including what kinds of contact to attempt and how much additional field effort may be useful. At some point, Census may determine that the available records are sufficient to count the housing unit without receiving a direct response.
In plain language, Census could use information collected by government agencies, private organizations, or other sources to determine whether the home is occupied, who probably lives there, and what information should be recorded about those residents. The Person Characteristic Frame, discussed in the next section, will help connect people to addresses and provide quality measures intended to show how much confidence Census should place in those records.
In-office enumeration is not the same as simply using administrative data to assist an enumerator. Outside records may be used throughout the census to help locate addresses, identify possible household members, or target field resources. In-office enumeration goes further: it allows those records to become the basis for the household’s completed census response.
Census also expects its models and matching systems to help resolve cases in which it receives more than one response from an address. Multiple responses can occur because two household members respond separately, a household responds through more than one mode, or more than one family lives at an address that Census treats as a single housing unit. Administrative data may help Census determine whether the responses are duplicates, whether one should receive priority, or whether they represent different people or households.
That decision is consequential. Two responses from the same address do not always mean someone made a mistake. They may reveal a basement apartment, an informally divided home, a doubled-up household, or another living arrangement that is not reflected accurately in the address list or administrative records. Census will therefore need rules that distinguish true duplicates from evidence that more people or housing units are present than its existing data suggest.
A simplified version of the process is:
Census invites the household to respond directly.
A household that does not respond moves into field follow-up.
Models and administrative data help shape the contact strategy and evaluate the available information.
Census decides whether additional direct collection is needed or whether the records are sufficient.
If the records are accepted as sufficient, Census counts the household through in-office enumeration without obtaining a completed response from a resident.
Matching systems may also use outside data to resolve multiple or conflicting responses.
The broad outline is public, but the rules that will govern these decisions are not. Census has not yet released the thresholds for deciding that administrative data are sufficient, the circumstances that will require additional field visits, or the procedures for resolving conflicts between a direct response and an outside record. Those details will determine whether in-office enumeration remains a carefully limited backstop or becomes a routine substitute for direct collection.
Where to look: 2030 Census Operational Plan, section 3.2.3, “In-Office Enumeration,” pp. 32–34; section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48; and Figure 14, p. 28. See also the discussion of contact strategies and administrative-data use in section 3.2.2, “In-Field Enumeration.”
B. What the Person Characteristic Frame Is, and What We Still Do Not Know
The Person Characteristic Frame, or PCF, is a new operational component for the 2030 Census. Census describes it as the source of person-level administrative and supplemental data for decennial census operations. It is expected to include lists of people linked to addresses, characteristics associated with those people, information about the composition of households at those addresses, and measures of the quality of the underlying data. It will support in-office enumeration, person matching, response processing, quality control, and other operational decisions.
The PCF is part of a broader Census Bureau initiative known as the Frames Program. That program is building enterprise-wide datasets organized around four basic units: addresses, businesses, jobs, and people. Its Demographic Frame is a person-level database assembled from Census Bureau surveys, administrative records, and third-party sources. It includes demographic characteristics, addresses associated with individuals, and identifiers used to link records across datasets and over time. The 2030 Census PCF will begin with the Demographic Frame and add any decennial-specific data Census determines it needs.
Census calls the PCF a “living frame” because it will be updated as new datasets and more recent versions of existing records are received and processed. This should reduce some of the problems created by relying on a single data snapshot. But “more recent” does not necessarily mean accurate as of Census Day, and combining many records does not guarantee that they contain the information needed to reproduce a census response.
The Operational Plan says that the PCF will contain four broad types of information:
People linked to addresses. Census will identify people who appear to be associated with each housing unit.
Characteristics associated with those people. These may include demographic information available in the underlying records.
Household composition. Census will attempt to represent which people make up the household at an address.
Quality measures. Census will assess the completeness, consistency, and apparent reliability of the data at the person, household, characteristic, and geographic levels.
That description sounds comprehensive, but it leaves some of the most important questions unanswered. Census has not publicly identified which characteristics will actually appear in the PCF, which source will supply each characteristic, or whether the frame will contain meaningful relationship information. The phrase “household composition” could mean a reasonably complete roster of people associated with an address. It could also mean a set of individual person-address links with little or no information about how those people are related to one another.
That distinction matters. A person frame can indicate that five people appear to live at an address without showing that the household includes a grandparent caring for a grandchild, two unrelated families sharing a home, an unmarried couple, or a same-sex married couple. Even where relationship information is present, it may reflect the rules of the program that collected the data rather than Census residence and relationship concepts.
Many administrative systems are designed to determine program eligibility, taxation, or receipt of a particular service. Their records may therefore represent an individual beneficiary, a tax unit, an insurance unit, or a program-defined family rather than everyone who lives together in a housing unit. They may omit household members who are not relevant to eligibility, divide one residential household among several program cases, or group people together who no longer live at the same address. A record can be accurate for administering a program while still providing an incomplete or misleading description of a census household.
Census has conducted research on building household relationships from linked records, but that work itself illustrates the challenge. One Census study of probabilistic parent-child links found meaningful differences between children who could and could not be connected to a parent. Children in lower-income and less-educated households and Hispanic children were less likely to be linked successfully. That does not establish how the 2030 PCF will perform, but it demonstrates that constructing family relationships from administrative data can produce differential gaps.
The quality of race and ethnicity information presents another major concern. The 2024 revision to Statistical Policy Directive No. 15 requires a combined race and ethnicity question, adds Middle Eastern or North African as a separate minimum category, and generally requires more detailed racial and ethnic information. Agencies have until March 2029 to update their collections and systems, meaning that many of the administrative records available during the development and testing of the PCF were created under the older standards.
Older records cannot provide information that was never collected. They may not include a MENA category, may contain only broad racial categories, and may record Hispanic origin separately from race. Census’s own current “Best Race and Ethnicity” administrative-record composite applies business rules to multiple sources and assigns each person a single race and a separate Hispanic-origin value. That structure is fundamentally different from the 2024 standard, which permits people to select multiple racial and ethnic identities within one combined question and generally calls for detailed group information.
This is particularly important for Hispanic and Latino households. The combined question was adopted in part because collecting Hispanic origin separately from race created confusion and incomplete reporting. An administrative record built under the prior format may therefore be both less detailed and less capable of representing how a person would answer the 2030 Census directly. Census may be able to use statistical bridging or information from other sources, but a reconstructed value is not equivalent to a current self-response.
The same problem could arise for same-sex couples and other less common household arrangements. Identifying a same-sex couple requires accurate information about both the gender of each partner and their relationship to one another. Simply locating two people at the same address does not establish that relationship. The Census Bureau has made specific changes to its direct relationship question to distinguish same-sex and opposite-sex spouses and unmarried partners because more general approaches created known measurement problems.
The Operational Plan does not say whether the PCF will attempt to reproduce all of those questionnaire characteristics. It may be designed primarily to establish a plausible population count and basic roster for an address, rather than to create a complete substitute for the demographic and relationship information collected through a direct census response. If so, Census could use the PCF to determine that a household has been “enumerated” even though important characteristics are missing, assigned from older records, or reconstructed through models.
This concern becomes more immediate because Census may remove a nonresponding household from further field follow-up after as little as one in-person visit, provided that other contact attempts have occurred and the available records are judged sufficient. The Operational Plan refers to “multiple contact attempts,” but also says that the Bureau will use the most cost-effective method to count nonrespondents after at least one in-person visit. Those attempts may therefore include mailings or other contacts that never reached a member of the household.
In some cases, the records may provide an accurate roster and useful characteristics. In others, Census could be turning from direct collection to information it already knows was collected under outdated standards, lacks important detail, or does not represent the household relationships measured by the census questionnaire. Whether that is acceptable will depend on how Census defines a sufficient enumeration and whether the quality standards apply separately to every required characteristic, rather than only to the population count.
Census says that the PCF will include quality measures assessing whether administrative information is the same as what would have been provided in a direct response. Those measures are expected to operate at the person, household, frame, and individual-characteristic levels. That is a promising commitment. But Census has not released the measures, thresholds, validation results, or rules governing what happens when one part of a record is strong and another is weak.
For example, Census might be highly confident that four people live at an address but have little confidence in their race and ethnicity or relationships. It is not yet clear whether the household would remain in the field workload until those questions are answered, whether the characteristics would be imputed, or whether older administrative values would be accepted as a response. It is also unclear whether a direct response will always take priority when it conflicts with the PCF.
These are not small technical questions. The PCF could help Census identify missing people and target fieldwork more effectively. But it could also become an incomplete administrative portrait that is treated as more reliable than the people and households it is intended to describe.
Current commitment
Census is developing a continuously updated PCF using the Demographic Frame, other enterprise datasets, and decennial-specific administrative and supplemental data. The plan says it will contain people linked to addresses, characteristics, household composition, and quality measures.
Open design questions
Census has not publicly identified the specific characteristics and relationship fields the PCF will contain; defined “household composition”; released its source-specific quality measures; or explained what happens when records are sufficient for a population count but insufficient for race, ethnicity, relationships, or other characteristics.
Credible risk
Census may count a household after limited field contact using records that accurately identify some residents but omit others, rely on outdated race and ethnicity standards, or fail to describe multigenerational families, same-sex couples, doubled-up households, and other living arrangements accurately.
Why advocates should care
The PCF may influence who receives additional fieldwork, when Census considers a case complete, where each person is counted, how conflicting responses are resolved, and whether the final census preserves the characteristics and household relationships that communities rely on.
Where to look: 2030 Census Operational Plan, section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48, and section 3.2.3, “In-Office Enumeration,” pp. 32–34. For the larger infrastructure behind the PCF, see the Census Bureau’s Frames Program overview, Demographic Frame page, and Continuous Count Study. Particularly relevant research includes Comparisons of Administrative Record Rosters to Census Self-Responses and NRFU Household Member Responses; Revisiting Methods to Assign Responses When Race and Hispanic Origin Reporting Are Discrepant Across Administrative Records and Third-Party Sources; Where Are Your Parents? Exploring Potential Bias in Administrative Records on Children; and the Frames Program’s research on public perceptions of a national demographic frame.
C. Potential Benefits
Administrative and supplemental data could improve the 2030 Census if Census uses them carefully and for the right purposes. The central question is not whether outside data have value. They clearly do. The question is whether they are used to support direct enumeration and resolve genuinely difficult cases, or to justify ending fieldwork before Census has made a meaningful effort to reach a household.
Some households remain unresolved even after repeated outreach and field visits. Residents may be away for an extended period, refuse to participate, or live in a place that is difficult for an enumerator to access. In those cases, reliable administrative records may provide better information than the alternatives traditionally available to Census.
One important potential benefit is reducing reliance on proxy responses. When an enumerator cannot reach a household member, Census may ask a neighbor, landlord, building manager, or another person for information about the home. A proxy may know that the unit is occupied and have a general sense of how many people live there but may not know everyone in the household or be able to report their characteristics accurately. Current and well-matched administrative records could sometimes provide a better household roster than a neighbor’s estimate.
Administrative data could also help Census avoid treating an occupied housing unit as vacant. An unresolved address presents a serious problem: Census must determine whether no one lives there or whether the residents simply did not respond. Records showing recent activity at the address could provide evidence that the home is occupied and prevent its residents from disappearing entirely from the count.
Outside data may also help identify possible omissions. If a household responds but leaves out a person who appears consistently in several recent records associated with the address, Census could flag the case for additional review or direct follow-up. Used in this way, administrative data would not replace the household’s response. They would help Census identify a possible problem while there is still time to resolve it.
The same tools may help Census identify duplicate responses. People sometimes respond more than once, use more than one response mode, or appear at multiple addresses. Matching systems can help Census determine when records probably describe the same person and prevent that person from being counted twice. This could improve both the national population total and the accuracy of local counts.
Administrative data and predictive models may also help Census use field resources more effectively. Instead of treating every nonresponding address in exactly the same way, Census could use available information to identify cases that are most likely to require additional visits or specialized assistance. For example, records might help distinguish an apparently vacant seasonal home from an occupied unit where the household has simply been difficult to reach.
These tools could be particularly useful late in the collection period, when Census must decide how to allocate limited time and staff. Good information could help the Bureau send enumerators back to cases with unresolved household rosters, questionable addresses, or signs that people may have been omitted. It could also reduce unnecessary visits to cases that have already been resolved through a reliable direct response.
The strongest case for in-office enumeration is therefore relatively narrow: Census has made meaningful efforts to obtain a direct response, the household remains genuinely unreachable, and several current and consistent records provide better information than a proxy response or an assumption that the home is vacant. In those circumstances, administrative data may prevent a clear error and improve the overall count.
A limited approach of this kind would also give Census its strongest argument that in-office enumeration satisfies the Constitution’s requirement for an “actual Enumeration.” Utah v. Evans did not establish a formal test, but the Supreme Court emphasized that the Bureau had exhausted its efforts to reach households, used imputation only as a last resort for a tiny share of the population, and selected a method expected to produce a more accurate result than counting unresolved cases as zero. Carefully limited administrative enumeration following meaningful field efforts would more closely resemble the method the Court upheld than a system that routinely replaces direct collection or ends fieldwork early.
The benefits are likely to be strongest for determining whether a housing unit is occupied and identifying a basic population roster. They may be weaker for detailed race and ethnicity, household relationships, or other characteristics that are not collected consistently in administrative systems. Census should therefore evaluate sufficiency separately for the population count and for each characteristic, rather than assuming that a record that identifies the correct number of people can serve as a complete substitute for a direct response.
Used as a carefully tested backstop, administrative data could help Census resolve difficult cases, reduce some forms of proxy error, identify omissions and duplicates, and concentrate field resources where they are most needed. Those potential benefits are real. They also depend on maintaining a clear preference for direct participation and on ensuring that administrative records supplement rather than prematurely replace meaningful efforts to reach the household.
D. Data-Quality Risks for Historically Undercounted Groups
The risks associated with administrative data are not limited to whether Census identifies the correct number of people. A household can receive an apparently plausible population count while still being represented inaccurately in important ways. Residents may be omitted, linked to the wrong address, grouped into the wrong household, or assigned characteristics that are incomplete or outdated.
These errors are especially concerning for historically undercounted communities because the underlying records are not equally complete or accurate for everyone. People who move frequently, live in informal or complex households, have limited interaction with government programs, or are not well represented in existing data systems may be more likely to experience several kinds of error at once.
This section organizes the risks by the form those errors may take. It considers whether administrative records may leave people out entirely, count them in the wrong place, misunderstand who lives together, or provide inaccurate information about their characteristics and relationships. It also examines whether broad national measures of data quality could conceal substantially worse performance for particular populations, household types, or geographic areas.
These distinctions matter because different errors require different safeguards. A method that performs reasonably well in identifying whether a housing unit is occupied may still perform poorly in determining who lives there or how household members are related. Census should therefore evaluate the quality of administrative enumeration at each level rather than treating an apparently accurate population total as evidence that the full household record is reliable.
1. Missing People
The most basic risk is that a person may be absent from the records Census uses to construct the household roster. Administrative data are often described as if the government already has a file containing everyone in the country. In reality, the Person Characteristic Frame must be assembled from many datasets, each of which includes only the people and information relevant to the program, transaction, or activity that created it.
A person can be “missing” from this system in several ways. They may not appear in any of the available records. They may appear in a source but lack enough identifying information to be linked reliably to other files. Census may be able to identify the person but not connect them to a current address. Or the person may be linked successfully but left off the administrative roster Census constructs for the housing unit where they actually live.
Using multiple sources can reduce these gaps, but it cannot eliminate them. Census research on administrative-record population estimates has identified several persistent challenges, including achieving comprehensive coverage, linking people to addresses, finding people without Social Security numbers, distinguishing current residents from former residents, and identifying people who were alive and living in the United States on the relevant date.
People with limited connections to government programs or formal economic systems may be particularly difficult to locate. Many federal administrative files arise from activities such as filing taxes, receiving benefits, earning reported wages, attending school, obtaining health coverage, or participating in another public program. Someone who has little interaction with those systems may appear in fewer sources, giving Census fewer opportunities to identify them, verify their address, or determine who lives with them.
Recent immigrants present one example. Census research found that coverage of the foreign-born population in administrative records was associated with characteristics of assimilation. Naturalized citizens, people with stronger English proficiency, people with more education, and people working full time were more likely to appear in the records studied. The results suggest that newer immigrants and those with weaker connections to the institutions producing administrative data may be less consistently represented, even where overall administrative-record coverage of the foreign-born population appears relatively high.
People experiencing homelessness or housing instability present a related but distinct challenge. People without a usual housing-unit address must be reached through specialized operations rather than treated as members of a conventional household. Others may move among shelters, temporary accommodations, relatives’ homes, and informal living arrangements. They may appear in administrative records but be connected to an earlier address, a mailing address, or the address of a program or service provider rather than the place where census residence rules say they should be counted.
Housing instability can therefore make someone effectively invisible to an address-based administrative enumeration even when Census can find a record for that person. A person who recently lost housing may remain associated with an old apartment. Someone temporarily staying with another household may appear in records at a previous residence and be omitted from the roster for the home where they are living on Census Day. Because the old address looks plausible, the error may not be obvious.
Young children are another major concern. Children under age five have long experienced one of the largest census undercounts, and they often live in households whose composition is difficult to reconstruct from administrative records. A child may not appear on a tax return, may be associated with a parent at another address, or may be present in a health or benefits record that Census cannot confidently connect to the correct household. Census research has used administrative data to help identify children missing from census responses, but that research also demonstrates that no single source contains every child or resolves where every child should be counted.
The problem is not limited to whether the child can be found somewhere in the data. Census must also connect the child to an adult and determine the correct household. Recent Census research on probabilistic parent-child linkages found that successful linkage differed across types of children and households. Children in lower-income and less-educated households and Hispanic children were less likely to be linked to a parent successfully. Those results do not establish how the final PCF will perform, but they show how administrative methods can reproduce existing disparities even when the underlying system contains records for many of the children involved.
People living temporarily with another household may be missed for similar reasons. Census residence rules generally seek to count people where they live and sleep most of the time, but administrative programs may use a legal residence, mailing address, tax address, custodial parent’s address, or address associated with program eligibility. Those concepts do not always identify where the person should be counted on Census Day.
This is particularly important for doubled-up households. A friend, relative, or second family staying in a home may not appear on a lease, utility bill, property record, or benefit case connected to the primary householder. Administrative records may confidently identify the established residents while omitting the additional people whose presence makes the household difficult to count in the first place.
These omissions may be hard to detect once Census has accepted an administrative roster as complete. If the records identify three consistent residents at an address, the system may view that consistency as evidence of quality. But it cannot measure the accuracy of people who are absent from every source used to construct the roster. Agreement among incomplete records can create a high-confidence result that is still wrong.
The relevant question is therefore not simply how many people in the country can be found in at least one administrative dataset. Census must determine whether the PCF includes everyone living at each address on Census Day, including people whose lives are less consistently reflected in official systems. Quality testing should measure omissions separately for recent immigrants, young children, highly mobile people, people in temporary or informal housing, and others with limited administrative-record coverage.
Census should also preserve a meaningful opportunity for direct field contact whenever the records may not reflect the full household. Administrative data can alert an enumerator that someone may have been omitted, but they should not be used to declare a roster complete merely because the people who do appear in the records are internally consistent.
Risk
The PCF may identify a plausible group of residents while omitting people who have limited administrative footprints, recently changed addresses, or live outside conventional household arrangements.
Why advocates should care
A household can appear successfully enumerated even though a young child, recent immigrant, temporarily housed relative, or other resident is missing. Because the accepted administrative roster looks complete, the omitted person may receive no further field follow-up and may be difficult to detect through later quality review.
2. Wrong Location
Administrative records may identify a person successfully without showing where that person should be counted on Census Day. This distinction is essential. The decennial census is not simply a national list of people. It must count each person once, only once, and in the correct place under the Bureau’s residence rules.
The addresses contained in administrative records reflect the purposes and schedules of the systems that created them. A tax file may contain the address used when a return was filed. A benefits program may retain an address until a participant reports a change. A school record may reflect the address used for enrollment, while a health or commercial record may contain a mailing or billing address. Any of these may be valid for the original program but outdated or inappropriate for determining where the person usually lives and sleeps.
Highly mobile people are especially vulnerable to this kind of error. A person may appear consistently in several records at the same old address because the systems have not yet incorporated a recent move. Agreement across records can make the location appear highly reliable even when every source is repeating the same outdated information. A continuously updated frame may reduce this problem, but it cannot eliminate delays in the underlying records.
Children can be particularly difficult to place correctly. A child may be associated with one parent’s address in tax, health, school, or benefits records even when the child usually lives with the other parent or divides time between households. The address used by a program may depend on custody, insurance coverage, school enrollment, or eligibility rules rather than Census residence rules. Records may therefore identify the child but place them in the wrong household or jurisdiction.
Temporary and complex living arrangements create similar problems. A person staying with relatives after losing housing may remain linked to a prior apartment. A college student may appear at a parent’s home even when Census rules require the student to be counted at college. Someone receiving mail at a stable address may actually live somewhere else. People who divide their time among several residences may appear in records at each location, leaving Census to determine which association reflects their usual residence.
People entering or leaving institutions also present significant placement challenges. Someone living in a correctional facility, nursing home, residential treatment center, military barracks, or other group quarters may remain connected in administrative records to a prior household address. Conversely, a person who has recently left an institution may still appear in facility data even after returning to a household. If records are not current as of Census Day, Census could count the person in the wrong type of living arrangement and the wrong community.
Wrong-location errors may not affect the national population total, but they matter greatly for how census data are used. A person counted in the wrong state can affect apportionment. Placement in the wrong district, county, municipality, tribal area, or neighborhood can affect redistricting, funding, planning, and the population statistics used to administer programs. One community may receive credit for a resident who does not live there, while the community where the person actually lives loses both population and associated characteristics.
Placement errors can also distort household information. A child assigned to the wrong parent’s address changes the apparent size and structure of both households. A person incorrectly linked to a former residence may create a household that no longer exists, while leaving their current household incomplete. Those errors can affect the reported number of children, multigenerational households, renters, families, and other populations in each location.
The use of multiple sources does not automatically resolve these questions. A large number of address associations may show that a person has connections to several places without identifying which location satisfies Census residence rules. Census will need methods for evaluating the recency, purpose, and reliability of each address, as well as procedures for resolving cases in which different records point to different locations.
Direct responses should ordinarily provide the strongest evidence of where a person lives, but even those can conflict. Two households may both include the same child, or a person may respond at both a household and a group quarters facility. Administrative data can help flag these cases for review. The risk arises when Census uses administrative records to resolve the conflict automatically, particularly if the records reflect legal, financial, or programmatic relationships rather than the person’s actual residence on Census Day.
Census should therefore evaluate location accuracy separately from person coverage. Showing that the PCF contains nearly everyone does not demonstrate that those people are attached to the correct addresses. Validation should examine how often administrative methods place people correctly across moves, shared-custody arrangements, temporary housing, institutional transitions, and other situations where residence is difficult to determine.
Risk
The PCF may identify a real person but associate them with an outdated, mailing, programmatic, parental, or institutional address rather than the place where Census residence rules require them to be counted.
Why advocates should care
Wrong-location errors can leave the national total unchanged while shifting population and political representation away from the communities where people actually live. They can also distort household composition, local demographic data, funding decisions, and measures of community need.
3. Incorrect Household Composition
Administrative records may identify several people at the correct address without accurately showing how they live together or how they are related. This is a different problem from missing a person or assigning someone to the wrong location. Census may find all of the residents but still construct the wrong household.
The decennial census treats a household as all the people who live together in a housing unit. It also asks how each person is related to a central householder, distinguishing among spouses, unmarried partners, children, parents, grandchildren, roommates, and other relatives and nonrelatives. Those relationships are important for understanding family structure, caregiving, housing needs, and the living arrangements of children and older adults.
Most administrative systems were not designed to create that kind of household picture. A benefits program may define a family according to whose income and resources count toward eligibility. A tax return reflects a tax unit, which may not include everyone living in the home. Health insurance records may group people according to coverage, while school records generally describe a child and a parent or guardian rather than the full household. Commercial data may show people associated with the same address without containing any reliable relationship information at all.
As a result, several people can appear together in administrative records without Census knowing whether they are one family, two families sharing a home, unrelated roommates, or some other arrangement. Conversely, people who form one residential household may be divided across multiple administrative cases because they participate in different programs or file taxes separately.
Doubled-up households are particularly difficult to reconstruct. Two families may share a home because of housing costs, eviction, a recent move, or another temporary circumstance. Administrative records may identify only the family on the lease or utility account, treat the two families as separate program units, or connect the second family to an earlier address. Even if everyone is found, Census may incorrectly merge them into one conventional family or fail to recognize that two distinct household groups are sharing the unit.
Unmarried partners may also be difficult to identify. Two adults living at the same address do not necessarily appear as partners in tax, benefits, employment, or commercial records. They may be classified as unrelated individuals, roommates, or separate program cases. If Census lacks reliable relationship information, it may not distinguish an unmarried couple from two unrelated adults who happen to share a home.
The risk is especially pronounced for same-sex couples. Identifying a same-sex married or unmarried couple requires accurate information about both the relationship between the partners and the sex or other relevant characteristics recorded for each person. Administrative records may not contain a relationship field, may contain outdated or inconsistent gender information, or may rely on program definitions that do not distinguish spouses and partners in the same way as the census questionnaire. A system that finds both people at the same address could therefore fail to recognize them as a couple or classify the relationship incorrectly.
Census has previously made deliberate changes to its relationship questions to improve the identification of same-sex couples. That experience demonstrates why simple address matching is not enough. Household relationships must be measured explicitly and accurately rather than inferred solely from who appears to live together.
Multigenerational households present another challenge. A home may include grandparents, adult children, grandchildren, siblings, and other relatives whose connections cannot be reconstructed from a single program record. Different members may appear in separate tax units, benefit cases, insurance arrangements, or school records. Census could identify the residents correctly but misunderstand which generations live together or how they are related.
The same issue arises in informal caregiving arrangements. A grandparent, aunt, family friend, or unrelated adult may care for a child without appearing as the child’s legal parent or formal guardian in every administrative system. An older adult may live with someone who provides daily care but is not identified as a relative or paid caregiver in the available records. These relationships may be central to how the household functions even though they are invisible to programs focused on legal responsibility, financial eligibility, or service receipt.
Household composition can also change more quickly than administrative systems update. A partner may move in, a family may take in a relative, or a child may begin living with a grandparent. Census needs to capture the household as it exists on Census Day, while administrative records may reflect arrangements from months or years earlier.
These errors matter even when the population count is correct. Misclassifying household relationships can distort statistics on marriage, partnership, child caregiving, multigenerational living, and family economic circumstances. It can make same-sex couples and informal families less visible, understate housing crowding or doubling up, and produce a misleading picture of how people share resources and care for one another.
The uncertainty surrounding the PCF makes this risk difficult to assess. Census says the frame will contain information about “household composition,” but it has not publicly explained whether that means a complete roster, explicit relationship links, modeled family structures, or simply a collection of people associated with the same address. Nor has Census identified which relationship variables will be included or how it will evaluate their accuracy.
If the PCF is designed primarily to establish a person roster rather than a family or relationship frame, Census may know who appears to live at the address but not how those people are connected. The Bureau could still consider the household sufficiently enumerated for population-count purposes while relying on imputation, older administrative values, or incomplete information for relationships and other characteristics.
Census should therefore distinguish among three separate questions:
Has it identified the correct number of people at the address?
Has it grouped those people into the correct residential household or households?
Has it accurately represented the relationships among the people in each household?
Confidence in the first question should not be treated as evidence that the second and third have been resolved. Census should evaluate household composition and relationship accuracy separately and preserve direct follow-up where administrative data cannot describe the living arrangement reliably.
Risk
The PCF may identify the correct people at an address while merging separate families, splitting one household into several administrative units, or misclassifying the relationships among residents.
Why advocates should care
Incorrect household composition could make doubled-up families, same-sex couples, multigenerational households, unmarried partners, and informal caregiving arrangements less visible in census data. It could also produce inaccurate information about children, families, housing needs, and the ways people share resources and care responsibilities.
4. Incomplete or Inaccurate Characteristics
Administrative records may provide enough information for Census to conclude that a housing unit is occupied and that a particular number of people live there. That does not mean the records contain accurate information about those residents or their household. A case may appear “sufficient” for population-count purposes while remaining incomplete or unreliable for race and ethnicity, relationships, housing tenure, or other characteristics collected by the census.
This distinction matters because the decennial census produces more than a national headcount. Its data show where people live and describe the demographic and housing characteristics of communities. Those characteristics support redistricting, civil rights enforcement, program administration, funding decisions, population estimates, and many other public uses. An administrative enumeration that counts the right number of people but records their characteristics inaccurately may protect the overall population total while still degrading the data communities rely on.
Administrative systems also vary substantially in what they collect. A source may contain a strong name, age, and address but no race or ethnicity information. Another may include a broad racial category but no detailed group. A property record may identify the legal owner of a home without establishing who occupies it or whether the residents consider themselves owners or renters. A benefits record may describe relationships according to program eligibility rules rather than the relationships measured by the census questionnaire.
Race and ethnicity data illustrate the problem particularly clearly. Most administrative systems were developed under the earlier federal standards, which collected Hispanic or Latino origin separately from race and did not include Middle Eastern or North African as a minimum category. Many records also contain only broad categories and lack the detailed racial and ethnic identities Census seeks to collect, such as Mexican, Puerto Rican, Cuban, Chinese, Filipino, Jamaican, Somali, or Lebanese.
The 2024 revision to Statistical Policy Directive No. 15 adopted a combined race and ethnicity question, added MENA as a co-equal minimum category, and generally called for more detailed reporting. Those changes were based in part on research showing that the combined format improves reporting, particularly for Hispanic and Latino respondents who often found the separate questions confusing or did not identify with the racial options provided.
Administrative records created before agencies implement the revised standards cannot supply information that was never collected. They may classify a MENA respondent as White, record a Latino respondent’s ethnicity separately from an incomplete or missing race response, or contain only a broad category that does not identify the person’s detailed community. Even where Census combines information from several files, it may be reconstructing a likely value from older categories rather than recording how the person would answer the 2030 Census directly.
Information can also become outdated. Race and ethnicity are self-identified characteristics, and the way a person reports their identity may change as question wording and response options improve. A record copied forward from an earlier government interaction may not reflect the person’s current response, especially after the addition of MENA and the shift to a combined question. Agreement among several administrative records may simply mean that the same older classification has been reused across systems.
Relationship information presents a similar challenge. As discussed in the prior subsection, administrative programs may identify a tax unit, benefits case, insurance family, legal guardian, or emergency contact rather than the full set of relationships among everyone living in the housing unit. Records may not distinguish a spouse from an unmarried partner, identify whether a couple is same-sex or opposite-sex, or describe informal caregiving and multigenerational relationships accurately.
Even where Census can construct a plausible household roster, it may lack enough information to assign every person a reliable relationship to the householder. The Bureau may then need to impute relationships, infer them from limited records, or leave the characteristic unresolved. Those approaches could produce a complete-looking dataset while making less common household arrangements systematically less visible.
Housing tenure may also be difficult to determine from outside records. The census asks whether a home is owned with a mortgage or loan, owned free and clear, rented, or occupied without payment of rent. Property and tax records may identify the legal owner, but the owner may not live at the address. A resident may rent informally from a relative, live in a family-owned home without paying rent, occupy a unit through an employer, or stay temporarily without appearing on a lease.
Administrative data could therefore classify a housing unit as owner-occupied because the property owner shares a surname with one resident, even though the household rents the home. It could identify renters from a lease that no longer reflects the current occupants or fail to distinguish a conventional rental from a household living without payment. These errors would affect measures of homeownership, renting, housing stability, and community wealth.
Other characteristics may present their own limitations. Age and date of birth are often available in administrative records, but conflicting or incomplete values still occur. Gender information may be outdated, inconsistently defined, or absent from particular sources. Records may not contain the information needed to determine whether people are related, whether a housing unit contains more than one household, or which respondent-provided value should be used when administrative files disagree.
The central problem is that data sufficiency is not a easily defined. Census could have:
high confidence that the housing unit is occupied;
high confidence in the number of residents;
moderate confidence in their identities and locations; and
little confidence in their race and ethnicity, relationships, or housing tenure.
A single overall quality score could conceal these differences. If Census considers the case complete because the population roster appears reliable, it may stop fieldwork even though the characteristic data remain incomplete. The resulting record may look finished in the system while depending heavily on imputation or outdated administrative information for the variables the public ultimately uses.
Census says the PCF will include quality measures at the individual-characteristic level. That is an important commitment, but the Bureau has not publicly explained how those measures will affect fieldwork. It is not clear whether a household with a reliable population count but weak characteristic data will receive additional contact, whether different standards will apply to different variables, or whether Census will accept lower-quality characteristics once the roster is judged sufficient.
Census should evaluate each major characteristic separately and distinguish between information reported directly by the household and information assigned from another source. Direct self-response should receive priority, particularly for characteristics based on identity or household relationships. Where outside records use outdated standards or lack meaningful detail, Census should not treat their internal consistency as proof that the information is accurate.
The Bureau should also report the extent to which final characteristics come from direct responses, administrative records, modeling, or imputation. Aggregate population accuracy could otherwise obscure substantial differences in the quality of the demographic and housing data produced for different communities.
Risk
Census may treat a household as sufficiently enumerated because its population roster appears reliable, even though its race and ethnicity, relationships, housing tenure, or other characteristics are missing, outdated, inferred, or recorded under standards that do not match the 2030 questionnaire.
Why advocates should care
Communities could appear in the population count while disappearing or being misrepresented in the characteristics data used for redistricting, civil rights enforcement, program administration, and resource allocation. The effects may be greatest for detailed racial and ethnic groups, MENA and Latino communities, same-sex couples, multigenerational households, informal families, and people in less conventional housing arrangements.
E. Decisions That Require Particular Scrutiny
The Operational Plan establishes the broad framework for in-office enumeration, but it does not yet provide the decision rules that will determine how the process works in practice. Those rules will decide how much effort Census makes to obtain a direct response, when administrative records are considered sufficient, and whether outside data can alter or displace information provided by a household.
These choices are especially consequential because they will be applied case by case, often through statistical models, matching systems, and automated processes. A small change in a threshold could affect hundreds of thousands of households. A rule intended to reduce unnecessary field visits could instead reduce outreach in communities that are more difficult to reach. A standard that appears reasonable nationally could produce very different results for highly mobile people, complex households, or populations that are poorly represented in administrative records.
The terminology used in the plan can also obscure the significance of these decisions. Phrases such as “multiple contact attempts,” “response sufficiency,” “optimal fieldwork,” and “data collection is complete” sound technical and neutral. In practice, each reflects a policy judgment about when Census has done enough to seek a direct response and what level of uncertainty it is willing to accept. The way Census defines these terms will help determine whether administrative data remain a limited backstop or become a routine substitute for field collection.
The subsections that follow examine several decisions that warrant particular scrutiny: what constitutes a meaningful attempt to reach a household; how Census will determine that administrative data are sufficient; whether direct responses will receive priority when sources conflict; how the Bureau will distinguish duplicate responses from multiple households sharing an address; and how predictive models will affect the distribution of fieldwork and outreach resources.
These issues should be addressed before the systems are finalized and deployed. Advocates will need access not only to general descriptions of the process, but also to the underlying thresholds, validation studies, business rules, and measures of differential quality. Without that information, it will be difficult to determine whether the design protects direct participation and improves accuracy for historically undercounted communities or simply allows Census to close difficult cases more quickly.
1. What Counts as Sufficient Field Effort?
The current Operational Plan says that Census will use administrative records to enumerate a nonresponding household only after “multiple contact attempts,” including at least one in-person visit. That commitment provides an important minimum safeguard. But it does not yet explain what Census will count as an attempt, what must happen during the in-person visit, or when the Bureau may conclude that further fieldwork is unlikely to improve the count.
Those details matter because one visit does not necessarily provide a household with a meaningful opportunity to respond. An enumerator may arrive while residents are at work, taking children to school, receiving medical care, or otherwise away from home. The visit may occur at a building the enumerator cannot enter, at an address that is difficult to locate, or at a time when residents are understandably unwilling to open the door. A record showing that an enumerator visited the address does not establish that anyone in the household knew Census was trying to reach them.
Census should therefore distinguish between an attempted visit and an opportunity for direct contact. At minimum, field-effort standards should consider:
whether the visit occurred at a reasonable time;
whether visits were attempted at different times or on different days;
whether the enumerator reached the correct housing unit;
whether anyone in the household received notice of the visit;
whether language or accessibility assistance was available; and
whether the address or building presented barriers requiring a different approach.
A single weekday visit during working hours, for example, should not be treated as equivalent to several attempts designed to reach residents at different times. Nor should Census consider fieldwork sufficient when an enumerator could not locate the unit, could not enter a building, or lacked the language skills needed to communicate with the household.
The meaning of “multiple contact attempts” is also unclear. The phrase may suggest repeated efforts to speak with a resident, but Census may count mailed invitations, reminder letters, text messages, telephone calls, emails, or digital advertisements toward the contact strategy. Those methods can be useful, but they do not all provide the same evidence that the household received or understood the request.
Mail may be delayed, discarded, or sent to an address with unreliable delivery. A telephone number may belong to a former resident. A text or email may look fraudulent. A QR code may be ignored because the recipient does not trust it. Counting these actions as separate attempts can make the outreach record appear extensive even when Census has never communicated directly with anyone in the household.
The contact strategy should therefore distinguish among:
Messages sent, such as mailings, calls, texts, or emails;
Contacts that reached someone, even if that person did not complete the census; and
Meaningful opportunities to respond, in which a knowledgeable household member received understandable information and had access to an appropriate response mode.
All three may be relevant, but they should not be treated as interchangeable. Before replacing direct collection with administrative data, Census should be able to show that it made a reasonable effort to create the third kind of opportunity.
Additional visits should generally be required when the first attempt produces no contact, especially where available information suggests that the housing unit is occupied. The number and form of those visits need not be identical for every household. A seasonal property, an apartment building with restricted access, and a rural home with a difficult-to-locate address may require different strategies. But flexibility should be used to improve the likelihood of direct response, not simply to reduce the number of visits.
Predictive models create a particular concern. Census plans to use information about a household’s expected likelihood of responding to help shape the contact strategy. That could be beneficial if it directs more resources, specialized staff, or trusted outreach toward households expected to face barriers. It could be harmful if a low predicted response rate becomes a reason to make fewer attempts and move more quickly to administrative enumeration.
Historically undercounted communities may be especially vulnerable to that outcome. A model may learn from prior response patterns that households in a particular neighborhood, language group, housing type, or demographic category are less likely to respond. If Census then reduces field effort for those households, the model will not merely predict unequal participation. It will help reproduce it.
Census should adopt the opposite principle: evidence that a household is difficult to reach should generally support a more appropriate or intensive contact strategy, not an earlier conclusion that direct collection is futile. That may mean bilingual enumerators, visits at different times, assistance from community partners, improved address information, or response options better suited to the household’s circumstances.
The constitutional implications also matter. In Utah v. Evans, the Supreme Court emphasized that Census used imputation after attempting to reach every household. Although the Court did not establish a formal checklist, meaningful efforts to obtain a direct response strengthened the conclusion that the method remained part of an “actual enumeration.” A system that turns to administrative records after minimal or largely indirect outreach would be more difficult to compare with the limited last-resort method the Court upheld.
Census should therefore establish and publish minimum field-effort standards before in-office enumeration begins. Those standards should explain which contacts count, when additional visits are required, what exceptions are permitted, and how the rules account for language, disability, building access, rural addressing, and other barriers. The Bureau should also report how often households move to in-office enumeration after one visit, two or more visits, or no successful contact with a resident.
Current commitment
The Operational Plan says that nonresponding households will receive multiple contact attempts, including at least one in-person visit, before Census uses administrative records to enumerate them.
Open design question
Census has not publicly defined a contact attempt, identified the circumstances requiring additional visits, or explained how response predictions will affect the amount of field effort a household receives.
Credible risk
Census may count mailings, unsuccessful calls, digital messages, and a single unanswered visit as sufficient effort, even though no household member received a meaningful opportunity to respond.
Why advocates should care
If households predicted to be difficult to reach receive less field effort, in-office enumeration could become most common in the same communities that have historically needed the most direct outreach. This could deepen differential undercounts and weaken Census’s legal argument that administrative enumeration is a limited last resort.
2. What Makes Administrative Data “Sufficient”?
The Operational Plan says Census may complete a nonresponding household using administrative records when the available information is considered sufficient. That word carries enormous weight. A finding of sufficiency may end fieldwork, establish the household roster, and determine which information enters the final census record. Yet Census has not publicly explained the thresholds or decision rules that will govern that judgment.
Administrative data are rarely simply complete or incomplete. Census may have strong evidence that a housing unit is occupied, moderate confidence that four people live there, conflicting evidence about one person’s current address, and little reliable information about the residents’ relationships or race and ethnicity. Whether that case is treated as complete will depend on how the Bureau weighs different sources, different kinds of uncertainty, and different census variables.
Census should therefore avoid a single, overall determination that the records are “good enough.” Sufficiency should be assessed separately across at least five dimensions.
Person coverage. Census must determine whether the records identify everyone who lives in the housing unit, not merely whether they identify several plausible residents. This requires rules for evaluating possible omissions, recent additions to the household, and conflicting rosters across sources. Agreement among the people who appear in several records does not establish that no one else lives there.
- The Bureau should disclose what level of confidence is required before it accepts an administrative roster as complete. It should also explain how the system treats evidence of a possible additional resident, such as a child found in one source but not others, a person newly connected to the address, or a direct response that names someone absent from the administrative roster.
Address confidence. Finding a person in administrative records is not enough. Census must determine that the person should be counted at that particular address under its residence rules. The relevant decision rules should consider the date, source, and purpose of each address; whether several records may be copying the same outdated information; whether the person has credible ties to multiple locations; and whether a mailing, legal, or program address reflects the person’s usual residence.
- Census should publish the confidence threshold required to assign a person to an address and explain what happens when that threshold is not met. It should also identify the circumstances in which uncertainty about location triggers additional fieldwork rather than an automated placement decision.
Household composition. A collection of person-address links does not necessarily establish a complete and accurate household. Census needs rules for determining whether everyone at an address forms one household, whether multiple families or household groups share the housing unit, and how residents are related.
- The Bureau should explain what evidence is required to distinguish spouses, unmarried partners, children, other relatives, roommates, and unrelated household members. It should also state whether reliable relationship information is necessary before a case is considered complete or whether the household can be closed with relationships later assigned through imputation or other processing.
Characteristic completeness. Census should separately evaluate whether it has reliable information for each characteristic included in the census record. A strong population roster should not automatically make incomplete race and ethnicity, relationship, sex, age, or housing-tenure information sufficient.
- This is especially important where administrative records use older categories or definitions that do not match the 2030 questionnaire. Census should explain whether a broad or outdated race and ethnicity value is considered complete, how conflicts among sources are resolved, and when missing detail requires renewed contact with the household. It should also disclose which characteristics may be imputed or modeled after a case is closed.
Data age and geographic accuracy. The value of a record depends in part on when it was created and what geographic information it contains. A source may be highly accurate for its original purpose but too old to describe the household on Census Day. A person may be associated with a valid address but lack the unit number needed to place them correctly within an apartment building. A record may identify the correct facility or parcel without identifying the correct housing unit.
- Census should publish standards for how recent a record must be, whether those standards differ by source and characteristic, and how conflicting dates are handled. It should also explain the minimum geographic precision required for enumeration and what happens when the available information identifies only a building, parcel, rural route, post office box, or former address.
These dimensions should not be collapsed into a single confidence score without clear safeguards. A high overall score could conceal a serious weakness in one component. Strong address evidence might offset missing race and ethnicity information mathematically, even though the two answer different census questions. Similarly, confidence that three identified people live at an address should not erase evidence that an unidentified fourth resident may be present.
The sources themselves also require scrutiny. Several records that agree with one another may not provide independent confirmation. One agency may have copied information from another, or several commercial files may trace back to the same transaction. This concern is heightened by the current Administration’s push to expand interagency data sharing and consolidate government data resources, together with Census’s own emphasis on reusing existing data to reduce respondent burden and collection costs. As more agencies exchange or derive information from common sources, datasets that appear independent may increasingly repeat the same underlying record or error. Census should disclose the origins and dependencies of the data it uses and explain how it will prevent duplicated information from creating a false impression of certainty.
The rules should also account for differences in source quality across populations and places. A threshold calibrated to national averages may produce less reliable decisions for people with limited administrative footprints, highly mobile households, rural addresses, or complex living arrangements. Where testing shows elevated error, Census should require stronger evidence or additional field contact rather than applying the same threshold everywhere.
Most importantly, the consequence of failing a sufficiency test should be clear. An unresolved question about one person, address, or characteristic should not simply disappear into a general quality score. Census should specify which deficiencies trigger additional visits, specialized follow-up, proxy collection, imputation, or another resolution process. Advocates need to know not only how a case passes, but what Census does when the evidence remains uncertain.
The thresholds and business rules should be published early enough for independent review. Census should also report how many households were administratively enumerated under each rule, how often sources conflicted, how frequently characteristics were missing or imputed, and how error rates varied across communities. Without this information, the public will be asked to trust a consequential decision process that it cannot examine.
Current commitment
Census plans to develop quality measures for people, households, characteristics, and geographic associations and to use those measures in determining whether administrative data can support enumeration.
Open design question
Census has not publicly released the thresholds, source-ranking rules, conflict-resolution procedures, or minimum characteristic requirements that will define sufficiency.
Credible risk
Census could use a broad confidence score or a lenient threshold to close cases with reliable occupancy information but uncertain rosters, outdated addresses, or incomplete demographic and household characteristics.
Why advocates should care
The definition of sufficiency will determine when Census stops trying to hear directly from a household. Unless the rules assess each component separately and account for differential error, an apparently technical standard could systematically accept lower-quality information for historically undercounted communities.
3. Will a Direct Response Always Receive Priority?
Direct response is generally the strongest evidence Census can receive about who lives in a household and how its members identify themselves. A current response from someone knowledgeable about the household is more closely tied to Census Day, Census residence rules, and the questions Census is actually asking than information collected earlier for taxes, benefits, health care, or another administrative purpose.
The move toward a more data-driven census nevertheless raises an important question: when a direct response conflicts with administrative records, which source will Census treat as authoritative? The answer will affect not only whether a person is counted, but where they are counted, how the household is constructed, and which demographic and housing characteristics appear in the final data.
Census should begin with a strong presumption in favor of a timely direct response. Administrative data may be useful for identifying possible omissions, duplicates, or inconsistencies, but they should ordinarily trigger review rather than automatically override information supplied by the household. Otherwise, the Bureau risks replacing a current answer to a census question with an older value collected under different definitions and for a different purpose.
That principle still leaves several kinds of conflict that Census will need to resolve.
Self-response and administrative data. A household may report four residents while administrative records identify only three, or the records may associate an additional person with the address whom the respondent does not include. Census will need to determine whether the difference reflects an omission, an outdated record, a recent move, a temporary resident, or a misunderstanding of the residence rules.
Administrative records can help flag these cases, but they should not be treated as inherently more reliable. A person may remain connected to an address long after moving, while a new resident may not yet appear in official files. The Bureau should disclose when a discrepancy leads to additional contact, when the direct roster is accepted, and under what limited circumstances an administrative record may alter a self-response.
The same issue arises for characteristics. A respondent may select a race or ethnicity category that differs from older government records, identify a spouse or unmarried partner differently, or report that the household rents a home listed in property records under a relative’s name. Current self-reported information should ordinarily control, especially for identity and relationship characteristics that administrative systems may collect inconsistently or under outdated standards.
Two self-responses from the same address. Multiple responses at one address may reflect a duplicate submission, but they can also reveal a more complicated living arrangement. Two people may each complete the census for the same household. A resident may respond online and later return a paper form. But two families sharing one housing unit may also submit separate rosters because neither respondent views the other family as part of their household.
Census should not assume automatically that one response is erroneous. The Bureau must compare the rosters, submission dates, response modes, and available information to determine whether the responses describe the same people, overlapping groups, or distinct households living at the same address. A system designed mainly to remove duplicates could inadvertently erase a doubled-up family or other residents who are not part of the primary householder’s family.
Where two responses contain different household characteristics, Census will also need rules for deciding which answer to retain. The most recent response is not always the most accurate, and the response associated with the apparent householder may not describe everyone in the housing unit. Significant conflicts should trigger review rather than being resolved solely by an automated ranking rule.
Administrative records pointing to different households. Outside records may produce competing rosters for the same address. Tax records may identify one family, benefits data another, and commercial sources a third group of people. These differences may reflect moves, separate program units within one household, multiple families sharing a home, or records that have not been updated on the same schedule.
Census will need to decide whether to merge the records, select one source, construct a combined roster, or seek additional direct information. Simply choosing the source with the highest general quality rating may not resolve the specific case. A source that is usually reliable may still contain an old address or reflect only the people relevant to a particular program.
The Bureau should disclose how it ranks sources, how it treats records with common origins, and what level of disagreement makes the available data insufficient for in-office enumeration. Conflicting household rosters should generally weigh in favor of further contact, not be averaged into an apparently complete answer.
Current responses and older government records. Timing should be a central consideration. The census seeks to measure the population and households as they exist on Census Day. Administrative records may have been collected months or years earlier and may continue to circulate across agencies after the original information becomes outdated.
A current direct response should therefore ordinarily receive greater weight than an older administrative value. This is particularly important for addresses, household membership, relationships, housing tenure, and race and ethnicity classifications recorded under previous federal standards. Older records can provide context or identify a possible inconsistency, but they should not displace current information merely because several agencies repeat the same earlier value.
Census will also need rules for cases in which the direct response itself is incomplete or appears unreliable. A respondent may omit a person, misunderstand a residence rule, provide conflicting answers, or submit a response that appears fraudulent. The existence of these possibilities does not justify treating all direct responses as provisional. Instead, the Bureau should define the specific evidence required before it modifies or rejects information supplied directly by a household.
Transparency will be particularly important because conflict resolution may occur during automated processing that the household never sees. Census should publish the hierarchy of evidence it uses, the circumstances in which a direct response can be overridden, and the procedures for resolving disagreements about rosters, addresses, relationships, and characteristics. It should also report how often final records differ from the information households submitted and which sources caused those changes.
A clear preference for direct response would preserve the proper role of administrative data: supporting enumeration, identifying cases that need attention, and resolving genuinely unreachable households. Without that preference, outside records could become a mechanism for silently rewriting current answers based on older or less relevant information.
Current commitment
Census intends to use administrative data and automated matching to identify duplicates, reconcile multiple responses, resolve inconsistencies, and support final census records.
Open design question
Census has not publicly established a clear hierarchy among current self-responses, multiple household responses, and conflicting administrative sources, or defined when directly reported information may be changed.
Credible risk
Automated conflict-resolution rules may favor administrative records, the apparent primary householder, or a single “best” response even when the conflict reflects a recent move, a doubled-up household, or outdated government information.
Why advocates should care
Direct responses are the principal way people describe their current households and identities in their own words. If Census can override those responses without strong evidence and transparent rules, historically undercounted communities may be counted according to what older systems say about them rather than what they report about themselves.
4. How Will Census Handle Multiple Households at One Address?
Census systems are designed to prevent people and housing units from being counted more than once. Duplicate detection is essential, particularly when households can respond online, by mail, by telephone, or through an enumerator. But two responses associated with the same address are not necessarily duplicates. They may indicate that the address itself does not fully describe the living arrangements at the property.
One street address can contain more than one housing unit. A basement or garage may have been converted into an apartment. A single-family home may have been subdivided without receiving separate unit numbers. An accessory dwelling unit (ADU) may share the primary home’s mailing address. Two buildings on the same parcel may use one address, or several apartments may receive mail through a common mailbox. In rural areas, tribal communities, mobile home parks, and informal settlements, an address may identify a general location without distinguishing every occupied unit.
A single housing unit can also contain more than one family or household group. Two families may share a home because of high housing costs, displacement, migration, or a temporary emergency. Residents may share a kitchen or entrance while managing their finances and family lives separately. Each group may complete a census response without knowing whether the other has responded or without viewing the other residents as part of its household.
In these situations, multiple responses are evidence that Census may not fully understand the property or household. They should not be treated simply as a data-cleaning problem.
Census must first determine whether the responses describe:
the same people responding more than once;
overlapping rosters for one household;
separate households sharing one housing unit; or
separate housing units that use the same formal address.
Each situation requires a different resolution. True duplicate responses should be combined or removed. Overlapping rosters may require additional information to determine who belongs in the household. Separate household groups must be reconciled into an accurate roster of everyone living in the housing unit. Separate housing units may need to be added to or corrected in the address frame so that each can be enumerated independently.
Automated matching may not be able to make these distinctions reliably. Two responses that include different names can still describe the same household if one respondent omitted some residents. Responses with overlapping surnames may represent extended families living in separate units. Two rosters with no names in common may be separate households, or one may describe former residents whose response was associated with an outdated address.
The physical address itself may offer little help. A system may assume that one address corresponds to one housing unit because the Master Address File contains only one listing. Property, postal, utility, or commercial records may reinforce that assumption because they also treat the property as a single address. Agreement among these sources does not prove that only one unit or household exists. It may instead reflect the fact that informal or newly created living quarters have not been recorded by any of the systems Census is using.
These cases are particularly important for accessory dwelling units and informally subdivided properties. An additional unit may be fully independent in practice but lack a separate postal address, utility account, or building permit. Its residents may respond using the primary address and add a description such as “basement,” “rear unit,” or “garage apartment.” A processing system focused on standardized addresses could discard that information, merge the response with the primary household, or treat it as an invalid duplicate.
Census should preserve and evaluate unit descriptions supplied by respondents, even when they do not match an existing address record. Multiple responses associated with one address should trigger an inquiry into whether the address frame is incomplete. The Bureau should not require residents of an informal unit to prove that the space is legally recognized before treating it as separate living quarters under census rules.
The appropriate response may also depend on whether the space meets Census’s definition of a housing unit. That determination generally turns on how people live in the space, not merely on zoning, ownership, or the labels used in administrative records. Census will need procedures for identifying separate living quarters where residents live independently but share an entrance, mailbox, or formal address.
Field verification may be especially valuable in these cases. An enumerator may be able to observe multiple entrances, speak with residents, or learn that a property contains an additional unit not shown in existing records. If Census instead relies on automated processing to select a single “best” response, it may eliminate precisely the evidence that should have prompted additional fieldwork.
The Bureau should also be cautious about treating the response associated with the owner, leaseholder, or apparent primary householder as authoritative. That person may not know everyone living elsewhere on the property. They may be unaware of who occupies an accessory unit, may not consider a doubled-up family part of their household, or may intentionally exclude residents who are not formally permitted to live there. Selecting the response with the strongest connection to property or administrative records could therefore privilege the most formally documented residents over the people whose living arrangements are less visible.
Fear may further complicate reporting. Residents of an unpermitted unit or overcrowded home may worry that identifying their living arrangement could lead to eviction, code enforcement, benefit consequences, or other harm. Census’s confidentiality protections are designed to prevent census responses from being used for those purposes, but the Bureau must communicate those protections clearly. A process that appears to compare responses against property, benefits, or other government records could make residents less willing to explain the arrangement.
Census should publish the rules it will use to identify and resolve multiple households or units at one address. Those rules should explain:
what patterns trigger review rather than automatic deduplication;
how respondent-provided unit descriptions will be preserved;
when Census will add a newly identified housing unit to the address frame;
when field verification will be required;
how the system will distinguish separate units from separate families sharing one unit; and
how residents will be counted when the available responses contain overlapping or incomplete rosters.
Testing should include properties and communities where formal address records are least likely to reflect actual living arrangements. That includes accessory dwelling units, converted homes, doubled-up households, rural properties, tribal lands, mobile home parks, and buildings with shared or inconsistent unit designations. Census should report how often multiple responses lead to a new housing-unit listing, a combined roster, additional fieldwork, or the removal of a response as a duplicate.
The governing principle should be that multiple responses are information, not merely noise. They may reveal a duplicate, but they may also reveal a household or housing unit that Census would otherwise miss. The process should be designed to investigate that possibility before eliminating any response.
Current commitment
Census plans to use matching and automated processing to reconcile multiple responses and prevent duplicate enumeration.
Open design question
Census has not publicly explained how it will distinguish true duplicates from separate households or housing units that share one formal address.
Credible risk
Automated systems may retain one response and discard another because the address frame assumes that only one unit or household exists, causing residents of accessory units, subdivided properties, or doubled-up households to be omitted.
Why advocates should care
The people living in arrangements that are least visible in formal address and property records are often the same people most vulnerable to being missed. Treating every additional response as a likely duplicate could erase evidence of hidden housing units and complex households rather than helping Census count them.
5. How Will Models Affect Field and Outreach Resources?
Models can help Census use limited resources more effectively. They may identify neighborhoods where mail delivery is unreliable, language assistance is needed, address records are weak, or households are likely to require in-person help. Used well, these tools could direct enumerators, mobile response teams, advertising, and community partnerships toward the places where additional support would make the greatest difference.
The same models could also be used in the opposite way. If Census predicts that a household or community is unlikely to respond, it may conclude that further outreach would have little value and move more quickly toward administrative enumeration. A tool intended to anticipate response behavior could therefore become a mechanism for deciding which households are worth continued effort.
That risk is especially important because response predictions are not neutral descriptions of individual willingness. They may reflect structural barriers such as limited language access, distrust of government, unstable housing, poor internet service, inaccessible response options, or a history of inadequate outreach. A low predicted response rate may indicate that Census needs a different strategy, not that residents are unwilling to participate.
Models may also rely on characteristics that are closely associated with race, ethnicity, income, immigration, age, disability, or geography, even if those characteristics are not included directly. Housing type, neighborhood, prior response patterns, internet access, mobility, and participation in government programs can act as proxies for the populations Census has historically struggled to count. Decisions based on those factors could systematically reduce resources in the same communities that already face the greatest barriers.
What matters is how Census acts on their predictions. A prediction that a community will have low self-response could lead Census to:
provide more bilingual staff and translated materials;
begin trusted-partner outreach earlier;
place additional questionnaire-assistance sites nearby;
schedule more field visits at varied times; or
improve address and contact information before fieldwork begins.
But it could instead lead Census to reduce mailings, shorten field follow-up, assign fewer enumerators, or rely more heavily on administrative records. The model itself does not determine which approach is appropriate. That is a policy choice embedded in the operational rules surrounding it.
Census should adopt a clear principle that predicted difficulty should ordinarily trigger additional or better-tailored assistance, not diminished effort. Models may reasonably help distinguish which form of outreach is most promising. They should not be used simply to label households as unlikely responders and close their cases more quickly.
Resource decisions made at the community level also require scrutiny. Census may use neighborhood response rates or model estimates to reassign staff during data collection. Moving enumerators away from places where response remains low could appear efficient if those areas produce fewer completed interviews per visit. But that approach would measure success by immediate productivity rather than by whether the people most at risk of being missed receive an adequate opportunity to participate.
This can create a self-reinforcing cycle. A community receives limited or poorly matched outreach, response remains low, and the model interprets the result as evidence that further effort is unlikely to succeed. Resources are then shifted elsewhere, producing even fewer direct responses and greater reliance on administrative data. The final outcome may appear to confirm the original prediction even though Census’s own resource decisions helped produce it.
Similar feedback loops can extend across census cycles and tests. If a model is trained on prior outcomes, it may learn patterns created by earlier operational choices. Communities that received insufficient language assistance or field staffing in one test may be classified as inherently difficult to enumerate in the next. Census should distinguish between barriers arising from the population and barriers created by the design of its own outreach.
Models may nevertheless offer important benefits when used to diagnose specific needs. A prediction that online response will be low could support additional paper questionnaires or in-person assistance. Evidence that an area contains many complex or unstable households could support more experienced enumerators and less reliance on administrative rosters. Models can also help identify emerging problems while there is still time to adjust staffing and outreach.
To realize those benefits, Census should disclose what each model is designed to predict and how its output affects operations. A general statement that models will “optimize” resources is not enough. The Bureau should explain whether a low predicted response results in more assistance, fewer contact attempts, a different response mode, earlier administrative enumeration, or some combination of these actions.
Census should also publish the factors used by the models, their relative importance, and their performance across populations and geographies. Independent reviewers should be able to assess whether the models systematically overestimate or underestimate response for historically undercounted groups and whether errors lead to unequal levels of service.
Testing should evaluate operational outcomes, not only predictive accuracy. A model could correctly predict that a household is unlikely to respond under the standard contact strategy but fail to identify that a different strategy would succeed. The relevant question is therefore not merely whether the prediction was accurate. It is whether using the prediction helped Census obtain a more complete and equitable count.
The Bureau should establish safeguards against resource decisions that deepen disparities. These could include minimum service levels that models cannot reduce, requirements for additional review before field resources are withdrawn from low-response areas, and explicit testing of whether modeled decisions increase differential undercounts. Census should also involve local officials, tribal governments, community organizations, and other trusted partners in interpreting patterns that the model may not fully explain.
Transparency should continue during data collection. Census should report how models are affecting the distribution of enumerators, language assistance, advertising, questionnaire-assistance centers, and partnership resources. Aggregate staffing totals will not show whether resources are being shifted away from the communities with the greatest unmet need.
The key distinction is between using models to identify barriers and using them to ration effort. The first approach could improve the census by matching resources to community needs. The second could turn predictions about low response into decisions that make low response inevitable.
Current commitment
Census plans to use models and near-real-time information to shape contact strategies, prioritize fieldwork, and allocate operational resources.
Open design question
Census has not fully explained what the models will predict, which factors they will use, or whether a low predicted response will lead to additional assistance or reduced field effort.
Credible risk
Models may direct resources toward households expected to respond readily while moving difficult cases more quickly to in-office enumeration, creating a feedback loop in which historically undercounted communities receive fewer opportunities to participate directly.
Why advocates should care
Models will not merely forecast census participation. They may influence where enumerators, language services, community partnerships, and other resources are deployed. Unless the rules are transparent and designed to reduce disparities, automated resource allocation could reproduce and deepen the patterns it is intended to manage.
Where to look: Begin with the 2030 Census Operational Plan. Section 3.2.2, “In-Field Enumeration,” pp. 30–32, describes the development of the contact strategy and the use of administrative records to reduce additional field visits. Section 3.2.3, “In-Office Enumeration,” pp. 32–34, discusses models for assessing data sufficiency, predicting response, determining contact strategies, resolving multiple responses, and deciding when collection is complete. Section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48, describes the development of quality measures and matching thresholds at the person, household, characteristic, and geographic levels. Sections 3.2.13 and 3.2.14, pp. 48–52, provide additional information on quality review and response processing.
Earlier Census materials on adaptive design, administrative-record enumeration, response unduplication, and the 2020 Nonresponse Follow-up operation provide useful background on how similar decisions were made in the last census. These documents do not establish the rules for 2030, but they illustrate the kinds of models, thresholds, source-ranking rules, and cost-quality tradeoffs that advocates should expect Census to address.
Note that the most important sources may not yet be public. Advocates should watch the 2030 memorandum series, research-and-testing materials, the 2026 Census Test and 2028 Dress Rehearsal reports, and future revisions to the Operational Plan. In particular, they should seek the actual rules governing minimum field effort, administrative-data sufficiency, conflicts between direct responses and outside records, multiple responses at one address, and the use of models to allocate field and outreach resources.
General assurances that Census will use “high-quality” data or “optimal” contact strategies will not be enough. The key information will be found in the thresholds, business rules, validation studies, differential-quality results, and operational consequences applied to individual cases.
F. The constitutional and statutory risk
The use of administrative records does not necessarily conflict with the Constitution or federal census law. Census has long used statistical methods and outside information to resolve difficult cases, and the Supreme Court has recognized that an “actual enumeration” does not require every person to be reached through a completed questionnaire or face-to-face interview. The legal question is therefore not simply whether Census uses administrative data, but how, when, and at what scale it uses them.
The risk increases if in-office enumeration moves beyond a limited last-resort method. A system is more legally vulnerable if Census ends direct collection after minimal field effort, relies on records that do not accurately reflect Census Day residence, uses models to determine large numbers of household rosters, or cannot demonstrate that the resulting count is more accurate than available alternatives. The same operational choices discussed in the preceding sections will therefore shape both data quality and legal defensibility.
The Supreme Court’s decision in Utah v. Evans provides the most important starting point. The Court upheld the Bureau’s limited use of imputation after emphasizing that Census had tried to reach the affected households, used the method for only a very small share of the population, selected a result expected to be more accurate than counting unresolved households as empty, and employed a method with little apparent potential for political manipulation. The decision did not establish a formal checklist, but those considerations provide a useful framework for evaluating the more extensive use of administrative records now being considered for 2030.
Statutory questions are also important. The Census Act gives the Secretary of Commerce substantial authority to determine how the census is conducted, but that discretion is not unlimited. Census methods must remain consistent with Congress’s requirement to conduct a decennial population census and with statutory language and regulations governing topics like residence, reporting, and confidentiality. Major design decisions must also be supported by evidence and a reasoned explanation, particularly where they depart from prior practice or create predictable differences in accuracy across communities.
This section examines how the constitutional requirement for an actual enumeration may apply to in-office enumeration, how Utah v. Evans both supports and limits Census’s legal argument, and which statutory and administrative-law questions deserve attention as the design develops. It also considers what evidence Census would need to defend the method, what circumstances could make a legal challenge more likely, and why uncertainty about a remedy could create serious operational and political consequences even before a court reaches a final decision.
The goal is not to predict that a court will invalidate the 2030 Census. Courts have historically given the political branches substantial latitude in census design. But that deference is strongest when Census can show that it made meaningful efforts to count people directly, tested its methods carefully, chose them because they improve accuracy, and applied them through transparent rules rather than political or arbitrary judgments. The farther the final design moves from those conditions, the greater the constitutional and statutory risk.
1. Applicable Precedent
The Constitution requires an “actual Enumeration” of the population every ten years, but it gives Congress broad authority to determine how that enumeration will be conducted. Article I says the census must occur “in such Manner as [Congress] shall by Law direct.” The Fourteenth Amendment further provides that representation is apportioned among the states according to their respective numbers, “counting the whole number of persons in each State.” Together, these provisions require a deliberate population count for apportionment while leaving substantial room to choose the methods used to produce it.
Congress has delegated much of that methodological authority to the Secretary of Commerce. Under the Census Act, the Secretary must conduct the decennial census “in such form and content as he may determine.” The Supreme Court has therefore generally reviewed census methods with considerable deference rather than requiring Census to use one particular approach.
In Wisconsin v. City of New York, the Court explained the broad constitutional standard. The case concerned the Secretary’s decision not to statistically adjust the 1990 Census to correct for an estimated undercount. The Court held that a census method is constitutionally permissible if it bears a reasonable relationship to accomplishing an actual enumeration, keeping in mind the census’s constitutional purpose of apportioning representation among the states. The Court also emphasized that the Constitution gives the federal government wide discretion over census design, including choices about how to balance different kinds of accuracy.
This is a deferential standard, but it is not an absence of legal limits. A census method must still constitute a genuine effort to determine the population rather than an arbitrary estimate. It must be consistent with the constitutional language and the goal of allocating representation according to population. Census must also remain within any additional limits Congress has imposed through the Census Act.
The most directly relevant precedent for in-office enumeration is the Supreme Court’s 2002 decision in Utah v. Evans. That case involved “hot-deck imputation,” a method Census used in 2000 when field visits failed to resolve whether a housing unit was occupied or how many people lived there. Census filled the remaining gap using information from a nearby housing unit of the same general type. The method added approximately 1.2 million people, or 0.4 percent of the national population, and its uneven effect across states changed the allocation of the final House seat between Utah and North Carolina.
Utah challenged the method on both statutory and constitutional grounds. It argued that imputation was a form of statistical sampling prohibited by 13 U.S.C. § 195 for determining the population used in congressional apportionment. It also argued that inferring the number of residents at an unresolved address was inconsistent with the requirement for an “actual Enumeration.”
The Supreme Court rejected both arguments. On the statutory question, the Court distinguished imputation from sampling. Sampling generally uses information collected from a selected portion of a population to estimate characteristics of a larger population. Hot-deck imputation instead filled a particular missing census record using information from a nearby unit. The Court concluded that the method was not “the statistical method known as sampling” and therefore did not violate the statutory prohibition.
On the constitutional question, the Court rejected the idea that an actual enumeration requires Census to obtain information through direct contact with every person. It explained that “enumeration” describes a counting process without prescribing every methodological detail. Census had also long relied on heads of household, neighbors, landlords, postal workers, and other proxies to determine how many people lived at a particular place. Inference was therefore not categorically outside the historical understanding of an enumeration.
The Court did not hold that Census may use any form or amount of inference it chooses. Instead, it emphasized the limited circumstances in which hot-deck imputation was used:
Census had made efforts to reach every household before turning to imputation.
The method involved inference for particular unresolved housing units rather than statistical sampling from part of the population.
Imputation affected only a very small share of the population.
The likely alternative was to count unresolved units as having no residents, which Census reasonably believed would be less accurate.
The limited and localized method presented little apparent opportunity for manipulation.
Under those circumstances, the Court held that the constitutional limits on census methodology had not been exceeded. It expressly declined to define precisely where those limits would lie in a different case.
These considerations are sometimes described as the Utah v. Evans “test,” but that term should be used cautiously. The Court did not establish a formal multi-part legal test in which every future census method must satisfy a fixed checklist. The factors are better understood as a framework for assessing why the particular use of imputation before the Court was permissible.
That framework offers some support for a limited use of administrative records in 2030. When Census has made meaningful efforts to obtain a direct response, cannot reach the household, and uses current person-specific records to avoid treating an occupied home as vacant, the method may resemble the narrow, accuracy-enhancing backstop upheld in Utah. Administrative records that directly identify particular people and addresses may also look less like statistical sampling than a method that estimates unobserved people from a representative group.
But the decision does not provide advance approval for the broader in-office enumeration system now under consideration. The comparison becomes weaker if Census relies on administrative data after only minimal outreach, uses models to construct or alter large numbers of households, accepts records with uncertain Census Day addresses, or applies the method to a substantial share of the population. The legal question would depend on the final design, its scale, its demonstrated accuracy, and the extent to which it remains part of an effort to enumerate particular people at particular places.
The distinction between inference and sampling will also require attention. Section 195 continues to prohibit the use of statistical sampling to determine state populations for congressional apportionment. Using an existing record that identifies a particular resident is not ordinarily the same as extrapolating from a sample. But a system that uses statistical models to infer the existence, location, or household characteristics of people not found in the underlying records could present a more complicated statutory question.
The central lesson from the precedent is therefore neither that administrative enumeration is clearly lawful nor that direct contact with every person is constitutionally required. Census has broad methodological discretion, including some authority to use inference. That discretion is most defensible when the method is a limited last resort, is tied to identifiable households, demonstrably improves accuracy, follows meaningful enumeration efforts, and is protected from arbitrary or political manipulation.
2. Differences Between Limited Imputation and Planned In-Office Enumeration
Census may argue that in-office enumeration is simply a modern version of the imputation method upheld in Utah v. Evans: when direct collection fails, the Bureau uses the best available information to avoid counting an occupied home as empty. That comparison is strongest when administrative records are used narrowly, after meaningful field efforts, to resolve a small number of otherwise unreachable households.
The comparison becomes less straightforward if in-office enumeration operates at a larger scale or plays a more central role in the overall census design. The Supreme Court evaluated a specific, limited method used in the 2000 Census. It did not decide whether Census could routinely replace direct responses and field interviews with administrative records, linked data, and predictive models.
Several potential differences deserve particular attention.
Scale. Hot-deck imputation affected approximately 0.4 percent of the population in 2000. The Court repeatedly described the method as limited and used only for the small number of cases that remained unresolved after census collection efforts.
The scale of in-office enumeration for 2030 has not yet been determined. The Operational Plan presents it as an integrated part of the census design, supported by a national Person Characteristic Frame and models that will evaluate large numbers of nonresponding households. If the method is ultimately used for only a small residual group, the comparison to Utah will be relatively strong. If it becomes a routine means of completing a substantial share of households, the Court’s emphasis on the limited reach of imputation will provide less reassurance.
Scale matters for more than the raw number of people affected. A method used widely can have systematic effects that would be difficult to detect in a small group of scattered cases. It can also influence field staffing, outreach planning, address development, and the overall willingness of Census to invest in direct collection.
Use before all reasonable collection efforts are exhausted. In Utah, Census used imputation after repeated attempts to obtain information through ordinary census operations. The method served as a last resort for cases the Bureau had been unable to resolve.
The 2030 plan requires at least one in-person visit before certain households are enumerated using administrative records, but it does not yet define what other efforts must occur or when further visits will be considered unnecessary. Models may help determine whether another contact attempt is likely to produce a response. In practice, administrative enumeration could therefore occur not only after Census has exhausted reasonable efforts, but after the Bureau predicts that additional efforts would not be efficient.
That distinction could be legally important. A court may view a method differently when it fills a gap that remains after meaningful enumeration efforts than when it helps determine how much effort the government will make in the first place. The more administrative records and response models are used to justify ending fieldwork early, the less the system resembles a limited backstop.
The source of the inferred information. Hot-deck imputation generally borrowed information from a nearby housing unit with similar characteristics. That method did not identify the actual residents of the unresolved unit, but it relied on information collected through the census itself and was designed to produce a more accurate estimate than treating the unit as vacant.
In-office enumeration will instead draw on administrative, commercial, and other supplemental records collected for different purposes and at different times. These sources may identify actual individuals, which in some respects makes the method more direct than hot-deck imputation. A current record that reliably connects a named person to a specific address may provide stronger evidence than borrowing a household size from a neighbor.
But administrative records introduce different uncertainties. They may contain an old address, omit a recent household member, reflect a mailing or benefits address rather than a usual residence, or describe only the people relevant to a particular program. Several sources may also repeat the same underlying information, creating apparent agreement without independent confirmation.
The legal comparison will therefore depend on what the records actually establish. Person-specific data may strengthen Census’s argument that it is enumerating identifiable residents rather than estimating from a sample. Older or incomplete records used to construct a probable household may look more like inference, particularly when the Bureau cannot validate the result through direct contact.
Models that help determine when fieldwork ends. In Utah, the contested method filled in cases that remained unresolved after the collection process. For 2030, statistical models may play an earlier and more influential role. They may evaluate the likelihood of response, the expected value of another visit, the quality of available records, and whether a case is ready for in-office completion.
Models will not merely supply a missing value. They may shape the sequence of enumeration itself by deciding which households receive continued field attention and which are moved to administrative processing. That creates a potential circularity: a model predicts that further outreach is unlikely to work, Census reduces the effort devoted to that household, and the resulting lack of direct response is then used to justify administrative enumeration.
The use of models does not automatically make the process unconstitutional or convert it into prohibited sampling. Models are already used throughout modern census operations. The relevant questions are what they predict, what decisions they control, whether their results have been validated, and whether they preserve a genuine effort to enumerate each household.
Use for characteristics as well as population count. The imputation challenged in Utah was relevant primarily because it affected the population totals used for apportionment. In-office enumeration may have a broader role. Census may use outside records not only to determine whether a unit is occupied and how many people live there, but also to assign or resolve age, sex, race and ethnicity, relationships, housing tenure, and other characteristics.
The constitutional requirements governing apportionment focus most directly on the population count. But the broader use of inferred or administrative characteristics raises related statutory and administrative-law questions. Census must be able to explain why the selected sources are fit for each use, how it handles missing or outdated information, and whether it is reasonable to treat a household as complete when only the population roster is reliable.
The inclusion of demographic characteristics may also increase the political and legal scrutiny the method receives. Race, ethnicity, citizenship, and related population measures are already central to disputes about representation, immigration, and demographic change. Critics may therefore portray the administrative assignment of these characteristics not simply as a technical method for completing missing records, but as a government decision that shapes how particular populations are defined, measured, and made visible in public data. Even where those claims rest on a misunderstanding of the method, they could make litigation more likely and place greater pressure on Census to demonstrate that its procedures are neutral, accurate, and firmly grounded in established standards.
The distinction could therefore matter in litigation over the design. A method may be defensible as the best way to avoid counting an occupied unit as having zero residents while being much harder to justify as a substitute for current, self-reported race, ethnicity, relationship, or housing information. The more politically contested the characteristic, the more important it will be for Census to preserve direct responses, document the source of assigned information, and show that administrative methods do not systematically alter the representation of particular communities.
Differential effects across communities. The imputation method reviewed in Utah was localized, and the Court noted its limited potential for manipulation. In-office enumeration could have broader and more uneven effects if the completeness and accuracy of administrative records vary across populations.
People with stable addresses and extensive connections to tax, employment, property, or benefits systems may be represented well. Young children, recent immigrants, people experiencing housing instability, and residents of informal or complex households may be less consistently represented. If those differences influence which households receive direct follow-up and the quality of the records ultimately accepted, in-office enumeration could affect communities differently even when the rules appear neutral.
Differential effects would not necessarily make the method unconstitutional. Census methods have never achieved identical accuracy for every population. But substantial and predictable disparities could weaken the Bureau’s claim that its design reasonably advances an accurate enumeration, especially if Census knew about those disparities and failed to adopt available safeguards.
The possibility of uneven effects also makes transparency especially important. A method applied through automated rules may appear resistant to political manipulation, but its design choices can still favor populations that are easiest to represent in existing data. Courts and the public would need information about source coverage, model performance, field-effort decisions, and error rates to evaluate whether the system functions as a neutral accuracy-enhancing tool.
None of these distinctions establishes that in-office enumeration will be unlawful. Some aspects of the proposed approach may be more defensible than the method upheld in Utah. Reliable person-level records could provide better evidence about an unresolved home than information borrowed from a nearby household. Administrative data may prevent clear omissions and allow Census to direct field resources toward cases where direct collection remains essential.
The legal risk will depend on the final balance. In-office enumeration will most closely resemble the method upheld in Utah if it is used after meaningful collection efforts, for a limited residual share of cases, with current and reliable person-specific data, and because testing shows it produces a more accurate result than the available alternatives. The comparison will be weaker if the method is used broadly, helps justify reduced field effort, relies heavily on uncertain or outdated records, or produces substantial disparities that national accuracy measures conceal.
Current commitment
Census plans to use administrative records and models as part of an integrated process for resolving nonresponding households after at least one in-person visit.
Open design question
The Bureau has not yet established the scale of in-office enumeration, the field efforts that must precede it, the characteristics it may supply, or the extent to which models will determine that additional outreach is no longer worthwhile.
Credible risk
The final design may differ from the method upheld in Utah v. Evans in several legally significant ways, including broader scale, earlier use, greater reliance on outside records and models, and more substantial differential effects across communities.
Why advocates should care
Utah v. Evans supports carefully limited inference used to improve an otherwise unresolved count. It should not be treated as blanket approval for any system Census describes as in-office enumeration. The strength of the precedent will depend on the operational details Census has not yet released.
3. Remedy and Institutional Risk
A successful challenge to in-office enumeration would not automatically require Census to repeat the entire decennial census. The appropriate remedy would depend on what the court found unlawful, how broadly the challenged method had been used, whether the affected records could be identified, and how far the census and apportionment processes had progressed.
Possible remedies could range from relatively contained corrections to major disruption. A court might prevent Census from using a particular model or decision rule, require the Bureau to revise its procedures, direct it to reconsider affected cases, or prohibit certain administratively generated results from being used in the apportionment or redistricting data. If the unlawful method had already affected final counts, Census might be required to revise the relevant records or recalculate population totals. But courts would also have to consider whether a proposed correction would improve the count or simply replace one known problem with another.
Federal law expressly anticipates litigation over statistical methods used in a decennial census. It allows an aggrieved person to seek declaratory, injunctive, and other appropriate relief against the use of a statistical method that violates the Constitution or federal law in determining population for congressional apportionment or redistricting. Whether every aspect of in-office enumeration would fall within the statute’s definition of a “statistical method” would itself depend on the design, but constitutional and administrative-law challenges could provide additional grounds for relief.
The timing of a challenge would be critical.
Before enumeration begins, a court could potentially block an unlawful method while Census still has time to change its systems, field procedures, or quality standards. Courts have previously reviewed major census methodology decisions before census operations were complete. Planned sampling for the 2000 Census, for example, was enjoined before implementation, and the Supreme Court resolved the statutory dispute in time for Census to proceed under a different design.
Early review could still be disruptive. Census systems are developed, integrated, and tested years in advance, and a late change could require new software, revised training, additional field staff, or increased funding. But a pre-enumeration ruling would generally leave the Bureau with more options than a decision issued after households had already been administratively enumerated.
During data collection, a court might prohibit further use of the challenged method or require Census to change the rules for unresolved cases. At that stage, however, the Bureau could face significant operational constraints. Field offices may already be reducing staff, scheduled visits may have ended in some areas, and the time remaining before statutory reporting deadlines may be short. Census might have to reopen cases while continuing to process the rest of the count.
After enumeration but before the final population totals are transmitted, Census might still be able to identify and reprocess affected records. The feasibility of that remedy would depend heavily on the Bureau’s data systems and documentation. Census would need to know which households were enumerated through the challenged method, what information came from each source, which models and thresholds were applied, and what alternative result would follow if those decisions were reversed.
This is one reason data provenance is not merely a matter of research transparency. Census must preserve an auditable record of how every administratively enumerated household was completed. Without that record, it may be impossible to correct an unlawful rule without either leaving the original result in place or discarding information for households that Census no longer has the capacity to contact.
After apportionment or redistricting data have been delivered, the consequences could extend well beyond the Census Bureau. The Secretary must report state population totals to the President within nine months of the census date. The President then transmits the apportionment and House-seat allocation to Congress. Census must also provide the states with the population tabulations used for redistricting within one year of Census Day.
Once those steps have occurred, states may have begun drawing congressional and legislative districts, local governments may be revising boundaries, and candidates and voters may be preparing for elections under the new maps. A correction that changes only a small number of individual records may have little practical effect. A correction that changes state totals, district populations, or the geographic distribution of particular communities could require apportionment calculations or redistricting work to be revisited.
The precise legal remedy at that point would be uncertain. A court could order revised calculations or other corrective action if it concluded that effective relief remained possible. In an earlier census case, a lower court ordered federal officials to recalculate an apportionment after excluding a disputed population, although the Supreme Court ultimately reversed the underlying legal ruling and the remedy never took effect. The case illustrates the kinds of relief litigants may seek without establishing that recalculation would be required in a future case.
Courts might also hesitate to impose a remedy that would cause greater disruption or produce a less accurate count. Removing administratively enumerated residents from the totals, for example, would not necessarily restore a lawful enumeration. It might instead count occupied households as empty. A court could remand the matter to Census to develop a lawful correction, but the Bureau would still have to identify an approach that was operationally possible, legally sufficient, and more accurate than the invalidated method.
The most serious scenario would be a ruling issued after enumeration that calls the validity of a substantial share of the count into question. If in-office enumeration has been used broadly, the problem may not be confined to a discrete set of easily corrected records. The challenged rules may have affected which households received additional visits, when fieldwork stopped, how conflicting responses were resolved, and which people and characteristics entered the final data.
In that situation, there may be no clean preexisting count to which Census can return. The direct-collection alternative was never completed because the model or administrative determination caused fieldwork to end. Enumerators may no longer be employed, temporary offices may be closed, and the opportunity to collect information close to Census Day may have passed. Attempting new enumeration months later would raise its own accuracy, cost, and legal problems.
Congress recognized this general danger when it created special procedures for challenges to census statistical methods. Its findings warned that meaningful relief could be impracticable after the enumeration had been conducted, and the law provides for expedited judicial consideration of covered claims. The point is especially relevant to in-office enumeration: legal review is far more useful before an untested or unlawful decision rule has been applied to millions of cases than after it has shaped the count in ways that cannot be fully reconstructed.
The institutional consequences would extend beyond any court-ordered correction. A late ruling could delay census products, create uncertainty about apportionment and redistricting, force emergency requests for funding, and divert staff from producing the data releases and quality assessments that normally follow the enumeration. States and communities could be left planning with provisional figures or litigating over whether revised data must be used.
It could also damage confidence in the Census Bureau. Administrative enumeration already asks the public to trust complex systems that operate largely outside public view. If Census cannot explain which records were affected, why particular households were closed, or how an unlawful rule can be corrected, the resulting controversy may reinforce concerns that the census is opaque, politically manipulable, or unreliable.
These risks do not mean Census should avoid administrative records altogether. They mean the Bureau should design the system with the possibility of review and correction in mind. It should publish consequential rules before they are operationally irreversible, test plausible alternatives, preserve the source and decision history for each case, and maintain contingency plans for households whose administrative enumeration is later found unreliable.
The safest legal and institutional course is to resolve these questions early. Courts, Congress, advocates, and outside experts should not first learn the scale and consequences of in-office enumeration after the field operation has ended and the final population totals are approaching certification.
Current commitment
Census intends to test in-office enumeration and develop quality measures and decision rules before the 2030 Census.
Open design question
Census has not explained how affected cases could be identified and corrected if a model, threshold, or administrative-enumeration method were later found unlawful or unreliable.
Credible risk
A legal ruling during or after enumeration could require Census to revise affected records while field capacity is declining and statutory apportionment and redistricting deadlines are approaching.
Worst-case scenario
A post-enumeration ruling calls a substantial and geographically uneven share of the count into question, but Census has no reliable direct-response alternative and cannot recreate the field operation. The resulting dispute could disrupt apportionment, redistricting, data releases, and confidence in the census without offering an obvious corrective method.
Why advocates should care
The risk is not limited to whether a particular method survives litigation. An opaque or legally vulnerable system could leave historically undercounted communities caught between an unreliable original count and a rushed remedy that does not restore the people or information the census failed to collect.
4. Documentation and Archiving
A data-driven census must leave a record of how each final result was produced. If Census completes households through administrative records, models, and automated processing, it must preserve enough information to reconstruct the path from the original sources to the final census record.
This is more than a general transparency concern. Documentation will be necessary to evaluate whether Census followed its own rules, measure how different methods performed, investigate possible disparities, and correct affected records if a method is later found unreliable or unlawful. Without a clear decision history, Census may know that a household was counted but be unable to explain why fieldwork ended, which information was accepted, or how a conflicting response was resolved.
At minimum, Census should preserve information showing how each household was enumerated. The record should identify whether the final result came from self-response, an enumerator interview, a proxy, administrative enumeration, imputation, or a combination of methods. Where different methods supplied different parts of the record, Census should preserve that distinction. A household roster may come from administrative records while race and ethnicity comes from a direct response and a missing relationship is later imputed. Labeling the entire case simply “administratively enumerated” or “complete” would conceal important differences in data quality.
Census should also retain the specific records used in each decision, including the source, date, and version of the information. It should be possible to determine whether a person-address link came from a current tax record, an older benefits record, a commercial source, or another agency’s derived dataset. Where several records supported the same conclusion, Census should document whether they were genuinely independent or originated from a common source.
The version of the underlying data matters. Administrative systems change over time, and the information available when Census first evaluates a household may differ from what is available later. Census should preserve the source vintage actually used rather than relying on a continually updated frame that may no longer reproduce the information before the decision-maker at the time.
The same principle applies to models and automated rules. Census should archive the exact model version, variables, thresholds, business rules, and source rankings used for each case. If a model is recalibrated during data collection, the Bureau should be able to identify which households were processed under the earlier and later versions.
Preserving only the model’s final recommendation would not be enough. Census should retain the relevant inputs, confidence measures, and output that led to the decision. A record stating that administrative data were “sufficient” would provide little basis for review unless it also showed whether the system had high confidence in occupancy, the population roster, the address, household relationships, and each characteristic.
Census must also document what efforts were made to obtain a direct response. The record should show the timing and method of mail, telephone, digital, and in-person contacts; whether a message reached anyone; whether an enumerator gained access to the correct unit; and whether language or accessibility assistance was available. A simple count of contact attempts would not reveal whether the household received a meaningful opportunity to respond.
For field visits, the documentation should distinguish among outcomes such as: no one home, refusal, inaccessible building, incorrect address, vacant unit, language barrier, and successful contact with someone who lacked knowledge of the household. These outcomes may all be coded as unsuccessful, but they have very different implications for whether additional fieldwork was reasonable.
The record should then explain why field collection was deemed complete. Census should preserve the rule that authorized closure, the evidence supporting it, and any exception or discretionary judgment applied by staff. If a model concluded that another visit was unlikely to succeed, the record should identify that recommendation and show how it affected the decision. If administrative data were accepted after the minimum number of visits, the documentation should state whether additional effort was considered and why it was rejected.
Conflict resolution requires similarly detailed records. When two self-responses, competing administrative rosters, or current and older information disagree, Census should retain each version rather than preserving only the winning value. The decision history should show:
which sources conflicted;
how each source was ranked;
whether the conflict triggered additional contact;
which rule selected the final result; and
whether any value was modeled, imputed, or overridden.
This is particularly important when Census changes information reported directly by a household. The Bureau should be able to identify how often direct responses were modified, which characteristics were affected, and what evidence justified the change.
Documentation should support analysis at both the individual-case and system-wide levels. Census needs case-level records to investigate a disputed household or reconstruct an affected set of cases. It also needs standardized data that allow researchers to determine how often each method was used, which rules produced the most closures, and whether particular populations or communities were disproportionately affected.
The Bureau should establish retention rules before collection begins. Important records should not be deleted when a temporary field system is shut down, overwritten when the PCF is updated, or lost when a contractor’s work ends. Model code, training data documentation, quality assessments, decision logs, source inventories, and operating procedures should be treated as part of the permanent record of the census design.
Census will need to balance documentation with confidentiality, cybersecurity, and data-minimization requirements. Preserving an auditable decision history does not mean that person-level administrative data should be released publicly. Some information may remain protected within secure systems or be available only to authorized reviewers. But confidentiality should not become a reason to preserve less information than Census itself needs to validate, defend, and, if necessary, correct the count.
Public documentation should be detailed enough for outside experts and affected communities to evaluate the system without exposing confidential records. Census can publish model descriptions, source inventories, decision rules, aggregate provenance statistics, validation results, and disaggregated error measures. Independent review may also require secure access arrangements for qualified researchers, inspectors, auditors, or courts.
Documentation is especially important because many decisions will occur inside systems that respondents cannot observe. A household may never know that an older record displaced its response, that a model ended additional visits, or that a second family’s questionnaire was classified as a duplicate. The Bureau’s internal record may be the only way to identify and assess those decisions later.
Good documentation will not eliminate legal or data-quality risk. It will, however, make the system more testable, more accountable, and more capable of correction. A census method that cannot be reconstructed cannot be meaningfully audited, and a result that cannot be traced to its sources may be difficult to defend as a reasoned and accurate enumeration.
Current commitment
Census plans to integrate data from multiple sources, use models and quality measures in operational decisions, and conduct near-real-time monitoring and review.
Open design question
Census has not publicly described the case-level provenance records, model archives, contact histories, or retention requirements that will allow those decisions to be reconstructed.
Credible risk
Continuously updated frames, changing model versions, contractor systems, and automated processing could leave Census with a final result but an incomplete record of how it was produced.
Why advocates should care
Without durable and detailed documentation, neither Census nor outside reviewers may be able to identify which households were affected by a flawed rule, measure differential impacts, resolve a legal challenge, or develop a reliable correction.
Where to look: The principal legal authorities are the Constitution’s Enumeration Clause; 13 U.S.C. §§ 141 and 195; and the Supreme Court’s decisions in Wisconsin v. City of New York, Utah v. Evans, Department of Commerce v. U.S. House of Representatives, which provides additional guidance on the statutory prohibition against sampling for apportionment. Franklin v. Massachusetts illustrates the legal and remedial issues that can arise when a census method affects state population totals.
For the planned 2030 design, see the 2030 Census Operational Plan, particularly section 3.2.2, “In-Field Enumeration,” pp. 30–32; section 3.2.3, “In-Office Enumeration,” pp. 32–34; section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48; and sections 3.2.13–3.2.14, pp. 48–52, on quality assurance and response processing. Census’s Statistical Quality Standards and federal records schedules provide the starting point for assessing whether the Bureau will preserve enough information to reconstruct and defend its decisions.
G. Privacy, Cybersecurity, and Misuse
The Person Characteristic Frame could create substantial value for the 2030 Census. Linking information about people, addresses, household composition, and demographic characteristics may help Census resolve difficult cases and improve statistical products. But concentrating those data also changes the nature of the privacy risk.
The concern is not limited to whether published tables reveal confidential information. Census must protect identifiable data while they are acquired, linked, stored, analyzed, and eventually archived or destroyed. It must also control who may use the information, for what purposes, and under whose authority.
The 2030 Census Operational Plan recognizes many of these responsibilities. It commits Census to encryption, multifactor authentication, access controls, continuous security monitoring, and protections throughout the data life cycle. It also says the Bureau will maintain required Privacy Impact Assessments and Systems of Records Notices and comply with Title 13, Title 26, the Privacy Act, and other applicable requirements.
These commitments are important, but they remain high-level. The plan does not yet explain who will have access to the PCF, how administrative sources will be separated or combined, which contractors may enter the systems, or how Census will respond to requests from political leadership or other agencies. Those decisions will determine whether the frame is protected as a narrowly governed statistical resource or becomes part of a broader government data infrastructure with less clearly defined boundaries.
Cybersecurity Risk
A detailed person-and-address frame would be an attractive target for cybercriminals, hostile governments, and other actors seeking personal information or the ability to disrupt the census. A successful attack could expose sensitive records. It could also corrupt person-address links, alter household information, interrupt collection, or undermine confidence in the final count.
Centralization can improve security by allowing Census to apply consistent protections and monitor a smaller number of systems. The Enterprise Data Lake, for example, is intended to provide centralized storage and processing, along with improved security and monitoring. But centralization also increases the consequences of a failure. A person who compromises a broadly connected platform may gain access to more data than would be available from a single program or isolated dataset.
The Operational Plan says Census will operate a Security Operations Center, conduct penetration testing, monitor cyberthreats, encrypt information, and continuously review permissions and privileged accounts. It also identifies ransomware, supply-chain attacks, and cloud threats as risks for which the program is preparing.
Advocates should nevertheless seek more information about how those protections will apply to the PCF and the systems surrounding it. Important questions include whether data will be separated according to sensitivity, whether identifying information will be available in readable form, and whether a compromise in one system could provide a path into other enterprise resources.
Security testing should also examine the integrity of the count, not only the possibility that information will be stolen. An attacker who changes addresses, inserts false residents, or disrupts the matching process could affect census accuracy without publicly releasing any data. Census should be able to detect unauthorized changes, restore earlier versions, and reconstruct who altered a record and when.
Insider Misuse and Access Controls
Not every misuse requires an outside attacker. Employees, contractors, system administrators, or other authorized users may already have some ability to enter the systems containing identifiable information. A person can misuse that access even if the underlying platform is technically secure.
Title 13 provides strong legal protections. It generally prohibits the use of information supplied under the Census Act for nonstatistical purposes and limits access to sworn personnel. Temporary staff from private organizations may assist Census only if they swear to observe Title 13’s confidentiality requirements, and wrongful disclosure can result in criminal penalties.
Those protections are essential, but an oath and a criminal penalty do not replace effective access controls. Census should follow an approach in which each person can reach only the information necessary for an approved task. Access should be limited in time, reviewed regularly, and removed promptly when an assignment ends.
Particularly sensitive actions may require additional safeguards. Census could require two-person approval to export identifiable data, alter source-ranking rules, or grant access to a new project. It should maintain durable logs showing who viewed, changed, copied, or transferred information. Monitoring should be designed to identify unusual searches, large downloads, repeated access to records unrelated to an employee’s work, and efforts to bypass normal review.
System maintenance also requires careful controls. A contractor responsible for backups or cloud performance may not need to see the contents of person-level records. Census should design systems so that technical administrators cannot automatically read the data they maintain. Where access to unencrypted information is unavoidable, the reason and duration should be documented.
Use by Political Leadership
The most obvious confidentiality protections focus on public disclosure or use of census information against an individual. A different concern is whether political officials could seek access to person-level data or attempt to direct how the frame is used.
Title 13 remains a significant legal barrier. Information protected by section 9 may be used only for the statistical purposes for which it was supplied and may not be made available to officials who lack authorized, sworn access. An executive order or agency instruction cannot override those statutory restrictions.
The current policy environment nevertheless makes the boundaries especially important. Executive Order 14243 directs agencies to expand the sharing and consolidation of unclassified government data for Administration priorities, while limiting that direction to actions consistent with existing law. The order does not eliminate Title 13, but it may generate new requests for access and new pressure to interpret data-sharing authority broadly.
Political influence may also occur without a direct demand for confidential records. Leadership can influence which sources Census acquires, which projects are approved, how long data are retained, and how broadly a “statistical purpose” is defined. Officials may also seek aggregate or derived information that approaches the boundary between statistical analysis and policy enforcement.
Census should establish procedures for handling requests from the Secretary of Commerce, the White House, and other agencies before a controversy arises. Requests involving identifiable or linked data should receive written legal and privacy review. The decision, governing authority, permitted purpose, and limits on access should be documented.
The Bureau should also protect the role of career privacy, security, legal, and statistical officials. A politically directed request should not be approved solely by the official making or sponsoring it. Inspector general and congressional oversight may be appropriate where senior officials seek unusual access or propose a use outside established Census programs.
Sharing or Repurposing Outside the Decennial Census
The PCF will not be created in complete isolation. The Operational Plan says it will draw from the Census Bureau’s broader Demographic Frame, while response data will be stored within enterprise systems such as the Enterprise Data Lake. These resources are designed to support work across Census programs, not only the decennial census.
Reusing data for legitimate statistical work can reduce collection costs and produce valuable research. It can also lead to gradual expansion beyond the purpose for which a dataset was originally assembled. A frame created to determine where people live for the 2030 Census might later be proposed for another survey, an experimental product, or a project sponsored by another agency.
Each new use should receive an independent review. The fact that Census lawfully possesses data does not mean every possible linkage or statistical project is necessary or appropriate. Census should ask whether the proposed use fits its statutory mission, whether it is consistent with the original data-use agreement, and whether the same purpose could be achieved with less identifiable information.
The Bureau should also avoid allowing a dataset to become permanent merely because future uses can be imagined. Retention periods should be tied to defined operational, research, and archival needs. Copies created for temporary projects should be tracked and destroyed when they are no longer necessary.
Repurposing becomes especially concerning if person-level information could flow back to the agencies that supplied it. Census has previously stated that administrative records received from another agency are not returned for enforcement use and that Census uses them for statistical purposes under Title 13. The PCF should not become an indirect mechanism through which another agency improves its records about particular people, verifies eligibility, or supports immigration, law-enforcement, or benefits decisions.
Cloud and Contractor Access
The 2030 Census will rely heavily on enterprise systems, cloud services, mobile devices, and contracts. The Operational Plan describes the Enterprise Data Lake as a cloud-based repository for Census data and says the program will use enterprise solutions and a cloud-first approach where possible. It also anticipates vendor support for mobile devices and other infrastructure.
Contracting does not remove Title 13 responsibilities. Contractor staff assisting Census must receive appropriate authorization and, where they have access to protected information, be sworn to comply with the law. Census’s statistical quality standards also apply privacy and confidentiality requirements to employees and Special Sworn Status individuals, including contractors working on covered activities.
Still, the use of contractors and cloud platforms expands the number of organizations, systems, and personnel that may support the data environment. Census should publish enough information to show:
which functions are operated by Census and which are contracted;
which contractor personnel can access identifiable data;
whether subcontractors have any system role;
how access to backups, logs, and test environments is controlled;
what happens when a contract or employee assignment ends; and
how breaches and suspected misuse must be reported.
Contract terms should require prompt incident reporting, preservation of audit records, secure deletion, and government access to information needed for an investigation. Census should also ensure that vendors cannot use system metadata, operational information, or respondent data to train commercial products or for any other independent purpose.
The Bureau’s move toward shared enterprise infrastructure may create efficiencies, but it also makes boundaries more important. A service used by several Census programs should not give every program access to every dataset. Technical integration should not be mistaken for legal permission.
Do the Same Protections Apply to Every Source and Use?
The PCF may combine information governed by several legal regimes. Title 13 protects census information and restricts its use to statistical purposes. Tax records carry additional protections under Title 26. Other administrative and commercial sources may be governed by the Privacy Act, program-specific statutes, contracts, or data-use agreements.
Census says that it acquires administrative records under Title 13 together with any additional authority identified by the provider, and that its agreements document the applicable privacy and stewardship requirements. The Operational Plan similarly recognizes that Title 13, Title 26, the Privacy Act, and the E-Government Act may apply to different parts of the data environment.
The result is a layered system of protection, not necessarily one identical rule for every source. Some laws may impose restrictions beyond Title 13. Agreements may limit the projects for which a source can be used, how long it can be retained, or which personnel can access it. Commercial records may carry different contractual limits from federal tax or benefits data.
Census should create a source-by-source legal and privacy inventory showing:
the authority under which each dataset was obtained;
the purposes for which it may be used;
the personnel and systems authorized to access it;
the applicable retention and destruction rules; and
whether restrictions continue to apply after the data are linked or transformed.
Linkage should not weaken the strongest protection attached to the source data. Nor should a derived characteristic lose protection because Census created it through a model rather than receiving it directly. If a person’s probable address or race and ethnicity is inferred from several protected records, the resulting value remains sensitive even though it did not appear in any one source.
Privacy Impact Assessments and Systems of Records Notices should describe the PCF and its surrounding systems in terms ordinary readers can understand. They should identify the categories of people and information included, routine uses, access groups, retention periods, contractors, and external sharing. The documents should be updated when new data sources or uses are added rather than treated as one-time approvals.
The larger concern is trust. People may be less willing to participate if they believe Census is building a permanent government file that can be accessed by political officials or reused against them. That harm can occur even without an actual breach. A system that is legally protected but poorly explained may still reduce response among communities with well-founded reasons to distrust government data collection.
Census should therefore treat privacy and security as part of census accuracy. Strong technical safeguards, clear legal limits, and visible accountability can help preserve participation. Unclear boundaries around a powerful person-level frame can undermine the direct response on which a fair census continues to depend.
Current commitment
Census says it will protect 2030 data throughout their life cycle through encryption, multifactor authentication, access controls, monitoring, privacy reviews, and compliance with Title 13 and other applicable laws.
Open design question
Census has not yet publicly described the detailed access structure for the PCF, the roles of cloud and contract personnel, the rules for secondary uses, or the process for responding to requests from political leadership and other agencies.
Credible risk
Centralized data and shared enterprise systems may expand the number of people and programs with technical or authorized access. Inadequate separation, inconsistent source restrictions, or weak oversight could permit misuse even without a conventional external breach.
Worst-case scenario
Identifiable census and administrative data are stolen, altered, disclosed, or repurposed for a nonstatistical government function. The immediate harm to affected people is compounded by a lasting loss of trust that reduces census participation and weakens other federal statistical collections.
Why advocates should care
The communities most vulnerable to misuse of government data may also be the least able to tolerate uncertainty about how their information will be handled. Census cannot rely on administrative records while treating the governance of those records as a technical issue outside public scrutiny.
Where to look: See the 2030 Census Operational Plan, particularly section 3.2.12, “Person Characteristic Frame Management,” pp. 46–48; section 4.2.1, “Census Engineering,” pp. 73–75; section 4.3.2, “Information Technology Infrastructure,” pp. 80–81; Appendix A’s discussion of the Demographic Frame and Enterprise Data Lake, pp. 84–86; and Appendix B, “Security, Privacy, and Confidentiality,” pp. 86–91. The 2030 Census IT Strategy and Roadmap and Acquisition Strategy provide additional context on enterprise systems and contracting.
For the governing protections, see 13 U.S.C. §§ 6, 9, 23, and 214; Census’s administrative-data agreements and Statistical Quality Standard B2; and the Bureau’s Privacy Impact Assessments and Systems of Records Notices. Advocates should watch for PCF-specific privacy documentation, access and retention rules, contractor requirements, and future GAO or inspector general reviews of the 2030 data environment.
H. Worst-Case Scenario
The most serious risk is not that any single model, administrative record, or legal decision fails on its own. It is that several weaknesses reinforce one another and become impossible to correct once the census has moved past field collection.
In this scenario, Census uses response-propensity models to identify households and communities it expects will be difficult to reach. Rather than directing additional language assistance, trusted outreach, or field resources toward those areas, the Bureau treats the predicted low response as evidence that further collection efforts are unlikely to succeed. Households receive limited in-person follow-up and are moved relatively quickly to in-office enumeration.
The administrative records used to complete those cases perform well for much of the population but less well for the communities receiving reduced field effort. They omit some residents, connect others to outdated addresses, and fail to capture doubled-up families, recent immigrants, young children, and people with unstable housing. Because the records identify plausible residents at many addresses, the cases appear complete even where the underlying household roster is wrong.
The resulting errors may be difficult to detect. National quality measures show high overall coverage because the method works well for the majority of households. The poorer results for particular communities are obscured by aggregate averages. Coverage Estimation may also miss some of the errors if the same people are absent or misplaced in both the census and the administrative sources used to evaluate it.
Census then applies administrative enumeration at a scale far beyond the limited imputation considered in Utah v. Evans. The method is no longer a narrow last resort used after extensive efforts to reach a small residual population. It has helped determine where fieldwork ends and has supplied the final records for a substantial and geographically uneven share of households.
The census proceeds to apportionment and the delivery of redistricting data. Only then do outside analysis, implausible local results, internal disclosures, or newly released quality studies reveal the possible scale and concentration of the errors. A state, local government, or other plaintiff challenges the method, arguing that Census did not make a sufficient effort to conduct an actual enumeration and relied too broadly on models and records that it knew performed unevenly.
At that point, the dispute would concern more than whether one technical rule was lawful. The challenged method may have influenced which households received additional visits, which communities lost field resources, and when hundreds of thousands or millions of cases were closed. There may be no unaffected set of direct responses that Census can substitute for the disputed results.
A successful legal challenge would not automatically require a new nationwide census, and the precise remedy would depend on the timing, scope, and nature of the violation. But the available options could all be unsatisfactory. Excluding the disputed records could treat occupied homes as empty. Reprocessing the same administrative data might reproduce the original errors. Conducting new fieldwork months after Census Day would be costly, difficult, and less able to determine where people lived on the relevant date.
By then, temporary field offices may have closed and much of the enumeration workforce may have been released. Apportionment may already have assigned seats among the states. States may be drawing or implementing congressional and legislative districts, and election deadlines may be approaching. Revising affected counts could disrupt those processes, while leaving the original results in place could mean allocating representation on the basis of a method a court has found unlawful.
This is the point at which a serious methodological failure could become a constitutional and institutional crisis. The government would be required to allocate political representation according to population but might lack a timely, lawful, and credible set of population figures with which to do so. Courts, Congress, the executive branch, and the states could disagree about which numbers should govern and what corrective authority each institution possesses.
The harm would extend beyond apportionment. Redistricting data, population estimates, federal surveys, program statistics, and planning decisions all depend on the decennial census. Uncertainty about the underlying count could spread through the federal statistical system and leave communities disputing whether later products reflect the same flawed records.
Historically undercounted communities could face the greatest harm at every stage. They may receive less field effort, experience more administrative-record errors, have those errors obscured by national quality measures, and then be asked to rely on a rushed remedy that cannot restore the information Census failed to collect directly. Political disputes about the validity of the census could further reduce trust and participation in future federal data collections.
This scenario is not inevitable. It depends on a series of choices that remain open: whether low predicted response leads to more assistance or less effort; whether administrative-data thresholds account for differential quality; whether direct responses receive priority; whether Census tests the method at realistic scale; and whether it preserves enough information to reconstruct and correct its decisions.
The purpose of identifying the worst case now is therefore not to predict failure. It is to show why the safeguards discussed throughout this section must be established before the design becomes operationally irreversible. A method that is carefully limited, transparently tested, and used only after meaningful collection efforts may improve the census and remain legally defensible. A system that uses administrative data to justify reduced effort in the communities those data represent least well could place both the accuracy and legitimacy of the 2030 Census at risk.
Worst-case scenario
Census reduces direct fieldwork in communities predicted to respond at low rates and uses incomplete or outdated administrative records to enumerate a substantial share of their residents. The resulting errors are concentrated but hidden by national quality measures. After apportionment or redistricting data are delivered, a court finds that the method exceeded the limits of an actual enumeration, but Census no longer has the field capacity, time, or reliable alternative data needed to produce a clean correction.
Why advocates should care
The danger is not simply that administrative records could produce errors. It is that the same design could reduce the opportunities to prevent those errors, obscure them during quality review, and leave no workable remedy once political representation has already been allocated.
I. Research Agenda
Many of the most consequential questions about in-office enumeration cannot be answered from the current Operational Plan. The Bureau has described the broad direction of the design, but the scale of administrative enumeration, the rules governing its use, and its effects across communities remain unsettled.
Research should begin early enough to influence those decisions. It should not be limited to determining whether administrative records improve national population totals or reduce costs. Census must also assess who is represented poorly, which kinds of information are unreliable, and how its operational rules affect the likelihood that historically undercounted communities receive a direct and accurate enumeration.
The research should be transparent, reproducible, and designed to test realistic operational conditions. Studies based only on households that successfully responded to the census may overstate administrative-record quality because the people most difficult to find in outside data may also be less likely to appear in the comparison sample. Results should therefore identify the limits of the available benchmarks and examine differential performance rather than relying primarily on national averages.
At minimum, the research agenda should include the following priorities.
1. A Full Constitutional and Statutory Analysis of In-Office Enumeration
Census should publish a detailed legal analysis of the final in-office enumeration design. That analysis should address the Constitution’s requirement for an “actual Enumeration,” the prohibition against using statistical sampling for apportionment, and the ways the proposed method resembles or differs from the limited imputation upheld in Utah v. Evans.
The analysis should examine the complete system rather than administrative records in isolation. Relevant questions include how much field effort precedes in-office enumeration, whether models help determine when that effort ends, how many households may be affected, and whether administrative data are used only for population counts or also for demographic and household characteristics.
Census should also address possible remedies before the method is deployed. The Bureau should explain whether affected cases can be identified, reconstructed, and reprocessed if a court later rejects a model, threshold, or decision rule. An operational design that cannot be corrected without recreating fieldwork presents a different level of legal and institutional risk from one that preserves a viable alternative.
Independent constitutional and statutory analysis would also be valuable. Congress, legal scholars, census experts, and civil rights organizations should evaluate the design before its use becomes operationally irreversible rather than waiting for litigation after the count.
2. Independent Evaluation of Administrative-Record Coverage by Community
Census should evaluate whether the Person Characteristic Frame identifies people and current addresses equally well across communities. National coverage rates are not sufficient. The research should measure omissions and weak linkages by race and ethnicity, age, nativity, housing stability, income, geography, and other factors associated with historical undercounts.
Where possible, results should include detailed racial and ethnic groups rather than only the minimum federal categories. A system may perform reasonably well for a broad category while performing poorly for a smaller community included within it. Research should also examine rural and tribal areas, places with nonstandard addresses, and neighborhoods experiencing high mobility or housing instability.
Independent evaluation is particularly important because Census is both developing the frame and deciding whether it is accurate enough for operational use. Outside researchers should be able to review the methodology, test alternative measures, and examine whether the Bureau’s thresholds conceal meaningful disparities.
Where confidentiality prevents public release of person-level data, Census should provide secure research access or publish sufficiently detailed results to permit meaningful outside assessment. A general assurance that administrative records have high national coverage would not establish that they are suitable for enumerating every community.
3. Evaluation of Characteristics, Not Only Population Counts
Research should evaluate the completeness and accuracy of each characteristic Census may obtain from administrative records. Correctly identifying the number of people at an address does not establish that the records accurately describe their race and ethnicity, relationships, sex, age, or housing tenure.
This evaluation should reflect the questions and standards Census intends to use in 2030. In particular, race and ethnicity research should assess whether older administrative records can support the 2024 SPD 15 categories, including the combined question, the MENA category, and detailed racial and ethnic identities. Testing should distinguish between a characteristic reported directly by a person and one reconstructed, bridged, modeled, or copied from an older source.
The Bureau should publish separate quality measures for each characteristic and explain what happens when the household roster is reliable but one or more characteristics are not. Research should test whether additional contact improves those data and whether accepting incomplete characteristics creates unequal effects for particular populations or household types.
4. Research on Address Mobility and Household Composition
Census needs more evidence about how administrative systems represent people who move frequently or live in complex households. Studies should examine how quickly source records reflect a move, how often several datasets repeat the same outdated address, and how accurately the frame identifies a person’s Census Day residence.
Research should include shared-custody arrangements, college students, people entering or leaving institutions, people experiencing homelessness, and residents temporarily staying with another household. It should also examine the geographic precision of the records, including whether they identify the correct apartment or housing unit rather than only a parcel, building, mailing address, or prior residence.
Household-composition research should test whether the PCF can distinguish a single family from roommates, doubled-up households, multigenerational families, unmarried partners, same-sex couples, and informal caregiving arrangements. Census should make clear whether its measures of “household composition” evaluate only the roster or also the relationships among residents.
These studies should reflect real living arrangements rather than limiting validation to households that are easy to represent in tax, property, benefits, or commercial records.
5. Comparison of Direct Responses, Proxies, and Administrative Enumeration
The relevant question is not whether administrative records are perfect. It is whether they produce a better result than the realistic alternatives available for a particular unresolved household.
Census should compare administrative enumeration with direct household responses, enumerator interviews, proxy responses, vacancy assumptions, and imputation. The comparison should assess population coverage, address placement, household composition, and characteristics separately.
Research should also identify the circumstances in which each method performs best. Administrative records may be more accurate than a neighbor’s estimate for a household with several strong and current records. A direct interview may be substantially better for a recently moved or doubled-up household. A proxy may know that a unit is occupied without knowing the residents’ identities or characteristics.
The results should inform operational rules rather than merely document average performance. Census should use the evidence to identify when administrative enumeration is an appropriate backstop, when further field contact is warranted, and when no available method meets an acceptable standard.
6. Simulations of Sufficiency Rules and Differential Undercounts
Before Census selects final sufficiency thresholds, it should simulate how competing rules would affect fieldwork, administrative enumeration, and population accuracy. Small changes in a confidence threshold could move large numbers of households from continued collection to in-office completion.
The simulations should compare scenarios such as:
requiring one in-person visit or several varied visits;
requiring agreement across different numbers and types of records;
applying separate thresholds for occupancy, person coverage, address, and characteristics;
giving current direct responses priority over older administrative information; and
using low response predictions to increase assistance or to reduce field effort.
For each scenario, Census should estimate both the national result and the effects across historically undercounted groups and small geographies. A rule that slightly improves overall cost or accuracy may still be unacceptable if it substantially increases omissions or placement errors for particular communities.
The simulations should also examine interacting errors. A person omitted from the administrative roster may also be absent from the comparison data used to measure coverage. A household assigned a high confidence score may have a correct population count but inaccurate race and ethnicity or relationships. Models should account for these dependencies rather than treating every error as independent.
The purpose of this research is not to identify a single risk-free threshold. No enumeration method will resolve every uncertain household perfectly. The goal is to make the tradeoffs visible and ensure that Census does not select rules based primarily on cost savings, national averages, or operational convenience.
Research results should be released before final operational decisions are made and in enough detail for independent replication. Census should also explain how each major finding changes the design. Testing has limited value if known disparities are documented but do not lead to stronger safeguards, additional outreach, or more cautious use of administrative enumeration.
A credible research program would give advocates, Congress, courts, and Census itself a clearer basis for judging the method. It would also help distinguish a carefully tested backstop that improves difficult cases from a broad substitution for direct collection whose weaknesses become visible only after the count is complete.
Where to look: The Census Bureau’s 2030 Census Research Project Explorer identifies several projects directly relevant to this research agenda. In Enhancement Area 1, Administrative Data Enumeration Applications compares administrative data with household and proxy responses and examines when outside records might replace additional field attempts. Targeted Quality Improvement considers coverage discrepancies, item nonresponse, and the use of administrative records to substitute or edit characteristics. Tailored Contact Strategies examines how administrative-data quality and predicted response propensity may affect contact methods and the number of attempts households receive.
Enhancement Area 5 includes the projects most directly focused on the Person Characteristic Frame. Administrative Data Quality and Usage examines coverage across populations, demographic groups, and geographies and compares administrative data with self-responses and household interviews. Decision Rules for Effort Allocation Leveraging Information from Administrative Data Sources calls for simulations of rules governing the division of cases between field and in-office enumeration. Research on the Undercount of Young Children Using Administrative Data, Improving Within-Household Coverage Using Administrative Data, and Person and Address Matching Research address omissions, household-roster changes, address placement, and record linkage.
These projects provide a useful starting point, but they do not fully answer the questions identified above. Advocates should look for the resulting reports and test findings, particularly disaggregated measures of coverage and characteristic quality, comparisons among direct responses, proxies, and administrative enumeration, and simulations showing how alternative sufficiency rules affect historically undercounted communities. The Explorer does not currently identify a project devoted to a full constitutional and statutory analysis of in-office enumeration, which remains an important gap.
J. Advocacy Priorities
Administrative records could help Census resolve difficult cases and avoid treating occupied homes as empty. The goal of advocacy should not be to prohibit all administrative enumeration. It should be to ensure that the method remains a carefully tested backstop rather than becoming a broad substitute for direct participation.
Many of the most important decisions remain open. Census has not yet published the thresholds for administrative-data sufficiency, the rules governing field effort, or the scale at which in-office enumeration may be used. Advocates therefore have an opportunity to seek safeguards before those choices are embedded in final systems and operating procedures.
1. Preserve Meaningful In-Person Follow-Up
Census should retain a strong commitment to direct collection, especially for households whose circumstances are not well represented in administrative systems. One in-person visit should be treated as a minimum starting point, not automatic proof that reasonable enumeration efforts have been exhausted.
Advocates should seek field standards that account for whether visits occur at varied and reasonable times, whether an enumerator reaches the correct unit, and whether language or accessibility barriers prevent meaningful contact. A failed attempt caused by a locked building, an incorrect address, or the absence of an appropriate-language enumerator should not count as evidence that the household is unwilling to respond or impossible to reach.
Models predicting low response should generally lead to more appropriate assistance rather than less effort. Census should use those predictions to improve the contact strategy through language support, better timing, trusted outreach, or additional field attention. It should not use them primarily to identify households that can be moved quickly to administrative enumeration.
2. Give Direct Responses Presumptive Priority
A timely response from someone knowledgeable about the household should ordinarily receive priority over administrative information collected earlier for another purpose. Administrative records can identify possible omissions or inconsistencies, but they should generally trigger review rather than silently overwrite a direct answer.
This presumption should be especially strong for characteristics based on self-identification or current household circumstances. Race and ethnicity, relationships, household membership, and housing tenure may be recorded incompletely or under outdated definitions in administrative systems. A current census response should not be displaced merely because several older records repeat a different value.
Presumptive priority does not mean Census must accept every response without question. Duplicate submissions, apparent fraud, and genuine residence conflicts will still require resolution. But Census should define the specific evidence needed to modify a direct response and preserve a record showing what changed, why it changed, and which source supplied the final value.
3. Publish Decision Rules Before Production
The rules governing in-office enumeration should be public before they are used in the 2030 Census. Advocates should seek the release of minimum field-effort requirements, administrative-data thresholds, source-ranking rules, model specifications, and procedures for resolving conflicting information.
Publication must occur early enough for meaningful review. Releasing a general description after systems have been finalized would not allow outside experts or affected communities to identify problems while changes remain feasible. Census should publish the rules before the 2028 Dress Rehearsal where possible, explain how they will be tested, and release revised versions before production.
The Bureau should also describe the operational consequence of each rule. It is not enough to know that a model produces a response-propensity score or that a source receives a quality rating. The public needs to know whether a particular score results in more assistance, fewer visits, administrative enumeration, or additional review.
Any material changes during data collection should also be documented. Census should not quietly alter sufficiency thresholds or model rules to meet cost, staffing, or schedule pressures without recording the change and assessing its effect on accuracy and differential undercounts.
4. Cap or Otherwise Constrain Administrative Enumeration
The method should have clear limits. A numerical cap is one possible safeguard, but it is not the only one. Census could also constrain administrative enumeration by requiring stronger evidence for particular uses, prohibiting it in cases with unresolved conflicts, or limiting it to households that have received a defined level of field effort.
A cap would help preserve the method’s character as a last resort and make the comparison to Utah v. Evans more credible. But a national cap alone could still permit concentrated use in particular communities. Any limit should therefore consider both the total scale and the geographic or demographic distribution of administrative enumeration.
Census should establish circumstances in which administrative enumeration is not permitted. These might include conflicting household rosters, uncertain Census Day addresses, evidence of additional residents, or administrative information that does not identify the correct housing unit. The Bureau should also consider stronger restrictions where testing shows that records perform poorly for a particular population or household type.
Separate limits may be appropriate for different purposes. Administrative records might be sufficiently reliable to determine that a unit is occupied while remaining inadequate for assigning race and ethnicity, household relationships, or tenure. Authorization to use the records for the population count should not automatically authorize their use for every characteristic.
5. Report Use by Geography and Population Group
Census should report how often in-office enumeration is used and where it is concentrated. A national total would not reveal whether the method was applied disproportionately in particular neighborhoods, rural areas, tribal communities, or places with high shares of historically undercounted residents.
Reporting should distinguish among major pathways. Census should identify households completed through direct self-response, enumerator interviews, proxies, administrative records, and imputation. Where different parts of a record come from different sources, the Bureau should provide information about characteristic-level provenance rather than assigning the entire household to one broad category.
Results should be disaggregated by geography and population characteristics to the extent confidentiality permits. Census should report not only how frequently administrative enumeration was used, but also the error rates, missing characteristics, and subsequent corrections associated with the method.
These reports should be released during testing as well as after the census. Waiting until all operations are complete would prevent the findings from shaping fieldwork and make it harder to correct emerging disparities.
6. Validate Characteristics Separately From Population Coverage
Census should not treat an accurate population total as proof that the full household record is reliable. Each characteristic supplied or altered through administrative data should receive separate validation.
Research should evaluate race and ethnicity, relationships, age, sex, and housing tenure against current direct responses wherever possible. The analysis should measure missing information, broad or outdated categories, and systematic differences across communities. It should also distinguish between values copied directly from a source and values constructed through matching, bridging, modeling, or imputation.
Characteristic quality standards should affect operations. If Census has high confidence in the number of residents but weak information about their characteristics, it should consider additional contact rather than simply closing the case and filling the remaining fields later.
Census should also make clear which characteristics are essential to declaring an administrative record sufficient. A household should not appear fully complete in internal or public reporting when the roster is supported by strong evidence but much of the demographic information is missing or inferred.
7. Require Independent Legal and Scientific Review
The final design should receive independent review before it is used at scale. Scientific review should examine the quality of the Person Characteristic Frame, model performance, sufficiency thresholds, and differential effects. Legal review should assess whether the design remains consistent with the requirement for an actual enumeration, the statutory prohibition on sampling for apportionment, and other applicable census law.
Reviewers should have access to more than high-level summaries. They need sufficient documentation about data sources, model design, validation results, and operational consequences to evaluate the system meaningfully. Confidential information can remain protected through secure access arrangements, but confidentiality should not prevent rigorous outside assessment.
The review process should include expertise in census operations, record linkage, civil rights, privacy, and constitutional law. It should also incorporate input from communities that may be poorly represented in administrative data, including those with experience of housing instability, language barriers, and complex household arrangements.
Census should respond publicly to major findings and explain which recommendations it accepts, rejects, or incorporates into further testing. Independent review will have little value if it occurs only after the operational design is fixed or if identified disparities do not lead to changes.
Taken together, these safeguards would not eliminate every error or legal risk. They would create a clearer boundary around administrative enumeration and make it more likely that the method improves difficult cases without reducing the government’s commitment to direct participation.
The central advocacy principle should be straightforward: administrative records should be used to strengthen the effort to count people, not to justify making less of an effort to reach them. Census should bear the burden of demonstrating that each expansion of in-office enumeration improves accuracy, protects historically undercounted communities, and remains consistent with the constitutional obligation to conduct an actual enumeration.