The 2018 census citizenship memo held on its data analysis and failed on how the request began

On March 26, 2018, the Commerce Secretary directed that a citizenship question be added to the 2020 census. The Supreme Court later found the decision reasonably explained on the evidence but its sole stated reason seemingly contrived. Agencies need both records to hold.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

On March 26, 2018, the Commerce Secretary directed that a citizenship question be added to the 2020 census. The Supreme Court later found the decision reasonably explained on the evidence but its sole stated reason seemingly contrived. Agencies need both records to hold.

  1. The week On Monday, March 26, 2018, the Secretary of Commerce signed a memorandum to the Under Secretary for Economic Affairs, who oversees the Census Bureau.
  2. The request: where the demand for data came from The memo presented the decision as a response.
  3. The alternatives: four ways to get the data The second record is the comparison of alternatives.
  4. The evidence: accuracy, response and cost The third record is the evidence on the quality consequences of each option.
  5. The record: what a reviewer could see The fourth record is the administrative record itself: the materials the decision maker considered, filed with a court when the decision is challenged.
  6. How this applies to federal, state and local agencies Federal agencies face Administrative Procedure Act review of informal actions, as the Census case showed.
  7. The strongest objections One objection is that the census question was a politically charged decision unlike ordinary data work.
  8. With hindsight The memo’s decision did not survive to the 2020 census questionnaire.

The week

On Monday, March 26, 2018, the Secretary of Commerce signed a memorandum to the Under Secretary for Economic Affairs, who oversees the Census Bureau. Its subject was the reinstatement of a citizenship question on the 2020 decennial census questionnaire. The memorandum directed that the question be asked.

The memo began with a request. It said that on December 12, 2017, the Department of Justice had asked the Census Bureau to reinstate a citizenship question. The stated purpose was to provide citizenship voting age population data at the census block level, which the memo said were not available from government survey data. It said the Justice Department and the courts use such data to determine violations of Section 2 of the Voting Rights Act.

The memo described the review that followed. It said the department had begun a review led by the Census Bureau, covering legal, program and policy considerations. It said the Secretary had met Census Bureau leadership several times, that staff had reviewed over 50 incoming letters and that he had spoken with over 24 interested parties. It noted that, by law, the list of census questions was due to Congress no later than March 31, 2018.

The memo set out the alternatives the Bureau had analyzed. Option A kept the status quo and relied on the American Community Survey, which has asked about citizenship since 2005 and, the memo said, is sent each year to a sample of approximately 2.6 percent of the population. Option B placed the ACS citizenship question on the decennial census. Option C added no question and instead gave the Justice Department citizenship data built from federal administrative records that the Bureau could access for statistical purposes.

The memo explained why each fell short in the Secretary’s view. It rejected Option A because the ACS sample could not supply complete block-level counts. It said the Bureau and stakeholders feared Option B would lower response rates among noncitizens, but that neither could document a material decline. It called Option C “a potentially appealing solution.” Then it noted that the Bureau could match 88.6 percent of the population to administrative records in the 2010 census, which would leave citizenship for more than 10 percent of the population to be imputed.

The Secretary asked the Bureau for a fourth alternative, Option D. It combined the census question with enhanced administrative records. The memo said that, in the Secretary’s judgment, Option D would give the Justice Department the most complete and accurate data. It directed the Bureau “to obtain as many additional Federal and state administrative records as possible.”

The memo cited Bureau figures on accuracy and cost. It said Bureau analysis found that between 28 and 34 percent of citizenship self-responses, for people whom administrative records showed to be noncitizens, were inaccurate. It reported a rough Bureau estimate that up to 630,000 additional households might need nonresponse follow-up if the question were added. It described that as a one-half percent increase in follow-up, within the 3 percent increase that the lifecycle cost estimate given to Congress had assumed.

In the same month, the Census Bureau issued its report to Congress, Questions Planned for the 2020 Census and American Community Survey. The report described the 2020 census as a short-form-only count asking about age, sex, race, Hispanic origin, relationship, tenure and citizenship. Its citizenship page said the question had been asked since 1820, in 1820, 1830, 1870 and 1890 to the present, and that the statistics were essential for enforcing the Voting Rights Act.

The report also restated the Bureau’s confidentiality promise. It said Title 13 of the US Code protects the confidentiality of all collected information and that violating it is a crime. It listed the Bureau’s privacy principles, beginning with necessity: before asking any question, the Bureau determines whether the information is truly necessary.

A census question can look like a matter for statisticians and lawyers. We read the March 26, 2018 memorandum as a data governance document that any agency could learn from. It is a full public account of how an agency weighed a new data collection: who asked for it, which alternatives were considered, what evidence on accuracy and burden existed, and what was decided. It was then tested in court against the agency’s full record.

The test produced a split result. The Supreme Court held in June 2019 that the Secretary’s decision was supported by the evidence before him and was reasonable and reasonably explained. It also held that the sole stated reason, enforcement of the Voting Rights Act, seemed to have been contrived, and it affirmed the district court’s remand to the agency.

Our argument is that the memo succeeded on its analysis and failed on its account of the request. For an agency planning any significant data collection, the two are separate records that must both hold. We take the decision in four parts, each a distinct record an agency produces. Together they make up the decision file:

  • The request: who asked for the data, why, and how the request came to be made.
  • The alternatives: which ways of obtaining the data were considered, and why one was chosen.
  • The evidence: what was known about accuracy, response and cost.
  • The record: what the agency kept, and what a reviewing court could see.

After the four parts, we look at how the four records apply to federal, state and local agencies, at objections, and at what a program director could check.

The request: where the demand for data came from

The memo presented the decision as a response. In its telling, the Justice Department asked on December 12, 2017, for block-level citizenship data to enforce the Voting Rights Act, and Commerce reviewed and granted the request. Every later section of the memo follows from that framing.

What the Court found in the fuller record

The Supreme Court’s opinion describes how the record grew. In June 2018, the government filed the department’s administrative record, including the Justice Department’s letter. Shortly afterward, at the Justice Department’s urging, it supplemented the record with a new memo from the Secretary. That memo stated he had begun considering a citizenship question in early 2017 and had asked whether the Justice Department would formally request it.

The district court then ordered the record completed. According to the Court’s syllabus, the parties stipulated to materials confirming that the Secretary and his staff began exploring the question shortly after his 2017 confirmation, tried to elicit requests from other agencies and eventually persuaded the Justice Department to make one. The Court’s opinion said the materials indicated Commerce “went to great lengths to elicit the request” from the Justice Department or any other willing agency.

Why the origin mattered legally

The Court did not hold that an agency must disclose every motive. It noted that a typical case may involve both stated and unstated reasons. What it found here was different: “the VRA enforcement rationale—the sole stated reason—seems to have been contrived.” It said the explanation was “incongruent with what the record reveals about the agency’s priorities and decisionmaking process.”

The Court tied this to the purpose of the reasoned explanation requirement. In its words, that requirement “is meant to ensure that agencies offer genuine justifications for important decisions, reasons that can be scrutinized by courts and the interested public.” It concluded that accepting contrived reasons “would defeat the purpose of the enterprise.”

What the request record is

Our reading is that a data collection decision rests on a request record distinct from its analysis. The request record answers who first proposed the collection, what they said it was for and how the formal request was produced. When an agency’s own leaders originate an idea and then seek a partner agency to request it, that sequence is part of the record. Presenting the partner’s request as the origin creates the gap the Court found.

For agencies that run data programs on behalf of others, the point is practical. A state data office building a dataset for a sister agency, or a federal statistical unit responding to a program office, receives requests regularly. The request record should show the actual sequence: the first discussion, the drafts, and the final request.

When the requester is also the user

The request in this case had a named user: the Justice Department, for Voting Rights Act enforcement. The Census Bureau’s March 2018 report to Congress described the same use. Its citizenship page said the statistics were essential for enforcing the Voting Rights Act and its protections against voting discrimination. The purpose appeared consistently across the memo and the report.

Consistency across documents is not the same as an accurate origin. Every document can repeat a stated purpose while the history behind it differs. The Court looked past the consistent statements to the materials showing how the request was produced. For agencies, the implication is that a purpose statement copied across several reports carries no more weight than the history it summarizes.

An agency that builds a dataset for another agency’s use should also record whether the user asked for that specific method. The Justice Department letter, as the memo describes it, asked for a question on the census. The memo’s Option D answered with a question plus records. Whether the user would have accepted records alone is the kind of fact an alternatives analysis can capture by asking.

Operational detail

The request record has an owner: the program or data office that receives the request. It should log the date and source of first contact, the stated purpose, any drafts exchanged and who signed the formal request. If the receiving agency suggested the request, the log should say so. That is not an admission of anything improper. It is an accurate history, which is what a reviewing body will compare the stated reason against.

The trade-off is candor against convenience. A clean request letter is tidier than a log of informal approaches. The Census case shows that a tidy letter which does not match the history can cost more than an untidy record that does.

The alternatives: four ways to get the data

The second record is the comparison of alternatives. Here the memo was detailed, and the Court found it adequate.

What the options were

The memo described three options the Bureau first analyzed and a fourth the Secretary requested. Option A relied on the existing ACS sample, with possible modeling. Option B asked every household the ACS question. Option C used administrative records alone. Option D combined the census question with expanded administrative records.

Each option is a different data architecture. Option A is sampling plus statistical modeling. Option B is a universal survey item. Option C is record linkage across agencies. Option D is a survey item used alongside linked records, with the records used to measure and correct survey error.

Modeling as an option

Option A bears directly on current analytics work. The memo said that keeping the ACS approach would require the Bureau to deploy a team of experts to develop model-based methods for block-level estimates. It said the Bureau did not assert and could not confirm that such modeling was possible with sufficient accuracy at the census block level.

That is the question any agency faces when it considers estimating a characteristic instead of collecting it. A statistical or machine learning model can produce values for every small area or every person. Whether those values are accurate enough for the decision they will support is an empirical question. The memo treated the absence of a validated model as a reason to reject the option.

For an agency weighing a model against a new collection, the alternatives record should include the validation evidence: how the model performs against known values, at what level of detail, and for which groups. If no such evidence exists, the record should say so and say what it would take to produce it. A model rejected for lack of validation can be reconsidered when validation exists.

Record linkage as a data product

Option C deserves attention from any agency data leader, because it describes a method that public data programs use. Records held by several agencies are linked to the people counted, and a characteristic is drawn from the records rather than asked. The memo said the Bureau had used administrative records to improve accuracy and reduce cost since the early twentieth century.

Linkage has its own quality profile. The match rate determines coverage: the memo cited 88.6 percent of the population matched to credible records in 2010. Those not matched need another source or a modeled value. The people least likely to match may differ systematically from those who do, which makes the unmatched share a bias question as well as a volume question.

The memo’s answer was to add a survey item to cover the gaps. An alternative would have been to improve linkage and model the remainder. Both are legitimate designs. The choice between them depends on how much error each introduces, for which groups, and how each affects the willingness of people to respond to anything at all. An alternatives analysis that states those trade-offs for each group is stronger than one that states them in aggregate.

Imputation as a disclosed method

Any design that leaves gaps relies on imputation. The memo framed Option D as reducing the need to impute citizenship for millions of people. It also said the Bureau would use the combined data to establish the ratio of citizen to noncitizen responses to impute where necessary. Imputation was therefore part of every option.

For agencies publishing data that include imputed values, the method is part of the product. Users need to know which values were observed, which were linked and which were modeled. A decision record that describes the imputation approach for each option lets a reviewer judge whether the comparison between options was fair.

How the choice was reasoned

The memo explained why it set aside each of the first three. It said the Bureau could not confirm that block-level modeling under Option A would be accurate enough. It said Option C would leave more than 10 percent of the population, which it put at some 25 million voting age people, needing imputed citizenship. It said Option D would let the Bureau compare survey answers against records for the whole population and estimate inaccurate response rates for citizens and noncitizens alike.

The Court accepted that reasoning as reasonable. Its syllabus says the Secretary examined the Bureau’s analysis of ways to collect better citizenship data. It says he explained why he chose both a question and administrative records to fill gaps. It said he weighed the value of more complete data against the uncertain risk of lower response and explained why he thought the benefits outweighed the risk.

The Bureau’s preference and the statute

The opinion also records disagreement inside the department. The opinion of the Court said the Secretary considered but rejected the Bureau’s recommendation to use administrative records alone. A separate opinion added that the Secretary had found that option potentially appealing but concluded it had shortcomings. The statute added weight to that choice. Section 6(c) of the Census Act requires the Secretary, to the maximum extent possible, to use administrative records instead of conducting direct inquiries.

The Court held that, assuming section 6(c) applied, the Secretary complied with it for essentially the same reasons his decision was not arbitrary and capricious. In his judgment, records alone would not provide the more complete and accurate data sought. For agencies, the reach of that holding is narrow. A statutory preference for administrative data can be satisfied by a reasoned explanation of why records alone fall short.

Operational detail

An alternatives record needs each option described in terms a reviewer can test: what data it produces, at what geographic or population level, with what known error, at what cost and burden. It needs the technical staff’s recommendation and the decision maker’s reasons for agreeing or departing from it. A departure is permitted. An unexplained one is the gap a reviewer will look for.

For a state or local agency, the same discipline applies to choices between a new form field and a data-sharing agreement. A benefits agency deciding whether to ask applicants a new question or match against another agency’s records faces the Census choice on a smaller scale. Its alternatives record will be read by auditors, legislators or courts, depending on the stakes.

The evidence: accuracy, response and cost

The third record is the evidence on the quality consequences of each option. The memo’s account of that evidence was candid about its limits.

What was known about response

The memo acknowledged a widely held concern that a citizenship question could lower response among noncitizens. It said that neither the Bureau nor stakeholders could document a material decline. It cited Bureau comparisons of 2000 census short and long forms, which showed self-response for noncitizen households falling 3.3 percent more than for citizen households. It said the Bureau could not isolate how much of that came from the citizenship question.

The memo also compared item nonresponse. It said ACS nonresponse to the citizenship question in 2013–2016 ranged from 6.0 to 6.3 percent for non-Hispanic whites, 12.0 to 12.6 percent for non-Hispanic blacks and 11.6 to 12.3 percent for Hispanics. It said these were comparable to nonresponse on other ACS questions, listing figures for the number of times married, education, utility costs, weeks worked, wages and property insurance.

What was known about accuracy

The memo’s accuracy evidence pointed in an interesting direction. It reported Bureau analysis that 28 to 34 percent of citizenship self-responses were inaccurate for people whom administrative records showed to be noncitizens. Survey answers on this item therefore had measurable error that records could detect. It also reported that only 88.6 percent of the population could be matched to credible records in 2010.

Those two figures frame a classic data quality trade-off. Survey answers cover everyone who responds but contain error. Linked records are more accurate for those matched but do not cover everyone. The memo chose to combine them. The disagreement was about whether adding the question would reduce response enough to outweigh what it added.

Testing a question versus testing its setting

The memo called the citizenship question well tested because the ACS had asked it since 2005. It also recorded why that testing might not carry over. It said response rates vary between decennial censuses and sample surveys. It reported ACS self-response 3.1 percentage points below the 2010 census, which the Bureau attributed to the census’s greater outreach and follow-up.

The memo noted the difference in form length as well. The 2000 census short form had eight questions; the 2000 long form had over 50 and took over 30 minutes on average; the ACS had over 45 questions. A question tested on a long survey may behave differently on a short one, where it is more prominent. The memo acknowledged the comparison was challenging.

That candor is a model for any agency adding an item that was tested elsewhere. Prior testing establishes that respondents understand the question. It does not establish how the question will affect response in a new instrument, for a new population or under new public attention. The evidence record should say which of those the prior testing covered.

Unknowns stated as unknowns

The memo also stated what nobody knew. It said there was no information available to determine how many people would not respond because a citizenship question was added, and that no one had identified a mechanism for making that determination. It said that while the belief that such residents existed might be true, no evidence quantified it.

An agency can treat an unknown in two ways. It can treat the absence of evidence as reassurance, or it can treat it as a risk to manage, through testing, monitoring or contingency plans. The memo weighed the uncertainty and judged the benefits greater. The Court held that weighing reasonable. A different decision maker could reasonably have weighed it differently. The evidence record’s job is to make that weighing visible, which the memo did.

What was known about cost

The memo said the Bureau considered the cost of preparing and adding the question minimal, because the ACS question had already undergone cognitive research and testing. It reported the Bureau’s rough estimate of up to 630,000 additional households needing follow-up, a figure it described as a one-half percent increase in follow-up operations. It said the department’s revised lifecycle cost estimate already assumed follow-up could rise 3 percent for several reasons.

How the Court treated the evidence

The Court held that the decision was supported by the evidence before the Secretary. It found the decision reasonable and reasonably explained, particularly given the long history of the question on the census. For standing purposes, it also accepted the district court’s finding that reinstating the question would likely lead noncitizen households to respond at lower rates. It noted states had shown that an undercount of as little as 2 percent would cost them federal funds.

Those holdings sit together without contradiction. A decision can be reasonable on uncertain evidence even if one of the risks it weighed later appears likely. Our reading is that the evidence record did its job: it showed what was known, what was not, and how the decision maker weighed the uncertainty.

Operational detail

For any agency adding a data element, the evidence record should include measured error rates where available, item nonresponse, coverage of any linked records and an honest statement of what is unknown. It should say whether the item was tested and how. The Census memo leaned on the ACS question’s prior testing to call the item well tested. An agency adapting a question from another survey should document that lineage the same way.

The record: what a reviewer could see

The fourth record is the administrative record itself: the materials the decision maker considered, filed with a court when the decision is challenged. It is where the first three records meet.

How the record was challenged

The Court’s syllabus describes the sequence. The government filed the record in June 2018. It then supplemented it with the Secretary’s new memo on the origin of the request. The challengers argued the record was incomplete, and the district court agreed and compelled its completion. It also authorized discovery outside the record, including a deposition of the Secretary, which the Supreme Court stayed.

The Court explained why it looked beyond the record. Ordinarily, a court is limited to the agency’s contemporaneous explanation in light of the existing record. The Court called it rare to review a record this extensive for informal agency action, “and it should be.” Having done so, it said it could not ignore the disconnect between decision and explanation.

Why the record mattered more than the memo

The memo was a strong document on alternatives and evidence. The record contained the history of the request. A reviewer comparing the two found they did not match. The analysis did not save the decision, because the remand turned on the explanation, not the analysis.

The Court was explicit that it did not hold the decision substantively invalid. It said agencies must pursue their goals reasonably and that reasoned decision making under the Administrative Procedure Act calls for an explanation for agency action. What the agency provided, it said, “was more of a distraction.”

Supplements change the story

The supplemental memo filed in June 2018 is a lesson in its own right about records. In the Court’s account, it stated that the Secretary had begun considering a citizenship question in early 2017 and had asked whether the Justice Department would request it. That changed the account of origin given in the March memo.

The challengers then argued the record was incomplete, and the district court agreed. A supplement that corrects the record is better than no correction. It also signals to a reviewer that the original file did not tell the whole story. The right time to record the actual sequence of a request is when the decision is made, before litigation makes every addition look tactical.

What the GAO decision added

A separate review, by the Government Accountability Office, addressed the memo’s legal form. On February 7, 2019, GAO concluded that the memo was not a rule under the Congressional Review Act. It said the memo was direction from a supervisor to a subordinate in connection with the statutory process for informing Congress of census questions. In our reading, that meant the memo did not have to be submitted to Congress as a rule under that act.

For agencies, the GAO decision is a reminder that the form of a data decision determines which review it faces. A memo to a subordinate avoided rule review. It did not avoid judicial review under the Administrative Procedure Act, which the Supreme Court held applied to the Secretary’s decision.

Operational detail

An agency cannot know in advance which decisions will be challenged. It can keep every significant data collection decision in a form that would survive being filed. That means the request log, the alternatives analysis, the evidence summary and the decision memo kept together, with drafts and correspondence retained under the records schedule. The test is whether the decision memo would read as an accurate summary of the file, not a separate account.

A practical step is a review before signature. Someone other than the drafter, such as counsel or a records officer, reads the decision memo against the file and flags any statement the file does not support. The reviewer asks, in particular, whether the stated purpose matches the request record. The step adds days to a decision. It is cheaper than a remand, a supplemental memo or a deposition fight, which are what the Census case produced instead.

The file should also note who saw what. A decision memo that relies on technical staff analysis should identify that analysis and its date. If the decision maker heard from outside parties, as the memo says the Secretary did, the file should list those contacts. A reviewer can then see the inputs the decision actually had.

How this applies to federal, state and local agencies

Federal statistical and program agencies

Federal agencies face Administrative Procedure Act review of informal actions, as the Census case showed. Statistical agencies also operate under confidentiality laws such as Title 13, which the Bureau’s March 2018 report restated. Program agencies that share administrative records with statistical agencies make data decisions of their own, with records to match.

The later executive order on citizenship data directed agencies to give Commerce access to specific record sets. Each such data-sharing decision has its own request, alternatives, evidence and record. The four-part file applies on both sides of the exchange.

State agencies

State agencies operate under their own administrative procedure laws, which vary. In our view, the principle that a stated reason should match the record is likely to matter under any of them, though the standards differ. Counsel should confirm how the state’s law treats informal data decisions. The four-part file is a sound practice under any standard.

Agencies that buy data or analytics

Agencies can face the census choice in procurement form. A vendor offers a dataset, a linked record service or a model that estimates a characteristic the agency does not collect. Each is one of the memo’s options, purchased instead of built. The four records apply in the same way.

The request record shows which program asked for the capability and why. The alternatives record compares the purchased option with collecting the data directly or using existing records. The evidence record holds the vendor’s accuracy and coverage claims and the agency’s own testing of them. The decision record explains the choice in terms that match the rest of the file.

A purchased model adds one difficulty: the agency may not be able to see how it works. An evidence record that rests only on the vendor’s description is thinner than one built on the agency’s own validation. Contracts can require the access needed to test a product before an agency relies on it.

Local governments

Counties and cities make data collection decisions through forms, applications and surveys. A local decision to add a question about immigration status, income or household composition can draw scrutiny from residents, advocates and courts. The same records apply on a smaller scale.

Trust as a data quality input

The confidentiality promise in the Bureau’s March 2018 report connects to the response question. The report said Title 13 protects all collected information and that violating it is a crime. The memo recorded stakeholder concerns that some communities distrusted government collection and feared census responses could be used for law enforcement.

For any agency collecting sensitive data, trust affects response, and response affects quality. A legal confidentiality guarantee is necessary and may not be sufficient. How an agency explains a new question, who it says asked for it and why, and how consistent that account is with its records all bear on whether people answer. Our view is that the request record is part of the trust a data collection depends on.

The strongest objections

That the case was political and unrepeatable

One objection is that the census question was a politically charged decision unlike ordinary data work. That is true of the stakes. The structure is general. Any data collection with consequences for funding, representation or eligibility can be challenged, and a challenger will compare the stated reason to the record.

That the analysis was sound, so the outcome was unfair

A second objection is that the Court found the analysis reasonable and still remanded. Four justices did not join the part of the opinion that affirmed the remand, according to the opinion’s list of who joined which parts. We draw no conclusion about which view was right. For agencies, the practical lesson is that the remand happened, and it happened because of the request record.

That every agency has mixed motives

A third objection is that every agency decision has stated and unstated reasons. The Court accepted that. Its concern was a sole stated reason that the record suggested was contrived. An agency can avoid that problem by stating its actual reasons, including the ones it would rather not lead with.

That the remand changed nothing

A further objection is that the question was dropped for practical reasons, so the remand had no effect on data. Executive Order 13880 said the Court’s ruling made it impossible as a practical matter to add the question given the census timetable. The remand therefore did affect the outcome, through timing. For agencies with fixed deadlines, a remand can decide a question even where the agency could, in principle, try again.

That data quality should decide such questions alone

A fourth objection, from the opposite side, is that technical staff should decide data collections on quality grounds. The Census Act gives the decision to the Secretary, and the Court held that the Enumeration Clause permits a citizenship question. Political leaders may decide. The record must show how and why they decided.

What a program director could check

  • Does the file for each significant data collection show who first proposed it and how any formal request was produced?
  • Does it describe each alternative in testable terms, with the technical recommendation and the reasons for any departure?
  • Does it state measured error, nonresponse and coverage where known, and say plainly what is unknown?
  • Would the decision memo read as an accurate summary of the whole file if both were filed together?
  • Is the file retained under the records schedule with drafts and correspondence, so it can be produced complete?

With hindsight

The memo’s decision did not survive to the 2020 census questionnaire. On February 7, 2019, the Government Accountability Office concluded the memo was not a rule under the Congressional Review Act. The decisive review came from the courts.

The Supreme Court heard the case on April 23, 2019, and decided it on June 27, 2019, in Department of Commerce v. New York. It held that at least some challengers had standing. It held that the Enumeration Clause permits a citizenship question and that the Secretary’s decision was reviewable under the Administrative Procedure Act. It held that the decision was supported by the evidence and reasonably explained, and that the Secretary had not violated sections 6(c) and 141(f) of the Census Act. It then held that the Voting Rights Act rationale, the sole stated reason, seemed to have been contrived, and affirmed the remand to the agency.

On July 11, 2019, the President issued Executive Order 13880. It said the Court’s ruling had made it impossible, as a practical matter, to include a citizenship question on the 2020 questionnaire. It said the Attorney General and the Secretary of Commerce had reported no practical mechanism for doing so, given the logistics and timing of the census and delays from litigation. The order directed all agencies to give Commerce the maximum assistance permitted by law in determining the number of citizens and noncitizens, including access to administrative records.

The order also restated the position at the time of the 2018 decision. It said the Census Bureau had then determined that administrative records it could access would let it determine citizenship status for approximately 90 percent of the population. It named record sets from the Department of Homeland Security and other agencies that Commerce had sought. In our reading, the policy moved to the memo’s Option C, enhanced by the records the memo had told the Bureau to pursue.

On January 20, 2021, Executive Order 13986 revoked Executive Order 13880 and a July 21, 2020 presidential memorandum on excluding people without lawful immigration status from the apportionment base. It stated a policy that apportionment be based on the total number of persons residing in the states, without regard to immigration status. It directed the Secretary to ensure the population information given to the President and the states was accurate.

Executive Order 13880 also gave its own historical account. It said every decennial census from 1820 to 2000, except 1840, asked at least some respondents about citizenship status or place of birth.

On January 20, 2025, Executive Order 14148 revoked Executive Order 13986 among other orders. We cite no later census decision here, and we do not assert what any future census will ask.

What held up from March 2018 was the memo’s structure: a request, four alternatives, evidence on accuracy and cost, and a decision. The Court found that structure adequate on the analysis. What did not hold up was the account of where the request began. For agencies making data decisions in 2026, the case remains a public example of an analysis the Court found reasonable, undone by its account of the request.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • 2020 census
  • citizenship question
  • administrative records
  • Administrative Procedure Act
  • data quality