The following Essay is part of a mini-symposium on the 2020 Census and its broader implications for our nation and democracy. Both essays rely on the perspectives of scholar-practitioners to interrogate the design of the enumeration process and outline its potential ramifications for the distribution of federal resources and political power in an evolving United States.

*  *  *

In early 2018, the federal government announced that it would ask every person in the country about their citizenship status on the 2020 Census. Controversy immediately followed. The Constitution makes the decennial census the federal government’s very first express responsibility; it drove existential questions about represen­tation and funding in 1790 and has become no less important in the centuries since. Many observers, including several former directors of the Census Bureau, believed that the 2018 decision to add a question on citizen­ship, in the existing political climate, would put the enumeration itself in jeopardy.

This piece is the first to interrogate the rationale of that decision and its consequences for the theoretical construction of American repre­sentative democracy and for the tangible distribution of clout and cash. Part I explores the decision’s likely impact, documenting the discarding of the Census’s normal evaluation process and presenting risks for the accuracy of the enumeration that the Census Bureau itself had previously deemed intolerable in less volatile climates. Part II turns to the proffered basis for this upheaval but finds the public rationale to be lacking adequate foundation. Part III investigates alternatives that might better explain—though not justify—the decision. It finds that the Census’s technocratic statistical infrastructure may have been co-opted as a weaponized instrument in a long-simmering battle over the reallocation of political representation.


    1. Past Efforts to Collect Citizenship Information
    2. Collecting Citizenship Information During the Enumeration in a Climate of Fear
    3. The Absence of Testing
    4. The Consequences of an Inadequate Enumeration
    1. Citizenship Data and VRA Enforcement
      1. The Use of Citizenship Data.
      2.  “Precision,” but Inaccuracy
    2. Evidence of Pretext
    1. Statistics on Immigration
    2. Interstate Competition for Representation and Power
    3. Intrastate Competition for Representation and Power
    4. Redistricting Population Base



On March 26, 2018, Secretary of Commerce Wilbur Ross determined that the Census Bureau would ask every person in the country about their citizenship status. 1 Letter from Wilbur Ross, Sec’y of Commerce, U.S. Dep’t of Commerce, to Karen Dunn Kelley, Under Sec’y for Econ. Affairs, U.S. Dep’t of Commerce (Mar. 26, 2018) [hereinafter Ross, Decision Letter], []. Specifically, the Census Bureau would add a question concerning citizenship status to the basic data sought in the decennial enumeration. The decision was immediately and enormously controversial; 2 Predictably, litigation followed. See, e.g., California v. Ross, 358 F. Supp. 3d 965, (N.D. Cal. 2019); New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.). This piece investigates the potential impact of, and rationales for, collecting citizenship information on the decennial enumeration, not the statutory or constitutional legality of that decision or the process that produced it. Still, some of the litigation, ongoing as this piece went to press, offers further confirmation of the factual research herein; that support is noted below. many believed that the decision could put the enumeration itself in jeopardy. 3 The Census Bureau reports that 137,695 comments pertinent to the citizenship question were received in response to a Department of Commerce notice about the change. See U.S. Census Bureau, USBC-2018-0005-79003, Summary of Comments Received on the 2020 Census Federal Register Notice 3 (2018) [hereinafter Comment Summary], []; Proposed Information Collection: Comment Request: 2020 Census, 83 Fed. Reg. 26,643, 26,646, 26,652–53 (June 8, 2018). The Bureau notes that 99.1 percent of those comments opposed the inclusion of a citizenship question on the decennial enumeration, including uniform opposition from leading civil rights organizations. See Comment Summary, supra, at 3; 140+ Civil Rights Groups Call on Commerce Department to Strike Unnecessary Citizenship Question, The Leadership Conference on Civil & Human Rights (Aug. 1, 2018), [] [hereinafter Leadership Conference]; Nick Brown, U.S. Census Citizenship Question Panned by Scientists, Civil Rights Groups, Reuters (Aug. 9, 2018), [].

It is impossible to overstate the constitutional significance of the decennial enumeration. The requirement to conduct an “actual Enumeration” of the population, once per decade, is embedded in the sixth sentence of the Constitution. 4 See U.S. Const. art. I, § 2, cl. 3. Because the enumeration drives the allocation of representation and taxation—establishing the basic ground rules for the composition of, and funding for, the governing entity that the Constitution created—the enumeration is the very first act that the Constitution prescribes as an express responsibility of the new federal government. It precedes the election of representatives and the selection of legislative process; 5 See, e.g., id. art. I, §§ 4–5; id. art. II, § 1. it precedes the power to coin and collect and borrow money; 6 See, e.g., id. art. I, §§ 7–10. it precedes the responsibility to establish defense forces and to declare and wage war; 7 See, e.g., id. art. I, § 8, cl. 11–16; id. art. I, § 10; id. art. II, § 2. it precedes the conduct of foreign relations and the establishment of a judiciary. 8 See, e.g., id. art. I, § 8, cl. 3–4, 9–10; id. art. I, § 10; id. art. II, §§ 2–3; id. art. III, §§ 1–2.

The terms of the initial enumeration, of course, reflected the deep stain of the compromise over slavery. The enumeration embraced a count of all of the people fully recognized as people and then added for representational purposes a fraction of each slave held as property. 9 See id. art. I, § 2, cl. 3. As Janai Nelson recounts in detail, this was not the last instance in which choices related to the Census created or perpetuated deep injustice. Still, with time and bloodshed, the constitutional command has since become a clear mandate to count each and every individual in the country. 10 Id. amend. XIV, § 2; Indian Citizenship Act of 1924, Pub. L. No. 68-175, 43 Stat. 253 (codified at 8 U.S.C. § 1401(b)). This enumeration remains the basis for apportioning seats in the House of Representatives—and, consequently, in the Electoral College as well. 11 U.S. Const. amend. XIV, § 2; id. art. II, § 1, cl. 2; 2 U.S.C. § 2a(a) (2012).

Congress has authorized the Census Bureau to seek, in several different instruments, a great deal of additional exceedingly valuable information en route to producing a periodic statistical American snap­shot. 12 See, e.g., 13 U.S.C. § 41 et seq. (2012). Yet whatever other goals the Census Bureau may pursue, its single indefeasible obligation is to ensure, to the best of its ability, the absolute and inviolate integrity of the constitutionally mandated decennial count.

The content of the basic enumeration questionnaire has varied some­what over time, but as of 2010, it was distilled to ten basic questions, designed to take ten minutes to complete. 13 See U.S. Census Bureau, U.S. Dep’t of Commerce, 2010 Questionnaire (2010) [hereinafter 2010 Questionnaire], []. The questionnaire attempts to capture the identity and location of each individual—that’s the mini­mum information required to meet the constitutional command. It also attempts to capture the sex, age (and date of birth), race, 14 The Census Bureau’s approach to determining racial categories has never been free of controversy. See generally, e.g., Off. Mgmt. & Budget, Standards for Maintaining, Collecting, and Preserving Federal Data on Race and Ethnicity, 81 Fed. Reg. 67,398 (Sept. 30, 2016); Off. Mgmt. & Budget, Revisions to the Standards for the Classification of Federal Data on Race and Ethnicity, 62 Fed. Reg. 58,782 (Oct. 30, 1997); U.S. Census Bureau, 2015 National Content Test: Race and Ethnicity Analysis Report (2017),
reports/2015nct-race-ethnicity-analysis.pdf []; Patrick F. Linehan, Thinking Outside of the Box: The Multiracial Category and Its Implications for Race Identity Development, 44 How. L.J. 43 (2000); Naomi Zack, American Mixed Race: The U.S. 2000 Census and Related Issues, 17 Harv. BlackLetter L.J. 33 (2001); Patricia Palacios Paredes, Note, Latinos and the Census: Responding to the Race Question, 74 Geo. Wash. L. Rev. 146 (2005); Measuring Race and Ethnicity Across the Decades: 1790–2010, U.S. Census Bureau,
2010.html [].
Hispanic or Latino origin, household or familial relationship of each individual, and whether the respondent owns, rents, or otherwise occupies the resi­dence. 15 U.S. Census Bureau, U.S. Dep’t of Commerce, Questions Planned for the 2020 Census and American Community Survey: Federal Legislative and Program Uses 5–19 (2018) [hereinafter Questions Planned for 2020], []; U.S. Census Bureau, U.S. Dep’t of Commerce, U.S. Census 2000 Short Form Questionnaire (2000), []. And now, the questionnaire will ask about the citizenship of each individual as well. 16 The precise question asks whether each individual was born in a state or a territory, born abroad to a U.S. citizen parent or parents, naturalized (and the year of naturalization), or is not a citizen. Questions Planned for 2020, supra note 15, at 7. The way that the question is phrased has also been controversial. First, the question distinguishes individuals who were “born in the United States” from those “born in Puerto Rico, Guam, the U.S. Virgin Islands, or Northern Marianas.” Id. Immigration law, at least, establishes that those born in Puerto Rico, Guam, the U.S. Virgin Islands, and the Northern Mariana Islands are born in the United States. See 8 U.S.C. § 1101(a)(38) (2012) (defining “the term ‘United States’ . . . when used in a geographical sense”). Second, the question appears to contain no category applicable to U.S. “nationals” born in American Samoa. See, e.g., Mark Joseph Stern, The Census’ New Citizenship Question Excludes an Entire Category of Americans, Slate (Mar. 30, 2018), []; see also Doug Mack (@douglasmack), Twitter (Mar. 29, 2018), [] (“There’s also the one (living) guy who was born on Palmyra Atoll, a U.S. possession. He can’t check any of these boxes.”).

This Essay attempts to parse the legal and political rationale and consequences of that decision. Part I examines the likely impact of the decision, demonstrating that Secretary Ross’s eleventh-hour decision presents risks for the decennial enumeration that the Census Bureau itself had previously deemed intolerable, and which have recently only intensified. Part II interrogates the rationale for the decision, finding that it is essentially unjustified by the public rationale offered. Part III probes alternative rationales beyond that pretext that might better explain why the decision was made. Several of these alternatives concern the allocation of political power in the decennial apportionment and redistricting cycle to come, including a long-simmering fight over the very nature of political representation.

I. The Consequences of the Citizenship Question

In the current political climate, asking a question about the citizen­ship of every individual in the country is no mere request for infor­mation. Those who work with communities skeptical about the role of the federal executive branch fear that the question will prove explosive. 17 See supra note 3. Secretary Ross made the determination to ask this question despite his own admission that the career staff of “[t]he Census Bureau and many stakeholders expressed concern that . . . [the decision] would negatively impact the response rate” for the enumeration, 18 Ross, Decision Letter, supra note 1, at 3; see also New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 530 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.). and despite the absence of any opportunity to test that impact before implementing the change. He did so despite the direct warning of six former Directors of the Census Bureau, whose collective twenty-five years of service as Director spanned eight presidential administrations (of both major political parties), that adding the question so late in the planning process would put the accuracy of the enumeration “at grave risk.” 19 Letter from Vincent P. Barabba et al., Former Dirs., U.S. Census Bureau, to Wilbur L. Ross, Sec’y of Commerce, U.S. Dep’t of Commerce 1–2 (Jan. 26, 2018), (on file with the Columbia Law Review). The Titanic was launched with less hubris and more preparation.

A. Past Efforts to Collect Citizenship Information

It is true that the Census Bureau has collected information on the citizenship of the American public for many years. But the context for how this information was collected is vitally important. The last time a question about citizenship was asked on the basic decennial enumeration questionnaire was 1950, 20 Bureau of the Census, U.S. Dep’t of Commerce, 1950 Census of Population and Housing (1950), []; see also New York, 351 F. Supp. 3d at 525. In 1950, enumerators asked about the naturalization status of foreign-born individuals. See Thomas P. Wolf & Brianna Cea, A Critical History of the United States Census and Citizenship Questions, 108 Geo. L.J. Online 1, 14 (2019). when  both  the  demographic  composition  of  the  country  and  the  political  climate  were  very  different. 21 See, e.g., Gary D. Sandefur et al., An Overview of Racial and Ethnic Demographic Trends, in 1 America Becoming: Racial Trends and Their Consequences 40, 43–44 (Neil J. Smelser et al. eds., 2001); infra text accompanying notes 37–38.

Beyond a special enumeration in New York and Puerto Rico, the 1960 Census did not ask about citizenship at all. 22 See, e.g., Bureau of the Census, U.S. Dep’t of Commerce, 1960 Censuses of Population and Housing: Procedural History 194 (1966); Bureau of the Census, U.S. Dep’t of Commerce, Notice of Required Information for the 1960 Census of Population and Housing (1960) [hereinafter 1960 Questionnaire], []. From 1970 on, ques­tions about citizenship were asked only in the context of surveys distinct from the enumeration and delivered to a smaller representative sample of the population. 23 From 1970 to 2000, questions about citizenship were asked in the context of a “long form” questionnaire; the question was asked of five percent of U.S. households in 1970, about nineteen percent of households in 1980, about sixteen percent of households in 1990, and about seventeen percent of households in 2000. See Bureau of the Census, U.S. Dep’t of Commerce, 1970 Census Questionnaire 6 (1970) [hereinafter 1970 Long Form], []; Bureau of the Census, U.S. Dep’t of Commerce, United States Census 2000, at 4, 12 (2000) [hereinafter 2000 Long Form], []; Nat’l Research Council, Modernizing the U.S. Census 188, 191–92 (Barry Edmonston & Charles Schultze eds., 1995); Bureau of the Census, U.S. Dep’t of Commerce, United States Census 2000, History: 2000 Overview, [] (last visited Mar. 10, 2019). In 2010, questions about citizenship were asked in the context of the American Community Survey, a rolling survey conducted each month that reports aggregate results in one-year and five-year increments. See U.S. Census Bureau, U.S. Dep’t of Commerce, American Community Survey: Information Guide 6, 12 (2017), []. The five-year compilation of the survey is designed to represent about 12.5% of households. See id. at 6. And when respondents to these surveys were asked about their citizenship, that question was lodged amidst a lengthy battery of other detailed and personal questions. 24 In 1960, this survey contained at least thirty-three questions (many with subparts) in addition to the basic questions in the decennial enumeration; in 1970, at least fifty-eight additional questions; in 1980, at least fifty-four additional questions; in 1990, at least fifty-eight additional questions; in 2000, at least forty-six additional questions; and in 2010, at least sixty-nine additional questions. See 1960 Questionnaire, supra note 22; 1970 Long Form, supra note 23, at 2–8; Bureau of the Census, U.S. Dep’t of Commerce, 1980 Census Long Form (1980), []; 2000 Long Form, supra note 23; U.S. Census Bureau, U.S. Dep’t of Commerce, The American Community Survey 2–11 (2010), []; Bureau of the Census, U.S. Dep’t of Commerce, Your Guide for the 1990 U.S. Census Form E-9 to E-15 (1990), [].

The length and detail of the American Community Survey (ACS), now the home for this supplemental inquiry, is significant. Not because the information is excessive—to the contrary, understanding who and where Americans are, what we do, and how we live is essential for the effectiveness and efficiency of public policy and private enterprise alike. Instead, the scope is significant because it mutes the impact of any one question. The Census Bureau estimates that the ACS, a twenty-eight-page instrument, will take the average household forty minutes to complete. 25 U.S. Census Bureau, U.S. Dep’t of Commerce, The American Community Survey 28 (2018), []. Because of the length and degree of intrusion, it is delivered only to a smattering of individuals on a rolling basis, eventually reaching 12.5% of households over a five-year span. 26 See supra note 23. In this context, a question about citizenship, while still salient, may not stand out overmuch. 27 This is not to suggest that the question is wholly uncontroversial even on a longer survey. There is reason to believe that a substantial portion of noncitizens lie when responding to the citizenship question on the ACS. See infra text accompanying notes 73–76. And others may just refuse to respond. Among those who responded generally to recent iterations of the ACS, 6–6.3% of Anglo individuals did not respond specifically to the citizenship question; 12–12.6% of non-Hispanic black individuals did not respond; and 11.6–12.3% of Hispanic individuals did not respond. See Ross, Decision Letter, supra note 1, at 3.
Moreover, in a survey like the ACS, statistical techniques can compensate to some degree for nonresponse. That is not the case for the decennial enumeration, both practically (because the enumeration is the basis for statistical adjustment of all other surveys) and legally (because there are legal limitations on the Census Bureau’s ability to statistically adjust the enumeration itself). See infra note 62.

B. Collecting Citizenship Information During the Enumeration in a Climate of Fear

The decennial enumeration questionnaire stands in stark contrast. It contains ten short questions for the head of household, and seven short questions for everyone else. 28 2010 Questionnaire, supra note 13. It is asked of every single household in the country, in a massive government effort that amounts to the country’s largest peacetime mobilization. 29 Office of Inspector Gen., U.S. Dep’t of Commerce, Final Report No. OIG-11-030-I, Census 2010: Final Report to Congress 17 (2011), []. It is designed to be short, simple, and minimally intrusive, to maximize response rates. 30 U.S. Census Bureau, U.S. Dep’t of Commerce, Census 2000 Operational Plan, at I-2 (2000), []; see also U.S. Census Bureau, U.S. Dep’t of Commerce, POL/02-MA, Measuring America: The Decennial Censuses from 1790 to 2000, at 128, 138 (2002), []; U.S. Census Bureau, U.S. Dep’t of Commerce, Preparing for Census 2000: Subjects Planned for Census 2000, Federal Legislative and Program Uses, at I-1 (1998), []. For households that do not respond by phone or the internet, enumerators will follow up by mail and in person. 31 U.S. Census Bureau, U.S. Dep’t of Commerce, 2020 Census Operational Plan: A New Design for the 21st Century 3.0, at 18–21 (2017) [hereinafter 2020 Census Operational Plan 3.0], []. Adding any content in this context—in the context of government officials sweeping the country door-to-door to ask a short series of questions—amounts to a substantial design change from asking the question in the ACS, with the potential to generate a substantial change in response behavior. 32 See Letter from Vincent P. Barabba et al., supra note 19, at 2 (“There is a great deal of evidence that even small changes in survey question order, wording, and instructions can have significant, and often unexpected, consequences for the rate, quality, and truthfulness of response.”). The change substantially elevates the prominence and salience of the question, magnifying its impact on the process of collecting data. 33 See New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 561 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.) (“A question concerning citizenship may take on added significance in the context of the much shorter decennial census questionnaire.”); J. David Brown et al., Understanding the Quality of Alternative Citizenship Data Sources for the 2020 Census 39 (Ctr. for Econ. Studies, Working Paper No. CES 18-38, 2018), (on file with the Columbia Law Review) (describing how the addition of a single question would likely stand out on the 2020 Census). Researchers have found that adding a “difficult” or “objectionable” question to an otherwise short survey—specifically, adding a request for Social Security number to a mailed government survey modeled on the Census and otherwise asking about name, age, gender, ethnicity, and race—may have a significant impact in driving down the survey’s response rate. See, e.g., Don A. Dillman et al., Effects of Questionnaire Length, Respondent-Friendly Design, and a Difficult Question on Response Rates for Occupant-Addressed Census Mail Surveys, 57 Pub. Opinion Q. 289, 292, 298–300 (1993).

A question about citizenship status is not merely any question. Even in a less turbulent era, noncitizens are—or just as important, perceive themselves to be—comparatively vulnerable members of American society. Many prefer to remain silent about their noncitizen status. 34 See Stella Burch Elias, Immigrant Covering, 58 Wm. & Mary L. Rev. 765, 802–06 (2017). Placing a citizenship question on the Census raises a serious concern that despite a legal duty to respond to the Census, 35 See, e.g., 13 U.S.C. § 221 (2012). noncitizens—even in an era of enlightened goodwill—will engage disproportionately in civil disobedi­ence rather than publicize their noncitizen status.

The Census Bureau has recognized this possibility. In 1980, the Bureau argued in federal court that adding a question about citizenship to the decennial enumeration would impair the enumeration itself:

[A]ny effort to ascertain citizenship will inevitably jeopardize the overall accuracy of the population count. Obtaining the cooperation of a suspicious and fearful population would be impossible if the group being counted perceived any possibility of the information being used against them. Questions as to citizenship are particularly sensitive in minority communities and would inevitably trigger hostility, resentment and refusal to cooperate. 36 Fed’n for Am. Immigration Reform v. Klutznick, 486 F. Supp. 564, 568 (D.D.C. 1980); see also New York, 351 F. Supp. 3d at 515, 525.

If the potential nonresponse anticipated from a citizenship question posed a risk to the accuracy of the enumeration in 1980, the volatility of the current political climate hardly provides less reason for concern. In 1950, when the Census last asked about citizenship in the enumeration, Americans had a very different relationship to the federal government. In the late 1950s, the National Election Study first regularly began asking Americans about their trust in the federal government; in those earliest surveys, seventy-three percent of the public trusted the government to do what is right. 37 Public Trust in Government: 1958–2017, Pew Research Ctr. (Dec. 14, 2017), []. By December 2017, that rate had plummeted to eighteen percent. 38 Id. Similarly, 74.5% of Americans in 2017 reported being “afraid” or “very afraid” of corrupt government officials, making it the single largest source of fear among the population. See America’s Top Fears 2017, Chapman Univ. (Oct. 11, 2017), [].

Moreover, the way that the federal government is currently per­ceived with respect to questions about citizenship is particularly fraught. 39 See, e.g., California v. Ross, 358 F. Supp. 3d 965, 980 (N.D. Cal. 2019) (“The macro-environment, particularly the political environment around immigration, has the potential to amplify the negative effect of the citizenship question on self-response rates.”). Immigration was one of the most salient issues in the 2016 election, and due to recent bitterly and prominently contested fights over immigration policy and enforcement, has grown even more salient since. 40 See generally Trump v. Hawaii, 138 S. Ct. 2392 (2018) (“travel ban”); id. at 2435–40 (Sotomayor, J., dissenting) (recounting the charged political backdrop to the “travel ban”); Philadelphia v. Sessions, 916 F.3d 276 (3d Cir. 2019) (“sanctuary cities”); E. Bay Sanctuary Covenant v. Trump, 354 F. Supp. 3d 1094 (N.D. Cal. 2018), appeal filed, No. 18-17436 (9th Cir. 2018) (eligibility for asylum); Ms. L. v. ICE, 310 F. Supp. 3d 1133 (S.D. Cal. 2018), appeal filed, No. 18-56151 (9th Cir. 2018) (“family separation”); NAACP v. Trump, 298 F. Supp. 3d 209 (D.D.C. 2018), appeals filed, Nos. 18-5243, 5245 (D.C. Cir. 2018) (DACA); Complaint, Pineda v. Trump, No. 1:18-cv-02534 (D.D.C. Nov. 1, 2018) (“caravan”); Memorandum from Dep’t of Def. on Support for Border Security to the President (Nov. 20, 2018),
decision-memorandum-department-of-defense-support-for-border-security.pdf [] (military deployment to ports of entry). See also New York, 351 F. Supp. 3d at 562 (“In particular, the undisputed evidence—including the Census Bureau’s own research—indicates that respondents are likely to react differently to a citizenship question in 2020 than they would have reacted only three years ago . . . .”).

Concern over the citizenship question is not merely an issue for indi­viduals whose own presence in the country may be in jeopardy, whether undocumented, 41 See, e.g., Elise Foley, Trump’s ICE Is Increasingly Arresting Immigrants Without Criminal Convictions, HuffPost (May 17, 2018), [https://perma.
lawfully present, 42 See, e.g., Inadmissibility on Public Charge Grounds, 83 Fed. Reg. 51,114 (proposed Oct. 10, 2018); Brittny Mejia, It’s Not Just People in the U.S. Illegally—ICE Is Nabbing Lawful Permanent Residents Too, L.A. Times (June 28, 2018), []. or—in this enforcement environ­ment—even citizens. 43 See, e.g., Brittny Mejia, Under Trump, the Rare Act of Denaturalizing U.S. Citizens on the Rise, L.A. Times (Aug. 12, 2018),
la-me-ln-denaturalization-20180812-story.html []; see also Masha Gessen, In America, Naturalized Citizens No Longer Have an Assumption of Permanence, New Yorker (June 18, 2018), [].
Many citizens and legal permanent residents have family or cultural connections to those perceived to be at risk or are uncertain about who is at risk. Citizen householders concerned for family and nonfamily members at home or in the broader community may resolve to avoid the enumeration or omit household members in their responses; citizen children living with noncitizen parents or caregivers are also at risk of being left out. 44 See California, 358 F. Supp. 3d at 981–83. And even minorities unconnected to any immigration controversy may feel unease in the present climate: Fear runs like contagion and infects friends and neighbors as well. Populations that are already among the most vulnera­ble generally are more likely to feel like they have more to lose when a government official shows up at their door asking for information, and may believe that the safest course is to keep their door firmly shut. 45 Id. at 982–83, 988.

This growing fear was clear to Census officials in recent attempts to administer surveys with far less prominence than the decennial enumeration, well before the Census Bureau announced that it would add a citizenship question to the decennial instrument. The rates at which individuals refused to respond to the American Community Survey were higher in 2015 and 2016 than ever before in the survey’s history, 46 See American Community Survey: Response Rates, U.S. Census Bureau, U.S. Dep’t of Commerce, [] (last visited Jan. 31, 2019). This effect was broadly distributed: In twenty-nine different states, the rate at which individuals refused to respond to the ACS was higher in 2016 than ever before; in forty-four different states, the rate at which individuals refused to respond to the ACS was higher in either 2015 or 2016 than ever before. Id. with the rate of nonresponse increasing faster in tracts with substantial concentrations of noncitizens. 47 See, e.g., Brown et al., supra note 33, at 11–12; see also infra note 63 (describing nonresponse rates that may be attributable to questions about citizenship). And even among those responding to the ACS generally, individuals have been refusing to respond to questions about their citizenship at a growing rate, 48 See William P. O’Hare, Georgetown Law Ctr. on Poverty and Inequality, Citizenship Question Nonresponse 11–12 (2018), []. with nonresponse rates growing most quickly for the Latino population. 49 See Brown et al., supra note 33, at 8–10.

By November 2015, nonpartisan career staff at the Census Bureau had already identified some of the significant challenges ahead. An oper­ational report noted “[d]istrust in government,” “[d]eclining response rates,” and an “[i]ncreasingly diverse population . . . who may have varying levels of comfort with government involvement” as important hurdles for the decennial enumeration. 50 U.S. Census Bureau, U.S. Dep’t of Commerce, 2020 Census Operational Plan: A New Design for the 21st Century 1.1, at 6–7 (2015), [].

By November 2017—months before Secretary Ross announced that he would add a citizenship question to the decennial instrument—the alarm bells for the enumeration had grown significantly louder. Nonparti­san career staff at the Census Bureau cited a “recent increase in respondents . . . spontaneously expressing concerns to researchers and field staff about confidentiality and data access related to immigration,” including the “perception that certain immigrant groups are unwelcome.” 51 Mikelyn Meyers, U.S. Census Bureau, Respondent Confidentiality Concerns and Possible Effects on Response Rates and Data Quality for the 2020 Census 2 (2017), []. They observed “increased rates of unusual respondent behaviors during pretesting and production surveys,” including “data falsification, item non-response, [and] break-offs.” 52 Id. at 3, 7. The flags were not merely raised by those responding to the surveys but also by those giving the surveys. As one interviewer put it: “Three years ago was so much eas­ier to get respondents compared to now because of the government changes  . . .  and  trust  factors . . . .  Three  years  ago  I  didn’t  have  problems  with  the  immigration  questions.” 53 Id. at 13.

And so a team at the Census Bureau conducted a brief qualitative study about the attitudes of respondents. The study reported that “[f]indings across languages [and] regions of the country, from both pretesting respondents and field staff[,] point to an unprecedented ground swell in confidentiality and data sharing concerns, particularly among immi­grants or those who live with immigrants.” 54 Id. at 15. The study noted that this “[m]ay present a barrier to participation in the 2020 Census,” and that it “[c]ould impact data quality and coverage for the 2020 Census.” 55 Id. See generally Memorandum for Associate Directorate for Research and Methodology (ADRM), U.S. Census Bureau (Sept. 20, 2017), [] (delineating similar concerns). See also California v. Ross, 358 F. Supp. 3d 965, 992–93 (N.D. Cal. 2019) (crediting the concerns).

Litigation over the addition of the question to the decennial enumera­tion has revealed that the civil-service leadership of the Census Bureau—the scientists tasked with delivering a valid enumeration—articu­lated similarly serious concerns to the Department of Commerce’s political leadership. For example, a draft memorandum by John M. Abowd, Chief Scientist and Associate Director for Research and Methodol­ogy at the Census Bureau, details the likely impact of the at­tempt to collect citizenship information on the decennial. 56 Memorandum from John M. Abowd, Chief Scientist and Assoc. Dir. for Research and Methodology, U.S. Census Bureau, to Wilbur L. Ross, Jr., Sec’y of Commerce (Jan. 19, 2018) [hereinafter Abowd Memorandum, Joint Appendix], Joint Appendix 104, at 104–22, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Mar. 6, 2019), 2019 WL 1114907. It notes that adding a citizenship question to the decennial risks “[m]ajor potential quality and cost disruptions.” 57 Id. at 106; see also id. at 105 (noting that the additional question “is very costly” and “harms the quality of the census count”).

Dr. Abowd understood that including a citizenship question on the decennial questionnaire would significantly depress the initial response rate. 58 See, e.g., California, 358 F. Supp. 3d at 977–78 (crediting Abowd’s testimony that the Census Bureau has produced quantitative evidence that adding a citizenship question to the 2020 Census will lower self-response rates). Cf. id. at *10 (citing a study predicting a decline of 6.3%–8.0% nationally, and 10.5%–14.1% in California, due to the citizenship question). He estimated that millions of dollars of follow-up would be neces­sary to attempt to compensate for this increased lack of response. Even after this follow-up, he thought that the question would yield hundreds of thousands of incorrect enumerations. And he stressed that this assess­ment represented a “lower bound” and a “conservative estimate.” 59 Abowd Memorandum, Joint Appendix, supra note 56, at 115; see also New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 516, 566, 580 (S.D.N.Y. 2019), cert. granted, 2019 WL 331100 (U.S. Feb. 15, 2019) (mem.); Brown et al., supra note 33, at 41–43.

It is difficult to know exactly how many individuals would refuse in 2020 to be counted by a national enumeration questionnaire with a citizenship question. But Dr. Abowd’s “lower bound” estimates of impact—with hundreds of thousands of errors, even after substantial follow-up—seem unrealistically low. Specifically, these “lower bounds” assume (1) that rates of nonresponse for a citizenship question on the decennial enumeration would be no larger than nonresponse rates for the American Community Survey, 60 See Abowd Memorandum, Joint Appendix, supra note 56, at 111, 114. One means to gauge the impact of the question on the ACS is the comparative dropoff in response rate from short form to long form in households with a noncitizen. There are many reasons, including the length and intrusion of the longer survey, why fewer households might respond to the long form than the short form. But there are few reasons beyond the citizenship question why households with a noncitizen might drop off to a greater degree than demographically similar households with a citizen.
In 2000, the dropoff rate from short form to long form was 3.3% greater in households with a noncitizen than in households comprising only citizens, suggesting reluctance to more than just the length of the long form. Brown et al., supra note 33, at 40; Ross, Decision Letter, supra note 1, at 4. That comparative dropoff rate increased in 2010, to about 6.1% (holding other demographic factors constant). See Brown et al., supra note 33, at 33–38; see also New York, 351 F. Supp. 3d at 578–80 (discussing similar estimates).
despite the dramatically increased prominence of the question on the decennial; (2) that nonresponse driven by the question should be expected only of households containing at least one noncitizen, despite fear within the broader community; (3) that the rates of nonresponse for a 2020 Census would be no larger than nonresponse rates in the 2010 Census, despite the substantial difference in overall political climate; and (4) that in-person follow-up visits will be able to cure a substantial amount of any initial nonresponse. There is considerable reason to doubt the validity of each of those four assumptions. 61 See, e.g., California, 358 F. Supp. 3d at 979, 985–92; New York, 351 F. Supp. 3d at 578–80, 590–92. If any are inaccurate, the disruption in cost and quality will only rise. If more than one of these assumptions is inaccurate, the disruption compounds further.

Moreover, on a survey like the ACS, analysts can partially compen­sate for known nonresponse rates with advanced statistical techniques, to preserve the accuracy of the instrument. But the Census Bureau is largely restricted from doing so with respect to the decennial enumeration. 62 At least, the Bureau is restricted in its ability to compensate for nonresponse in the enumeration generated to fulfill the Bureau’s lone constitutional obligation: a count for the apportionment of congressional seats. See 13 U.S.C. § 195 (2012) (“Except for the determination of population for purposes of apportionment of Representatives in Congress among the several States, the Secretary shall, if she considers it feasible, authorize the use of the statistical method known as ‘sampling’ in carrying out the provisions of the this title.”); Dep’t of Commerce v. U.S. House of Representatives, 525 U.S. 316, 334–44 (1999) (“[T]he proposed use of statistical sampling to determine population for purposes of apportioning congressional seats among the States violates [13 U.S.C. § 195].”); cf. Utah v. Evans, 536 U.S. 452, 479 (2002) (holding that sampling would violate 13 U.S.C. § 195 while the statistical method known as “hot-deck imputation” does not). For a description of the limits more generally, see Nathaniel Persily, The Law of the Census: How to Count, What to Count, Whom to Count, and Where to Count Them, 32 Cardozo L. Rev. 755, 758, 764 (2011) [hereinafter Persily, Law of the Census]. If the response to the decennial enumeration is broken, it cannot be repaired.

Even in 2016 and 2017, there were signs of a serious brush fire jeopardizing the accuracy of the enumeration, based on underfunding of Census operations and a climate fostering fear of interaction with government officials. 63 See, e.g., U.S. Census Bureau, U.S. Dep’t of Commerce, U.S. Census Bureau’s Budget: Fiscal Year 2017, at 9–11 (2016), []; Tara Bahrampour, Census Watchers Warn of a Crisis if Funding for 2020 Count Is Not Increased, Wash. Post (Apr. 18, 2017),
20d6-11e7-a0a7-8b2a45e3dc84_story.html (on file with the Columbia Law Review); Danny Vinik, Trump’s Threat to the 2020 Census, Politico (Apr. 9, 2017), [].
Many public and private actors were attempting to combat the fear as best they could. 64 See, e.g., Letter from Advancement Project California et al., to Governor Edmund G. Brown Jr. et al. (May 24, 2017), []. But that was all before the Commerce Secretary’s decision to add a question on citizenship to the decennial enumeration, profoundly magnifying the prominence of the issue. This decision pours gasoline on the preexisting fire.

C. The Absence of Testing

It would be one thing if the Census Bureau (or if the Bureau’s professional staff had lost the confidence of the political leadership, some other wing of the Department of Commerce), having tested deploy­ment of a potentially explosive topic on the decennial enumeration, had solid evidence that adding the question would cause no actual damage despite the widespread concern of those most active in the communities most affected. It is standard operating procedure for the Census Bureau to relentlessly and rigorously test, iteratively and over years, every change to the decennial enumeration—including changes much less controversial than this question. 65 See generally 2020 Census Operational Plan 3.0, supra note 31 (“The 2020 Census has been designed and developed in an iterative fashion, incorporating results from various tests conducted over the decade.”); see also California, 358 F. Supp. 3d at 1019; New York, 351 F. Supp. 3d at 526–27, 558–60.

Unfortunately, the government agencies normally responsible for this testing have offered no such assurances to the public. Instead, in his memorandum announcing the change, Secretary Ross flipped the standard burden of proof, citing the absence of reliable data about the question’s potential negative impact as a factor in favor of making the change. 66 Ross, Decision Letter, supra note 1, at 3. The only response data Secretary Ross mentioned were drawn from longer Census surveys or from a single private survey, without any evaluation of the impact of the question in the far more prominent context of the decennial enumeration. Id. at 3–4, 6. Litigation has revealed reason to doubt even the accuracy of Secretary Ross’s invocation of that private survey. See New York, 351 F. Supp. 3d at 563–65 (detailing the discrepancies between Ross’s account of his conversation with the Senior Vice President of Data Science for the Nielsen Company and the account of the Senior Vice President). It is impossible to overstate how much of a departure this represents from the way that the Census Bureau normally conducts business. In the movies, the Census Bureau is the sterile room where lab officials in goggles, latex gloves, and white booties work with impeccable care to prep the space probe or prepare the antidote. “We don’t know how much damage the bomb will do, but we’ll be able to figure it out after we drop it” is not how the Census Bureau normally executes its constitutional responsibility. 67 See supra note 65 and accompanying text; see also New York, 351 F. Supp. 3d at 526–27.

There has not been, and will not be, any opportunity for public testing of the citizenship question in an environment approximating the decennial enumeration before the 2020 enumeration is conducted. 68 See New York, 351 F. Supp. 3d at 541 (emphasizing the dismay of six former Directors of the Census at the lack of opportunity to test the citizenship question adequately). Secretary Ross claims that the citizenship question has already been evaluated because it was tested before it was included in the ACS. Ross, Decision Letter, supra note 1, at 7. The precise phrasing of the question may well have been tested in the context of the ACS. But it is shoddy science to simply assert that testing decades ago on a much longer instrument delivered to a much smaller sample population will accurately anticipate the impact of the question on a ten-question survey delivered in the present climate to every household in the country. See New York, 351 F. Supp. 3d at 560–61. The question, added in 2018, was simply inserted too late in the process. Testing for the 2020 enumeration began thirteen years earlier, in 2007, including real-world experiments in 2010 followed by further rigorous evaluation and retesting in the decade before liftoff. 69 See Nat’l Research Council, Experimentation and Evaluation Plans for the 2010 Census: Interim Report 8 (Lawrence D. Brown et al. eds., 2008); 2020 Census Operational Plan 3.0, supra note 31, at 34–49. The culmination of this relentless research process was the 2018 “End-to-End Census Test,” also known as the “dress rehearsal.” 70 Patrick R. Potyondy, The One and Only End-to-End 2020 Census Test, Nat’l Conference of State Legislatures (Nov. 28, 2017), []; see also 2018 Census Test: Frequently Asked Questions for the 2018 Census Test, U.S. Census Bureau, [] (last updated Apr. 6, 2018) (explaining the purpose and structure of the 2018 Census Test). This dress rehearsal, meant to be the final full test of the enumeration before execution of the 2020 Census, was planned, prepared, and begun before Secretary Ross made his decision. As a result, it contained no questions about citizenship in a decennial enumeration setting. 71 See U.S. Census Bureau, U.S. Dep’t of Commerce, 2018 Census Test Questionnaire (2018), []. Even without the ability to conduct valid tests on the enumeration instrument containing the citizenship question, locals in Providence, Rhode Island—where the “dress rehearsal” is being conducted—have expressed concern about depressed participation based on the controversy. See Jon Kamp & Janet Adamy, Citizenship Question Rankles in a Trial Run of 2020 Census, Wall St. J. (Apr. 13, 2018), (on file with the Columbia Law Review).
Several years ago, the Census Bureau articulated as a “lesson learned” from the 2010 Census that it was necessary to “[e]nsure sufficient time for testing the questionnaire content.” 2020 Census Operational Plan 3.0, supra note 31, at 69. The lesson does not appear to have been learned well enough. See Letter from Vincent P. Barabba et al., supra note 19, at 2 (“Adding a citizenship question without a testing opportunity in a contemporary, census-like environment will invalidate the results and lessons learned from the End-to-End test.”).
The question is now less than a year from exploding on the launchpad.

D. The Consequences of an Inadequate Enumeration

The most serious potential consequence of adding the citizenship question to the decennial enumeration is the accuracy of the enumera­tion. As discussed above, this is mostly driven by the concern that individuals will refuse to respond to the enumeration in this climate if a citizenship question is added. 72 See supra section I.B.

There is an additional risk, however, that the results of those who do respond will be contaminated by fabrication. In assessing the likely quality of the data to be gained from asking about citizenship in the enumeration, the Census Bureau compared answers given by individual respondents on the ACS against administrative data 73 For example, the Census Bureau acquires data pertinent to citizenship from the Social Security Administration, Internal Revenue Service, and U.S. Citizen and Immigration Services. See Questions on the Jan 19 Draft Census Memo on the DoJ Citizenship Question Reinstatement Request [hereinafter Questions on DoJ Citizenship Request, Joint Appendix], Joint Appendix 123, at 133–34, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Mar. 6, 2019), 2019 WL 1114907. —and found that “nearly one third of [individuals listed as] noncitizens in the administrative data respond to the questionnaire indicating they are citizens.” 74 Summary Analysis of the Key Differences Between Alternative C and Alternative D, Joint Appendix 145, at 147, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Mar. 6, 2019), 2019 WL 1114907. That is, about one-third of individuals listed as noncitizens in bureaucratic administrative data said in ACS responses that they were citizens. 75 See id.; see also Brown et al., supra note 33, at 21, 23 (finding a 37.6% discrepancy). The discrepancy rate may be significantly higher among more recent arrivals. See Jennifer Van Hook & James D. Bachmeier, How Well Does the American Community Survey Count Naturalized Citizens?, 29 Demographic Res. 1, 3 (2013) (noting that a late-1990s study found that “[a]mong new arrivals (those in the U.S. fewer than five years) from all national origins, about 75% of those who were reported [on the Census long form] as naturalized were probably not”); id. at 26 (finding a discrepancy in updated data as well). It is possible that the administrative data are outdated and do not capture more recent naturalization, and it is possible that the administrative data are incorrect (or incorrectly linked to census responses). But it seems at least as likely that some of the respondents were falsely presenting themselves as citizens in their responses to the questionnaire. And again, there is reason to believe that the error rate triggered by citizenship questions on the ACS may be more pronounced in the more prominent context of the decennial enumeration. 76 See New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 536–37 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.). To be clear, the decennial enumeration has likely always been inaccurate to some degree. See, e.g., id. at 577; see also Gaffney v. Cummings, 412 U.S. 735, 745 & n.10 (1973) (“[T]he results of the United States census . . . may be as accurate as such immense undertakings can be, but they are inherently less than absolutely accurate.”). After each enumeration, the Census Bureau conducts an analysis of its own performance to attempt to measure the error. In 2010, for example, the Census estimated “a net overcount of 0.01%.” See Memorandum from Patrick J. Cantwell, Assistant Div. Chief, Sampling and Estimation, Decennial Statistical Studies Div., to David C. Whitford, Chief, Decennial Statistical Studies Div. 1 (May 22, 2012) [hereinafter Census Coverage Measurement Estimation Report], [].
But errors in the enumeration are not borne equally. See generally, e.g., Janai Nelson, Counting Change: Ensuring an Inclusive Census for Communities of Color, 119 Colum. L. Rev. 1399 (2019) (discussing the historical undercount of people of color). For example, 3.8% of non-Hispanic white residents were omitted in 2010, but imputation and error led to a 0.83% overcount. See Census Coverage Measurement Estimation Report, supra, at 16 & tbl.8, 17 tbl.9. In contrast, 7.7% of Latino residents and 9.3% of African-American residents were omitted, and even after accounting for imputation and mistake, the enumeration still resulted in a 1.54% undercount for Latinos and a 2.06% undercount for African Americans. See id. The concern is that the presence of a citizenship question in the decennial enumeration could cause these inaccuracies—and the disparities in the inaccuracies—to increase exponentially.

These errors have several consequences. First is the simple default on a constitutional obligation. If burdening the decennial enumeration with an extra question on citizenship drives down the response rate or distorts the count in this political climate, the decision to include the question amounts to a dereliction of constitutional duty. The Census Bureau has no lawful authority to prioritize any goal over its constitu­tional mandate to ensure an “actual Enumeration” of every person in the country. 77 U.S. Const. art. I, § 2, cl. 3; id. amend. XIV, § 2.

The second consequence is to our informational infrastructure. The enumeration not only provides a rough outline of the country’s popula­tion but also serves as the underlying basis for all of the other infor­mation collection that brings the picture to life. Government surveys like the American Community Survey depend on the enumeration to ensure that the surveyed populations contain a representative sample of the public; a skewed enumeration skews all of the government’s other statistical collection. 78 See 2020 Census Operational Plan 3.0, supra note 31, at 5; see also Andrew D. Reamer, Counting for Dollars: The Role of the Decennial Census in the Geographic Distribution of Federal Funds 6–7 (2010) [hereinafter Reamer, 2010], []. Private surveys and political polls, and the decisions they inform, further depend on the enumeration and the additional government surveys based on the enumeration. 79 See Weighting, Am. Ass’n for Pub. Op. Research, [] (last visited Feb. 18, 2019) (describing post-stratification weighting using census data); see also 2020 Census Operational Plan 3.0, supra note 31, at 5. If the enumeration is off, it will skew policy and business decisions until 2030. We are less able to confront challenges and seize opportunities when the basic facts and figures at the core of our shared understanding are simply incorrect.

The third consequence is for political representation and funding. The enumeration drives the apportionment of congressional districts; redistricting for federal, state, and local districts of all kinds; and the distribution of billions of dollars in government grants tied by formula to population. 80 See U.S. Const. amend. XIV, § 2; id. art. II, § 1, cl. 2; 2 U.S.C. § 2a(a)(2012); see also Marisa Hotchkiss & Jessica Phelan, U.S. Census Bureau, Uses of Census Bureau Data in Federal Funds Distribution 3–10 (2017), []; see generally Reamer, 2010, supra note 78 (describing the various uses of census data in the distribution of funds). Communities that are undercounted lose political voice and substantial government aid.

The nature of this loss, however, has substantial ripple effects. When vulnerable populations do not respond to the Census, it is not just the vulnerable populations who suffer. Everyone in an area with an undercount loses clout and cash. 81 See Reamer, 2010, supra note 78, at 18–19 (“[A]n undercount, particularly among the hard-to-count, will serve to drive down the flow of federal funds to states and local areas. Consequently, a high participation rate and an accurate count are in all communities’ fiscal interests.”). Immigrant farmworkers lose—and so do the rural agricultural communities dependent on those farmworkers. Urban minorities lose—and so do the suburbs that depend on those cit­ies’ economic engines. Border communities lose—and so do those who live in or trade with those border communities, including members of the border patrol. We are firmly tethered to each other by the enumeration.

Indeed, some of the most visible effects will be statewide. 82 See, e.g., California v. Ross, 358 F. Supp. 3d 965, 1002–04 (N.D. Cal. 2019) (finding that any “measurable differential undercount of households containing noncitizens” would cause California to lose federal funding and political representation in Congress). Congressional representation and federal budget dollars are both finite resources with allocations driven by the enumeration, which means that states compete with each other for these resources based on the Census population count. But in this competition, with the citizenship question skewing real results, it is not the absolute number of vulnerable people that matters most, nor the growth of the population, nor even the demographic distribution of that growth. Among large, swiftly growing states where the scale creates serious leverage, what matters most in the competition for resources is the comparative level of government distrust. If minorities in Texas feel less safe than minorities in California and respond to the decennial census at lesser rates, Texas loses power and funding to California. And vice versa.

This comparative impact is worth repeating, if only because it may appear to conflict with the conventional wisdom. Texas is America’s most rapidly growing state. 83 Specifically, of relevance to apportionment and financing, Texas is gaining more population than any other state, even if other states have a slightly higher growth rate off of a smaller population base. See U.S. Census Bureau, Annual Estimates of the Resident Population: April 1, 2010 to July 1, 2018 (2018), (on file with the Columbia Law Review) [hereinafter Annual Estimates]. If the Census count is accurate, most projections suggest that Texas will accrue billions of dollars in additional federal cash driven by the Census count. 84 See, e.g., Andrew Reamer, Counting for Dollars 2020: The Role of the Decennial Census in the Geographic Distribution of Federal Funds 5 tbl.2 (2017), [] (finding that Texas received more than forty-three billion dollars from federal assistance programs driven by the Census in Fiscal Year 2015). Similarly, if the Census count is accurate, most projections suggest that Texas will gain an extra three congressional seats—and given recent electoral patterns, those seats are likely to be drawn by Republican legislators. 85 See, e.g., Press Release, Election Data Servs., Some Change in Apportionment Allocations with New 2017 Census Estimates; But Greater Change Likely by 2020, at 2 (Dec. 26, 2017),
NR_Appor17c3wTablesMapsC2.pdf [].

Like most other growing states, Texas’s population boom has been fueled by growth in historically vulnerable minority communities. Approximately 55% of Texans were racial and ethnic minorities in 2010, and projections from past growth patterns indicate that 59.1% of Texans are likely to be racial and ethnic minorities in 2020. 86 See Tex. Demographic Ctr., Texas Population Projections 2010 to 2050, at 4 tbl.1 (2019), []. Based on the local climate, if those minorities are substantially less likely to complete the decennial enumeration than are residents of other states, those funds and those seats vanish. And they vanish for all Texans, to be picked up by states where the population is less afraid.

The most swiftly growing big states are, in order, Texas, Florida, California, Arizona, North Carolina, Washington, Georgia, and Colorado. 87 See Annual Estimates, supra note 83. Some of these states were already feverishly working to bol­ster Census response before the Secretary’s decision to include a ques­tion on citizenship. 88 See Nat’l Conference of State Legislatures, The Census Is Coming. States Are Preparing., Medium (Apr. 27, 2017), []. They  will  likely  redouble  their  outreach  now. 89 See, e.g., California v. Ross, 358 F. Supp. 3d 965, 999–1000 (N.D. Cal. 2019) (finding that the citizenship question was one of the driving forces behind the California Legislature’s increased appropriation for census outreach). Some states may be less proactive. If adding a citizenship question to the decennial enumeration is likely to depress participation, the jurisdictions least responsive to their minority communities may have the most to lose.

II. The Ostensible Justification for the Citizenship Question

None of the tumult described above is necessary. Secretary Ross has claimed that the decision was motivated by the request of the Department of Justice; 90 See, e.g., Hearing with Commerce Secretary Ross: Hearing Before the H. Comm. on Ways & Means, 115th Cong. 51 (2018) (testimony of Wilbur Ross, Sec’y of Commerce) (“Department of Justice, as you know, initiated the request for inclusion of the citizenship question.”), []; see also New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 546–57 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.). the Department of Justice has claimed that it needs a citizenship question on the decennial enumeration in order to better enforce the Voting Rights Act (VRA). 91 See Ross, Decision Letter, supra note 1, at 2 (“DOJ states that the current data collected under the ACS are insufficient in scope, detail, and certainty to meet its purpose under the VRA.”); Letter from Arthur E. Gary, Gen. Counsel, Justice Mgmt. Div., U.S. Dep’t of Justice, to Dr. Ron Jarmin (Dec. 12, 2017) [hereinafter DOJ Letter, Joint Appendix], Joint Appendix 276, at 276, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Jan. 25, 2019), 2019 WL 1470266 (“This [citizenship] data is critical to the Department’s enforcement of Section 2 of the Voting Rights Act and its important protections against racial discrimination in voting.”). Neither Secretary Ross nor the Department of Justice offered any reason for the change beyond its purported utility in enforcement of the VRA. But there is substantial reason to believe that these claims are pretext.

I had the privilege of serving in the Civil Rights Division of the Department of Justice, supporting and supervising the Division’s voting rights docket, among other areas. I do not believe that the DOJ during my tenure was unduly shy about bringing VRA cases when the facts and the law indicated a violation, and I do not believe that we were unduly shy about asking for additional legal or evidentiary authority when that additional authority would enhance our ability to enforce civil rights law. 92 For example, after Shelby County v. Holder, 133 S. Ct. 2612 (2013), DOJ repeatedly urged Congress to repair a damaged VRA. See, e.g., Press Release, U.S. Dep’t of Justice, Attorney General Loretta E. Lynch Delivers Remarks at White House Event Commemorating 50th Anniversary of the Voting Rights Act (Aug. 6, 2015), https://www. []. Cf. Department of Justice Proposes Legislation to Improve Access to Voting for American Indians and Alaska Natives, U.S. Dep’t of Justice, [] (last updated May 21, 2015).
DOJ also requested additional data from the Census—for inclusion on the ACS, not the decennial enumeration questionnaire—when the additional data would likely enhance enforcement. Letter from Arthur E. Gary, Gen. Counsel, Justice Mgmt. Div., U.S. Dep’t of Justice, to John H. Thompson, Dir., Econ. & Statistics Admin., U.S. Census Bureau, U.S. Dep’t of Commerce (Nov. 4, 2016), [] (requesting inclusion of a topic “relating to LGBT populations” in the ACS); Hansi Lo Wang, Census Bureau Caught in Political Mess Over LGBT Data, NPR (July 18, 2017), [] (finding that the Obama Justice Department asked the Census Bureau to add questions about sexual orientation and gender identity to the ACS and “laid out the ‘legal authority supporting the necessity’ for collecting that information”).
Despite a deep commitment to enforcing the VRA, we never requested that the decennial enumeration include a question relating to citizenship. Nor had the Civil Rights Division of any Justice Department, under any administration, for the previous fifty-three years—that is, for the entire lifespan of the VRA.

Similarly, consider the position of civil rights groups most intimately engaged in extensive private enforcement of the VRA and most fiercely advocating for vigorous public enforcement of the Act. Each and every one has expressed vigorous opposition to the Commerce Secretary’s decision to include a question related to citizenship on the decennial enumeration in this political climate. 93 See, e.g., Leadership Conference, supra note 3. If the information were really necessary to enforce the VRA, this unified opposition by the private organizations most frequently litigating cases enforcing the VRA would be exceedingly odd.

There are two main reasons for the absence of pre-2017 clamor to place a citizenship question on the decennial enumeration questionnaire in order to enforce the VRA. First, since the VRA was enacted in 1965, existing survey data on citizenship—originally from the “long form” of the Census and now from the successor American Community Survey—have been largely sufficient to bring and win VRA cases. And second, for any additional data to be incrementally useful as an enforcement tool, they must be not only more precise, but more accurate. The Census Bureau’s action is not likely to meet this basic standard.

A. Citizenship Data and VRA Enforcement

1. The Use of Citizenship Data. — There are three main ways in which citizenship data are relevant in enforcing the VRA. These arise primarily in the redistricting context, when evaluating vote dilution claims. 94 In some circumstances, citizenship data might also be relevant in VRA cases concerning barriers to casting and counting a ballot. For example, VRA disputes over the discriminatory placement of polling places or the discriminatory impact of electoral restrictions may turn in part (though only in part) on the comparative demographic composition of the voters affected. Cf. Sanchez v. Cegavske, 214 F. Supp. 3d 961, 973–975 (D. Nev. 2016) (addressing the discriminatory siting of in-person registration and early voting sites in areas with substantial Native American populations). For these sorts of claims, Census-based data will generally be useful on similar terms as for the redistricting claims discussed in the text: The claims will turn on approximate demographic ranges rather than precise point estimates. First, before a redistricting plan or at-large structure can be said to dilute electoral opportunity on the basis of race or language-minority status, the affected communities must prove that they could exercise effective electoral opportunity with districts drawn in a different fashion. 95 See Justin Levitt, Quick and Dirty: The New Misreading of the Voting Rights Act, 43 Fla. St. U. L. Rev. 573, 586 (2016) [hereinafter Levitt, New Misreading]. In the name of litigation efficiency and administrability, the Supreme Court has set a bright-line threshold for this standard: Plaintiffs must show that they could constitute more than half of the electorate in a district-sized population. 96 Thornburg v. Gingles, 478 U.S. 30, 50 (1986); see also Bartlett v. Strickland, 556 U.S. 1, 14–15, 18–20 (2009) (plurality opinion). This showing, in turn, requires information about the electorate in a given area, by race or language-minority status—and the most readily available such data are census data about the citizen voting-age population (CVAP).

Second, before a redistricting plan or at-large structure can be said to dilute electoral opportunity on the basis of race or language-minority status, the affected communities must prove that they are comparatively unified, and that other voters in the area are also sufficiently cohesive to deny the affected groups equitable electoral opportunity most of the time. 97 See Levitt, New Misreading, supra note 95, at 586. That is, plaintiffs must show that voting in the area is racially polarized. 98 Gingles, 478 U.S. at 51. Because no cast ballot, on its own, is identified by race or language-minority status, researchers must impute electoral preferences to groups using several well-established methods of inference from the ecological population characteristics of voters within each precinct. 99 See id. at 52–53; J. Gerald Hebert et al., The Realist’s Guide to Redistricting: Avoiding the Legal Pitfalls 45–48 (2d ed. 2010) (“[T]hese methods allow[ ] experts, using aggregate data, to draw certain conclusions about voting preferences of groups of voters—specifically, how members of particular racial or ethnic groups cast their ballots.”). These assessments are more accurate when the population characteristics more closely mirror the active electorate—and in some states, the most readily available such data are often data, by race or language-minority status, about the precinct’s CVAP. 100 Census-based CVAP data may be the most readily available in some states, but there are alternatives, including data based on actual registrants or actual voters, that may in certain circumstances be even more useful for estimating cohesion and polarization in the active electorate. One such data source is an assessment of predicted race and ethnicity based on names and surnames of registrants or voters. See infra text accompanying notes 121–127.

Finally, if plaintiffs can establish a violation of the VRA based on vote dilution, that violation must be remedied by implementing a system in which race or language-minority groups have an equitable opportunity to elect candidates of their choice. 101 See, e.g., United States v. Brown, 561 F.3d 420, 435 (5th Cir. 2009); Bone Shirt v. Hazeltine, 461 F.3d 1011, 1022 (8th Cir. 2006). Testing whether the remedy will actu­ally provide an equitable electoral opportunity requires an assessment of the local electorate and local voting patterns by race and ethnicity. As above, in many states, the most readily available such data are usually derived from, or incorporate, data concerning the CVAP of the precinct, by race or language-minority status. 102 See supra note 100.

In each of these three areas, Census-based information about citizen­ship rates has—for the entire history of the VRA—come from a survey of a representative American sample. 103 See, e.g., Citizen Voting Age Population by Race and Ethnicity, U.S. Census Bureau (Feb. 1, 2018), []. First, it came from the Census “long form,” then from the American Community Survey. 104 See supra note 22 and accompanying text; see also American Community Survey: Sample Size, U.S. Census Bureau, [] (last visited Jan. 29, 2019).
Census geography boundaries change in connection with each decennial enumeration. In the past, it had been challenging in multi-year ACS compilations ending just before a decennial year to ensure that ACS responses from old Census geography were comparably aligned with redistricting information tied to new Census geography. See Persily, Law of the Census, supra note 62, at 777. But that limitation of the ACS should no longer present substantial difficulty. The Census Bureau has indicated that in 2020, they will be able to translate the most recent five-year ACS results into new 2020 geography for the release of ACS–CVAP data contemporaneous with the release of data from the 2020 decennial enumeration. U.S. Census Bureau, U.S. Dep’t of Commerce, 2020 Census Operational Plan: A New Design for the 21st Century 4.0, at 143 (2018), [].

Like any information from a survey, these data are not perfectly precise. First, because the most accurate ACS results are aggregated over a five-year span, the results present a rolling average rather than a count on a given day. For populations manifesting consistent growth (or consistent decline), ACS results will lag behind the population’s present value. 105 See Nathaniel Persily, Who Counts for One Person, One Vote?, 50 U.C. Davis L. Rev. 1395, 1410 (2017). Second, while enumeration data are released at the level of a census “block,” ACS survey data are released only in aggregate “block groups”; it is difficult to assess ACS data in geographies smaller than an aggregate group of multiple census blocks. 106 A census block is the smallest unit of geography measured by the Census. A census “block group” is a larger aggregation of multiple blocks; on average, there were fifty blocks for every block group in the 2010 Census, and the Census Bureau expects that most block groups will contain approximately 600 to 3,000 people. See U.S. Census Bureau, Citizen Voting Age Population (CVAP) Special Tabulation from the 2012–2016 5-Year American Community Survey (ACS) 1–4, [] (last visited Jan. 31, 2019); Geographic Terms and Concepts—Block Groups, U.S. Census Bureau, [] (last visited Jan. 29, 2019); Geography, 2010 Census Tallies of Census Tracts, Block Groups & Blocks, U.S. Census Bureau, [] (last visited Jan. 31, 2019). Citizenship rates may occasionally be translated to the census block level, by imputation, though this translation inevitably involves some further guesswork. See Brief of Nathaniel Persily et al. as Amici Curiae in Support of Appellees at 23–24, Evenwel v. Abbott, 136 S. Ct. 1120 (2015) (No. 14-940), 2015 WL 5719746 [hereinafter Brief of Nathaniel Persily et al.]. Third, because the ACS is a survey of a sample, it arrives with an associated margin of error. 107 Of course, the decennial enumeration data are also not perfect. Because they are not drawn from a survey, they do not have a similar statistical “margin of error,” but there are known lapses and imperfections. And while the Census enumeration generally has become more accurate over time, it continues to miss—to disproportionate degrees—members of racial and ethnic minority groups, young children, and poorer populations living in rental housing or who are homeless. See, e.g., U.S. Census Bureau, Investigating the 2010 Undercount of Young Children—Analysis of Census Coverage Measurement Results 1 (2017), []; Census Coverage Measurement Estimation Report, supra note 76, at 1–3, 15–17. The margin of error is larger in smaller geographies and for shorter time periods, and smaller in larger geographies and for longer time compila­tions. 108 The longer time compilations may introduce complications at the margins, as younger citizens reach voting age and older citizens pass away, but the extent of this error is likely small in most jurisdictions. See Persily, Law of the Census, supra note 62, at 777. But for the last fifty-three years, when Census-based data on citizenship have been necessary at all, the survey estimates have been sufficient for enforcing the VRA. 109 See, e.g., Montes v. City of Yakima, 40 F. Supp. 3d 1377, 1393 (E.D. Wash. 2014) (“Although U.S. Census data may not be perfectly accurate, it is routinely relied upon in § 2 cases.”).
Indeed, for some purposes, the survey estimates of citizenship are superior to any decennial tally, even in a counterfactual world in which the decennial solved more problems than it created. Recall that the ACS is a rolling monthly survey, updating information constantly throughout the decade; the enumeration, in contrast, is fixed at April of a decennial year. See supra notes 25–26 and accompanying text. For swiftly growing populations, the rolling ACS estimates may be the only means to establish when racial or language minority communities reach threshold size in the middle of the decade. VRA litigation is not only pursued in a decennial year.

Estimates of the citizen voting-age population are usually sufficient because each VRA calculation using these data is itself an estimate. For example, VRA plaintiffs must show that the relevant racial or language-minority communities could constitute more than half of the electorate in a district-sized population. But “district-sized” is a range, rather than a point: The Supreme Court has repeatedly held that district sizes may vary. 110 The Constitution generally grants more latitude in district size to state and local districts and less to congressional districts, but in both cases, a modest level of deviation is permissible if that deviation is the result of legitimate redistricting purposes. See Tennant v. Jefferson Cty. Comm’n, 567 U.S. 758, 759–60, 765 (2012); see also Harris v. Ariz. Indep. Redistricting Comm’n, 136 S. Ct. 1301, 1306–07 (2016). Given the permissible range of district size, the fact that infor­mation concerning the relevant size of the minority population may also represent a range—a point estimate along with a margin of error—is rarely of concern.

Similarly, both the extent of racially polarized voting and the effectiveness of any particular remedy involve assessments of the electoral strength and cohesion of relevant racial or language minorities. But these evaluations involve patterns, not points. The goal is not to predict the precise vote count in a future election based on ironclad certainty about each individual’s anticipated turnout likelihood and race-based voting preferences. Instead, the purpose of the analysis is to determine whether past voting behaviors generally indicate that racial or language minority communities would vote similarly most of the time, and whether they would be likely presented with effective equitable electoral oppor­tunity more often than not. 111 See Joseph Fishkin, The Administration Is Lying About the Census, Balkinization (Mar. 27, 2018), [] [hereinafter Fishkin, The Administration Is Lying]. If the assessments of district size above con­template a range, these assessments of voting strength add estimates and trends, relying on multiple past elections to anticipate likely approximate future behavior. These qualitative legal assessments must rely on rigorous analysis, 112 See Levitt, New Misreading, supra note 95, at 587–89. but to the extent they are informed by quantitative data, they tolerate a degree of imprecision. 113 See Fishkin, The Administration Is Lying, supra note 111. Where citizenship data from the ACS have been used, they have largely been sufficient. 114 See California v. Ross, 358 F. Supp. 3d 965, 1035–36 (N.D. Cal. 2019) (citing unrebutted expert testimony to this effect).

I reviewed eighteen years of the most recent cases brought by the Department of Justice to enforce the VRA against claimed vote dilution (where citizenship data have been most relevant), across both Republican and Democratic administrations, spanning two decades’ worth of “long form” and ACS data. To the best of my knowledge, not one of these cases represents an instance in which a decennial enumera­tion would have enabled enforcement that the existing survey data on citizenship did not permit. 115 These cases included: United States v. Euclid City Sch. Bd., 632 F. Supp. 2d 740, 745 (N.D. Ohio 2009); United States v. City of Euclid, 580 F. Supp. 2d 584, 593–95 (N.D. Ohio 2008); United States v. Vill. of Port Chester, No. 7:06-cv-15173, 2008 WL 190502, at *8–11 (S.D.N.Y. Jan. 17, 2008); United States v. Osceola Cty., 475 F. Supp. 2d 1220, 1230–31 (M.D. Fla. 2006); United States v. Alamosa Cty., 306 F. Supp. 2d 1016, 1028–29 (D. Colo. 2004); United States v. Charleston Cty., 316 F. Supp. 2d 268, 276–77 (D.S.C. 2003); Complaint at 4, United States v. City of Eastpointe, No. 4:17-cv-10079 (E.D. Mich. Jan. 10, 2017); Complaint at 5, United States v. Texas, No. 5:11-cv-00360 (W.D. Tex. Sept. 25, 2013); Consent Judgment and Decree at 3, United States v. Town of Lake Park, No. 9:09-cv-80507 (S.D. Fla. Oct. 26, 2009); Consent Judgment and Decree at 3, United States v. Sch. Bd. of Osceola Cty., No. 6:08-cv-00582 (M.D. Fla. Apr. 23, 2008); Consent Judgment and Decree at 4, United States v. Georgetown Cty. Sch. Dist., No. 2:08-cv-00889 (D.S.C. Mar. 21, 2008); Consent Decree at 3, United States v. Crockett Cty., No. 1-01-1129 (W.D. Tenn. Apr. 17, 2001); Consent Judgment and Decree at 4, United States v. City of Morgan City, No. 6:00-cv-01541 (W.D. La. Aug. 17, 2000); Order at 2, United States v. South Dakota, No. 00-3015 (D.S.D. Aug. 10, 2000); Complaint at 3, United States v. Upper San Gabriel Valley Mun. Water Dist., No. 2:00-cv-07903 (C.D. Cal. July 21, 2000); Complaint at 3, United States v. City of Santa Paula, No. 00-03691 (C.D. Cal. Apr. 6, 2000); Consent Decree at 2–3, United States v. Roosevelt Cty., No. 00-50 (D. Mont. Mar. 24, 2000); and Consent Judgment and Decree at 4, United States v. Benson Cty., No. A2-00-30 (D.N.D. Mar. 10, 2000). Indeed, not one of these cases has realisti­cally been close to the line. 116 Cf. New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 652 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.) (noting that in its request for a citizenship question on the decennial enumeration, DOJ did not “identify a single VRA case that DOJ failed to bring or lost because of inadequate block-level CVAP data”). See also California, 358 F. Supp. 3d at 1037 (citing testimony that “no reported Section 2 case has ever failed on account of the purported inadequacy of ACS data . . . as a measure of CVAP”).

Adding private litigation expands the sample set, but even in that context, it is exceedingly rare for plaintiffs enforcing the VRA to run into trouble based on the adequacy of the Census’s survey data, in any way that asking a citizenship question on the decennial enumeration might possibly cure. 117 There have been some cases in which ACS data were not put forward, and which failed for lack of proof that the population was sufficiently sizable, but those cases do not indicate that the ACS data would have been inadequate had they been provided. See, e.g., Cisneros v. Pasadena Ind. Sch. Dist., No. 4:12-cv-02579, 2014 WL 1668500, at *5–9 (S.D. Tex. Apr. 25, 2014).
There have also been some cases brought on behalf of minority populations that had been growing quite rapidly; though the decennial enumeration did not show that the minority population was sizable enough to be the majority of a district-sized population, plaintiffs hoped that multiple ACS surveys from the middle of the decade would show that they were on the cusp of meeting this litigation threshold. In these cases, the power of the particular mid-decade survey evidence that plaintiffs produced was not up to the task for the geographies in question. See, e.g., Benavidez v. Irving Ind. Sch. Dist., 690 F. Supp. 2d 451, 456–61, 464 n.18 (N.D. Tex. 2010). Indeed, as seasoned litigator Bob Heath has noted, the Benavidez case above was the lone case identified by a DOJ official in which ACS survey data were insufficiently precise for private plaintiffs to enforce the VRA as they preferred. Gopoversightandreform, Progress on the 2020 Census—Continued, at 1:15:00–1:16:35, YouTube (May 18, 2018), (on file with the Columbia Law Review) (showing the testimony of John M. Gore, Acting Assistant Attorney General of the Civil Rights Division in the Department of Justice, before the House Committee on Oversight and Government Reform). But Benavidez would not have been remedied by a more precise decennial enumeration of citizenship. At the start of the decade, existing data were sufficiently precise to show that the plaintiffs had not yet reached the appropriate threshold. The only question in the case concerned mid-decade growth. See Benavidez, 690 F. Supp. 2d at 453–54 (stating that the plaintiff’s argument “relies on the 2007 one-year American Community Survey data . . . rather than data from the 2000 Census”). Such growth cannot be measured by the decennial enumeration. Nothing in Secretary Ross’s proposal would have given these plaintiffs any comfort.
I am familiar with only one such case. And rather than demonstrating a need for a citizenship question on the decennial enumeration, it instead demonstrates that such a step is unnecessary.

The case in question is Fabela v. City of Farmers Branch. 118 No. 3:10-cv-01425, 2012 WL 3135545 (N.D. Tex. Aug. 2, 2012). In 2010, Latino plaintiffs claimed that the at-large election system in Farmers Branch, Texas, unlawfully diluted their right to vote. In furtherance of that VRA claim, they presented evidence, including data aggregated from the ACS, that Latino citizens constituted a majority of the voting-age citizens in four different illustrative districts. 119 Id. at *4–5; see also supra text accompanying note 96.

In a thorough opinion, the court reviewed some of the limitations of the ACS, including the limitations above. 120 See supra text accompanying notes 105–109. And the court recognized that ACS data offered some challenges in geographies as small as illustra­tive districts within Farmers Branch, which split “block groups” and where ACS estimates arrive with a margin of error. 121 Fabela, 2012 WL 3135545, at *5–7. Accurate data from the decennial enumeration would have offered less inherent uncertainty. But the court also recognized that precedent permitted flexibility in the data available to prove a VRA violation. 122 Id. at *4, *7 (citing, inter alia, Thornburg v. Gingles, 478 U.S. 30, 46, 73 (1986); Westwego Citizens for Better Gov’t v. City of Westwego, 906 F.2d 1042, 1046–47 (5th Cir. 1990); Citizens for a Better Gretna v. City of Gretna, 834 F.2d 496, 502 (5th Cir. 1987)). And it specifically noted the availability of alternative evidence to establish the size of the Latino elec­torate: Plaintiffs also presented data directly from the voter files, tallying registered voters with surnames separately identified by the Census as of Latino or Hispanic origin (also known as “SSRV” data). 123 Id. at *6–7. “SSRV” stands for “Spanish-surname registered voters.” Id. at *6. In that case, said the court, “the SSRV data strongly corroborate the accuracy of the Hispanic CVAP estimates [from the ACS].” 124 Id. at *8. And so the court found, with the combination of ACS data and SSRV information, that plaintiffs had met their burden of proving by a preponderance of the evidence that the Latino electorate in the area was sufficiently sizable to comprise a majority for purposes of the VRA threshold. 125 Id.

Indeed, in many smaller jurisdictions, litigators regularly deploy similar analysis based on the likely racial or ethnic identity of registered voters’ names or surnames—sometimes on its own, and sometimes in concert with Census-based survey data—in order to help enforce the VRA. The Department of Justice has used this sort of information in its own VRA litigation. 126 See, e.g., United States v. Alamosa Cty., 306 F. Supp. 2d 1016, 1022 (D. Colo. 2004); see also United States’ Brief in Opposition to Defendants’ Motion to Exclude Bayesian Improved Surname Geocoding (BISG) Evidence, United States v. City of Eastpointe, No. 4:17-cv-10079, 2018 WL 4144225 (E.D. Mich. Apr. 26, 2018). It is a practice directly endorsed in the legislative history of the VRA itself. 127 Congress explained, when it expanded VRA coverage to include language minorities, that “persons . . . of Spanish heritage” under the statute, see 52 U.S.C. § 10310(c)(3) (2017), included “‘persons of Spanish language’ as well as ‘persons of Spanish surname’ in Arizona, California, Colorado, New Mexico, and Texas.” S. Rep. No. 94-295, at 24 n.14 (1975). Analysis of minority electorate size by the names or surnames of registered voters is also not perfectly exact. But particularly in concert with ACS survey data, it has been sufficient. The ostensible need for citizenship data on the decennial enumeration does not arise from any meaningful limitations revealed by past litigation.

2. “Precision,” but Inaccuracy. — Existing data have been largely sufficient to enforce the VRA. It is, of course, possible that there exists a reservoir of marginal potential enforcement actions, just outside of the searchlight of extant litigation, for which ACS survey data are not up to the task, and for which other means of establishing the electorate (like SSRV data) are similarly inadequate. Yet even in these cases, it is exceed­ingly unlikely that adding a question on citizenship to the decennial enumeration in this climate would improve any entity’s ability to enforce the VRA.

These theoretical cases would all depend on the limitations of the ACS as a survey. First, because of the limited ACS sample, ACS data cover block groups rather than individual blocks: They represent molecules, rather than atoms. 128 See supra note 106 and accompanying text. Particularly in smaller jurisdictions, the ability to draw an effective remedial district might depend on lines drawn to include certain individual blocks and excise others, relying on distinctions more finely calibrated than ACS data coverage, as a fine-tipped pen permits more precision than a blunt Sharpie. Second, because the ACS is a survey, it has a margin of error. In hamlets with small minority populations, that margin of error might be fairly large, such that it is not clear whether the minority population of interest constitutes 40% or 60% of the relevant CVAP, or somewhere in between. 129 The fact that ACS data are estimates, subject to a margin of error, may also affect the examination of patterns of polarization in election results, if the demographic composition of precincts is sufficiently imprecise to drive effective analysis of broad voting patterns by race or ethnicity. And even in large cities with large minority populations, a community may be just on the cusp of the requisite size, such that it is not clear whether the minority population of interest constitutes 49.9% or 50.1% of the relevant CVAP. As described above, the fact that the test for district size is more about a range than a point diminishes the likelihood that these margins of error prove determinative, but it is at least theoretically possible that the margin of error would materially impact a determination at the extreme lower bound of a district-sized range. 130 See supra text accompanying note 110. In these three instances, a jurisdiction could conceivably be liable at the same time that litigants lacked the data adequate to bring an enforcement proceeding.

Adding a question to the decennial enumeration may give the illusion of increased precision and greater statistical power. The enumeration is a household-by-household count, rather than a survey, and so it arrives without a margin of error derived from statistical sampling. But this does not mean that the enumeration is error-free. To the contrary, even with imputation, systematic errors in the enumeration are often substantial. 131 See supra note 76. In addition to instances of undercounting and overcounting, the Census Bureau itself specifically introduces margins of uncertainty in the course of publicly reporting enumeration data. For example, in some small geographies, the Census Bureau intentionally introduces this uncertainty to preserve the confidentiality of responses to the enumeration, when the uniqueness of particular responses might otherwise permit the identification of individuals. See generally, e.g., Amy Lauger et al., Disclosure Avoidance Techniques at the U.S. Census Bureau: Current Practices and Research (2014), [] (discussing the various techniques of “disclosure avoidance” that the Census Bureau uses to maintain the confidentiality of individual respondents). See also California v. Ross, 358 F. Supp. 3d 965, 1045 (N.D. Cal. 2019) (acknowledging that these techniques lead to an unavoidable margin of error even in reporting citizenship data obtained through the decennial enumeration process). Adding a citizenship question will only increase the error. In this climate, the illusion of precision likely arrives at the cost of significant accuracy.

That inaccuracy may take two forms. First, vulnerable communities—including minority communities seeking protection from the VRA—may decline to respond to the enumeration entirely. 132 See supra text accompanying notes 47–48. This is the most significant risk, and the one that, contrary to the claims of the Administration, wreaks the most substantial havoc on the enforcement of the VRA in practice. The very minority communities most likely to need VRA protection are already chronically undercounted. 133 See Census Coverage Measurement Estimation Report, supra note 76, at 1–3, 15–17. With the addition of a citizenship question, they will be that much more likely to go untallied, which means that the enumeration is likely to systematically undercount precisely the people who most need the VRA. If the problem with the ACS survey is that it occasionally leaves doubt whether a population is sufficiently sizable to merit VRA protection, asking the question on the decennial enumeration may drive down participation so that it appears certain that the population is not sufficiently sizable to merit VRA protection. And because of the undercount, that certainty will be false. The communities on the margin, on whose ostensible behalf the information is sought in the enumeration, will inaccurately appear to be pushed below the threshold for enforcement.

The second potential inaccuracy runs in the other direction but also works to the detriment of minority communities. If vulnerable communities are not sufficiently scared to shut the door on the enumeration, they may be sufficiently scared to lie. That is, noncitizens fearing the repercus­sions of their answers may respond to the enumeration by claiming to be citizens. 134 See supra text accompanying notes 73–75. There is no way to know whether this error will be larger or smaller than the nonresponse error, but there is also no reason to expect that they will systematically cancel each other out in the same geographic area. In some jurisdictions, false answers may exaggerate the size of the minority electorate. As a result, a remedial district will be banking on the perceived muscle of an electorate that is actually smaller than it appears—and which, inevitably, will fail to live up to turnout expectations. VRA enforcement ostensibly aimed at electoral opportunity would lead to districts designed to leave that opportunity just out of reach, given the real facts on the ground.

Absent miraculous statistical happenstance, 135 It is hypothetically possible—though extremely unlikely—that the number of noncitizens falsely claiming citizenship in a particular geography would precisely equal the number of citizens declining to be enumerated. adding the citizenship question to the decennial enumeration in this climate is likely to lead to either significant undercount or overcount of the true Latino citizen population of jurisdictions like Farmers Branch, and other immigration-sensitive minority communities on the cusp of VRA protection. Any case currently out of the reach of enforcement due to ACS data would likely be placed even further out of the reach of meaningful remedy due to predictable errors in data returned by the decennial enumeration. An ostensible performance-enhancing drug that cripples the patient does not enhance performance.

B. Evidence of Pretext

The analysis above leaves plentiful reason to question the legitimacy of the proffered rationale for the addition of a citizenship question to the decennial enumeration. The adverse effect is untested but potentially huge, and the demonstrated need is virtually nonexistent when it is not actively counterproductive. Census Bureau career staff even noted that if citizenship information were required for more potent enforcement of the VRA, the Census Bureau could provide the necessary detail by consulting administrative records, avoiding the substantial detrimental impact of placing the question on the decennial enumeration question­naire. 136 See Questions on DoJ Citizenship Request no. 21, Joint Appendix, supra note 73, at 137 (noting that decennial enumeration data—without the citizenship question—combined with administrative records on citizenship would be “sufficient for DoJ’s request”); see also California, 358 F. Supp. 3d at 1015, 1018; New York v. U.S. Dep’t of Commerce, 351 F. Supp. 3d 502, 530, 532–36 (S.D.N.Y. 2019), cert. granted, 139 S. Ct. 953 (2019) (mem.). That suggestion was cast aside. 137 See Ross, Decision Letter, supra note 1, at 4 (“Option C, the use of administrative records rather than placing a citizenship question on the decennial census, was a potentially appealing solution to the DOJ request . . . . However, the Census Bureau . . . does not yet have a complete administrative records data set for the entire population.”). Something didn’t seem to add up.

Litigation discovery has revealed substantial further reason to believe that the Department of Justice’s request to collect citizenship data in the decennial enumeration for purposes of VRA enforcement was pretext. 138 See California, 358 F. Supp. 3d at 973 (“What ensued was a cynical search to find some reason, any reason, or an agency request to justify that preordained result.”); see also id. at 1013, 1024, 1026, 1040, 1044. As just one example: More than seven months before the DOJ requested decennial collection of citizenship information, Secretary Ross discussed in a May 2, 2017, email his “months old request that we include the citizenship question.” 139 See DOJ Letter, Joint Appendix, supra note 91, at 276 (requesting, in mid-December 2017, the reinstatement of the citizenship question on the 2020 Census); Email from Earl Comstock to Wilbur Ross (May 2, 2017) [hereinafter Comstock Email, Joint Appendix], Joint Appendix 276, at 276, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Jan. 25, 2019), 2019 WL 1470266 (demonstrating that email from Ross to Comstock was sent on May 2, 2017, more than seven months prior to the DOJ’s formal request); see also New York, 351 F. Supp. 3d at 550, 567–68. Earl Comstock, his Director of the Office of Policy and Strategic Planning at Commerce, responded that “we will get that in place. . . . We need to work with Justice to get them to request that citizenship be added back as a census question . . . . I will arrange a meeting with DoJ staff this week to discuss.” 140 Comstock Email, Joint Appendix, supra note 139, at 276 (emphasis added); see also New York, 351 F. Supp. 3d at 550, 567–68.

It is difficult to read this exchange without concluding that the Secretary had already developed a desire to include the citizenship question in the decennial enumeration, several months before May 2017. And it is difficult to read this exchange without concluding that the Department of Justice’s request, nine months later, was designed to accommodate this preexisting desire. 141 See New York, 351 F. Supp. 3d at 548–72 (recounting the extensive evidence demonstrating that Secretary Ross pursued DOJ involvement to generate a public post-hoc rationale for his preexisting decision to include the citizenship question in the enumeration); see also California, 358 F. Supp. 3d at 1013 (concluding that the evidence “belie[d] any notion that Secretary Ross’s motivation was to meet DOJ’s VRA enforcement needs,” indicating instead that “it was DOJ that was meeting his preferences”).

III. The Real Rationale(s)?

If Secretary Ross’s stated rationale for adding the citizenship ques­tion appears inadequate, it is worth considering whether there are more plausible alternatives that point to additional consequences well beyond the VRA. We may never know the real reason. But adding a question to the decennial enumeration might serve at least four purposes, or several in combination. 142 In addition to the purposes discussed here, it is conceivable that asking about citizenship on the decennial enumeration could potentially serve an immigration enforcement interest, but the modest incremental informational gain makes this possibility unlikely. Though the question on the decennial enumeration will inquire into a respondent’s citizenship or noncitizenship, it will not reveal noncitizens’ legal immigration status, and therefore does not identify any violation of law. There are also strict limitations on the use of census data in anything other than statistical aggregations: The identification to immigration officials of particular households containing noncitizens, for example, is not only contrary to existing law, but clearly so. See, e.g., 13 U.S.C. § 8(b)–(c) (2012) (“[T]he Secretary may furnish copies of . . . statistical materials which do not disclose the information reported by, or on behalf of, any particular respondent . . . . In no case shall information furnished under this section be used to the detriment of any respondent or other person to whom such information relates . . . .”); id. § 9(a) (prohibiting the use of census information “for any purpose other than the statistical purposes for which it is supplied”); id. § 214 (“Whoever . . . publishes or communicates any information . . . which comes into his possession by reason of his being employed (or otherwise providing services) [by the Census Bureau] . . . shall be fined not more than $5,000 or imprisoned not more than 5 years, or both.”); Fourteenth and Subsequent Decennial Censuses: Hearing Before the H. Comm. on the Census, 65th Cong. 169–71 (1918) (emphasizing the Census Bureau’s understanding that the statutory prohibition on using census information “to the detriment of the person or persons to whom such information relates” should insulate noncitizens from detrimental proceedings based on truthful information) (on file with the Columbia Law Review). Moreover, the Census Bureau will mask any data that could be used to identify any individual for any purpose. See generally, e.g., Lauger et al., supra note 131 (describing disclosure avoidance techniques employed by the Census Bureau). And statistical information on citizenship patterns is already available via the ACS in block groups. See Brown et al., supra note 33, at 4–5. In order to believe that the question was meant for enforcement, immigration personnel targeting their enforcement operations would have to find that a block-by-block count of noncitizens (with indistinguishable lawful and nonlawful status) specifically designed to mask any identification of select individuals, and deployed despite a legislative history prohibiting such use, were considerably more valuable than the same information at a slightly higher block group aggregation.
Of course, it is hypothetically possible that there is no justification for placing the question on the decennial enumeration beyond the notion that it simply makes sense to seek additional basic information about the American population when the opportunity arises. Such a rationale would either ignore or discount the above-outlined risk of serious error introduced by a question about citizenship in this climate. And it would fail to explain the presence of a question about citizenship in particular, as opposed to any other demographic question that might usefully inform public policy.

A. Statistics on Immigration

First, it may be that the question was added to the decennial enu­mer­ation precisely in order to achieve the likely effects explored above. In 1976, Justice Stevens recognized that “normally[,] the actor is presumed to have intended the natural consequences of his deeds.” 143 Washington v. Davis, 426 U.S. 229, 253 (1976) (Stevens, J., concurring). His maxim may no longer be a sufficient standard of the evidence required to prove constitutionally illegitimate intent, 144 See, e.g., Personnel Adm’r of Mass. v. Feeney, 442 U.S. 256, 279 (1979) (holding that discriminatory intent is constitutionally cognizable only when “the decisionmaker . . . selected or reaffirmed a particular course of action at least in part ‘because of,’ not merely ‘in spite of,’ its adverse effects upon an identifiable group”). but it is still a reasonably good guide to human behavior.

Adding a question about citizenship to the decennial enumeration in this climate is likely to depress participation, particularly but not ex­clusively among noncitizens. It is also likely to cause some noncitizens to state inaccurately that they are citizens. Both errors can be expected to substantially understate the number of noncitizens present in the country.

Given the prominence of immigration enforcement as a campaign issue for this Administration, and the degree to which the Administration has courted nativism, an artificially low noncitizen count may be a feature rather than a bug. Supporters of reducing the size of the immigrant population, lawfully present and unlawfully present alike, will be able to cite an undercount of noncitizens as evidence that noncitizen numbers are dwindling, and hence as evidence of an immigration policy deemed successful by their terms. The fact that the data are likely to be inaccurate will be unlikely to reduce the potency of the talking point.

B. Interstate Competition for Representation and Power

Second, it may be that the question was added to the decennial cen­sus to achieve the results above, but for a different underlying rationale. The decennial enumeration is the basis for apportioning congressional seats and Electoral College votes, and billions of dollars of federal funds allocated by statutory formula. 145 See supra note 80 and accompanying text. States with a larger undercount will receive fewer congressional seats, fewer Electoral College votes, and fewer federal funds; in states that are themselves larger and more swiftly growing, that effect will be exaggerated. 146 See New York, 351 F. Supp. 3d at 594–96. Because the premise of congressional apportionment is, explicitly, the whole number of “persons” in each state, rather than “citizens” or “residents,” it does not matter for apportionment purposes whether a respondent is a citizen or not. Compare U.S. Const. amend. XIV, § 2 (apportioning Representatives according to the number of “persons”), with id. art. I, § 2, cl. 2 (requiring that Representatives be “citizens”), and id. art. II, § 1, cl. 5 (requiring that the President be a “citizen”); and id. art. II, § 1, cl. 5 (requiring that the President be a “resident”). That is, the only material error flowing from the Census in the congressional apportionment context would stem from refusal to participate in the enumeration; any error in assessing whether respondents are citizens or noncitizens would not impact the apportionment.
Some individuals have sought to use the decennial enumeration to catalog not only citizens and noncitizens, but also noncitizens who are lawful permanent residents and those who are not. See Email from Kris Kobach, Kan. Sec’y of State, to Wilbur Ross, U.S. Sec’y of Commerce (July 14, 2017), Joint Appendix 185, at 186–87, Dep’t of Commerce v. New York, No. 18-966 (U.S. filed Jan. 25, 2019), 2019 WL 1470266. Despite the textual reference to “the whole number of persons in each State,” U.S. Const. amend. XIV, § 2, some have advocated that noncitizens temporarily (or unlawfully) present are not “in” each state for apportionment purposes. See, e.g., Complaint for Declaratory Relief at 2, Alabama v. U.S. Dep’t of Commerce, No. 2:18-cv-00772 (N.D. Ala. May 21, 2018); Charles Wood, Losing Control of America’s Future—The Census, Birthright Citizenship, and Illegal Aliens, 22 Harv. J.L. & Pub. Pol’y 465, 473–93 (1999). But I am not aware of any commentator suggesting that lawful permanent residents may lawfully be excluded from the apportionment count, and so without the additional distinction, the question could not affect the distribution of congressional seats among the states.

As explained above, the competition for interstate representation and funding is zero-sum, and so comparative undercount is more important than absolute undercount. If vulnerable Texan communities are more wary of the enumeration than vulnerable Californians, Texas—all Texans—suffer more in the interstate competition. 147 See supra text accompanying note 83. Ultimately, this impact may harm states controlled by the Trump Administration’s fellow Republican partisans most. 148 See supra text accompanying note 87. Specifically, many of the states most likely to forego additional congressional seats if significant portions of the population do not respond to the enumeration are states in which it is likely that enti­ties controlled by the Republican party will be responsible for redrawing the congressional lines in 2021. Whether that is widely understood—or whether the other consequences discussed above and below are perceived to compensate for that impact—is unclear.

C. Intrastate Competition for Representation and Power

Similarly, it may be that the errors were desired, but for their impact on the intrastate rather than interstate allocation of political power and funding. The Constitution requires federal, state, and most local legisla­tive districts to be of substantially similar size. 149 See Avery v. Midland Cty., 390 U.S. 474, 484–86 (1968); Reynolds v. Sims, 377 U.S. 533, 577 (1964); Wesberry v. Sanders, 376 U.S. 1, 17–18 (1964). Part III.D, infra, discusses the population base used to determine this equality in considerably more detail. After the decennial enumeration reveals population shifts, state and local governments will recalibrate their districts to achieve equality based on the latest figures, and reallocate their dollars to where the people live.

If vulnerable populations refuse to respond to the enumeration, and those refusals are not evenly geographically dispersed, they will lose political power (and funding) as federal, state, and local districts are divvied up within state lines. 150 See New York, 351 F. Supp. 3d at 594–95, 597–98. Funding will appear to be distributed to where the people are, but the dollars will be stretched far more thinly in areas of a substantial undercount. And districts will appear to embrace similar numbers of people, but a substantial undercount will mean that some districts are in fact far more populous than others, forcing more residents to compete for representation than similarly situated residents elsewhere. While the Constitution tolerates some size disparity among districts, larger disparities amount to constitutional harm. 151 Generally, congressional seats must be precisely equally populated, with minor variances permissible when those variances are necessary to achieve legitimate, consistently applied state criteria. Karcher v. Daggett, 462 U.S. 725, 730 (1983); see also Tennant v. Jefferson Cty. Comm’n, 567 U.S. 758, 759–61, 765 (2012). State and local districts have more constitutional latitude to be smaller or larger than the mean, if there is a sufficiently good reason for the deviation, but a ten percent deviation from largest to smallest district creates the rebuttable inference of a violation. See, e.g., Evenwel v. Abbott, 136 S. Ct. 1120, 1124 (2016). If the undercount is sufficiently severe, the inaccurate enumeration may be masking what is in fact unconstitutional malapportionment.

The intrastate deprivation of political power, in particular, will likely have predictable partisan impact depending on the local political demography—and if the desire for partisan punishment is indeed motivating the addition of the citizenship question to the enumeration questionnaire, that has serious constitutional ramifications. 152 See Justin Levitt, Intent Is Enough: Invidious Partisanship in Redistricting, 59 Wm. & Mary L. Rev. 1993, 2014–18 (2018). Those who live in the areas of an undercount will see their political power wane. 153 See supra section I.D. But to acknowledge that the local partisan ramifications of an under­count are predictable is not to say that they will always match conven­tional wisdom. Undercounts will not be confined merely to noncitizens, and the impacts of undercounts will not be confined merely to those households undercounted. And though Democrats in urban areas with a significant undercount would see a diffusion of power, for example, so too may Republicans in agricultural communities dependent on immigrant labor. 154 See supra text accompanying note 81.

D. Redistricting Population Base

The rationales above all hinge on errors in the enumeration that the citizenship question will likely significantly aggravate. A final alternative rationale for placing the citizenship question on the decennial does not depend on this potential disruption to the enumeration. Instead, it is based on a desire to fundamentally rewrite the terms of American representation.

As mentioned above, the Constitution requires districts to be of substantially similar size. 155 See Avery v. Midland Cty., 390 U.S. 474, 484–86 (1968); Reynolds v. Sims, 377 U.S. 533, 577 (1964); Wesberry v. Sanders, 376 U.S. 1, 17–18 (1964). This mandate was imposed in a series of cases deemed the “reapportionment revolution” 156 See generally Gordon E. Baker, The Reapportionment Revolution: Representation, Political Power, and the Supreme Court (1966) (chronicling the judicial and political repercussions of a group of landmark Supreme Court cases on legislative apportionment). —the cases Chief Justice Earl Warren famously branded “the most important” of his tenure on the Court. 157 See Earl Warren, The Memoirs of Earl Warren 306 (1977). Chief Justice Warren specifically identified Baker v. Carr, 369 U.S. 186 (1962), as the most important of his tenure—it was the first case in the reapportionment revolution, and the rest followed swiftly thereafter. Id. They recognized fundamental equality interests in the draw­ing of political districts. 158 Warren, supra note 157, at 306–08. But they left some ambiguity in determining equality of what.

When the Constitution requires the apportionment of congressional seats based on the “whole number of persons in each state,” 159 U.S. Const. amend. XIV, § 2. that is a standard unambiguously rooted in the representation of all persons, including the virtual representation of those who may not vote. 160 Joseph Fishkin, Taking Virtual Representation Seriously, 59 Wm. & Mary L. Rev. 1681, 1683–86 (2018) [hereinafter Fishkin, Virtual Representation]. It im­plies that congressional representatives have the authority (and perhaps obligation) to represent all of their constituents; that all individuals within a polity deserve to be represented, and represented proportionate to their numerosity; that the wishes of the minority of the governed should not prevail over the wishes of the majority; and that constituents’ various concerns have roughly equal claims on the representatives’ time and attention from district to district, such that 50,000 individuals aren’t jostling for one representative’s efforts in one state while 500,000 individuals seek face time from one representative in the state next door. 161 See, e.g., Garza v. County of Los Angeles, 918 F.2d 763, 781 (9th Cir. 1990) (Kozinski, J., concurring and dissenting in part) (“[Apportionment by raw population] assures that constituents have more or less equal access to their elected officials, by assuring that no official has a disproportionately large number of constituents to satisfy.”). The fact that congressional districts are apportioned based on total population does not, of course, make population equality the only theory of representation endorsed in the Constitution. At least while the country’s total population divided by the tally of the least populous state exceeds the total number of federal representatives, the Constitution’s allotment of at least one representative to each state, U.S. Const., art. I, § 2, cl. 3, ensures that congressional districts will not share precisely the same population from state to state. And even more directly, the fact that the Constitution guarantees two senators to each state, id. art. I, § 3, cl. 1, famously achieves neither equality of representation nor equality of voting power, nor any approximation of either, in the Senate.

In the first of the substantive cases concerning district size, the Court determined that congressional redistricting must maintain the same principle as congressional apportionment. 162 Wesberry v. Sanders, 376 U.S. 1, 13, 18 (1964). Following the allocation of congressional districts among states, the construction of congressional districts within states must ensure that each district contains a roughly equal number of people. 163 Id.

For state and municipal legislative districts, the law is admittedly more muddled. 164 See, e.g., Garza, 918 F.2d at 780–85 (reviewing the ambiguity). The cases refer to equality of representation, achieved through equalizing total population, following the model for congres­sional apportionment and the model embodied in the constitutions of most states. 165 See, e.g., Bd. of Estimate v. Morris, 489 U.S. 688, 693–94 (1989) (emphasizing the importance of population equality for representation in elected multimember municipal legislatures); Gaffney v. Cummings, 412 U.S. 735, 748–49 (1973) (same for state legislative office); Avery v. Midland Cty., 390 U.S. 474, 478, 480, 485–86 (1968) (same for county commission); Burns v. Richardson, 384 U.S. 73, 91 (1966) (same for state legislative office); Reynolds v. Sims, 377 U.S. 533, 559, 562–63, 565, 567–68, 573, 575–77, 579, 581 (1964) (same for state legislative office). But the cases also refer—sometimes in the same sentence 166 See, e.g., Reynolds, 377 U.S. at 576 (“The right of a citizen to equal representation and to have his vote weighted equally with those of all other citizens in the election of members of one house . . . would amount to little if States could effectively submerge the equal-population principle in the apportionment of . . . the other house.” (emphasis added)); id. at 577 (“We realize that it is a practical impossibility to arrange legislative districts so that each one has an identical number of residents, or citizens, or voters.”). —to the need to maintain equality among citizens 167 See, e.g., Morris, 489 U.S. at 693–94; Gaffney, 412 U.S. at 748; Avery, 390 U.S. at 481 n.6, 484; Richardson, 384 U.S. at 91–92, 94–95 (“At several points [in Reynolds], we discussed substantial equivalence in terms of voter population or citizen population, making no distinction between the acceptability of such a test and a test based on total population.”); Reynolds, 377 U.S. at 564 n.41, 565, 568, 576. and an equally weighted vote. 168 See, e.g., Morris, 489 U.S. at 693–94; Gaffney, 412 U.S. at 746–48; Avery, 390 U.S. at 478, 480; Richardson, 384 U.S. at 91; Reynolds, 377 U.S. at 554–56, 559–63, 564 n.41, 565, 567–68, 576, 579, 581. Indeed, the substantive discussion in Reynolds, which seems to require districts of equal population, begins with a broad encomium to suffrage and the right to an undiluted vote. See id. at 554–55. Despite their conflation, these are distinct con­cepts, 169 Not all citizens are voters. Many citizens cannot vote, including children, those disenfranchised by conviction, or those blocked by regulation or mistake; others may choose not to vote. See Asma Khalid et al., On the Sidelines of Democracy: Exploring Why So Many Americans Don’t Vote, NPR (Sept. 10, 2018), [https://].
and  under­justified  to  the  extent  they  have  any  meaning  at  all. 170 The right to an equally weighted vote, like the “one-person, one-vote” slogan, Evenwel v. Abbott, 136 S. Ct. 1120, 1130–31 (2016), is an enormously slippery concept in the redistricting context, too often deployed as a substitute for analysis rather than a thoughtful shorthand. Even in Reynolds, the concept of an equally weighted vote is built upon undertheorized handwaving.
[I]t is inconceivable that a state law to the effect that . . . the votes of citizens in one part of the State would be multiplied by two, five, or 10, while the votes of persons in another area would be counted only at face value, could be constitutionally sustainable. Of course, the effect of state legislative districting schemes which give the same number of representatives to unequal numbers of constituents is identical.
Reynolds, 377 U.S. at 562–63; see also Morris, 489 U.S. at 698 (equating the ability of one constituent to vote for two representatives with the ability of one constituent to vote for one representative while in a district half the size of another).
The “one person, one vote” slogan is not completely devoid of meaningful content. It is perhaps most cogent as a principle for the casting and counting of ballots: In an election for any given office in which the winner gains the largest number of votes, one elector’s ballot should not be counted multiple times while another similarly situated elector’s ballot is counted once. The principle naturally extends to an election for a single office in which votes from similarly situated electors are tallied differently depending on where they are cast, like the “county unit” structure invalidated in Gray v. Sanders, 372 U.S. 368, 371–73, 379–81 (1963).
Beyond those applications, however—and particularly in the redistricting context, which involves decisions about which individuals to put where—the precise content of the powerful slogan turns out to be remarkably difficult to pin down in a manner that is not merely tautological. See Joseph Fishkin, Weightless Votes, 121 Yale L.J. 1888, 1898 (2012) [hereinafter Fishkin, Weightless Votes]; Sanford Levinson, One Person, One Vote: A Mantra in Need of Meaning, 80 N.C. L. Rev. 1269, 1270–72, 1277–78, 1281–82 (2002). “One person, one vote” is a sensible metaphor for equal legislative gravitas based on the size of the constituency metaphorically behind each legislator’s actions, helping to ensure that power in the policymaking body is distributed to the majority. But it does not purport to resolve what the majority is the majority of. As Professor Joseph Fishkin has examined in considerably more depth and sophistication, the notion of an equally “weighted” vote yields no further clarity. See Fishkin, Weightless Votes, supra, at 1892. What, exactly, does it mean to draw districts such that each person’s vote is of equal “weight,” when each vote is valued the same as every other in the contest for electing a given representative? Does it imply that the political composition of districts must be designed such that every voter’s ballot has an equally decisive chance of determining the winner of an election? See id. at 1893–95. Or an equally decisive chance of determining policy? But see Morris, 489 U.S. at 697–98 (declining to adopt the Banzhaf index, a measure of the influence of a voter on a policy outcome, as a basis for constitutional equality of representation). If it represents the notion that each voter should compete only with an equal number of other voters across districts—and what is the representational principle behind such a claim?—does it imply that the value of the franchise varies with turnout variance? See, e.g., Fishkin, Weightless Votes, supra, at 1896–97.
Moreover, the doctrine reveals internal contradiction as well as imprecision: Just a few sentences after suggesting that states might be free to choose voter population as a basis for district equality, the Court explicitly held that the number of registered voters in each district would not suffice as a permissible apportionment base. 171 More specifically, the Court held that equalizing the number of registered voters in each district would be permissible only as a proxy for a permissible population basis, and only if it did not diverge substantially from a permissible population count. Richardson, 384 U.S. at 93. The Court’s theoretical conception of constitutional requirements for equality of representation in structures of state government, or of the permission that the Constitution grants to state governments in this arena, is hazy at best.

At the moment, per the Supreme Court’s apparent holding in Reynolds v. Sims, each state draws state legislative lines to equalize total popula­tion. 172 See Evenwel, 136 S. Ct. at 1124; 377 U.S. at 568 (“We hold that, as a basic constitutional standard, the Equal Protection Clause requires that the seats in both houses of a bicameral state legislature must be apportioned on a population basis.”).
A few states adjust individuals’ location, tallying persons who are incarcerated in the location of their preincarceration residence, rather than where they are imprisoned. See Cal. Elec. Code § 21003(b) (2018); Del. Code tit. 29, § 804A(b) (2018); Md. Code Ann., State Gov’t Code § 2-2A-01(2) (West 2018); N.Y. Legis. Law § 83-m(13) (McKinney 2018). These states do not presumptively exclude any individual from the apportionment base, if the prior in-state residence can be determined: The question is not whether individuals are to be counted, but where. Similarly, these states and a few others will adjust Census tallies for residency, excluding those who are not state residents from the apportionment count. See Haw. Const. art. IV, § 4; Kan. Const. art. X, § 1(a); Cal. Elec. Code § 21003(b)(2); Del. Code tit. 29, § 804A(a); Md. Code Ann., State Gov’t Code § 2-2A-01(1); N.Y. Legis. Law § 83-m; see also Fishkin, Virtual Representation, supra note 160, at 1684 n.11. In theory, if other states reciprocated this practice, each person would be counted somewhere for redistricting purposes. In practice, nonresidents may end up excluded from the base of the states above.
In 2014, a group of plaintiffs in Texas sued under the Equal Protection Clause, in an attempt to force the state to equalize the number of voters in each district. 173 See Evenwel v. Perry, No. 1: 14-cv-00335, 2014 WL 5780507, at *1 (W.D. Tex. Nov. 5, 2014), aff’d sub nom. Evenwel, 136 S. Ct. at 1120. In the course of that litigation, amici defending the propriety of the total population base urged that the existing data—including ACS data—were insufficient for equalizing the number of voters in each district. They argued that, even if the goal were desirable, ACS data were not up to the task. 174 See Brief of Nathaniel Persily et al., supra note 106, at 12–27. The purported precision of the decennial enumeration, by contrast, made equalizing total population feasible. 175 Or, at least, the Court has agreed that the precision of the decennial enumeration is a sufficiently stable legal fiction to support legal equal population requirements. See League of United Latin Am. Citizens v. Perry, 548 U.S. 399, 421 (2006) (plurality opinion) (noting the legal fiction as such); Reynolds, 377 U.S. at 583–84 (recognizing that the need for stability does not require redistricting more frequently than a decennial cycle).

And so it may be that adding a citizenship question to the decennial census becomes a vehicle for a block-by-block dataset of citizen popula­tion, ostensibly suitable for a novel redistricting population base. Some states might seek to depart from their historic norm and use this dataset to draw their state legislative districts with equal numbers of citizens. Nebraska law may already contain such a provision, unenforced for years, and still of uncertain legality given the Court precedent described above. 176 Neb. Const. art. III, § 5 (“The basis of apportionment [for state legislative districts] shall be the population excluding aliens . . . .”). A bill introduced in the Nebraska legislature in 2018 would ensure that state and local districts were drawn to equalize the number of citizens. See Leg. 1115, 105th Leg., 2d Sess. (Neb. 2018). A Missouri state constitutional amendment along similar lines passed the state House in 2018. 177 See H.J. Res. 100, 99th Gen. Assemb., 2d Sess. (Mo. 2018); H.R. Journal, 99th Gen. Assemb., 2d Sess. 2920 (Mo. 2018). And Texas’s own litigation position in Evenwel v. Abbott was that although it had not drawn districts to equalize citizen population, it might, and it could. 178 See Brief for Appellees at 18–23, Evenwel v. Abbott, 136 S. Ct. 1120 (2016) (No. 14-940), 2015 WL 5562189.

It is not clear whether such a choice is available under federal law. Evenwel refrained from holding that total population is the only permissible population base, 179 136 S. Ct. at 1133. in part because to do so would have been a substantial expansion of the question presented. The issue on the table was whether challengers could force Texas to depart from a total popula­tion base against the state’s wishes, not whether Texas was free to choose a different base on its own. 180 See Brief for Appellants at i, Evenwel, 136 S. Ct. 1120 (2016) (No. 14-940), 2015 WL 4624625. But the case—and its predecessors—contains some strong language hinting toward the singular propriety of an equal-representation theory for state legislative redistricting. 181 See Evenwel, 136 S. Ct. at 1128–29, 1131, 1132 (briefly canvassing the history of the debate, noting the Fourteenth Amendment’s choice of total population at least for congressional districts, and explaining the representational merits of districts based on total population); see also, e.g., Fishkin, Virtual Representation, supra note 160, at 1686 (noting that the Evenwel Court “elevat[ed] the use of total population . . . to a sort of default baseline”); Derek T. Muller, Perpetuating “One Person, One Vote” Errors, 39 Harv. J.L. & Pub. Pol’y 371, 393 (2016) (“And the majority opinion [in Evenwel] suggested that total population is the only acceptable basis for redistricting, formally reserving that question for another day.”). For example, the Court firmly declared that “[a]s the Framers of the Constitution and the Fourteenth Amendment comprehended, representatives serve all residents.” 182 Evenwel, 136 S. Ct. at 1132. If that is true, 183 The Framers of the 1789 Constitution may have contemplated that representatives would serve most residents. See, e.g., Fishkin, Virtual Representation, supra note 160, at 1706–07. But it is difficult to believe that they thought representatives would serve slaves as well. See id. at 1706 n.72. In contrast, the Framers of the Fourteenth Amendment put a higher premium on representation of, and responsibility for, all persons within the jurisdiction. See U.S. Const. amend. XIV, §§ 1–2. it necessarily im­plies that equality of rep­resentation in the districting context depends on total population. In contrast, drawing districts based on equal numbers of citizens depends on the legitimacy of exclusion. 184 Moreover, this theory of exclusion is at least uncomfortably juxtaposed with the clause requiring districts of approximately equal size: The Equal Protection Clause applies not only to citizens, but to “any person within [the state’s] jurisdiction.” U.S. Const. amend. XIV, § 1. It hinges upon a theory that though noncitizens present in a state—including noncitizens present at the express invitation of the federal government—are subject to the state’s regulation and taxation, they may be deprived of representation in state government.

It is also important to note that if the Supreme Court’s case law in this area seems to vacillate between theories of equality of the governed based on representation of all people and theories of equality of the political community based on representation of voters, a redistricting base of citizens accomplishes neither. Citizen populations contain large numbers of people who may not or will not vote in the district, including temporary sojourners (including some members of the military), children, individuals disenfranchised by conviction, individuals ineligible by means of mental capacity, and people who do not cast valid ballots, either by choice or by happenstance.

Even if the Constitution does not prevent a jurisdiction from choosing a redistricting base other than total population for its state legislature, the VRA might. The  VRA  prohibits  dilution  of  the  right  to  vote  on  the  basis  of  race  or  language-minority  status. 185 See, e.g., Levitt, New Misreading, supra note 95, at 574–75, 584–86, 587–89. A jurisdiction with a troubled history of discrimination or danger signs of present discrimination may not select freely from among multiple electoral options if one or more of those options leads to dilution of opportunities for minorities to effectively exercise the franchise. 186 See, e.g., Justin Levitt, Race, Redistricting, and the Manufactured Conundrum, 50 Loyola L.A. L. Rev. 555, 576–77 (2017). A state’s choice of a redistricting population base should be no different in this regard from any other election-related choice. And in many jurisdictions with trou­bled histories of race relations, it is likely that excluding all noncitizens from the redistricting base will cause district sizes to swell in areas with significant minority populations; as district sizes grow, minority electoral opportunity will tend to shrink. The VRA may well constrain that choice. 187 See, e.g., Fishkin, Weightless Votes, supra note 170, at 1907–08.

Finally, it is worth noting that courts and commentators have long observed the potential for manipulation in choosing a redistricting base other than total population. In Burns v. Richardson, for example, the Court confronted Hawaii’s system of equalizing the number of registered voters in each district. 188 384 U.S. 73, 93–94 (1966). In Hawaii, this was done only as a proxy for the state’s desire to apportion by citizen population, for which statistics were not readily available, and by total population, which was determined along lines of census geography that did not at the time respect traditional local boundaries. Id. The Court approved the practice, but only as an interim measure, and only to the extent that it approximated other permissible population bases. Id. at 94–97. The Court noted not only that such a base may fluctuate substantially over the course of a decade but that incumbents would have the incentive to drive fluctuation for political advantage: “[A] registered voter or actual voter basis . . . is thus susceptible to improper influences by which those in political power might be able to perpetuate underrepresentation of groups constitutionally entitled to participate in the electoral process . . . .” 189 Id. at 92. The locus for manipulation of a citizen-population base is different, but perhaps no less notable. First, there is the decision to use a citizen-population base in the first instance, and the likelihood that such a decision would be based more on the potential for partisan gain than on deep representational theory. And second, there is the construction of the citizenship base: Federal govern­ment actors control the naturalization process, and in the pursuit of political advantage might improperly increase or decrease the pace of naturalization in advance of a decennial census year to put another thumb on the scale of a state legislative map. 190 See, e.g., Fishkin, Weightless Votes, supra note 170, at 1906. It is, regrettably, not implausible to believe that an immigration process could be used as the tool for partisan electoral gain. 191 Cf., e.g., Greg Jaffe & Dan Lamothe, Former Generals Worry that Trump’s Border Mission Uses Troops as a Political Tool, Wash. Post (Nov. 2, 2018), [] (describing former military officials’ public criticism regarding the possible political motivation of the Trump Administration’s deployment of the military to the southern United States border just prior to the 2018 midterm elections).


Each of the ends discussed in Part III above appears more likely to be furthered by the addition of a citizenship question to the decennial enumeration questionnaire than will vigorous enforcement of the VRA. As for the official justification put forward in support of the proposal, the analysis above indicates that the eleventh-hour decision to add a citizen­ship question to the decennial enumeration is both unnecessary and counterproductive. And as bipartisan former Census Bureau officials have explained, in this climate, the decision poses a substantial risk to the Bureau’s ability to undertake its one constitutionally mandated duty. A distorted enumeration will have profound and lingering consequences for funding and political power. Indeed, that might well have been the point.